Encyclopedia of Nuclear Magnetic Resonance, Volume 9

ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI

Anisotropically Restricted Diffusion in MRI Michael E. Moseley and Alex de Crespigny Stanford University, CA, USA

diffusion is rarely restricted but is rather slowed by the presence and extent of permeable or semipermeable membranes. The complete concept of proton diffusion then is that of a tensor, where the diffusion rate or ADC value is a function of the diffusion direction in three-dimensional space. As the tissue water protons constantly `experience' the three-dimensional microenvironment through the processes of permeability, exchange, ¯ow, and thermally-induced diffusion, their traveled path hLi along any one given direction (along the magnet bore direction, for example) can be estimated by:9 hL2 i ˆ 2
T…ADC†

1 INTRODUCTION A dominant source of proton MR signal loss following radiofrequency excitation is caused by dephasing. While coherent ¯ow will cause dephasing, other motions can contribute. One important mechanism of dephasing is that of a random diffusion of protons. This random phase shifting leads to signal loss because of phase cancelation. While T1 and T2 relaxation times re¯ect complicated frequency-dependent rotational and proton exchange processes, diffusion is caused solely by random proton displacements or translations through a tissue. Since the phenomenon of proton diffusion in vivo is complicated, the term `apparent diffusion' described by the apparent diffusion coef®cient, ADC, has been used to denote signal loss induced by random proton ¯uctuations. This applies to protons in water, phosphorus in high-energy metabolites, indeed to any MR-observable nucleus.1±3 Considering water proton displacements in pure isotropic solutions, where the probability that a water proton will diffuse in any given direction is equal, the rate of random proton phase cancellation due to diffusion will be same regardless of the direction observed. This is termed `isotropic' diffusion and is seen in homogeneous environments. In ordered environments, such as that found in liquid crystals for example, the resulting diffusion coef®cient will depend on the direction of the motion being considered and on any geometric anisotropy created by the nature of the microenvironment.4,5 For water protons diffusing or moving within a tissue matrix, the observed diffusion rate and direction will re¯ect the molecular and macromolecular barriers or hindrances that the proton experiences during the translational process. 2 APPARENT DIFFUSION COEFFICIENTS AND OBSERVATION DIRECTION Simply stated, proton apparent diffusion is slowed if the protons are hindered in their random motion.6±8 Hindrances can be impermeable barriers that completely restrict diffusion. In living in vivo tissues, however, water protons are probably slowed solely by the presence of membranes, cell walls, and macromolecules due to differences in water proton permeability. More speci®cally, myelin ®bers and neuro®brils in white matter tracts possess a special hindrance for water proton movement which is greater in one direction than another. This causes protons to diffuse faster along any direction of least resistance, in this case along the long axis of the tracts, causing the apparent diffusion coef®cients to be anisotropic or directionally-dependent. It is important to note that in vivo proton

1

…1†

where over an observation time of
T (s) the mean square displacement is expressed by an apparent diffusion coef®cient, ADC (usually given in m2 sÿ1 or cm2 sÿ1). Typical proton diffusion coef®cients, ADC, of 110ÿ5 cm2 sÿ1 are observed in cat cerebral gray matter,10 the average mean square displacement along that one given observation direction hLi becomes 1 (2  40 ms  10ÿ5 cm2 sÿ1)2 or 9 m, observed over an effective observation time, (
T), of 40 ms. To date, most diffusion-weighted MR imaging and spectroscopy sequences have been based upon a spin echo Stejskal±Tanner (ST) sequence,11±13 either as the sequence itself or as a magnetization preparation set of prepulses, prior to another MR sampling scheme. Since the two ST `diffusionsensitizing' gradient pulses are symmetrical, all proton spin dephasing caused by the ®rst diffusion-sensitizing gradient pulse will be refocused (spin-rephased) by the second diffusion-sensitizing gradient pulse for stationary spins. Randomlymoving spins (because of ¯ow, perfusion, diffusion, etc.) will not completely refocus and will attenuate the observed echo or echo train. Note that mild coherent motions will result in phase shifts and not in signal attenuation without phase cancellation. The observed echo or echo train signal intensity S(Gi) can be expressed as: S…Gi† ˆ S…0†eÿbD

…2†

where the b value is de®ned12,13 by equation (3): b ˆ 2 2 G2 …
ÿ =3†

…3†

and S(0) is the signal obtained without applied diffusion-sensitizing gradient pulses, is the gyromagnetic ratio,  and G are the duration and the amplitude of the diffusion-sensitizing ST gradient pulses along a given direction,
is the time interval between the leading edges of the diffusion-sensitizing gradients pulses and D is the water apparent diffusion coef®cient (often expressed as the ADC value).12,13 Measurement of apparent diffusion from MR images would require plotting the natural logarithm of the image intensities against the b value. The slope of this observed monoexponential decay directly leads to a diffusion coef®cient.11 There exist three major advantages to using a `pulsed' gradient approach for measuring in vivo diffusion. One is that the stronger pulsed gradient diffusion-sensitizing pulses overcome poor magnet homogeneity (which upon re¯ection, is a sort of residual magnetic ®eld gradient itself).11 Also, because the diffusion-sensitizing gradients are relatively short pulses, the time during which the sequence is sensitive to diffusion, (
ÿ /3), can be accurately controlled. From a knowledge of this obser-

2 ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI vation time, (
ÿ /3), the mean path length can be estimated from the above Einstein displacement equation given the measured diffusion coef®cient [equation (1)]. A ®nal advantage is that the direction of the diffusion-sensitizing gradient pulses can be controlled accurately. All MRI scanners possess three orthogonal and linear magnetic ®eld gradient coils, any or all of which can be pulsed to produce the diffusion sensitization, the effective angle of which is then the geometric sum of the pulsed gradients. This opens the path to mapping of the apparent diffusion tensor.14,15 The complete concept of proton diffusion is then that of a tensor14,15 where the diffusion rate or ADC value is a function of the diffusion direction in three-dimensional space. Thus the tensor for free, isotropic, diffusion can be visualized as a sphere, with the same ADC in all directions, while the anisotropic diffusion of a white matter tract can be viewed as an ellipsoid tilted at some angle to the three principal axes, depending on the particular ®ber orientation. Diffusion is then fastest along the principal or long axis of the ellipsoid, and takes a lower (but equal) value along the two orthogonal axes. The `trace' of this ellipsoidal diffusion tensor is just the average of the ADC values measured along any three orthogonal directions, and is independent of ®ber orientation thus yielding a useful measure of the average rate of diffusion without the complications of ®ber directionality.

3 MAPPING AND THE ORIGIN OF THE APPARENT DIFFUSION ANISOTROPY The b value-dependent diffusion effects on image intensity can be displayed in different ways. A diffusion-weighted image contains typically T1-, T2-, and diffusion-weighting, the amount of the diffusion-weighting being dependent on the b value. Within diffusion-weighted images, relative faster intravoxel motion will be observed as increased attenuation of the signal from that voxel, and consequently an observed hypointensity from that voxel as the b value is increased. Conversely, slower motions or diffusion leads to smaller signal attenuation. The observed result is that of regional relative hyperintensity. This is best seen in lipid regions outside the brain where lipid protons experience little or no diffusion because of the molecular size and environment of the long chain hydrocarbons. A pure diffusion of ADC image is calculated from the acquisition of two or more images of varying b values. ADC images eliminate T1- and T2-weighting; voxel intensities re¯ect only the apparent diffusion.12,13 The ADC images are produced by ®tting equation (2) to a series of two or more diffusionweighted images of varying b value. The ADC images are then displayed with the voxel intensity related or scaled to the rate of diffusion. Visualization of cerebral details is sometimes easier from diffusion-weighted images, which are devoid of cerebrospinal ¯uid (CSF) hyperintensity because of the relatively faster apparent diffusion of water protons in CSF. In the corresponding ADC images, the pronounced CSF hyperintensity (re¯ecting faster apparent diffusion) can often limit the visualization of more subtle ADC differences in gray and white matter. An appropriate example of ordered or anisotropic diffusion can be seen in proton diffusion-weighted images of ®brous vegetables or fruits. Apparent diffusion along the direction of

ordered ®bers (measured by application of the diffusion-sensitizing gradient parallel to the long axis), renders the diffusion-weighted intravoxel intensity within the ®bers hypointense, implying fast diffusion of water protons along the ®bers (since rapid diffusion produces signal attenuation on diffusion-weighted images). Corresponding images acquired with the diffusion-sensitizing gradient perpendicular to the long axis of the ®bers result in very hyperintense image intensity within the ®bers, implying relatively slow proton diffusion (little echo loss) along this direction. The translational distances traveled by water protons perpendicular and parallel to ®brous matter during the MR observation time (
ÿ /3) of roughly 20±50 ms is estimated to be 5±13 m (respectively). If water proton displacements in any given direction are signi®cantly shorter that the statistical diameter of a structure of barriers along that direction, then little or no anisotropy will be observed. Anisotropy is apparent when the water protons within a voxel can sample the microenvironment adequately. The rationale behind diffusional anisotropy in an ordered structure is that water protons within a matrix statistically prefer to move along ordered structures rather than across them (by the principle of least resistance). In other words, average proton displacements along the ®bers are greater due to fewer barriers being present in this direction. Proton displacements are shorter across the ®bers due to barriers presented by the very orientation of the ®bers themselves. When the diffusionsensitizing gradient direction coincides with the direction of least hindrance, the apparent diffusion appears greater and the observed ADC is larger. Anisotropy of diffusion was ®rst suggested as a possible explanation for the large regional differences in in vivo human white matter apparent diffusion coef®cients by Thomsen et al.16 More recent descriptions of the existence of and characterization of in vivo anisotropy have subsequently appeared.10,17,18 In axial and coronal diffusion-weighted images in animal models,10 the directional dependence of the white matter apparent diffusion is not observed in regions of gray matter. Within the corpus callosum, however, apparent diffusion can be as slow as 0.4 ‹ 0.1  10ÿ5 cm2 sÿ1 (across the short axis of the tract) and as high as 1.3 ‹ 0.1  10ÿ5 cm2 sÿ1 (along the long axis of the corpus callosum). This suggests faster proton diffusion in white matter along the direction of the long axis of the tract. From the observed image intensities and knowledge of the diffusion-sensitizing gradient direction, one can then ascertain the orientation of individual white matter tracts. The translational distances traveled by water protons perpendicular and parallel to the white matter orientation during (
ÿ /3 =) 40 ms can be estimated to be 6 and 11 m respectively, suggesting that axonal diameters may be determinable from such measurements.

4

CLINICAL IMPLICATIONS

The in vivo directional diffusion-weighted images observed in human volunteers indicate that water proton diffusion anisotropy is best observed in the large diameter, fast-conducting motor and somatosensory nerve ®bers (Figures 1±5). Signi®cant anisotropy is observed in adult volunteers in all white matter tracts such as the corpus callosum, splenium, and in-

ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI

Figure 1 Axial, multislice diffusion-weighted images of a human volunteer acquired on a GE Signa at 1.5 T. Top row: minimal diffusion-weighting (b = 12 s mmÿ2). Second row: diffusionsensitizing gradient applied along AP direction (up±down in images), b = 512 s mmÿ2. Third row: diffusion-sensitizing gradient applied along LR direction (left±right in images), b = 512 s mmÿ2. Bottom row: diffusion-sensitizing gradient applied along IS direction (through-plane in images), b = 512 s mmÿ2. All images were acquired with TR = 2 cardiac cycle (~1500 ms), TE = 102 ms and 124 ms (second echo used for navigation), 1 NEX. Gradient durations  = 45 ms, gradient separation
= 62 ms, gradient strengths = 1 mT mÿ1 (1 G cmÿ1), FOV = 24 cm, 128  256 matrix size, and scan times = 4.3 min for eight slices per TR. Navigation achieved by correction of navigational echo magnitude. CSF suppression not applied. All images are scaled except top row. On these images, containing T1-, T2-, proton density-, and diffusion-weighting, relative regions of hypointensity denote rapid apparent diffusion of protons. Regions of relative hyperintensity demarcate slower diffusion. Note little or no anisotropic effects in gray matter or lipids (which display very slow apparent diffusion). TR, repetition time; TE, echo time; NEX, average; FOV ®eld of view

ternal capsules. In addition, myelinated tracts in spinal cord,10 optic (A. de Crespigny, unpublished data), and peripheral nerves20 appear to be associated with signi®cant anisotropic effects. Several pioneering studies verifying white matter diffusional anisotropy in normal and abnormal human and neonate subjects have been presented.18,21±25 It has been shown that the newborn brain demonstrates isotropic diffusion of water molecules.22,23 As the brain matures, diffusion becomes anisotropic, being greater along the longitudinal axes of the major axonal bundles of the cerebral white matter than perpendicular to them. This anisotropic ordering of water motion most likely occurs along the long axis of the individual neuro®brils as well as along or within the axons isolated by the myelin sheath. Studies indicate that the presence of myelin is not necessary for ordering of proton apparent diffusion since anisotropic proton mobility is seen in nerves lacking microtubules and fast axonal

3

Figure 2 Diffusion-weighted images chosen from one slice and taken from images in Figure 1. Top left: minimally diffusion-weighted; top right: diffusion-sensitizing gradient applied along AP direction (up± down in images), b = 512 s mmÿ2. Bottom left: diffusion-sensitizing gradient applied along LR direction (left±right in images), b = 512 s mmÿ2. Bottom right: diffusion-sensitizing gradient applied along IS direction (through-plane in images), b = 512 s mmÿ2. Note regional hypointensity in white matter tracts when applied diffusion-sensitizing gradient direction is parallel to white matter orientation. This is best seen in the image at top right, in the lower limbs of the splenum of the corpus callosum

transport.26 It is reasonable to assume that the development of anisotropy is related to the development of myelin sheaths; however, the temporal evolution of anisotropy seems to precede that of myelination as described by most authors.27 Additionally, diffusional anisotropy has been proved to occur in nonmyelinated nerve bundles.26 Another possibility is that diffusion anisotropy develops in relation to the maturation of the microtubules, intraaxonal organelles that facilitate axonal transport.28,29 Thus, diffusion-weighted imaging (DWI) seems to be useful in the real time assessment of cerebral maturation. The use of diffusion-weighted imaging for the determination of water proton apparent diffusion anisotropy may aid in the study of peripheral CNS tissues, which exhibit water proton diffusional anisotropy effects. One novel application, termed `MR neurography'20 can increase tissue contrast by using MR diffusion-weighting in a variety of directions to evaluate the diffusion tensor found in nerve tissue. An important adjunct to neurography is the use of fat saturation to reduce hyperintense artifacts due to the slow mobility of protons in lipid tissues and in the use of maximum intensity projection (MIP) methods to track nerve tissue in a MR angiography-like fashion. Most clinical usage of diffusion-weighted MRI has been hampered by two signi®cant problems, gradient strength and motion artifacts. Many clinical scanners possess gradients capable of strengths of no more than 10 mT mÿ1 (1 G cmÿ1), limiting b values to roughly 300 s mmÿ2 along any given direction. These limitations typically require pulsing two or all

4 ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI three gradients together for longer durations (with increases in TE to well over 100 ms) to push the b value to above roughly 500 s mmÿ2, above which diffusion effects can be clearly seen above the inherent T2-weighting. This in turn limits the signalto-noise ratio (SNR) and degrades image quality, an important factor in diffusion imaging since good diffusion-weighting will itself attenuate SNR exponentially. More importantly, longer TE values lead to signi®cant motional artifacts. Linear excursions of just a few millimeters can easily ruin images where diffusion-weighted tissue contrast arises from proton displacements on the order of microns. The solutions employed to date have included restraint of the head with rigid holders or vacuum cushions that seal the head in the coil. These are uncomfortable and are dif®cult to implement on

Figure 4 Enlargements from the ADC and anisotropy maps for one slice taken from the images shown in Figure 3. Top left: ADC image with diffusion-sensitizing gradient applied along AP direction (up± down in images). Top middle: ADC image with diffusion-sensitizing gradient applied along LR direction (left±right in images). Top right: ADC image with diffusion-sensitizing gradient applied along IS direction (through-plane in images). Bottom left: trace image. Bottom middle: standard deviation image

uncooperative patients, particularly when several images are required to map the apparent diffusion tensor or to synthesize ADC images. Fast-scan and high-speed MRI capable of acquiring diffusion-weighted images in seconds or less requires typically large gradient strengths (not currently present on most clinical scanners) and has been limited to relatively low-resolution images which make proper visualization of the individual white matter tracts very dif®cult. The use of multiple shots or echo trains in

Figure 3 Series of three diffusion-weighted images were acquired at b values of 12, 350, and 512 s mmÿ2 to produce the pure diffusion (ADC) images shown here for the three gradient axes and for the same four slices shown in Figure 1. Top row: diffusion-sensitizing gradient applied along AP direction (up±down in images). Second row: diffusion-sensitizing gradient applied along LR direction (left±right in images). Third row: diffusion-sensitizing gradient applied along IS direction (through-plane in images). Note that regional hyperintensity on ADC images indicates rapid apparent diffusion. CSF hyperintensity is an example of fast apparent motion or diffusion, so much so that smaller variations in gray and white matter are masked out and dif®cult to appreciate. For this reason, diffusion-weighted or other image synthesis is needed. Fourth row: images of the `trace' of the diffusion tensor determined for each of the slices shown in the top three rows. The trace of the tensor is essentially an average of the diffusion images from the three principal axes and is useful for removing orientation effects from diffusion-weighted or ADC images. The trace image is useful for discrimination of ADC alterations due to stroke or other cerebrovascular diseases,19 which is often dif®cult because of regional ADC differences due to white matter anisotropy. Bottom row: the synthesized `standard deviation' images19 taken from the ADC images and the trace image. The SD image is essentially a pixel-by-pixel map of the deviations from the trace and can be very useful for the identi®cation of white matter tracts and discrimination from stroke or regional gray matter

Figure 5 Enlargements of another slice from the image set in Figure 3; image layout is as for Figure 4. This slice clearly shows the genu of the corpus callosum in the center of the ADC images (top row), which is not seen in the trace image (bottom left) but stands out in the anisotropy image (bottom middle) due to the strong orientation effects of the corpus callosum

ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI

high-speed imaging to improve resolution or SNR will in turn render the images again sensitive to motional artifacts. Phase navigation30,31 combined with a conventional single or multi echo imaging sequence has great potential for clinical application. It involves simple modi®cations of standard clinical pulse sequences using additional spin- or gradient-echoes acquired without phase encoding to map and correct each phase-encoded line or train of k-space. This can yield relatively motion-free images of excellent resolution with limited gradient strengths at the expense of conventional scan times. Uncooperative patients can create motion artifacts not easily correctable with linear phase navigation methods (A. de Crespigny, unpublished data). However, with motion correction or navigation, a multishot approach (using EPI, RARE, or spiral k-space coverage) could become the optimal apparent diffusion imaging technique, using conventional gradients to build resolution in much shorter scan times. This would be fast enough to acquire a series of images with increasing b value or varying gradient direction in order to generate an ADC or an orientation map. Utilization of this unique MRI technique for determining the orientation of white matter through the measurement of water proton apparent diffusion has the potential greatly to improve our understanding and assessment of demyelination disorders, white matter infarcts with associated edema development,32 neoplasms involving white matter tracts, and of neonatal brain and spinal cord development. Signi®cant proton anisotropic effects have also been observed in human arm muscle, 33 opening up the potential of using this technique to map performance or pathology. Finally, images devoid of white matter orientation effects can be synthesized by mapping the trace of the tensor enabling better visualization and earlier detection of ischemia in the CNS.19 5 RELATED ARTICLES Diffusion: Clinical Utility of MRI Studies; Image Formation Methods; Intracranial Infections; Magnetic Resonance Imaging of White Matter Disease; Spin Warp Data Acquisition. 6 REFERENCES 1. G. E. Wesbey, M. E. Moseley, and R. L. Ehman, Invest. Radiol., 1985, 19, 484. 2. C. T. W. Moonen, P. C. M. van Zijl, D. LeBihan, and D. Despres, Magn. Reson. Med., 1990, 13, 467. 3. M. E. Moseley and P. Stilbs, Chem. Scr., 1980, 15, 215. 4. M. E. Moseley and A. Loewenstein, Mol. Cryst. Liq. Cryst., 1982, 90, 117. 5. M. E. Moseley, J. Phys. Chem., 1983, 87, 18. 6. E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 43, 3579. 7. J. E. Tanner and E. O. Stejskal, J. Chem. Phys., 1968, 49, 1768. 8. R. L. Cooper, D. B. Chang, A. C. Young, C. J. Martin, and D. Ancker-Johnson, Biophys. J., 1974, 14, 161. 9. A. Einstein, `Investigations on the Theory of the Brownian Movement', Dover Publications, New York, 1956. 10. M. E. Moseley, Y. Cohen, J. Kucharczyk, J. Mintorovitch, H. S. Agari, N. F. Kenland, J. Tsurula, and D. Norman, Radiology, 1990, 176, 439.

5

11. E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288. 12. D. LeBihan, E. Breton, D. Lallemand, P. Grenier, E. Cabanis, and M. Laval-Jeantet, Radiology, 1986, 161, 401. 13. D. LeBihan, E. Breton, D. Lallemand, M. Aubin, J. Vignaud, and M. Laval-Jeantet, Radiology, 1988, 168, 497. 14. P. J. Basser, J. Mattiello, and D. LeBihan, J. Magn. Reson. Ser. B, 1994, 103(3), 247. 15. J. Mattiello, P. J. Basser, and D. LeBihan, Magn. Reson., Med. B, 1997, 37(2), 292. 16. C. Thomsen, O. Henriksen, and P. Ring, Acta Radiol., 1987, 28, 353. 17. D. Chien, R. B. Buxton, K. K. Kwong, and B. R. Rosen, J. Comput. Assist. Tomogr., 1990, 14, 514. 18. W. M. Chew, J. Tsuruda, and M. E. Moseley, Radiology, 1990, 177, S121. 19. P. van Gelderen, M. M. de Vleeschouwer, D. DesPres, J. Pekar, P. C. M. van Zijl, and C. T. W. Moonen, Magn. Reson. Med., 1994, 31, 154. 20. F. A. Howe, A. G. Filler, B. A. Bell, and J. R. Grif®ths, Magn. Reson. Med., 1992, 28, 328. 21. T. L. Chenevert, J. A. Brunberg, and J. Pipe, Radiology, 1990, 177, 401. 22. H. Sakuma, Y. Nomura, K. Takeda, T. Tagami, T. Nakagawa, Y. Tamagawa, Y. Ishi, and T. Tsukomoto, Radiology, 1991, 180, 229. 23. M. A. Rutherford, F. M. Cowan, A. Y. Manzur, L. M. Dubowitz, M. Pennock, J. V. Majnal, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1991, 15, 188. 24. M. Doran, J. V. Hajnal, N. Van Bruggen, M. D. King, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1990, 14, 865. 25. J. V. Hajnal, M. Doran, and G. M. Bydder, in `Magnetic Resonance Imaging', eds. D. D. Stark and W. G. Bradley, Mosby, St. Louis, MO, 1991, Vol. 2, p. 1081. 26. C. Beaulieu and P. S. Allen, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 1728. 27. A. J. Barkovich and D. M. Wimberger, in `Magnetic Resonance Neuroimaging', eds. J. Kucharczyk, M. Mosley, and A. J. Barkovich, CRC Press, Boca Raton, FL, 1993. 28. F. Espejo and J. Alvarez, J. Comp. Neurol., 1986, 250, 65. 29. V. Faundez and J. Alvarez, J. Comp. Neurol., 1986, 250, 73. 30. R. A. Ordidge, J. A. Helpern, Z. Qing, and R. A. Knight, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 1902. 31. R. Asato, T. Tsukamoto, R. Okumura, Y. Miki, E. Yoshitome, and J. Konishi, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 1226. 32. T. Ebisu, S. Naruse, Y. Horikawa, S. Ueda, C. Tanaka, M. Uto, M. Umeda, and T. Higuchi, J. Magn. Reson. Imaging, 1993, 3, 863. 33. M. E. Moseley, and M. F. Wendland, Proc. 10th Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 108.

Biographical Sketches M. E. Moseley. b 1951. B.S., 1973, North Carolina State University, Ph.D., 1980, Uppsala University. Postdoctoral work at Weizmann Institute of Science, Rehovot (with Z. Luz). Faculty at UCSF, 1985± 93; Stanford, 1993±present. Approx. 150 publications. Research interests include MR methods of mapping diffusion, perfusion and functional physiology. A. de Crespigny. b 1965. Ph.D., 1991, Cambridge University. Postdoctoral work at UCSF. Approx. 22 publications. Research interests include MR methods of mapping anisotropy in neonates, high-speed MRI, and image display and fusion.

BRAIN PARENCHYMA MOTION OBSERVED BY MRI

Brain Parenchyma Motion Observed by MRI Van J. Wedeen Harvard Medical School, USA

and Brigitte Ponceleti Massachusetts General Hospital, MA, USA

which descend at velocities as great as 5±10 mm sÿ1. By contrast, the cerebral mantle is almost immobile with respect to the calvarium. Velocities of intermediate size and complex 3D form are present in central deep gray and midbrain structures, among them the compressive motion of the thalami postulated by DuBoulay. Superimposed on the cardiac pulsation is a respiratory pulsation of a much smaller size.2 Observations of brain parenchymal strain patterns (see Section 5.1) suggest that variation in vascular turgor, particularly of the large arteries, may in¯uence the observed parenchymal motion. MRI measurement of CNS motion may be confounded by head movements of musculoskeletal or cardiac-ballistic origins.3

3 1 INTRODUCTION MRI studies of brain motion afford a unique window onto the mechanical homeostasis of the central nervous system (CNS). Velocity-sensitive MRI techniques may be based upon modulation of the magnitude or phase of the NMR image data, and have the capacity to detect the distributions of velocities, the displacements, or the material strainrates within the brain parenchyma. Having the capacity to resolve tissue displacements on the scale of micrometers, limited in practice by proton diffusion, NMR movies have mapped the physiological pulsation of the brain and spinal cord with the heartbeat and have measured brain parenchymal deformations due to everyday cranial accelerations. Clinical applications are projected to include the prediction of response to surgical treatment for a tethered spinal cord, and, potentially, the diagnosis of normalpressure hydrocephalus. Brain motion studies will be of value in elucidating the mechanisms of CNS trauma of both exogenous and endogenous forms.

1

OVERVIEW OF MRI OF CNS MOTION

CNS motion has been assessed by a variety of NMR imaging techniques. The earliest MRI demonstration of CNS motion was offered by Feinberg and Mark, who in 1987 presented echocardiogram (ECG)-gated movies of Fourier velocity spectra along selected 1D lines within the head, as shown in Figure 1.6,11 (For pioneering work on Fourier transform imaging spectra, antecedent to the present techniques and to phase contrast MRA, see Moran.12) Later investigators have estimated mean velocity using reduced numbers of velocity phase encoding steps and increased spatial or temporal resolution in conventional 2D multishot MRI,13±15 echo planar MRI,10 and realtime 1D-selective line MRI (`M-mode' NMR).16 The technique of NMR `interferography' of Hennig is a variant type of velocity encoding, based on a spatial modulation of local signal magnitude rather than phase.17 A quite different approach to CNS pulsation was taken by Lee, Wang, and Mezrich, who measured pulsatile changes in the volume of the lateral ventricles using conventional MRI.18

2 BRAIN MOTION Physiological brain motion re¯ects the reaction of the brain parenchyma, spinal cord, and cerebrospinal ¯uid (CSF) to changes in arterial and venous pressure and volume with the cardiac pulse and respiration.1 According to the Monro±Kellie doctrine, cranial entry of an arterial blood volume bolus with each systole must be matched by an equal egress of blood, CSF, and brain parenchyma from this incompliant compartment.2±4 Ultimately, parenchymal egress is effected by the movement of the cervical cord caudally through the foramen magnum. A detailed picture of the physiological brain motion has gradually emerged. DuBoulay postulated that CSF pulsation is pumped by the brain parenchyma, speci®cally the compression of the third ventricle by a medial motion of the thalami.5 MRI has since con®rmed this and has been instrumental in the emergence of a more complete view. Anatomically, brain parenchymal pulsation may now be seen chie¯y as a radial in¯ow of brain tissue inward and downward toward the foramen magnum: a herniation in miniature.6±10 For a superb treatment of multiaxial cine velocity data, see Greitz et al.8 Most rapid motion is seen in the cervical cord and brain stem,

Figure 1 Fourier-encoded cine velocity study for a selected line of tissue at the level of the foramen of Monroe. Cardiac gate delays from the R-wave are denoted. Cranio caudal velocities are depicted on the vertical axis, with single-pixel resolution of 0.4 mm sÿ1. Line selection was performed using orthogonal slice selects. The basic downward ``piston'' motion of the midbrain is demonstrated. (Reproduced by permission from Feinberg and Mark6)

2 BRAIN PARENCHYMA MOTION OBSERVED BY MRI 4 MOTION-SENSITIVE ECHO PLANAR MRI OF BRAIN MOTION The displacements associated with brain motion range from a typical maximum of 1±2 mm in the cervical cord down to unmeasurably small values in the cerebral mantle. Consequently, brain motion justi®es the highest feasible levels of motion sensitivity. MRI can achieve high velocity sensitivities by the method of magnetic gradient phase encoding, familiar in magnetic resonance angiography (MRA) and diffusion imaging and more recently in the imaging of myocardial motion.19,20 A pulsed gradient G is applied such that $0TEG(t) dt = 0 to induce a phase shift  for velocity V of  = kv . V, where kv = $0TE G(t) t dt. The velocity resolution of a phase-sensitive MRI experiment is approximately
V = N/kvS, where N is the desired number of standard deviations of discrimination and S is the signal-tonoise ratio (SNR) of the MRI phase over the region in question, typically equal to  times the SNR of the MRI magnitude.21 If the velocity changes over time, the phase shift  represents the time-average velocity weighted by the integral of the gradient: E(T) = $0TG(t) dt.10 4.1 Subtraction of Head Motion Rigid motion of the brain and skull can be an important confounding factor in high-sensitivity studies of brain motion.10 Head motion is caused by cardiac pulsation, respiration, and involuntary activity of skeletal muscles seen even in cooperative subjects with good head ®xation. Echo planar MRI has demonstrated that head motions of 1 mm sÿ1 may occur even under relatively optimal conditions. Using ordinary means of head ®xation, extrinsic skull motion and intrinsic CNS motion in the supratentorial region appear to be of similar magnitudes, but may be out of phase. In the cervical region, the craniocaudal motion of the cord is several-fold greater than extrinsic motion. The effects of rigid head motion in studies of brain parenchyma motion can be conveniently removed by mathematical means.10 Using the fact that a rigid motion de®nes a linear function, velocities of the skull are measured, and brain motion due to this rigid motion estimated by linear interpolation (brain parenchymal strain rate may be corrected for rigid motion by subtracting from each component qVi/qXj times its mean value over the skull).22 When this rigid component of motion is subtracted out, the speci®c motion of the brain relative to the skull is revealed. 4.2 Limitations MRI velocity sensitivity is ultimately limited by molecular diffusion and by the spatiotemporal characteristics of the motion pattern under study. Signal attenuation due to tissue water selfdiffusion limits the intensities and durations of the gradient pulses that may be usefully applied. A gradient pulse of intensity G and duration  establishes in the tissue parallel isophase planes of spatial period G diffusional attenuation becomes severe as the Einstein diffusion length (2D)1/2, where D is the diffusivity, approaches the isophase lamellar spacing.23±25 A second limit to velocity sensitivity is imposed by signal attenuation due to velocity gradients within each voxel. Velocity gradients may result from rigid rotation or material shear. In the simplest case, velocity and phase are linear functions of position.

In this case, image magnitude will be attenuated by a Fourier coef®cient of the spatial pro®le of the voxel.22 Typically, the voxel pro®le has the form of a rectangular function in the slice direction (z) and a sincfunction in each in-plane coordinate (x and y). Then the attenuation due to shear, ASH, becomes simply ASH ˆ rect …@=@x† rect …@=@y† sinc …@=@z†

…1†

when the spatial coordinates (x, y, z) are expressed in units of pixel diameters,  is in cycles and rect (u) = 1 if | u | < 12, 0 otherwise. Equation (1) implies that attenuation due to in-plane velocity shear is all-or-none: 0% attenuation when | q/qxi | < 1 1 2 cycle/pixel and 100% attenuation when | q/qxi | > 2 cycle/ pixel. No de®nite evidence of shear attenuation has to date been demonstrated in brain parenchyma, although such attenuation has been seen in the CSF in zones of high-shear such as the aqueduct.26 4.3

Normal Pulsatile Brain Motion

Qualitatively, the intrinsic brain motion consists of a primary systolic downstroke of the brainstem, which moves like a plunger within the relatively ®xed cerebral mantle, followed by a slower diastolic recovery toward the initial con®guration. In synchrony with this axial motion, the thalami move inward bilaterally to compress the third ventricle. Anteroposterior velocities are typically less than 25% of maximum craniocaudal or mediolateral velocities. Figure 2 shows the three components of intrinsic brain parenchyma velocity in axial slices in a normal volunteer. The slices are at the midventricular level, and cine data were acquired with 50 ms temporal resolution. Gray scale images show the magnitude and mediolateral, craniocaudal, and anteroposterior velocity component images 100 ms after the ECG Rwave. The curves plot the velocity components as functions of time for the elliptical regions of interest indicated. The net craniocaudal displacement 100 ms after the R-wave is shown as a 3D surface in Figure 3. Craniocaudal velocity in a midline sagittal plane is shown in Figure 4. The velocity curves [Figure 2(b)] show a monophasic or biphasic motion burst 50±200 ms after the ECG R-wave, synchronous with the systolic arterial pulse wave arrival, and subsequently in diastole much smaller velocities of recovery (diastolic RMS velocity < 20% systolic). The velocity images demonstrate the rapid caudal descent and of central structures, and the medial compressive motion of the thalami; anteroposterior velocities are relatively small. In normal volunteers, absolute CNS motion proves to be reproducible between heartbeats ‹15% (SD). It follows that conventional multishot MRI with an ECG trigger would produce reasonably accurate maps of CNS motions. 4.4

Tethered Spinal Cord

Under normal conditions, the spinal cord participates in the pulsatile movement of the CNS, with signi®cant vertical motion seen throughout cervical and upper thoracic levels. Physiological cord motion is normally rapid, with velocities of up to 10 mm sÿ1, and may exhibit several cycles of damped oscillation, possibly related to elastic recoil of the dentate ligaments.27

BRAIN PARENCHYMA MOTION OBSERVED BY MRI

Figure 2(a)

3

4 BRAIN PARENCHYMA MOTION OBSERVED BY MRI (b) Cephalic anterior right

1.5

Velocity (mm s–1)

1 0.5 0 –0.5

–1 Caudal posterior –1.5 0 left

100

200

300 400 500 600 Delay from R-wave (ms)

700

800

Figure 2 (a) Images of intrinsic brain velocity components for axial slices at the midventricular level in a normal subject 100 ms after the ECG Rwave, the approximate moment of peak parenchymal pulsatile velocities. To the left are the magnitude images; to the right are the mediolateral, craniocaudal and anteroposterior velocity component images with scales. The craniocaudal velocity map shows the rapid central descent (dark gray to black), and the mediolateral image shows the rapid bilateral medial velocities of the thalami (white±black dipole). The bottom image shows the anteroposterior velocity component. (b) The velocity±time curves for the left thalamic ROI designated by the circle. Parenchymal velocities are highest between 50 and 200 ms after the ECG R-wave during cardiac systole; much slower velocities of parenchymal recovery are seen beyond about 200 ms after the R-wave, during diastole. ~, cephalocaudal; *, anteroposterior; &, left±right

During normal development, the vertebral column undergoes greater craniocaudal growth than the spinal cord, and so the distal cord appears to ascend within the spinal canal. If the ascent of the cord is hindered by a mass, bone spur, or ®brous band, the cord may become stretched, which may lead by ischemia to symptoms of paresis, paresthesias, and loss of sphincter control. This syndrome is called the `tethered cord',

and is treated surgically. Given the normal physiological motion of the cord, it is reasonable to inquire whether changes in the normal pattern of motion could be a diagnostic marker of tethering. Investigators at The Johns Hopkins University School of Medicine and Georgetown have shown that motion of the cervical cord is a good predictor of clinical response to surgery

Figure 3 Net craniocaudal displacement of an axial slice 300 ms after the R-wave. Displacement is represented by the surface height, with a full scale of ‹3 mm. Image magnitude has been texture mapped onto the 3D surface. The predominant displacement is a monophasic downward movement of central structures; this ®gure corresponds to maximal displacement. Bidirectional CSF motion is seen in the horns of the lateral ventricles

BRAIN PARENCHYMA MOTION OBSERVED BY MRI

5

Figure 4 Craniocaudal velocity in a midline sagittal slice. A magnitude image is at the top, and gray-coded images of craniocaudal velocity triggered at various times after the R-wave are shown at the bottom. Velocities in general increase with proximity to the foramen magnum, this downstroke systolic and is synchronous with the known systolic egress of CSF from the calvarium. Maximal velocities in the upper cervical cord are about 3 mm sÿ1 caudally

for a tethered spinal cord.7,13 If cord motion is normal, the patient is unlikely to bene®t from surgical untethering, and an alternative etiology for neuropathic symptoms should be sought. Based on an experience of over 400 patients, MRI CNS motion studies are now standard at the Massachusetts General Hospital in cases of suspected tethering.

4.5 Normal-Pressure Hydrocephalus (NPH) NPH is a syndrome of uncertain etiology whose classic form includes dementia, ataxic gait, and urinary incontinence.28 Accurate diagnosis is imperative, as treatment can fully restore neurological function, whereas failure to treat may lead to progressive neurological dysfunction and death. At present, NPH diagnosis remains problematic, with no set of clinical or laboratory criteria clearly able to predict the individual response to CSF shunting.29,30 Prompted by recognition of the abnormal CSF dynamics in NPH,29,31 MRI studies of brain motion in NPH have been undertaken. Wahkloo et al. found brain motion in NPH patients to have greater amplitude and complexity than in age-matched controls.32 This ®nding is particularly interesting because of

the support it lends to hypotheses linking the neurophysiology of the NPH syndrome with endogenous mechanical trauma.33

5

NMR IMAGING OF PARENCHYMAL STRAIN RATES

Strain de®nes the degree of contraction and elongation at each point in a material during deformation. Velocity-sensitive NMR imaging can be used to reconstruct images of the distribution of physiological strain rates within tissues, including the myocardium and the brain parenchyma, as described here. By de®ning the pattern of tissue deformation at each location, strain imaging may shed light on both the mechanical causes and the speci®c neural effects of disorders of the mechanical homeostasis in the CNS. In an elastic material, strain is proportional to stress, which is the internal force per unit area, the constant of proportionality being the elasticity (in general, a fourth-order tensor). Strain rates are described by a strain rate tensor S whose components Sij are de®ned by the directional derivatives q/qxj of velocity components Vi according to

6 BRAIN PARENCHYMA MOTION OBSERVED BY MRI  Sij ˆ 1=2

@Vi @Vj ‡ @xj @xi

 …2†

The rate of tissue elongation or contraction in the direction of a unit vector v is simply: vT . S . v. Measurement of the complete 3D strain rate tensor S by equation (2) requires acquisition of complete 3D spatial and velocity data. As such complete acquisitions have not yet been performed in the brain, images have been produced that reconstruct tensor components based upon one or two components of velocity.20 To portray strain rate tensor ®elds usefully, novel display methods are needed. Let i and Ei denote the eigenvalues and eigenvectors of the strain rate tensor, respectively, and let f be a positive constant such that | fi | << 1 for all observed S. Then the value of S at each location can be depicted by means of a small rhomboid whose axes are given by the four vectors …1  f i †Ei

…3†

Intuitively, this rhomboid shows the con®guration of a unit square after S has acted upon it for a time equal to f. These rhomboids show the magnitude and direction of strain rate at each location; their long axes indicate the directions of most rapid stretch. Alternatively, a color code may be used to display the values of tensor invariants of the strain rate.

5.1

Normal Brain Parenchymal Strain Rates Due to Physiological Pulsation

Normal systolic parenchymal in-plane strain rate patterns are shown in Figure 5. In-plane (x and y) components of velocity were encoded in a gated cine echo planar acquisition, and strain rates were computed according to equation (2). Data were acquired using a velocity sensitivity | kv | = 3.7 rad s mmÿ1 (gradient pulse intensity G = 0.085 mT mÿ1, duration  = 35 ms, separation
= 40 ms). Given an MRI magnitude SNR per pixel of 25: 1, this experiment should have the capacity to resolve velocity differences of 0.02 mm sÿ1 per pixel (P < 0.05). De®ning the `strain rate' sensitivity ks so that the phase shift per pixel equals ks . S; then ks = Rkv where R refers to the dimensions of the voxel. In the experiments described, the strain sensitivity for in-plane strains is | ks | = 3.7 rad sÿ1 mmÿ1 (for 3 mm = 11 rad s). MRI motion sensitization in effect magni®es strain rates by the ratio of the strain rate sensitivity | ks | to the motion observation period
; here, | ks | /
= 275: 1. The display shows a strain rate rhomboid, constructed according to equation (3), and uses color to depict the trace, Tr (S) = 1 + 2. The trace represents the rate of compression or dilation of area within the 2D image plane. Because tissue is incompressible, in-plane area change implies compression or extension in the through-plane direction (i.e. Tr (S) + qVz/qz =

Figure 5 Images of in-plane 2D strain rates of the brain parenchyma in a normal subject in (a) coronal±oblique and (b) sagittal planes 150 ms after the ECG R-wave, corresponding to the rapid parenchymal motion of cardiac systole. A rhomboid graphic at each pixel shows the magnitude and orientation of the 2D strain rate tensor S at that location [according to equation (3) with f = 4 s]. Color indicates the rate of change of in-plane area, the trace of S, with red denoting expansion [Tr(S) > 0] and blue shrinkage [Tr(S) < 0]. The predominant in¯ow of brain parenchyma toward the foramen magnum is striking in both image planes. The coronal slice shows prominent downward tissue elongation in and near the internal capsule; the sagittal slice shows downward and forward deformation of the cerebellum and midbrain, and vertical stretching of the upper cervical cord, also seen in the coronal image. Compression of the thalami upon the region of the third ventricle is evident in the strain rhomboids in the coronal image and is also evident in the sagittal image, where red hues of in-plane expansion near the third ventricle (at center) suggest compression in the through-plane direction

BRAIN PARENCHYMA MOTION OBSERVED BY MRI

0). Color hue indicates the sign of Tr (S) at each location, with red corresponding to positive and blue to negative. Color saturation is proportional to | Tr (S) | , with full saturation at | Tr (S) | 5 0.05 rad sÿ1. Figure 5 shows coronal and sagittal images of strain rate 150 ms after the ECG R-wave. Strain rate rhomboids demonstrate the predominant elongation of parenchyma toward the foramen magnum. In the coronal image, there is seen a dramatic elongation in the region of the upper cervical cord, internal capsule, and corona radiata; in the sagittal image such elongation is seen in the cerebellum and midbrain. Maximal strain rates in these zones are on the order of 0.1 rad sÿ1. The red color in the region of the third ventricle in the sagittal image, denoting Tr (S) > 0, demonstrates the through-plane compressive motion of the thalami in this zone. Small strain rates are seen in the cerebral mantle. Cine velocity studies have shown, and strain rate cine studies have con®rmed, that motion in the territory of the basilar circulation precedes that in the middle cerebral territory cerebellar motion.9 Of interest, we see in the coronal image dilational strain patterns deep in the Sylvian ®ssures bilaterally. We can speculate that this re¯ects pulsatile motion of the middle cerebral vessels due to pulsatile turgor, the stiffening of the `vascular endoskeleton'.34 5.2 Brain Parenchymal Shear Due to Small Cranial Accelerations The human brain is an organ uniquely vulnerable to acceleration injury. Acceleration injury is associated with a variety of clinical syndromes, including acute concussion syndromes of mild to profound severity and delayed effects, as in the dementia pugilistica or `punch drunk' syndrome. An attractive unifying hypothesis is that these syndromes may be caused by acceleration-induced shear injury to deep cerebral white matter. This idea was advanced on theoretical grounds by Holburn, and soon received strong support in the experiments of Pudenz and Shelden, who, placing transparent windows into the skulls of monkeys, directly observed the swirling motion of the cerebral cortex due to cranial acceleration.35,36 Clinicopathological correlations in patients with severe concussion have demonstrated prominent damage to deep white matter tracts including the corpus callosum, cerebellar peduncles, and corticospinal tracts.37 Particularly intriguing were the ®ndings of hemorrhage at gray±white matter boundaries, suggesting elastic impedance mismatch as a cause of trauma. Similar lesion patterns have been seen in animal models of concussive nonpenetrating head injury.38 These experiments, however, have become increasingly dif®cult to justify due to concerns for animal welfare. MRI may afford the opportunity to characterize the response of the human brain to mild acceleration noninvasively, and, from this, perhaps to extrapolate the phenomenology of trauma. To investigate this, coronal images sensitive to anteroposterior velocity were acquired while a normal subject shook the head `No' twice per second for 10 s. Contributions of rigid head motion were subtracted as described above, and data were processed to demonstrate the qVz/qx component of parenchymal shear, where z is the through-plane coordinate and x is the horizontal coordinate. Figure 6 shows intrinsic parenchymal strain rates at three points in the head shake cycle. Accelera-

7

Figure 6 Brain parenchymal strain rates due to voluntary head shake. Images acquired during (a) clockwise cranial acceleration, (b) zero acceleration, and (c) counter-clockwise acceleration. The magnitude images are shown to the left and the strain rate images are to the right. The strain rate gray scale is proportional to qVz/qx, where x is the horizontal and z is the through-plane direction. Locally increased qVz/ qx, indicative of localized rotation or shearing strain, are seen adjacent to the falx, in bilateral vertical bands deep to the temporal lobes, and at the spinomedullary junction where the brainstem seems to twist slowly upon the upper cervical cord

tion-induced shear rates in this experiment are at most 0.1 rad sÿ1. This study demonstrates concentrations of strain adjacent to the falx, consistent with transmission of angular acceleration from the falx to the hemispheres. Shearing strain is seen to concentrate in parasagittal bands deep to the temporal lobes, and torsional strain to concentrate at the spinomedullary junction.

6

EFFECTS OF PARENCHYMA MOTION ON MRI OF PERFUSION AND DIFFUSION

Motion of the head and brain can interfere with studies of cerebral diffusion and cerebral function. Functional NMR studies of cerebral hemodynamic activation based on T1 in¯ow or T2* deoxyhemoglobin effects must detect changes in signal magnitude of 1%.39 These measurements, whether made with conventional or echo planar MRI, can be confounded by parenchymal in¯ow40±42 related to cardiac and respiratory motions.43,44 Cerebral diffusion measurements are exquisitely motion-sensitive; when acquired using conventional multishot MRI, these studies are easily degraded by uncontrolled cerebral motion leading to phase inconsistency among Fourier data lines. This has been addressed by cardiac gating,45 data postprocessing,46 compensation for measured motion using `navigator pulses',47 and by the adoption of single-shot acquisition techniques. Interestingly, whole-head motion has

8 BRAIN PARENCHYMA MOTION OBSERVED BY MRI overshadowed the effect of intrinsic brain motion in studies to date, suggesting the relative ineffectiveness of head ®xation. 7 PROSPECTS Motion-sensitive MRI opens a novel window onto the complex phenomena of brain parenchyma motion. This emerging picture of brain motion, however, represents only a ®rst step of a program to understand the underlying biophysics of brain motion. At issue is the desire to know intracerebral distributions of mechanical stress. At a basic level, the distributions of stress may reveal the sites of action of the cardiovascular impulse and of resistance to it. Clinically, stress concentrations may account for the speci®city of the neurological syndromes related to derangements of CNS dynamics (e.g. the relation of NPH to gait disturbance), and help predict the outcomes of intervention (e.g. how and when does fenestration of the optic nerve sheath alleviate the visual de®cit of pseudotumor cerebrii). One promising avenue of study is exempli®ed by the work of Pelc and Enzmann, who have measured concurrent intracerebral volume changes of brain parenchyma, blood, and CSF,15 an important step toward the uni®cation of brain motion with CSF dynamics in the spine and with the physiology of arterial pulse wave propagation.48±50 8 RELATED ARTICLES Assessment of Regional Blood Flow and Volume by Kinetic Analysis of Contrast-Dilution Curves; Methods and Applications of Diffusion MRI; Marker Grids for Observing Motion in MRI; Phase Contrast MRA; Time-of-Flight Method of MRA. 9 REFERENCES 1. J. E. A. O'Connell, Brain, 1943, 66, 204. 2. A. Monro, `Observations on the Structure and Function of the Nervous System', Creech and Johnson, Edinburgh, 1783. 3. G. Kellie, Trans. Med. Chir. Soc. Edinburgh, 1824, 1, 84. 4. L. B. Flexner, J. H. Clark, and L. H. Weed, Am. J. Phys., 1933, 101, 292. 5. G. H. DuBoulay, Br. J. Radiol., 1966, 39, 255. 6. D. A. Feinberg and A. S. Mark, Radiology, 1987, 163, 793. 7. D. C. McCullough, L. M. Levy, G. DiChiro, and D. L. Johnson, Pediatr. Neurosurg., 1990, 16, 3. 8. D. Greitz, R. Wirestam, A. Franck, B. Nordell, C. Thomsen, and F. Stahlberg, Neuroradiology, 1992, 34, 370. 9. D. R. Enzmann and N. J. Pelc, Radiology, 1992, 185, 653. 10. B. P. Poncelet, V. J. Wedeen, R. M. Weisskoff, and M. S. Cohen, Radiology, 1992, 185, 11. D. A. Feinberg, J. C. Hoenninger, L. E. Crooks, L. Kaufman, J. C. Watts, and M. Arakawa, Radiology, 1985, 156, 734. 12. P. Moran, JMIR, 1982, 1, 197. 13. L. M. Levy, G. Di Chiro, D. C. McCullough, A. J. Dwyer, D. L. Johnson, and S. S. L. Yang, Radiology, 1988, 169, 773. 14. R. Wirestam, D. Greitz, C. Thomsen, F. StaÊhlberg, F. Nordell, and M. Stubgaard, Proc. IXth Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 1107. 15. D. E. Enzmann, M. R. Ross, and N. J. Pelc, Proc. Xth Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 155.

16. S. E. Maier, C. J. Hardy, and F. A. Jolesz, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 151. 17. J. Henning, D. Ott, T. Adam, and H. Friedburg, JMIR, 1990, 8, 543. 18. E. Lee, J.-Z. Wang, and R. Mezrich, Am. J. Neuroradiol, 1989, 10, 1145. 19. N. J. Pelc, R. J. Herfkins, and L. R. Pelc, JMIR, 1991, 1, 181. 20. V. J. Wedeen, Magn. Reson. Med., 1992, 27, 52. 21. N. J. Pelc and M. A. Bernstein, Proc. IXth Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 475. 22. V. J. Wedeen, R. M. Weisskoff, and B. P. Poncelet, Magn. Reson. Med., 1994, July. 23. V. J. Wedeen, B. R. Rosen, and T. J. Brady, `The Magnetic Resonance Annual', Raven Press, New York, 1987, p. 174. 24. P. T. Callaghan and C. D. Eccles, JMIR, 1988, 71, 1. 25. D. G. Cory and A. N. Garroway, Magn. Reson. Med., 1990, 14, 435. 26. W. G. Bradley, Jr, A. R. Whittermore, K. E. Kortman, A. S. Watanabe, M. Homyak, L. Teresi, and S. J. Davis, Radiology, 1991, 178, 459. 27. D. J. Mikulis, M. L. Wood, B. Poncelet, O. A. M. Zerdoner, and D. E. Bohning, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 626. 28. R. D. Adams, C. M. Fisher, S. Hakim, R. G. Ojemann, and W. H. Sweet, N. Engl. J. Med., 1965, 273, 117. 29. R. P. Friedland, J. Am. Med. Assoc., 1989, 262, 2577. 30. N. R. Graff-Radford, J. C. Godersky, and M. P. Jones, Neurology, 1989, 39, 1601. 31. H. D. Portnoy, M. Chopp, C. Branch, and M. B. Shannon, J. Neurosurg., 1982, 56, 666. 32. A. K. Wahkloo, F. Jungling, and J. Hennig, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 627. 33. C. M. Fisher, Neurology, 1982, 32, 1358. 34. K. R. Davis, Personal communication, 1992. 35. A. H. S. Holburn, Lancet, 1943, ii, 438. 36. R. H. Pudenz and C. H. Shelden, J. Neurosurg., 1946, 3, 487. 37. S. J. Strich, Lancet, 1961, 443. 38. T. A. Genarelli et al., Acta Neuropath. Suppl., 1981, 7, 23. 39. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, and R. Turner, Proc. Natl. Natl. Acad. Sci., USA, 1992, 89, 5675. 40. J. R. Singer, Science, 1959, 130, 1652. 41. L. Axel, Am. J. Roentgenol., 1984, 143, 1157. 42. F. W. Wehrli, J. R. MacFall, L. Axel, D. Shutts, G. H. Glover, and R. J. Herfkins, Noninvasive Med. Imaging, 1984, 1, 127. 43. P. Jezzard, D. Le Bihan, C. Cuenod, L. Pannier, A. Prinster, and R. Turner, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1392. 44. R. M. Weisskoff, J. Baker, J. Beleiveau, T. L. Davis, K. K. Kwong, M. S. Cohen, and B. R. Rosen, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 7. 45. D. Chien, R. B. Buxton, K. K. Kwong, T. J. Brady, and B. R. Rosen, J. Comput. Assist. Tomogr., 1990, 14, 514. 46. J. J. M. Cuppen, J. P. Groen, J. J. E. In den Klef, and H. A. Tuithof, Proc. IVth Ann Mtg. Soc. Magn. Reson. Med., London, 1985, p. 962. 47. W. S. Kim, C. W. Mun, D. J. Kim, and Z. H. Cho, Magn. Reson. Med., 1990, 13, 25. 48. M. Chopp and H. D. Portonoy, J. Neurosurg., 1980, 53, 516. 49. B. Berman and G. Agarwal, Surg. Neurol., 1984, 22, 83. 50. M. Kosteljanetz, Acta Neurolog. Scand., 1987, 75, 5.

Biographical Sketches Van J. Wedeen. b B.A. (Mathematics), Harvard College, USA, 1977. M.D., Albert Einstein College of Medicine, USA, 1981. Certi®cation

BRAIN PARENCHYMA MOTION OBSERVED BY MRI in Nuclear Medicine, 1993. Introduced to NMR by T. J. Brady, B. R. Rosen, and I. Pykett. Department of Radiology, Massachusetts General Hospital, 1983±present. Approx. 30 publications. Research specialties: the theory and application of MRI to the study of motion and ¯ow, cardiovascular MRI, mathematical issues in neurocognitive functional MRI.

9

Brigitte P. Poncelet. b B.S. (Physics), Free University of Brussels, Belgium. Introduced to NMR by C. Segebarth and P. van Dijk. Ph.D. candidate, Free University of Brussels, 1989±present. Research fellow in Radiology, Massachusetts General Hospital, USA, 1990±present. Approx. 10 publications. Research specialties: the MRI study of motion and ¯ow, functional MRI.

Diffusion and Perfusion in MRI

resulted in the recent and fascinating incursion of MRI into the field of functional brain imaging. With MRI it is now possible to monitor completely noninvasively and with unmatched combined spatial and temporal resolution the activity of the brain during mental processes and the execution of cognitive tasks.

Denis Le Bihan Warren G. Magnuson Clinical Center, Bethesda, MD, USA

2 DIFFUSION MRI

1 2 3 4 5

Introduction Diffusion MRI Perfusion MRI Related Articles References

1

INTRODUCTION

1 1 6 11 11

The effects of diffusion on the NMR signal have been deeply analyzed and powerful diffusion measurement techniques using NMR have been extensively used in physics and chemistry. The recent application of such principles, coupled to NMR imaging (MRI) in the field of medical sciences, represents a spectacular and somewhat unpredicted development. The measurement of molecular displacements in biological tissues in vivo is of enormous interest in terms of its potential applications—from the determination of the molecular organization in tissues, to the emergency management of stroke patients, to the monitoring of laser surgery. MRI is the only means available today of approaching the molecular diffusion process in vivo noninvasively. In this article the basic principles of diffusion measurements made using NMR are presented, together with a description of the various methods that have been proposed for producing images of diffusion. Some technical considerations are discussed, followed by a description of the effects on diffusion measurements of the microdynamics and microstructure of biological tissues. Emphasis will be given to the difficulties encountered when implementing diffusion imaging, and the significance of diffusion measurements in biological tissues. It has been proposed that, because of its relative safety, its high spatial and temporal resolution, its ability to acquire images in any orientation, and its greater availability compared to other modalities such as positron emission tomography (PET), MRI can be used to image perfusion. Two approaches have been developed that allow MRI to be used to image quantities related to perfusion. One approach mimics conventional nuclear medicine methods, but uses radioactively inert external tracers. The other approaches are more original and MRI specific, using blood directly as an endogenous natural tracer. Such methods offer new insights, for instance valuable information can be obtained by monitoring variations in blood oxygenation in human brain cortex during activation tasks. This article covers the physical principles and fields of applications of these techniques, both tracer and blood pool agent based approaches and methods specific to NMR where flowing blood is magnetically labeled. Finally, a large section of this article is devoted to the ‘blood oxygen level dependent’ (BOLD) contrast approach which has

2.1 Diffusion and NMR

The effect of diffusion on the NMR signal can be understood from the bipolar pulsed gradient spin echo experiment (echo time TE ) suggested by Stejskal and Tanner1 (Figure 1). The first gradient pulse induces a phase shift ϕ 1 of the spin transverse magnetization, which depends on the spin position z 1 : ϕ1 = γ Gδz1

(1)

where γ is the gyromagnetic ratio, G is the gradient strength along the z axis, δ is the gradient duration. The time interval between the pulse onsets is denoted by . After the 180◦ rf pulse, ϕ 1 is inverted to −ϕ 1 . The second pulse will produce a phase shift ϕ 2 : ϕ2 = γ Gδz2

(2)

where z 2 is the spin position during the second pulse. The net transverse magnetization is: Mxy /M0 = exp[i(ϕ2 − ϕ1 )] = exp[iγ Gδ(z2 − z1 )]

(3)

where all relaxation effects have been included in M 0 . Obviously, for ‘static’ spins z 1 = z 2 , so the bipolar gradient pair is easily evaluated. For ‘moving’ spins, however, there is a net dephasing which will depend on the spin history during the time interval  between the pulses. Indeed, we must now consider a population of spins, which may have different motion histories. If P (z 2 , z 1 , ) dz 2 is the conditional 90˚

180˚

Echo

G

G

d

d ∆

Figure 1 Stejskal and Tanner diffusion spin echo sequence. Gradients pulses G are separated by a time interval ; their duration is δ

2 DIFFUSION AND PERFUSION IN MRI probability of finding a spin initially at z 1 between positions z 2 and z 2 + dz 2 after a time interval , the amplitude attenuation is:
Mxy /M0 = exp[iγ Gδ(z2 − z1 )]P (z2 , z1 , ) dz2 (4)

most widely used sequence in NMR diffusion measurements. Taking into account spin diffusion during δ, equation (6) becomes:

In the case of the diffusion process, one has:

The diffusion time T d can be formally defined as ( − δ/3), but its physical significance is clear only when the gradient pulse duration δ is sufficiently short compared with . Tissues with short T 2 values require short TE s which may not allow sufficiently long diffusion times to produce a large enough diffusion effect.

P (z2 , z1 , ) = (4πD)−1/2 exp[−(z1 − z2 )2 /4D]

(5)

where D is the diffusion coefficient. Combining equations (4) and (5), the signal attenuation is: Mxy /M0 = exp[−(γ Gδ)2 D]

(6)

In practice, as δ is usually not negligible compared with , movement during the application of the gradient pulses can no longer be neglected (see below). Also, the contribution of other gradient pulses, such as the imaging pulses, must be taken into account.2 Solving the Bloch–Torrey equation,3 one obtains:  Mxy /M0 = exp


− γ2

TE 0




t

G(t  ) dt 

2

  dt D

0

≡ exp(−b · D)

(7)

This relationship is valid for any gradient pulse combination, provided the sign of G is inverted for all gradient pulses following the 180◦ pulse. This expression may become analytically complex when many gradient pulses are involved, as in MRI. Therefore, it is useful to characterize the sensitivity of MRI sequences to diffusion using the ‘gradient factor’ b.2,4 One must remember, however, that this equation is valid only for diffusion in an infinite and homogeneous medium. If diffusion is restricted or anisotropic, the signal attenuation must be calculated using a more general formalism based on diffusion tensors.5 2.2

M/M0 = exp[−γ 2 G2 Dδ 2 ( − δ/3)]

(9)

2.2.2 Stimulated Echo Technique

A stimulated echo is generated from a sequence consisting of three rf pulses separated by time intervals τ 1 and τ 2 . Gradient pulses must be inserted within the first and the third periods of the stimulated echo sequence (Figure 2). The diffusion time now includes τ 2 and can be much longer than with the spin echo sequence, because only longitudinal relaxation occurs during this time interval. The Stejskal–Tanner relationship [equation (9)] still applies, provided that the period τ 2 is included in .8 The longer diffusion time is useful for studying very slow diffusion rates, or in compensating for the unavailability of large gradients.9 Unfortunately the signal is reduced by one-half compared with the spin echo signal. 2.2.3 Gradient Echo Technique

The effect of diffusion on the amplitude of a gradient echo formed by a bipolar gradient pulse pair of reversed polarity does not differ from that of a spin echo sequence.2 However, if the gradient echo is part of a steady state free precession (SSFP) sequence, where some degree of phase coherence is propagated throughout successive cycles, multiple echo paths with different diffusion times and different diffusion weightings must be considered.10 In MRI, the

The Different Approaches

Almost any NMR sequence can be designed to measure diffusion.

90˚

90˚

Stimulated echo

90˚

t1

t2

t1

2.2.1 Constant Field Gradient Spin Echo Method

G

G

In the presence of a simple constant linear gradient G 0 , the echo attenuation is:6,7

d

d

M(nT E)/M0 = exp(−γ 2 G20 DT E 3 /12n)

(8)

where n is the number of echoes in a multiple echo experiment. This simple approach is difficult to combine with NMR imaging because the presence of a constant gradient during the rf pulses of the imaging sequence severely impairs the selected slice profile. In addition, the relaxation time T 2 of the medium should not be too short, so that enough diffusion attenuation occurs during TE . 2.2.1.1 Bipolar Gradient Pulse Spin Echo Technique. This technique, suggested by Stejskal and Tanner,1 is by far the

∆

Figure 2 Diffusion-weighted stimulated echo sequence. The stimulated echo results from the spin excitation by three rf pulses separated by time intervals τ 1 and τ 2 . During τ 2 , relaxation is driven by T 1 and not T 2 . Diffusion effects can be enhanced by placing gradient pulses within the τ 1 periods, where transverse magnetization is sensitive to field inhomogeneities

DIFFUSION AND PERFUSION IN MRI

contrast-enhanced (CE-fast) scheme has been proposed.11 – 13 Unfortunately, this sequence remains very sensitive to motion artifacts.14 Moreover, the effects of diffusion and relaxation are intermingled so that diffusion measurements are always contaminated to some degree by relaxation effects.15

Diffusion measurements can also be achieved by means of the rf field (B1 ) produced by an NMR rf coil oriented perpendicularly to the main transmit/receive NMR coil.16,17 With rf gradients, extremely short switching times can be achieved, since there are no eddy currents. Furthermore, substantial gradient strength may be produced from the rf transmitter, allowing the measurement of very low diffusion coefficients.

2.3

0 –0.1 –0.2 log (S/S0)

2.2.4 Diffusion Measurements with B1 Field Gradients

3

–0.3 –0.4 –0.5 –0.6 –0.7 0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 b (×103 s mm–2)

Figure 3 Plot of the logarithm of the signal intensity in brain white matter against the gradient factor b. The plot is linear, and the slope is equal to the diffusion coefficient (D = 0.60 ± 0.01 × 10−3 mm2 s−1 )

Combining Diffusion NMR with MRI 90˚

‘Diffusion-weighted’ images are obtained by inserting gradient pulses into any NMR imaging sequence. For quantification it is necessary to determine the degree of diffusion weighting, which is achieved by calculating the factor b [equation (7)] associated with the acquisition parameters. Without additional gradients, typical spin echo imaging sequences have intrinsically very low b values, typically less than 1 s mm −2 , so that diffusion effects are completely negligible (for water, D = 2 × 10−3 mm2 s−1 at room temperature and the attenuation is less than 1%).2,18 When inserting gradient pulses for diffusion, however, the combination of the imaging and diffusion gradients produce cross terms in the b factors.2,19 These cross terms will depend on the time course of these gradient pulses and may be significant, so that the Stejskal–Tanner relation [equation (9)] is incorrect in most cases. Using the b factor values associated with each image, it is possible to compute diffusion images, i.e. images where the diffusion coefficient is determined for each pixel according to equation (7).2,4,9,18 The computation of such diffusion images is obtained by fitting the signal intensity of each pixel of a series of images differently sensitized to diffusion (i.e. with different b values) with equation (7) (Figure 3). Using the spin echo sequence, it is possible to vary the strength or the duration of the diffusion-sensitizing gradients, or their direction, in order to enhance anisotropic diffusion effects.2 The spin echo two-dimensional FT method is certainly the simplest to implement.4,18,20 Other sequences2 that have been investigated are stimulated echo sequences; line-integral projection reconstruction; variants of the SSFP technique; ‘turbo’ sequences; and echo planar imaging.21,22 These schemes have been suggested to overcome, at least partly, some of the problems encountered with the spin echo two-dimensional FT method. With echo planar imaging (EPI), the entire set of echoes needed to form an image is collected within a single acquisition period (single shot) of 25–100 ms. This is obtained by switching the echo signal formation in a train of gradient echoes, by means of a large gradient the polarity of which is very rapidly inverted as many times as is required to achieve the desired image resolution.23 EPI may easily be sensitized to diffusion21,22 (Figure 4).

180˚

Echoes rf Slice select Readout Phase encode

Figure 4 Diffusion echo planar imaging sequence. The spin echo is split into a series of gradient echoes by quickly switching the readout gradient. Sensitization to diffusion of an echo planar imaging sequence can be easily achieved by additional gradient pulses (hatched boxes)

Sensitization consists of providing a pair of large compensated gradients for a period of time before rapid gradient switching and data acquisition. The refocusing may be achieved either by simply reversing the polarity of the gradient halfway through the period over which it is applied, or by inserting a 180◦ rf refocusing pulse at the midpoint without reversing the gradient polarity. With EPI, motion artifacts are virtually eliminated. The accuracy on diffusion achieved with EPI is generally extremely good, as many images differently sensitized to diffusion can be generated due to the very short acquisition time (typically less than 100 ms). The EPI technique is the method of choice for in vivo diffusion imaging, although it is very vulnerable to susceptibility artifacts, which are responsible for image distortion or signal dropout, and to chemical shift artifacts which require efficient fat suppression. Diffusion-EPI has been successfully used to measure the diffusion of water in the human brain in volunteers and patients21 (Figure 5).

2.4 Experimental Considerations in Diffusion MRI 2.4.1 Gradient Requirements

Any mismatch between the diffusion sensitizing gradient pulses may cause artifactual signal losses, due to improper spin

4 DIFFUSION AND PERFUSION IN MRI

Figure 5 Diffusion images calculated from a set of 16 diffusion-weighted coronal echo planar images of the brain of a normal volunteer. These are 64 × 64 pixel images with a field of view of 16 cm and thus an in-plane resolution of 2.5 × 2.5 mm. The section thickness was 10 mm. The diffusion gradients varied from 0 to 38 mT m−1 in the z direction, the duration of each lobe was 20 ms, and their separation was 6 ms (b = 0 to 872 s mm−2 ). The acquisition time for each image was about 50 ms. Note that diffusion is highest in ventricular cavities containing cerebrospinal fluid (free water). Because of diffusion anisotropy, diffusion in corpus callosum is very low, while vertical frontal white matter tracts have higher diffusion (see Table 1)

rephasing. Two major sources of problems are commonly seen. One is gradient instability, which may arise when gradient amplifiers are driven hard for fast switching of large gradient intensities. Variations from shot to shot of the bipolar gradient balance results in widely distributed ghost artifacts. While such an artifact does not occur when a single-shot technique such as EPI is used, it is still essential to have high quality gradient amplifiers with some reserve capacity for accurate diffusion imaging. The second problem results from eddy currents generated mainly in the cryostat when switching rapidly large gradient pulses. Eddy currents may also be a major cause of gradient pulse mismatch, leading to signal loss. The best solution to this problem is undoubtedly to remove eddy currents at source by using actively shielded gradient coils, which have no fringe fields and therefore do not generate eddy currents. The use of a gradient coil of small dimensions, which remains at a fair distance from the magnet core and with which it is easier to generate large gradient amplitudes, is certainly an attractive alternative.21 2.4.2 Motion Artifacts

As the sequences used are deliberately sensitized to motion by the addition of large gradients, a major problem occurring with in vivo imaging of diffusion arises from irregular motion of the object. Artifacts result from discontinuities that occur between successive cycles of an imaging sequence. Results of such temporal incoherence are commonly visible as ‘ghosts’ along the phase-encoding direction. These ‘ghosts’ are particularly intense in the presence of the diffusion gradients and render the diffusion measurements meaningless. Cardiac gating has been used to mitigate this problem, but even this motion is not strictly cyclic. It is difficult to compensate diffusion imaging sequences for motion, since the use of successive bipolar gradient pulses considerably reduces the

value of the gradient factor b.2 Chenevert et al.24 have suggested eliminating any phase encoding from the acquisition by sacrificing one dimension of the image, which is then reduced to a single line. Ultimately, however, the best way to avoid motion artifacts is to use a single-shot technique, such as EPI, and to secure patients comfortably within the magnet using cushions or inflating devices. 2.5 Diffusion in Biological Systems

The diffusion coefficient of water in tissues has been found to be 2–10 times less than that of pure water (Table 1). This can be largely understood considering that water molecules are obliged to divert tortuously around obstructions presented by fibers, intracellular organelles, or macromolecules. Biological systems thus differ greatly from an ‘infinitely large medium’; they are very heterogeneous and made of multiple subcompartments (microstructure). Depending on the permeability of the barriers that limit these compartments, we have to consider exchanges and transport Table 1 Diffusion Coefficients of Water in Human Braina ×103 mm2 s−1 Cerebrospinal fluid Gray matter White matter Corpus callosum Axial fibers Transverse fibers

2.94 ± 0.05 0.76 ± 0.03 0.22 ± 0.22 1.07 ± 0.06 0.64 ± 0.05

a Measurements obtained in normal volunteers using a diffusion sensitized EPI sequence.21 Direction of fibers refers to z axis (direction of the diffusion sensitizing gradient pulses).

DIFFUSION AND PERFUSION IN MRI

5

between them (microdynamics). A classical treatment of the NMR signal may not then reflect properly the tissue structure or properties. In these conditions diffusion coefficients may become meaningless if the measurement timescale or the measurement direction are not provided. The main difficulty is that the structure of the medium is generally unknown in detail. 2.5.1 Restricted Diffusion

Diffusion is restricted when boundaries in the medium prevent molecules from moving freely.2,25,26 When measurement times are very short, most molecules do not have enough time to reach boundaries, and so they behave as if diffusing freely. Once the diffusion time increases, an increasing fraction of molecules will strike the boundaries, and diffusion will deviate from the free behavior. The diffusion distance, calculated as for free diffusion using Einstein’s relation ζ 2  = 2 DT d , shows a curvature and, finally, a levelling-off when the diffusion distance reaches the size of the restricting compartment. The effects of restriction, which depend on the shape of the restricting volumes and the types of restriction,2 will thus appear in the NMR signal for diffusion times such that molecular displacements are of the order of the size of the restricting volumes. When the restrictive barriers become permeable to diffusing molecules, the restricted diffusion pattern changes. An example has been given by Tanner27 for the case of equally spaced, plane, and permeable barriers. When the diffusion time is increased, the apparent diffusion coefficient decreases, as expected for restricted diffusion, but saturates at a level which depends on the permeability constant. 2.5.2 Anisotropic Diffusion

Diffusion in tissues may also be anisotropic if hindrance or restriction is not the same for different directions of motion. The measured diffusion coefficient then varies according to the direction of measurement. Examples include muscle28 and brain white matter (see Figure 6).29 To consider anisotropy, the diffusion ‘coefficient’ must be replaced by a tensor, D.1,5 The diagonal terms of the diffusion tensor D xx , D yy , and D zz represent molecular mobility along the axes x , y, and z of the laboratory frame of reference, while the off-diagonal terms indicate how displacements are coupled in directions which are mutually perpendicular. The echo attenuation must be rewritten as:     bij Dij (10) M/M0 = exp − i=x,y,z j =x,y,z

where b ij are components of a matrix b and D ij are components of the diffusion tensor.5 Off-diagonal terms in the diffusion tensor may be important when the gradient directions do not coincide with the orthotropic directions of the tissue. The echo attenuation then depends on a combination of the diagonal and nondiagonal terms of the diffusion tensor. The three orthotropic axes of the tissue coincide with the eigenvectors of D. The diffusivities along these principal directions are the eigenvalues of D. The main direction of the tissue, e.g. the fiber-tract direction for muscle of white

Figure 6 Diffusion image showing anisotropic diffusion in white matter. Excellent contrast is achieved, although T 1 and T 2 effects have been removed. Corpus callosum and temporal white matter fibers, which are horizontal, are dark. Vertical corona radiata frontal fibers and internal capsule fibers are bright. In brainstem, vertical fast-conducting motor and somatosensory tracts are also bright

matter, is given by the eigenvector with the largest principal diffusivity. Also associated with D are scalar invariants, such as Tr(D), which are independent of the orientation of the sample with respect to the laboratory frame of reference; they depend only on the tissue microstructure.5 In white matter, the results of the measurements depend on the respective orientation of the myelin fiber tracts and the gradient direction at each different image location. Diffusion coefficients are significantly lower when the myelin fiber tracts are perpendicular to the direction of the magnetic field gradient used to measure molecular displacements (Figure 6). Considering that most NMR visible water is in the axons, a simple model would be to consider that water molecules are enclosed in the axonal spaces and that water diffusion outside the axons is prevented by the myelin sheath. Recent diffusion experiments have not been able to show complete restricted diffusion, since the diffusion distance increases linearly with the square root of the diffusion time, as for free diffusion30 (Figure 7). The reduced value of the diffusion coefficient across myelin fibers can thus only reflect a decreased water mobility through the successive lipid layers. 2.5.3 Diffusion in Multiple Compartment Systems

Most diffusion measurements in biological tissues refer to an ‘apparent’ diffusion coefficient (ADC), and it is generally considered that diffusion in the measurement volume (the ‘voxel’, in imaging) has a unique diffusion coefficient: Da =



ρi Di

(11)

i

This simplification may not always be legitimate, since most tissues are made of multiple subcompartments—at the least, intracellular and extracellular compartments. Assuming measurement times are short, so that diffusion is unrestricted

6 DIFFUSION AND PERFUSION IN MRI 37

12 10 8 6 Gray matter White matter, x White matter, z

4 2 0 0

2

4

6

8

10

Td (ms1/2)

Figure 7 Plot of the diffusion distance against diffusion time in gray and white matter. The diffusion distance was obtained from the diffusion coefficient using Einstein’s relationship. There is no leveling-off of the diffusion distance, suggesting that water diffusion is not completely restricted

Temperature (MRI) (˚C)

Diffusion distance (m m)

14

32

T1 Diffusion

27

22 22

in each subcompartment i , and there is no exchange, the signal attenuation is:  ρi exp(−bDi ) (12) M/M0 =

27 32 Temperature (Luxtron) (˚C)

37

Figure 8 Comparison of temperature measurements using T 1 and diffusion MRI, and using probes (Luxtron) in a phantom. The accuracy was found to be 0.2◦ C with diffusion and 0.5◦ C with T 1

i

where ρ i is the density of molecules diffusing in compartment i , and D i is the associated diffusion coefficient. In this case, the ‘apparent’ diffusion coefficient that would be measured would depend on the range used for b and would not reflect properly the diffusion in the voxel. Measurements with low b values would then be more sensitive to fast diffusion components. The ideal approach would be to separate all subcompartments by fitting the data with a multiexponential decay. Unfortunately, the values for D i are often low and not very different from each other, so that very large b values and very high signal-to-noise ratios would be required.

2.6

Artery

Microcirculation exchanges

Vein

Blood flow

Conclusion

Diffusion thus appears to be a powerful source of contrast for tissue characterization or functional studies. It has been shown that there is no correlation between the diffusion coefficient and the relaxation times T 1 and T 2 . T 1 and T 2 may be normal or elevated in diseased states, while diffusion is lowered, as shown in early brain ischemia.31 – 33 Other applications include temperature imaging, a technique based on the sensitivity of diffusion coefficients to temperature34 (Figure 8). Temperature imaging is especially useful for monitoring interventional procedures.

3

Capillaries

PERFUSION MRI

Classical techniques used to measure and image perfusion can be split into three categories: methods based on diffusible tracers (Figure 1); methods based on intravascular tracers (blood pool tracers); and deposition techniques (microspheres) (Figure 9). Conventional perfusion measurement techniques based on radionuclide tracers have led to the identification

Tissue (mass M) Microspheres Diffusible tracers Pure intravascular tracers

Figure 9 Principles of conventional perfusion measurements. Microspheres are particulates which, because of their size, are trapped at the entry of the capillary network; a count of their deposition pattern gives the flow rate. Diffusible tracers cross the capillary wall; their concentration in the tissue reflects blood flow. Intravascular tracers remain inside the capillary compartment and may be used to evaluate blood flow (under some assumptions)

of ‘perfusion’ with ‘blood flow’ in the tissue. However, perfusion has also often been used simply to denote the degree (density) of normal or abnormal microvasculature in a tissue as seen, for instance, with conventional catheter angiography.

DIFFUSION AND PERFUSION IN MRI

Diffusible Tracers

0.2

Blood flow measurements using diffusible tracers were first introduced by Kety and Schmidt.35 Blood flow can, in principle, be determined from two parameters: the arterial input C a (t), which is assumed to be uniform across all voxels, and the tissue tracer concentration, measured by means of the external imaging detector. The tracer tissue concentration at time T , C 0 (T ), is related to the normalized arterial flow F (in mL min−1 per 100 g of tissue):
C0 (T ) = (F /λ) exp[−(F /λ)T ]

pCO2 = 48.5 T CBF = 72 ml min–1 per 100 g

0.1 pCO2 = 23.5 T CBF = 29 ml min–1 per 100 g

T

Ca (t) exp[(F /λ)T ] dt 0

(13) where λ is the blood/tissue partition coefficient of the tracer. With NMR, the tracers are nonproton nuclei, such as 19 F compounds, D2 O or H2 17 O; so far these have been used in animal studies. Tissue concentration is determined from integrated spectral line areas with spectroscopy techniques or from pixel signal intensities with MRI methods after tracer has been administered, and the arterial input function is derived from blood samples or, in some cases, from monitoring the NMR signal in a peripheral artery (either on-line or off-line). In general, one has to deal with multiple compartment fitting, e.g. gray/white matter in the brain, signal contamination from surrounding muscles, and tracer recirculation. 19 F has a sensitivity and a resonant frequency which are close to those of the proton, and is thus easily observed by NMR. Most studies have used inhalation of inert Freon gases, such as FC22 (CHClF2 )36 or FC23,37 – 39 which appear to be less toxic and induce fewer changes in brain physiology. Deuterium is used as a tracer injected intraarterially in the form of deuterated water (D2 O), which quickly exchanges protons with the surrounding water to become HOD.40 HOD behaves very similarly to normal water, and the potential toxicity is limited at low concentration. The low sensitivity of the nucleus and its short T 2 result in low signal-to-noise ratios. 17 O is a stable isotope of natural abundance of 0.0037%, is chemically equivalent to 16 O and has a very low toxicity, even at somewhat high concentrations. The current limitation to its use is its low availability and high cost. While the 17 O NMR signal has been observed directly,41 other workers have based their approach on the scalar coupling of 17 O which relaxes water protons efficiently.42,43 Much higher signal-tonoise ratios may thus be reached by observing 17 O water through proton–hydrogen MRI (Figure 10). 17 O may also be administered as H2 17 O and used similarly to H2 15 O with PET to evaluate perfusion, or as a gas to evaluate tissue oxygen extraction.41 3.2

Relative concentration (AR2)

3.1

7

0.0 0

2

4

6

Minutes

Figure 10 Perfusion measurements based on relaxivity effects of [17 O]water. Cerebral blood flow (CBF) was found to increase with pCO2 , as expected (from ref. 43, with permission)

amplification effect over the simple relaxivity effect. These gradients are responsible for spin dephasing which leads to significant signal loss, especially with long echo time gradient echo sequences and high field strengths (Figure 11). At ‘clinical’ concentrations, i.e. 0.1 mM kg−1 , the relationship between the tissue agent concentration and the NMR signal loss is experimentally linear:44 R2∗ = (1/T2∗ ) = k[Gd]

(14)

The constant k has been empirically determined in the brain, but potential variations between tissues, and especially abnormal tissues, remains an important issue.

‘Blood Pool’ Agents

If a nondiffusible NMR paramagnetic agent, such as gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA), is injected into the blood stream in the form of a bolus, the difference in magnetic susceptibility between the tissue and the blood where the contrast agent is compartmentalized results in internal magnetic field gradients, which extend far beyond the capillary limits and, therefore, represent a significant

Figure 11 Set of 16 echo planar images acquired after injection of a bolus of Gd-DTPA. Images were obtained at 1-s intervals. The transit of the contrast agent bolus through the brain capillaries produces a signal drop which lasts a few seconds. This drop is higher in gray matter than in white matter. The limited signal drop in the low grade astrocytoma suggests that the tumor has a low capillary density (blood volume)

8 DIFFUSION AND PERFUSION IN MRI Capillaries

0.2 +

log (Signal)

0

++ ++ +

+ ++ ++ +

+++

++ + +

++ ++

++ + ++ + ++ ++

+ + ++ ++ + ++ + ++ + + + + ++ ++ + ++ + +++ ++

Artery

Vein

–0.2

v

–0.4 +

–0.6

Gray matter SR2 = 150 Tumor 1 SR2 = 121 White matter SR2 = 44 Tumor 2 SR2 = 456

L

–0.8 60

Arterial input

70

Figure 12 Typical magnetic resonance signal time course study after injection of a bolus of Gd-DTPA in a patient with a high-grade glioma. Each curve corresponds to a given region of interest. The necrotic center of the tumor is completely avascularized. The tumor wall has a very large blood volume. Note that, due to the breakdown of the blood–brain barrier in the tumor, the contrast agent starts leaking inside the tumor, where T 1 effects (which counteract the susceptibility-related signal drop) may lead to underestimation of the blood volume SR2, area under first-pass curve [proportional to cerebral blood volume (CBV)].

As with other blood pool tracer approaches,45 information on perfusion (i.e. blood flow, blood volume, or mean transit time) may, in theory, be derived from the tissue concentration–time curve. In practice, the local blood volume V derived from the magnetic resonance signal time course is proportional to the area (integral) under the tissue concentration–time curve:
(15) V ∝ Fa /Q · S(t) dt

Tissue residue, t = L/v

CT (t)

30 40 50 Time after injection (s)

Time

Time

Actual bolus

Actual tissue concentration

CT (t)

20

CA(t)

10

CA(t)

0

Time

Time

Figure 13 Diagram showing that the actual tissue concentration time course is the product of convolution of the tissue residue function and the arterial input function. (Adapted from Axel45 )

where Q is the total amount of tracer injected as a tight bolus, and F a is the arterial blood flow. Only the relative blood volume is determined but, assuming the same blood comes from a common feeding artery, comparison can be made between different parts of the brain (Figure 12). Several difficulties have to be overcome when integrating the tissue concentration over time, such as the recirculation of the tracer.46 As for the blood flow, in principle the recirculation of the tracer could be determined using the Stewart–Hamilton theorem: Fa = V /MTT

(16)

where MTT is the mean transit time, i.e. the average time required by the tracer to pass through the tissue.47 Ideally, the tracer should be introduced into the vascular compartment as an impulse bolus. In practice, the mean transit time could be determined by deconvolving the tissue concentration–time curve by the arterial input function obtained by monitoring the contrast agent concentration in a feeding artery using MRI (Figure 13). Without the knowledge of the arterial input function, the area under the tissue tracer concentration–time course only gives the regional blood volume, and not blood flow.48 This technique has been used clinically to produce semiquantitative maps of the regional cerebral volume, for instance in brain tumors (Figure 14). Under some circumstances, however, such as brain activation studies, blood volume evaluation may provide useful information on hemodynamic changes.49

Figure 14 Blood volume map calculated from a set of echo planar images obtained during the transit of a bolus of Gd-DTPA. The tumor shows a rim with very high blood volume, while the necrotic center remains black. The blood volume is 2–4 times larger in gray matter than in white matter

3.3 Magnetic Labeling

3.3.1 Methods Based on Phase Shifts

Spins moving in the presence of a magnetic field gradient get out of phase. Depending on the capillary geometry and circulation conditions on the one hand, and the MRI acquisition parameters on the other, perfusion can be viewed

DIFFUSION AND PERFUSION IN MRI

as a predominantly incoherent motion (intravoxel incoherent motion, IVIM)50 or a predominantly coherent motion (intravoxel coherent motion, IVCM).51 3.3.1.1 IVIM. When the capillary bed is sufficiently tortuous, phase shifts of flowing spins are distributed within each voxel and result in signal attenuation. A simple approach considers microcirculation as a pseudodiffusion process.50 The signal attenuation A depends then on a ‘pseudodiffusion coefficient’ D*: A ∝ exp(−bD ∗ )

(17)

D* = lv /6, where l is the mean capillary segment length, and v is the average blood velocity; b is the gradient factor defined in equation (7). Pseudodiffusion has, in practice, to be separated from bulk water diffusion, which requires a high signal-to-noise ratio52 and hardware stability, given the very small blood volume fraction f in the brain (2–4%). The use of EPI allows one to obtain a large number of images in a short time and significantly reduces the occurrence of motion artifacts21 (Figure 15). The IVIM parameters, f and D*, may be related to actual perfusion in terms of blood flow F under some assumptions about capillary network organization:53 ∗

F = 6f D /(ρlt l)

(18)

where ρ is tissue density and l t is the average total capillary length (between the arteriole and the venule). An interesting and successful approach is to use 19 F blood substitutes (perfluorocarbons), which remain in the vascular compartment.54 Paramagnetic contrast agents may also be used to enhance signal from flowing blood.55 3.3.1.2 IVCM. Considering now the coherent part of microscopic flow, even echo rephasing has been used in an attempt to observe perfusion in first and second echo difference images,56 but variations in T 2 tend to mask the Intercept [f (flowing blood volume) = 8 ml per 100 g] Initial slope (D* = 11 × 10–3 mm2 s–1)

4.7 4.5

ln (S)

4.3 Slope at high b values (D = 1.1 × 10–3 mm2 s–1)

4.1

9

effects of perfusion. A more powerful technique uses velocity compensation. It is possible to devise two sequences with the same echo time, one of which refocuses the effects of blood moving with constant velocity along capillaries (‘flow compensated’), while the other gives an additional attenuation resulting from this motion. A difference image then highlights the effects of perfusion. This effect has been shown in a model of rat brain ischemia, where the perfusion deficit is clearly visible as the disappearance of the difference between the two images.57 This technique, however, requires excellent hardware stability and high signal-to-noise ratio. The order of flow compensation by gradient nulling may be increased to take into consideration the tortuosity of the capillary network. 3.3.2 Arterial Water Spin Labeling

In the experiment proposed by Williams et al.58 to look at brain perfusion, a series of rf inversion pulses is used to saturate continuously or invert spins of arterial blood water protons. The inflow of these inverted spins in the imaging volume results in a small signal loss S /S 0 , which depends on blood flow F : F = (λ/T1app )(S/S0 − 1)/2α 

(19)

where λ is the brain/blood partition coefficient, α  is the effective degree of arterial inversion, and T 1app is the apparent spin–lattice relaxation time in the presence of arterial inversion. In practice, one has also to consider T 1 recovery of labeled flowing spins between the excitation and the detection slices, and cross relaxation between tissue water and macromolecules. A critical parameter is the degree of arterial spin labeling α  , which depends on the flow velocity and the technique used to label spins. Steady-state spin inversion has been successfully achieved using adiabatic fast passage (AFP) spin inversion. Quantitative perfusion maps have been obtained with this technique at high magnetic fields in animals, which have a short distance between neck and brain. High magnetic fields may be required to achieve a sufficient signal-to-noise ratio and dispose of longer T 1 values. A variant of this technique is to invert or saturate spins in the imaged slice and to monitor entry of fresh spins. Results may not be quantitative, but variations in blood flow during brain activation tasks may be observed.59

3.9 Gray matter

3.7

3.3.3 Blood Oxygenation Level Dependent (BOLD) Contrast

White matter 3.5 3.3 0

200

400 600 b (s mm–2)

800

1000

Figure 15 Plot of the logarithm of the signal decay versus gradient b factor in the presence of IVIM (experiment carried out at 1.5 T on a 70year-old human volunteer using IVIM EPI). A series of 16 differently diffusion-sensitized spin echo images was collected. For white matter, the plot follows a straight line, as expected for a pure diffusion process. The slope of the straight line gives the diffusion coefficient D. For gray matter, the plot clearly shows a curvature effect for low b values. This curvature has been ascribed to microcirculation effects. The straight line obtained for large b values gives the diffusion coefficient. The deviation from the diffusion asymptote, as measured at the intercept, gives the perfusion fraction f . D*, blood velocity

Although the approach based on Gd-DTPA was a significant breakthrough for functional brain imaging, it had several limitations, in particular the requirement to use an external contrast agent. The number of measurements is therefore limited (for instance, to one baseline and one activation image) and brain function cannot be evaluated in real time. Based on earlier works describing the magnetic properties of hemoglobin60 and their effect on the magnetic resonance signal,61 it was recently suggested that deoxyhemoglobin (which is paramagnetic, although oxyhemoglobin is not) in red blood cells could be used as an endogenous contrast material.62,63 Kwong et al.,59 closely followed by others,64,65 showed that MRI could be used to monitor in real time the local modulation of the level of blood oxygenation associated

10 DIFFUSION AND PERFUSION IN MRI 4

Signal change (%)

3 2 1 0

–1 Stimulus –2 On

Off

–3 0:06

0:36 Time (min:sec)

1:06

Figure 16 Typical magnetic resonance (MR) signal time course study during brain activation by a visual stimulus using the BOLD method (1.5 T). Gradient echo EPI images were obtained at 3-s intervals. During the stimulus the MR signal increases. The amplitude of the signal change is usually small, but largely exceeds baseline fluctuations. After stimulation, the MR signal sometimes goes below baseline. There is a delay of several seconds between the beginning of the activation task and the MR signal change, due to the delayed hemodynamic response over the neuronal response

with brain activity on an individual basis, without the need for intersubject averaging. During brain activation tasks, there is a large local increase in blood flow which overcomes a small increase in tissue oxygen uptake. There is, therefore, a surplus of oxyhemoglobin over deoxyhemoglobin, resulting in a reduction of the susceptibility effects and thus in increased MRI signal (Figure 16). The effect is usually small (a few per cent) and depends on the sequence parameters, the voxel size, and the field strength.66,67 Recent articles have demonstrated the tremendous potential of MRI to aid in our understanding of how the brain works, with unmatched combined spatial and temporal resolution68 – 71 (Figure 17). Besides exploration of the normal brain, potential clinical applications include presurgical mapping,72 monitoring of recovery after stroke or head trauma, and monitoring the fate of neuropharmacological agents.73 Presurgical mapping consists in functional evaluation of brain tissue to be removed or spared during surgery, for instance in cases of intractable complex partial seizures if language is to be spared during lobectomy. So far, hemispheric language dominance is determined from invasive and risky tests, such as the Wada test which is based on intracarotid amobarbital injection, and electrical cortical recording through electrodes positioned intraoperatively. It is clear that functional MRI (FMRI) language studies have the potential to replace such invasive tests.68,74 On the other hand, FMRI could also be used to visualize epileptic foci with, or even without, associated clinical seizure. The use of FMRI for monitoring of functional recovery after stroke or head trauma appears especially important for patients with hemiplegia or aphasia. Studies may be repeated at will on individuals during the recovery period without concerns about irradiation. MRI could also be used to visualize regions of the brain where receptors have been activated or inhibited by neuropharmacological agents through effects on blood flow and blood oxygenation, as we recently suggested using

Figure 17 Activation map of the visual cortex. Activation maps can be calculated by identifying pixels where significant magnetic resonance signal change occurred during stimulation. The degree of activation is shown using a color scale. The raw BOLD images were obtained using a spoiled-GRASS gradient echo (SPGR) sequence with a spatial resolution of 0.8 mm × 1.6 mm. The activation map is overlayed on a high resolution anatomical image to show the exact location of activation

neurochemical activation by arecholine in monkey brain.73 Also, the modulation by such drugs of the response of the brain to specific stimuli or cognitive tasks could be evaluated. There are, of course, some limitations when using FMRI. Patient cooperation is particularly necessary. Not only is their active participation required for the activation tasks, but they are also required to stay absolutely immobile during scanning in order to avoid misregistration artifacts. Registration algorithms are being evaluated, as well as sophisticated statistical analysis packages. Another concern regards the size of the vessels (veinules or feeding arteries) responsible for the effect. If the vessels are not small enough, they may correspond to a fairly large territory, so that the area of activation may differ from the location where activation is seen on the images.

3.4 Conclusion

MRI appears to be a very promising imaging modality for looking at microcirculation and perfusion. Compared with other approaches, MRI may provide images of high temporal and spatial resolution totally noninvasively. However, today, methods which give the best spatial and/or temporal resolution are not always quantitative, at least in terms of blood flow or other physiological parameters, often due to the lack of an adequate model. On the other hand, there are techniques, most of them using the basic principles of more conventional methods, which are potentially quantitative, but remain difficult to implement and give images with limited spatial and/or temporal resolution. It is likely that progress

DIFFUSION AND PERFUSION IN MRI

will be made in both directions, so that physiologists and clinicians will soon have available to them methods which give quantitative information on perfusion with unmatched spatial and temporal resolution.

4

RELATED ARTICLES

Anisotropically Restricted Diffusion in MRI; Brain: Sensory Activation Monitored by Induced Hemodynamic Changes with Echo Planar MRI; Diffusion: Clinical Utility of MRI Studies; Whole Body Magnetic Resonance Angiography.

5

REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9.

10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.

E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288. D. Le Bihan, Magn. Reson. Q., 1991, 7, 1. H. C. Torrey, Phys. Rev., 1965, 104, 653. D. Le Bihan, E. Breton, D. Lallemand, P. Grenier, E. A. Cabanis, and M. Laval-Jeantet, Radiology, 1986, 161, 401. P. J. Basser, J. Mattiello, and D. Le Bihan, J. Magn. Reson., 1994, in press. E. L. Hahn, Phys. Rev., 1950, 80, 580. H. Y. Carr and E. M. Purcell, Phys. Rev., 1954, 94, 630. J. E. Tanner, J. Chem. Phys., 1970, 52, 2523. K. D. Merboldt, W. Hanicke, and J. Frahm, J. Magn. Reson., 1985, 64, 479. R. Kaiser, E. Bartholdi, and R. R. Ernst. J. Chem. Phys., 1974, 60, 2966. D. Le Bihan, Magn. Reson. Med., 1988, 7, 346. D. Le Bihan, R. Turner, and J. R. MacFall, Magn. Reson. Med., 1989, 10, 324. K. D. Merboldt, W. Hanicke, M. L. Gyngell, J. Frahm, and H. Bruhn, J. Magn. Reson., 1989, 82, 115. K. D. Merboldt, H. Bruhn, J. Frahm, M. L. Gyngell, W. Hanicke, and M. Deimling, Magn. Reson. Med., 1989, 9, 423. E. X. Wu and R. B. Buxton, J. Magn. Reson., 1990, 90, 243. D. Canet, B. Diter, A. Belmajdoub, J. Braondeau, J. C. Boubel, and K. Elbayed, J. Magn. Reson., 1989, 81, 1. G. S. Karczmar, D. B. Twieg, T. J. Lawry, G. B. Matson, and M. W. Weiner, Magn. Reson. Med., 1988, 7, 111. D. Le Bihan and E. Breton. CR Acad. Sci. Paris, 1985, 301, 1109. M. Neeman, J. P. Freyer, and L. O. Sillerud, J. Magn. Reson., 1990, 90, 303. D. G. Taylor and M. C. Bushell, Phys. Med. Biol., 1985, 30, 345. R. Turner, D. Le Bihan, J. Maier, R. Vavrek, L. K. Hedges, and J. Pekar, Radiology, 1990, 177, 407. R. Turner and D. Le Bihan, J. Magn. Reson., 1990, 86, 445. P. Mansfield, J. Phys. C: Solid State Phys., 1977, 10, L55. T. L. Chenevert, J. G. Pipe, D. M. Williams, and J. A. Brunberg, Magn. Reson. Med., 1990, 17, 197. E. O. Stejskal, J. Chem. Phys., 1965, 43, 3597. R. L. Cooper, D. B. Chang, A. C. Young, C. J. Martin, and D. Ancker-Johnson, Biophys. J., 1974, 14, 161. J. E. Tanner, J. Chem. Phys., 1978, 69, 1748. G. G. Cleveland, D. C. Chang, C. F. Hazlewood, and H. E. Rovschach, Biophys. J., 1976, 16, 1043.

11

29. M. E. Moseley, Y. Cohen, J. Kucharczyk, J. Mintorovitch, H. S. Asgari, M. F. Wendland, J. Tsuruda, D. Norman, and P. Weinstein, Radiology, 1990, 176, 439. 30. D. Le Bihan, R. Turner, and P. Douek, NeuroReport, 1993, 4, 887. 31. M. E. Moseley, J. Kucharczyk, J. Mintorovitch, Y. Cohen, J. Kurhanewicz, N. Derucin, H. Asgari, and D. Normou, Am. J. Neuroradiol., 1990, 11, 423. 32. S. Warach, D. Chien, W. Li, M. Ronthal, and R. R. Edelman, Neurology, 1992, 42, 1717. 33. D. Le Bihan, R. Turner, P. Douek, and N. Patronas, Am. J. Radiol., 1992, 159, 591. 34. D. Le Bihan, J. Delannoy, and R. Levin, Radiology, 1989, 171, 853. 35. S. S. Kety and C. F. Schmidt, Am. J. Physiol., 1945, 143, 53. 36. N. M. Bolas, A. J. Petros, D. Bergel, and G. K. Radda, Proc. Soc. Magn. Reson. Med. 1985 Meet., p. 315. 37. S. Eleff, M. Schnall, L. Ligetti, M. Osbakken, H. Subramanian, B. Chance, and J. S. Leigh, Magn. Reson. Med., 1988, 7, 412. 38. D. Barranco, L. N. Sutton, S. Florin, J. Greenberg, T. Sinnwell, L. Ligetti, and A. C. McLaughlin, J. Cereb. Blood Flow Metab., 1989, 9, 886. 39. J. R. Ewing, C. A. Branch, S. C. Fagan, J. A. Helpern, R. T. Simkins, S. M. Butt, and K. M. A. Welch, Stroke, 1990, 21, 100. 40. J. J. H. Ackerman, C. S. Ewy, N. N. Becker, and R. A. Shalwitz, Proc. Natl. Acad. Sci. USA, 1987, 84, 4099. 41. J. Pekar, L. Ligeti, Z. Ruttner, R. C. Lyon, T. M. Sinnwell, P. van Gelderen, D. Fiat, C. T. W. Moonen, and A. C. McLaughlin, Magn. Reson. Med., 1991, 21, 313. 42. A. L. Hopkins, E. M. Haacke, J. Tkach, R. G. Barr, and C. B. Bratton, Magn. Reson. Med., 1988, 7, 222. 43. K. K. Kwong, A. L. Hopkins, J. W. Belliveau, D. A. Chesler, L. M. Porkka, R. C. McKinstry, D. A. Finelli, G. J. Hunter, J. B. Moore, R. G. Barr, and B. R. Rosen, Magn. Reson. Med., 1991, 22, 154. 44. C. R. Fisel, J. L. Ackerman, R. B. Buxton, L. Garrido, J. W. Belliveau, B. R. Rosen, and T. J. Brady, Magn. Reson. Med., 1991, 17, 336. 45. L. Axel, Radiology, 1980, 137, 679. 46. J. W. Belliveau, B. R. Rosen, H. L. Kantor, R. R. Rzedzian, D. N. Kennedy, R. C. McKinstry, J. M. Vevea, M. S. Cohen, I. L. Pykett, and T. J. Brady, Magn. Reson. Med., 1990, 14, 538. 47. K. L. Zierler, Circ. Res., 1962, 10, 393. 48. N. A. Lassen, J. Cereb. Blood Flow Metab., 1984, 4, 633. 49. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, R. M. Buchbinder, R. M. Weisskoff, S. M. Cohen, J. M. Vevea, T. J. Brady, and B. R. Bosen, Science, 1991, 254, 716. 50. D. Le Bihan, E. Breton, D. Lallemand, M. L. Aubin, J. Vignaud, and M. Laval-Jeantet, Radiology, 1988, 168, 497. 51. I. R. Young, A. S. Hall, D. J. Bryant, D. G. T. Thomas, S. S. Gill, M. S. Dubowitz, F. Cowan, and J. M. Pennock, J. Comput. Assist. Tomogr., 1988, 12, 727. 52. J. Pekar, P. C. M. van Zijl, and C. T. W. Moonen, Magn. Reson. Med., 1992, 23, 122. 53. D. Le Bihan and R. Turner, Magn. Reson. Med., 1992, 27, 171. 54. J. J. Neil and J. J. H. Ackerman, J. Magn. Reson., 1992, 97, 194. 55. J. J. Neil, L. A. Scherrer, and J. J. H. Ackerman, J. Magn. Reson., 1991, 95, 607. 56. C. B. Ahn, S. Y. Lee, O. Nalcioglu, and Z. H. Cho, Med. Phys., 1987, 14, 43. 57. J. H. Maki, J. R. MacFall, and G. A. Johnson, Magn. Reson. Med., 1991, 17, 95.

12 DIFFUSION AND PERFUSION IN MRI 58. D. S. Williams, J. A. Detre, J. S. Leigh, and A. P. Koretsky, Proc. Natl. Acad. Sci. USA, 1992, 89, 212. 59. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H. M. Cheng, T. J. Brady, and B. R. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 60. L. Pauling and C. D. Coryell, Proc. Natl. Acad. Sci. USA, 1936, 22, 210. 61. K. R. Thulborn, J. C. Waterton, P. M. Matthews, and G. K. Radda, Biochim. Biophys. Acta, 1982, 714, 265. 62. S. Ogawa, T. M. Lee, A. S. Nayak, and P. Glynn, Magn. Reson. Med., 1990, 14, 68. 63. R. Turner, D. Le Bihan, C. T. W. Moonen, D. Despres, and J. Frank, Magn. Reson. Med., 1991, 22, 159. 64. P. A. Bandettini, E. C. Wong, R. S. Hinks, R. S. Tikofsky, and J. S. Hyde, Magn. Reson. Med., 1992, 25, 390. 65. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellerman, S. G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1992, 89, 5951. 66. J. Frahm, H. Bruhn, K. D. Merboldt, and W. Hanicke, J. Magn. Reson. Imag., 1992, 5, 501. 67. R. Turner, P. Jezzard, H. Wen, K. K. Kwong, D. Le Bihan, T. Zeffiro, and R. S. Balaban, Magn. Reson. Med., 1993, 29, 277. 68. G. McCarthy, A. M. Blamire, D. L. Rothman, R. Gruetter, and R. G. Schulman, Proc. Natl. Acad. Sci. USA, 1993, 90, 4952.

69. D. Le Bihan, R. Turner, T. A. Zeffiro, C. A. Cuenod, P. Jezzard, and V. Bonnerot, Proc. Natl. Acad. Sci. USA, 1993, 90, 1 1802. 70. S. Kim, J. Ashe, K. Hendrich, J. M. Ellermann, H. Merkle, K. Urgubil, and A. P. Georgopoulos, Science, 1993, 261, 615. 71. W. Schneider, D. C. Noll, and J. D. Cohen, Nature, 1993, 365, 150. 72. C. R. Jack, R. M. Thompson, R. K. Butts, F. W. Sharbrough, P. J. Kelly, D. P. Hanson, S. J. Rieckrer, R. L. Ehmon, N. J. Hangianoreou, and G. D. Cascino, Radiology, 1994, 190, 85. 73. C. A. Cuenod, M. C. J. Chang, T. Arai, L. Pannier, S. Posse, D. DesPres, J. A. Frank, S. Rapoport, and D. Le Bihan, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 1387. 74. L. Rueckert, I. Appollonio, J. Grafman, P. Jezzard, R. Johnson, and D. Le Bihan D, J. Neuroimag, 1994, 4, 67.

Biographical Sketch Denis Le Bihan. b 1957. Ph.D., 1987, Physics, M.D., 1984, Radiology, University of Paris, France. Chief, Diagnostic Radiology Research Section, National Institutes of Health, Clinical Associate Professor of Radiology, Georgetown University, USA. Approx. 250 articles, book chapters and abstracts. Research specialties: diffusion, temperature, interventional, perfusion and functional MRI.

DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

Diffusion: Clinical Utility of MRI Studies Ole Henriksen Danish Research Center of Magnetic Resonance, Hvidovre University Hospital, Denmark

1 INTRODUCTION Tissue characterization by MRI has been a subject of great interest because the multiparameter dependence of the MRI signal makes it possible to improve the differentiation between various pathologies in vivo, which may lead to an improved speci®city of clinical MRI investigations. So far, these approaches have mostly been based on signal contrast behavior due to differences in proton densities, and T1 and T2 relaxation behavior between different tissue types. In recent years, methods for studying diffusion processes in vivo by means of MRI have been developed, and possible clinical applications are currently under investigation. The studies published so far have mostly been directed to imaging of diffusion processes of water in the human brain. The results are very promising with respect to future clinical utility. These studies will be described in this article, outlining possible major areas of application for diffusion studies in clinical practice.

2 MRI MEASUREMENTS OF WATER DIFFUSION IN BIOLOGICAL TISSUES IN VIVO A detailed description of the factors which in¯uence the diffusion of water molecules in biological tissues is given in other articles (see the list of related articles). Only those aspects relevant for the interpretation of clinical in vivo diffusion studies will therefore be considered here.

The water diffusion in the cytosol may be hindered by the presence of barriers such as cell membranes, intracellular organelles, the presence of macromolecules, electrical charges, etc. The consequence is that the diffusion path of the water molecule becomes geometry-dependent, and the concept of a diffusion coef®cient depending only on the properties of the diffusion medium breaks down. The practical consequence is that the diffusion coef®cients obtained in MRI studies also include prolonged diffusion paths encountering such barriers (tortuosity), and they should thus be regarded as `apparent diffusion coef®cients', and are denoted D* in the following discussion. The signal contrast in diffusion-weighted imaging depends on the design of the pulse sequence because D* depends on the time (diffusion time) in which the diffusion processes in¯uence the signal obtained by diffusion-sensitized pulse sequences. If the diffusion time is very short, the probability that a water molecule meets a barrier is low compared with that of diffusion over a longer period of time. Thus, D* decreases monotonically from values close to D for short diffusion times to a constant smaller value as the diffusion time is extended to very long periods. The fractional signal attenuation, A, due to diffusion in a given time,
, in pulsed gradient experiments, neglecting the small effects of the imaging gradients, can be written as A ˆ exp‰ÿbD …
†Š

Translational molecular diffusion processes are a phenomenon in which molecules in a ¯uid of in®nite extent move along random paths, colliding with and moving past each other. Diffusion can be described in statistical terms. The diffusion coef®cient, D, describes the mobility at the molecular level, and has the unit of length squared per unit of time. D is speci®c for a given molecule, and varies with viscosity and temperature. For the free diffusion of water molecules in water the diffusion coef®cient D is approximately 3  10ÿ9 m2 sÿ2 at 37  C. If the diffusion coef®cient of water in a given tissue is less than this value at 37  C, and it varies with the diffusion time, the diffusion is said to be restricted, and depends on a number of factors related to the microstructure of the tissue in question, which will be described brie¯y below.

…1†

where b is the diffusion attenuation factor in units of time per length squared [equal to 2G22(
ÿ /3)] and D*(
) is the apparent diffusion coef®cient at the diffusion time,
( is the effective gradient duration and
is the time between the leading edges of the pulses). Thus, in images displaying D*, tissues with high D* will appear with a high signal intensity. In diffusion-weighted images the situation is the opposite, as tissues with high D* values tend to appear with low signal intensity, but the interpretation is more dif®cult because the signal will also be in¯uenced by T1- and T2-weighting factors. Even though the D* in cerebrospinal ¯uid (CSF) is very high and equals D, the signal appears bright in diffusionweighted images using a diffusion-sensitized spin echo sequence because the T2 of CSF is very long compared with that of the brain tissue.

2.2 2.1 The Concept of Restricted Diffusion

1

Sources of Errors

Apart from imperfections in the MRI equipment, macroscopic movements are a main source of error in diffusion studies in vivo. Patient movements may be minimized by careful ®xation, but one still has to take physiological movements induced by arterial pulsation, respiration, etc., into account. Studies of brain motion using the phase contrast method1 have indicated an average velocity of about 0.5 mm sÿ1 during the cardiac cycle, corresponding to distances of at least 10 m in 20 ms, which are very similar to the expected values for the apparent average diffusion distance with
of 20 ms. Cardiac and respiratory synchronization and compensation for ®rst order bulk movements by gradient moment nulling techniques reduce the movement problems considerably when spin echo sequences are used.2,3 Another way is to use singleshot sequences with very short imaging times.3

2 DIFFUSION: CLINICAL UTILITY OF MRI STUDIES Table 1 Clinical MRI studies of diffusion in human brain tissues Tissue CSF White matter Perpendicular Parallel Gray matter

Apparent diffusion coef®cient D*/(10ÿ9 m2 sÿ1) 2.9±3.5 0.2±1.7 0.2±0.9 0.8±1.5 0.7±2.3

References 3,4,5,6 4,7,10,11 2,7 2,7 3,6,8,9,10

3 CLINICAL MRI STUDIES OF DIFFUSION IN BRAIN 3.1 Normal Subjects So far the vast majority of in vivo diffusion studies applied to human beings have been carried out on the brain. The reasons are that the brain movements are small, and the relatively long T2 values of brain tissue make it possible to achieve a reasonable signal-to-noise ratio in high-resolution images (voxel size 8 mm3). Relatively few studies where quantitative measurements of the apparent diffusion coef®cient have been performed in healthy humans have been published so far. The results are summarized in Table 1. The main ®ndings are that the apparent diffusion coef®cient is reduced in gray and white matter compared with the diffusion coef®cient of water molecules in water at 37  C (Figure 1). Studies of white matter

indicate that the diffusion is anisotropic, as the apparent diffusion coef®cient is smaller for diffusion perpendicular to than parallel to the direction of the axons.11 Another important ®nding is that the diffusion appears to be restricted in white matter perpendicular to the direction of the nerve ®bers as D* decreases for diffusion times above 26 ms, which corresponds to an average diffusion distance of about 4±5 m.5 No sign of restricted diffusion is seen in the direction of the ®bers up to distances of 9 m.5 The same seems to be true for gray matter.5 The apparent diffusion coef®cients obtained in CSF may be regarded as a kind of internal control of the validity of the quantitative measurements, since the expected values should be equal to the diffusion coef®cient of water molecules in water at 37  C (D = 3  10ÿ9 m2 sÿ1). In the early study of Thomsen et al.,8 D* in CSF is higher than the expected value of free diffusion in water, indicating that bulk movements contributed to the measurement. In a later study where compensation for movements of constant velocity is incorporated, D* in CSF corresponds to the expected value.2 3.1.1

Age Dependency

Gideon et al.6 have observed that the apparent diffusion coef®cient for water in deep white matter increases with age from 0.7  10ÿ9 m2 sÿ1 (age ~20 years) to approximately 1.0  10ÿ9 m2 sÿ1 (age 60 ±75 years), using a ¯ow-compensated, diffusion-sensitized spin echo sequence with a diffusion time of 19 ms (Figure 2). This relationship between water diffusion and age has not been seen in gray matter.6 Measurements on

Figure 1 Images of the calculated apparent diffusion coef®cient of water in the brain of (a) an adult healthy volunteer and of (b) a premature neonate (gestational age 33 weeks). The calculated images are based on diffusion-weighted spin echo sequences with compensation for ®rst-order bulk movements. ECG triggering was performed as well. The color scale shown to the right of the images is linear, corresponding to D* = 0 (magenta) to D* = 3.5  10ÿ9 m2 sÿ1 (red) which corresponds to the free diffusion coef®cient of water in water at 37  C. The voxel size is 1  1  8 mm3. Note the relatively low D* in white matter perpendicular to the ®ber direction in the adult brain and that D* is considerably less than in the CSF, which corresponds to the free diffusion of water molecules in water. In contrast to the adult brain, the D* in white matter is higher than in gray matter in the neonatal premature brain

DIFFUSION: CLINICAL UTILITY OF MRI STUDIES 1.4

ADC (10–9 m2 s–1)

1.2 1 0.8 0.6 0.4 0.2 0 0

10

20

30 40 50 Age (years)

60

70

80

Figure 2 D* of water in periventricular white matter of healthy subjects plotted against age. Regression analysis shows a signi®cant correlation between D* and age. (Adapted from Gideon et al.6)

neonates and small infants suggest that the degree of water diffusion anisotropy in frontal white matter and optic radiation is less in neonates than in young adults.12 No clear difference is seen between the neonates and adults when D* is measured parallel to the direction of the ®bers.12 3.1.2

Diffusion-weighted Imaging

The studies mentioned so far in which the apparent water diffusion in human brain has been estimated in quantitative terms indicate that the water diffusion in white matter is restricted and there is an appreciable degree of anisotropy. The possible mechanisms underlying the diffusion anisotropy in brain white matter is discussed in Anisotropically Restricted Diffusion in MRI, but the observations in neonates12 suggest that the myelin sheaths may at least be partly responsible. The practical consequence is that anisotropically restricted diffusion creates a signal contrast between nerve ®ber bundles running in different directions in relation to those of the diffusion-sensitizing gradients. This has been shown in a number of studies where major ascending and descending projection tracts are visualized.5,13 It should be recalled that nerve ®bers running perpendicular appear brighter than those running in parallel with the direction of the diffusion-sensitizing gradients. 3.2 Brain Pathologies 3.2.1

Brain Ischemia

A number of animal studies show that the apparent diffusion of water is reduced in experimentally induced ischemic areas of the brain.14±16 These alterations seem to be detectable within the ®rst half hour after the onset of brain ischemia, and occur a long time before changes are detectable in T2-weighted images.15,17 In diffusion-weighted images, focal ischemic areas appear brighter than normal brain tissue in a gray±white scale, because the smaller apparent diffusion coef®cient of water causes less reduction of the signal due to diffusion processes. The animal studies also indicate good agreement between the anatomical spreading of focal brain ischemia as visualized by diffusion-weighted imaging, perfusion-related imaging of the passage of a gadolinium compound (input function is not taken into account), and histological evaluation based on staining

3

with Evans blue.14,17 Diffusion-weighted imaging of photochemically induced ischemia in rat brain shows hyperintense signals corresponding to the ischemic area on the ®rst day. During the following days the signal intensity fades, indicating that the diffusion of water becomes less impeded in the subacute stage.16 Taken together, the animal studies mentioned above clearly demonstrate that the apparent diffusion of water is reduced in ischemic brain tissue and that these alterations occur in the very early phase. The reason for the impeded diffusion of water is unclear at present. It cannot be explained by a decrease in tissue temperature because this has to drop about 10  C in order to explain the observed decrease in the apparent diffusion coef®cient of water of about 30%. It seems to be associated with energy failure secondary to severe ischemia.18 The impediment of diffusion probably re¯ects alterations of the water balance between tissue compartments, swelling of membranes, and accumulation of macromolecules. Only a few studies of water diffusion in the brain of stroke patients have been published. As early as 1987 Thomsen et al.8 had presented evidence suggesting that the apparent diffusion coef®cient of water is reduced in subacute brain infarcts. Warach et al.19 have studied 32 patients with stroke using a highspeed diffusion-weighted stimulated echo sequence (turboSTEAM). The time after the onset of symptoms ranged between 1 h and about 4 months. Diffusion-weighted images con®rm the results of the animal studies described above, as a hyperintense signal is seen in the ischemic area, and these changes can be seen in the very acute stage where no changes are seen in T2-weighted images (Figure 3). In accordance with this, parametric images of the calculated apparent diffusion coef®cient of water show that this is reduced compared with the contralateral side in the acute and subacute stages, whereas the apparent diffusion coef®cient in infarcted tissue is increased in the chronic stage (Figure 4). The absolute values of D* may have been in¯uenced as bulk movements due to gross head movements, brain movements caused by arterial pulsation,1,2,20 and respiration may not have been suf®ciently controlled. However, another study using a diffusion-sensitized spin echo pulse sequence and cardiac triggering also suggests a reduced apparent diffusion of water in ischemic areas following acute stroke.20 These results show a considerable interindividual variation, and they do not indicate a signi®cant correlation between D* and the duration of ischemia. Taken together, the experimental animal and human studies have established that the apparent diffusion coef®cient of water is reduced in brain tissue subjected to severe acute ischemia, and that these alterations can be visualized at a very early stage. The evidence obtained suggests that the apparent water diffusion coef®cient increases within the lesion to supranormal values in the chronic stage of stroke. As mentioned before, the exact pathophysiological events underlying these observations are still poorly understood. When the diffusion values obtained in these often elderly patients are compared with those obtained in healthy volunteers, the latter should be carefully age matched because the apparent diffusion coef®cients in white matter have been shown to increase with age6 (cf Figure 2). Further studies on anisotropy as well as on the relationship between water diffusion characteristics and regional cerebral perfusion are warranted in order to obtain a better understanding of the underlying pathophysiological mechanisms in acute

4 DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

Figure 3 T2-weighted (left) and corresponding diffusion-weighted (right) images obtained (a, b) 105 min, (c, d) 13 h, (e, f ) 9 days after an acute stroke incident. Note that the ischemic tissue located in the left hemisphere can already be clearly visualized approximately 1.5 h after the clinical attack, where no changes are seen in the T2-weighted image. (Reproduced by permission from Warach et al.19)

DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

by using a ¯ow refocusing sequence and cardiac triggering. It is of interest that the apparent diffusion coef®cient is higher in acute lesions than in chronic but stable lesions, and that there is a signi®cant increase in D* in apparently normal white matter in the conventional spin echo images21 (Figure 6). Finally, no relationship between the apparent diffusion coef®cient and the T2 relaxation rate measured by means of a multi-spin echo technique is seen, stressing that these phenomena depend on different mechanisms.21 Diffusion-weighted imaging of a patient with chronic multiple sclerosis shows a high signal in the lesion compared with apparently normal white matter.21 Since D* of water is increased, as shown by Christiansen et al.,21 a low signal should be expected in the diffusion-weighted images. The most likely reason for this is that the T2 weighting dominates compared with the effect of increased diffusion in lesions with high T2 values when pulsed gradient, diffusionsensitized spin echo pulse sequences are used. Even though our experience is relatively limited, the results obtained so far are promising with respect to differentiation between acute and chronic multiple sclerosis lesions.21 The higher D* value in acute than in chronic lesions may re¯ect an increase in water content due to edema and demyelinization in acute lesions, whereas gliosis dominates in the chronic stage. To what extent diffusion studies may contribute to the differential diagnosis between lesions due to multiple sclerosis and other diseases including, for example, subcortical infarcts and viral infections, is still too early to say, as systematic comparative diffusion studies are still lacking. A single case study suggests that diffusion-weighted imaging may give further information in patients with progressive multifocal leukoencephalopathy.13

ADC ratio, lesion/control

 ,
,
, ,
,, 

2

5

1

0

0 –12 h

12–24 h

24–48 h

5–12 d

24–60 d >4 months

Time

Figure 4 Calculated apparent diffusion coef®cients (ADC) ratios of water in ischemic and contralateral nonischemic brain tissue plotted against time after a stroke incident. Note that the ratios seem to increase from values <1 in the acute and subacute stages to values >1 in the chronic stage. (Adapted from Warach et al.19)

cerebral ischemia, and to be able to evaluate the prognostic information in diffusion-weighted images of brain ischemia with respect to tissue survival. 3.2.2

White Matter Disorders

Quantitative water diffusion studies indicate that the apparent diffusion coef®cient is increased in multiple sclerosis lesions as compared with healthy age-matched volunteers,21 in agreement with preliminary results published by Larsson et al.,22 (Figure 5). Possible in¯uences by movement due to brain pulsation and ¯ow phenomena have been taken into account

Figure 5 (a) T2-weighted and (b) calculated apparent water diffusion coef®cient images obtained in a patient with an acute attack of multiple sclerosis. The color scale shown to the right is similar to that described for Figure 1. Note the increase in diffusion corresponding to the acute lesion seen at the left posterior horn of the lateral ventricles. There also seems to be an increased diffusion in the apparently normal periventricular white matter. (Reproduced by permission from Christiansen et al.21)

6 DIFFUSION: CLINICAL UTILITY OF MRI STUDIES Controls

Multiple sclerosis patients

2.00 1.90 1.80 1.70 1.60 1.50 1.40 1.30 1.20 1.10 1.00 0.90 0.80 0.70 0.60 0.50

Normal white matter

White matter without plaques

Chronic multiple sclerosis plaques

Acute multiple sclerosis plaques

Figure 6 Apparent diffusion coef®cient of water in white matter obtained in age-matched controls (left) and patients with chronic and acute multiple sclerosis lesions. Note that the average diffusion coef®cient obtained in apparently normal white matter of multiple sclerosis patients is signi®cantly increased compared with the control group. (Adapted from Christiansen et al.21)

3.2.3

Intracranial Tumors

Only a few studies on a limited number of patients with intracranial tumors have been reported so far.5,23,24 The ®ndings suggest both an increase and decrease in D* within, as well as between, tumors compared with normal brain tissue

Figure 7 (a) T2-weighted and (b) calculated apparent water diffusion coef®cient images, obtained from a patient with a cystic astrocytoma. Note that D* in the central part of the tumor is close to the free diffusion of water in water, indicating a cystic component of the lesion containing free ¯uid

(Figure 7). The increase in D* is most prominent in cystic parts of the tumor approaching that of CSF and water. Diffusion measurements seem to make it possible to differentiate between solid tumor tissue and cystic lesions having the same appearance on conventional T1- and T2-weighted images due to a high content of paramagnetic protein. In the

DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

latter case, diffusion studies will reveal D* values corresponding to cyst ¯uid.25 Diffusion studies may also improve the assessment of the delineation between epidermoid tumor and adjacent CSF.23 The extent to which diffusion studies may improve the tissue characterization of solid brain tumors is as yet unclear. Another important clinical aspect is whether diffusion studies allow differentiation between solid tumor and surrounding edema. Anecdotal evidence suggests that the signals from presumed edema in diffusion-weighted images are isointense with the solid tumor components but the edema may be associated with a higher degree of diffusion anisotropy.13 It is, however, much too early to draw any de®nite conclusions. 3.2.4

Disturbances in CSF Hydrodynamics

Preliminary studies indicate that the apparent diffusion coef®cient of water is signi®cantly higher in white as well as gray matter of patients with normal-pressure hydrocephalus than in a group of healthy age-matched controls26 (Figure 8). Surprisingly, the diffusion coef®cients in the CSF of the patients seem also to be higher than in the controls, even though the diffusion-sensitive pulse sequences were compensated for ®rst-order motion, and ECG triggering was performed. Since pulsatile motion of CSF is increased in these patients,27 a higher degree of complex ¯ow patterns may have contributed to the signal loss. This may explain the apparent elevation of the diffusion coef®cients in CSF of patients with normal-pressure hydrocephalus. Examination of two patients with high-pressure hydrocephalus also suggests a higher apparent diffusion coef®cients in both white and gray matter compared with normal subjects.26 The increase in the apparent diffusion of water in brain white matter of patients with normal-pressure hydro-

7

cephalus may re¯ect the loss of myelin sheaths, especially in the periventricular regions. Soelberg Sùrensen et al.28 have observed a pronounced increase in the apparent diffusion coef®cient of water in patients with benign intracranial hypertension or pseudotumor cerebri. The alterations in water diffusion are most pronounced in the periventricular white matter. In some cases the calculated apparent diffusion coef®cient exceeds that of free diffusion of water in water, indicating that the obtained diffusion values may to some extent have been in¯uenced by pulsatile movements of the brain as well as by convective water movements in the brain tissue. In a later study where ®rst order movement was corrected for in the pulse sequence, Gideon et al.26 con®rm that the apparent diffusion is increased in periventricular white matter of patients with benign intracranial hypertension, whereas no signi®cant changes are seen in gray matter (Figure 9). Thus, it seems that the increased apparent diffusion coef®cient of water in gray matter observed by Soelberg Sùrensen et al.28 can most likely be ascribed to pulsatile brain movements, which may be more pronounced in these patients than in healthy volunteers. Quantitative measurements of T1 and T2 of gray and white matter, respectively, show no differences between patients with benign intracranial hypertension and healthy volunteers.28 In accordance with this, T1- and T2weighted images do not show any abnormalities, indicating that quantitative diffusion imaging is much more sensitive to alterations in intra- and extracellular water balance and dynamics. 3.2.5

Brain Edema

Sevick et al.29 have studied the in¯uence of cytotoxic edema secondary to hyponatremia on the apparent diffusion

Figure 8 (a) T2-weighted and (b) calculated apparent water diffusion coef®cient images obtained from a patient with normal-pressure hydrocephalus. There is an increased diffusion in the periventricular white matter as compared with normal subjects. (Reproduced by permission from Gideon et al.6)

8 DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

Figure 9 (a) T2-weighted and (b) calculated apparent water diffusion coef®cient images obtained in a patient with benign intracranial hypertension. Note the marked increase in D* in the periventricular white matter in the diffusion images, whereas the T2-weighted image appears normal. (Reproduced by permission from Gideon et al.6)

coef®cient of water in rat brain. The results suggest that cytotoxic edema reduces the apparent diffusion coef®cient of water in the brain by about 10%, and there seems to be an inverse relationship between the calculated D* value and the total water content of the brain tissue. It cannot, however, be totally excluded that the results may at least partly have been in¯uenced by a concomitant reduction in pulsatile movements of the edematous brain, since no cardiac triggering or compensation for ®rst order movements was performed. Anyway, the relatively small decrease in the apparent diffusion coef®cient observed in cytotoxic brain edema does not explain the alterations seen in acute brain ischemia described in a previous section. A number of studies indicate that the apparent diffusion coef®cient of water is higher in tumor-induced vasogenic brain edema than in normal gray and white matter13 (cf Figure 6). Taken together, the preliminary evidence suggests that quantitative diffusion studies may make it possible to differentiate between cytotoxic (D* reduced) and vasogenic brain edema (D* increased). 3.2.6

Pediatric Disorders

The experience with diffusion studies of infants is very limited at present. Rutherford et al.30 have recently published a study in which they examined 31 neonates and infants with various disorders of the brain. They did not calculate the apparent diffusion coef®cient. Cysts and ¯uid collections are recognized by their hypointense signals, indicating an increase in the D* value of water. Low signal intensities are seen in chronic infarcts, consistent with a relatively free isotropic diffusion, whereas the signals are low in intracerebral

hematomas, suggesting restricted isotropic diffusion. Patients with leukodystrophy associated with congenital muscle dystrophy reveal an anisotropic diffusion pattern. In three cases, abnormalities can be seen in the diffusion-weighted images despite conventional imaging being normal. Finally, abnormalities in the corticospinal tract are frequently seen in children with a history of birth asphyxia. Even though the retrospective character of the study does not allow any de®nite conclusions to be drawn, the results are very promising, indicating that diffusion studies may contribute to the diagnosis of a number of pediatric brain disorders. Before reaching that point, however, a large number of systematic studies with quantitative measurements of the apparent diffusion coef®cient of water are warranted in order to interpret the signal changes which are seen in the diffusion-weighted images.

4

OTHER ORGANS

Only a very limited number of diffusion studies of organs other than the brain are available at present. The reason is that diffusion studies may be hampered by considerable gross movements of the organ, caused by respiration and arterial pulsation. Evidence of anisotropic water diffusion has been obtained in human skeletal muscle. The apparent diffusion coef®cients of water range from 1.1  10ÿ9 to 1.5  10ÿ9 m2 sÿ1 perpendicular to and are about 2  10ÿ9 m2 sÿ1 parallel to the muscle ®bers, giving an anisotropic ratio of about 0.7.31,32

DIFFUSION: CLINICAL UTILITY OF MRI STUDIES

The ability to map changes in tissue temperature noninvasively in vivo in been investigated in human skeletal muscle during hyperthermia.33±35 Noninvasive mapping of tissue temperatures in vivo may offer great clinical potential in the control of hyperthermia therapy. The question is whether diffusion measurements as described above are suf®ciently accurate and reproducible. A study by Morvan et al.35 suggests that it is possible to measure temperature differences of approximately 2  C. This approach should be compared with an alternative method where changes in resonance frequencies induced by alterations in tissue temperature are measured.36 The latter method may turn out to be more sensitive and reproducible. Preliminary studies indicate that it may be possible to obtain meaningful diffusion-weighted images of human kidneys in vivo by means of an ultrafast diffusion-sensitized echo planar imaging sequence.37 It is much too early to speculate on the clinical signi®cance, however, because water transport in the kidney is very complex. Furthermore, substantial sources of error due to motion have to be taken into account. Ultrafast techniques combined with breath holding may be a part of the solution. 5 DIFFUSION-WEIGHTED SPECTROSCOPY Recent developments in localized spectroscopic techniques have made it possible to take advantage of chemical shift information to extend diffusion measurements to include molecules other than water. By means of 31P spectroscopy it is possible to study the diffusion of various phosphorus compounds, including phosphocreatine, in skeletal muscle.38 Diffusion characteristics of N-acetylaspartate, total creatine, and cholinecontaining compounds have been measured in anesthetized rat brain,39 and recently also in human brain,40 using a diffusionsensitized localized proton spectroscopic sequence with water suppression. The calculated apparent diffusion coef®cients (mean ‹ 1 SD) of N-acetylaspartate, choline-containing compounds, and total creatine in normal human brain are (0.18 ‹ 0.02)  10ÿ9, (0.13 ‹ 0.03)  10ÿ9, and (0.15 ‹ 0.03)  10ÿ9 m2 sÿ1, respectively.40 Applications of diffusion-weighted spectroscopic techniques are very interesting because compounds which do not penetrate the cell membranes can be used to study the intracellular medium, including viscosity, cell dimensions, etc. At present, the drawback is relatively poor spatial resolution, which means that the calculated D* values represent a weighted average of diffusion coef®cients in a mixture of tissue types consisting of many different cell types. 6 CLINICAL UTILITY OF MRI DIFFUSION STUDIES: A GENERAL VIEW Although clinical experience with diffusion imaging is rather limited, it is possible to outline some general trends with respect to clinical utility. So far the method has almost exclusively been applied to the brain, because problems due to movement can be dealt with, and secondly, the relatively high T2 values make it possible to achieve a reasonable signal-tonoise ratio at the long echo times which are necessary in order to obtain suf®cient diffusion sensitivity.

9

One of the most striking areas for the use of diffusionweighted imaging is early visualization of brain ischemia. This will most likely have a great impact on treatment strategies, where success depends critically on a very early start to intervention. Further research is needed in order to evaluate the prognostic information of alterations in diffusion characteristics, with particular emphasis on questions related to the penumbra zones.41,42 Studies on anisotropy during the course of infarction may give new information about pathophysiological mechanisms operating during severe brain ischemia. Secondly, diffusion studies may contribute to tissue characterization in the brain and thus improve the speci®city of MRI investigations, taking advantage of the information related to anisotropy. Such studies may contribute to the separation of solid tumors and edema, identi®cation of ¯uid compartments, and maybe also to differentiation between cytotoxic and vasoactive edema. Quantitative diffusion imaging is very sensitive to subtle changes in water balance and viscosity, as observed in patients with benign intracranial hypertension, where imaging by all other imaging modalities, including radio isotope techniques, is normal. Thirdly, it is most likely that diffusion studies will contribute signi®cantly to the characterization of brain white matter disorders. Alterations in diffusion anisotropy may give information of lesions in speci®c nerve tracts. Quantitative measurements of the apparent diffusion coef®cients may contribute, for example, to the differentiation between acute and chronic multiple sclerosis lesions. Diffusion-weighted spectroscopy may offer unique information about speci®c molecules. The study of chemical compounds which do not penetrate cell membranes and other structures makes it possible to obtain information about cell dimensions, intracellular ¯uid viscosity, etc. It is important to avoid errors due to motion in diffusion studies. Cardiac gating and compensating for ®rst-order movements are mandatory unless single acquisition techniques such as echo planar imaging are used. Quantitative measurements of apparent diffusion coef®cients in pathologies are needed in order to obtain the knowledge necessary for interpretating the diffusion-weighted images where the signal intensity is also in¯uenced by proton density, and T1 and T2 relaxation. Thus, diffusion-weighted imaging is a very challenging ®eld which offers great potential for biomedical research and diagnosis. Systematic clinical studies are now required in order to de®ne the clinical indications for diffusion-weighted imaging by magnetic resonance.

7

RELATED ARTICLES

Anisotropically Restricted Diffusion in MRI; Brain MRS of Infants and Children; Susceptibility and Diffusion Effects in NMR Microscopy.

8

REFERENCES

1. D. Greitz, R. Wirestam, A. Franck. B Nordell, C. Thomsen, and F. StaÊhlberg. Neuroradiol., 1992, 34, 370.

10 DIFFUSION: CLINICAL UTILITY OF MRI STUDIES 2. H. B. W. Larsson, M. Stubgaard, C. Thomsen, and The Scandinavian Flow Group, Proc. 9th Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 389. 3. D. Le Bihan, R. Turner, C. T. W. Moonen, and J. Pekar, JMRI, 1991, 1, 7. 4. D. Chien, R. B. Buxton, K. K. Kwong, T. J. Brady, and B. R. Rosen, J. Comput. Assist. Tomogr., 1990, 14, 514. 5. D. Le Bihan, R. Turner, and P. Douek, and N. Patronas, Am. J. Roentgenol., 1992, 159, 591. 6. P. Gideon, C. Thomsen, and O. Henriksen, JMRI, 4, 185. 7. P. Douek, R. Turner, J. Pekar, N. Patronas, and D. Le Bihan, J. Comput. Assist. Tomogr., 1991, 15, 923. 8. C. Thomsen, O. Henriksen, and P. Ring, Acta Radiol., 1987, 28, 353. 9. D. Le Bihan, E. Breton, D. Lallemand, P. Grenier, E. Cabanis, and M. Laval-Jeantet, Radiology, 1986, 161, 375. 10. K. Harada, N. Funita, K. Sakurai, Y. Akai, S. Kim, N. Nakanishi, and T. Kozuka, Proc. 9th Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, 375. 11. T. L. Chenevert, J. A. Brunberg, and J. G. Pipe, Radiology, 1990, 177, 401. 12. H. Sakuma, Y, Nomura, K. Takeda, T. Tagami, T. Nakagawa, Y, Tamagawa, Y Ishii, and T. Tsukamoto, Radiology, 1991, 180, 229. 13. J. V. Hajnal, M. Doran, A. S. Hall, A. G. Collins, A. Oatridge, J. M. Pennock, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1991, 15, 1. 14. M. E. Moseley, J. Kucharczyk, J. Mintorovitch, Y. Cohen, J. Kurhanewicz, N. Derugin, H. Asgari, and D. Norman, Am. J. Neuroroentg., 1990, 11, 423. 15. M. E. Moseley, Y. Cohen, J. Mintorovitch, L Chileuitt, H. Shimizu, J. Kucharczyk, M. F. Wendland, and P. R. Weinstein, Magn. Reson. Med., 1990, 14, 330. 16. N. van Bruggen, B. M. Cullen, M. D. King, M. Doran, S. R. Williams, D. G. Gadian, and J. E. Cremer, Stroke, 1992, 23, 576. 17. M. Maeda, S. Itoh, H. Ide, T. Matsuda, H. Kobayashi, T. Kubota, and Y. Ishii, Radiology, 1993, 189, 227. 18. A. L. Busza, K. L. Allen, M. D. King, N. van Bruggen, S. R. Williams, and D. G. Gadian, Stroke, 1992, 23, 1602. 19. S. Warach, D. Chien, W. Li, M. Ronthal, and R. R. Edelman, Neurology, 1992, 42, 1717. 20. D. Chien, K. K. Kwong, D. R. Gress, F. S. Buoanno, R. B. Buxton, and B. R. Rosen, Am. J. Neuroroentg., 1992, 13, 1097. 21. P. Christiansen, P. Gideon, C. Thomsen, M. Stubgaard, O. Henriksen, and H. B. W. Larsson, Acta Neurol. Scand., 1993, 87, 195. 22. H. B. W. Larsson, C. Thomsen, J. Frederiksen, M. Stubgaard, and O. Henriksen, Magn. Reson. Imag., 1992, 10, 7. 23. J. S. Tsuruda, W. M. Chew, M. E. Moseley, and D. Norman, Am. J. Neuroroentg., 1990, 11, 925.

24. J. S. Tsuruda, W. M. Chiew, M. E. Moseley, and D. Norman, Magn. Reson. Med. 1991, 19, 316. 25. M. Maeda, Y. Kawamura, Y. Tamagawa, T. Matsuda, S. Itoh, H. Kimura, T. Iwasaki, N. Hayashi, K. Yamamoto, and Y. Ishii, J. Comput. Assist. Tomogr. 1992, 16, 514. 26. P. Gideon, C. Thomsen, F. StaÊhlberg, O. Henriksen, P. Soelberg Sùrensen, and F. Gjerris, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 1609. 27. P. Gideon, F. StaÊhlberg, C. Thomsen, F. Gjerris, P. Soelberg Sùrensen, and O. Henriksen, Neuroradiology, 1994, 36, 210. 28. P. Soelberg Sùrensen, C. Thomsen, F. Gjerris, J. Schmidt, L. Kjaer, and O. Henriksen, Neurol. Research, 1989, 11, 160. 29. R. J. Sevick, F. Kanda, J. Mintorovitch, A. I. Arieff, J. Kucharczyk, J. S. Tsuruda, D. Norman, and M. E. Moseley, Radiology, 1992, 185, 687. 30. M. A. Rutherford, F. M. Cowan, A. Y. Manzur, L. M. S. Dubowitz, J. M. Pennock, J. V. Hajnal, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr. 1991, 15, 108. 31. M. E. Moseley and M. F. Wendland, Proc. Xth Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, 108. 32. D. Morvan, A. Leroy-Willig, P. Jehenson, C. A. Cuenod, and A. Syrota, Radiology, 1992, 185, 871. 33. D. Le Bihan, J. Delannoy, and R. L. Levin, Radiology, 1989, 171, 853. 34. A. S. Hall, M. V. Prior, J. W. Hand, I. R. Young, and R. J. Dickinson, J. Comput. Assist. Tomogr., 1990, 14, 430. 35. D. Morvan, A. Leroy-Willig, A. Malgouyres, C. A. Cuenod, P. Jehenson, and A. Syrota, Magn. Reson. Med., 1993, 29, 371. 36. J. de Poorter, C. de Wagter, C. Thomsen, Y. de Deene, F. StoÊhlberg, and E. Acten, J. Magn. Reson., 1995, 33(1), 74. 37. R. Turner, D. Le Bihan, J. Maier, R. Vavrek, and J. Baich, Proc. 9th Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, 1136. 38. C. T. W. Moonen, P. C. M. van Zijl, D. Le Bihan, and D. Des Pres, Magn. Reson. Med., 1990, 13, 467. 39. K.-D. Merboldt, D. HoÈrstermann, W. HaÈnicke, H. Bruhn, and J. Frahm, Magn. Reson. Med., 1993, 29, 125. 40. S. Posse, C. A. Cuenod, and D. Le Bihan, Radiology, 1993, 188, 719. 41. M. J. Quast, N. C. Huang, G. R. Hillman, and T. A. Kent, Magn. Reson. Imag., 1993, 11, 465. 42. J. Seega and B. Elger, Magn. Reson. Imag., 1993, 11, 401.

Biographical Sketch Ole Henriksen. b 1944. M.D., 1973. Specialist, Clinical Physiology and Nuclear Medicine, 1982. Doctor, Medical Science, 1977. Professor, MRI, 1990. Director, The Danish Research Center of Magnetic Resonance, 1985±present. Approx. 100 publications on MRI since 1985. Research interests are mainly in vivo studies of physiological and pathophysiological mechanisms in humans using MRI techniques including diffusion, ¯ow perfusion, and spectroscopy measurements.

1

Methods and Applications of Diffusion MRI

will be given to the dif®culties encountered when implementing diffusion imaging and the potential of diffusion measurements for clinical applications.

Denis Le Bihan Service Hospitalier FreÂdeÂric Joliot, CEA, Orsay, France

1 INTRODUCTION Diffusion is the process by which matter is transported from one part of a system to another as a result of random molecular motion, also called Brownian motion. This motion, of thermal origin, results in the macroscopic ¯ux of different molecular species, which can be observed in nonuniform systems and is characterized by a diffusion coef®cient. Classically, diffusion coef®cients may be determined by measuring the concentration of molecular species at different times using either physical or chemical methods based on the classical ®rst Fick Law.1 Such approaches rely on the introduction of a tracer in the medium, as similar as possible to the studied molecular species, and then monitoring the concentration of the tracer in the medium by chemical or radiotracer techniques.2 Microscopic displacements on the scale of millimeters can be seen with such tracers over diffusion times of minutes. Such tracer methods have been successfully applied to biological systems, such as the brain3 but are, of course, extremely invasive. An alternative in studying diffusion is to monitor the random walks of molecules. On a statistical scale, the diffusion re¯ects the meansquare distance traveled by molecules in a given interval of time (m2 sÿ1). Diffusion NMR, which relies on these principles, is the only method today available to provide noninvasively such information on molecular displacements which occur over diffusion distances that extend largely beyond elementary molecular jumps, justifying the considerable success of diffusion NMR in physics and chemistry, and more recently in biology.4 Taking typical values for water diffusion ~110ÿ3 mm2 sÿ1) and diffusion time (~20 ms) achievable on conventional MRI equipment free water molecules diffuse over distances on the order of 6 m, which is about the size of many tissue structures. NMR has, in principle, a displacement sensitivity of around 100 nm.5 Because of its noninvasive nature, it is especially suited to probing the molecular dynamics and structural information of biological systems, as well as transport processes. The recent combination of such principles with MRI6±8 represents a spectacular and somewhat unpredicted development in the ®eld of medical sciences. Measuring molecular displacements in biological tissues in vivo may have enormous value, from the determination of the molecular organization in tissues to the emergency management of stroke patients or the monitoring of laser surgery. In this chapter, the basic principles of diffusion measurements with NMR will be presented, as well as the various methods that have been proposed to produce images of diffusion. Some technical considerations will be discussed, followed by a description of the effects on diffusion measurements of microdynamic and microstructure in biological tissues. Emphasis

2

DIFFUSION MRI

Although the effects of diffusion on the NMR signal have been described very early,9,10 most diffusion NMR studies started after the seminal work of Stejskal and Tanner, who introduced the bipolar pulse ®eld gradient method (Figure 1).11 This approach not only provides much better control over what is actually measured but also simpli®es the understanding of the encoding of the diffusion process in the NMR signal. The purpose of these gradient pulses, the duration and the separation of which are classically represented, respectively, as  and
, is to label spins carried by diffusing molecules magnetically. To some extent, these labeled spins are like the endogenous tracers that can be monitored with classical diffusion measurement tracer methods. However, the behavior of these labeled spins in an NMR experiment leads to more peculiar and speci®c results that are unique to diffusion NMR, and diffusion data obtained with NMR may not always be directly comparable with those obtained by other means. The ®rst gradient pulse induces a phase shift '1 of the spin transverse magnetization, which depends on the spin position z1: …1†

' ˆ Gz1

where is the gyromagnetic ratio and G is the gradient strength (here, along the z axis). After the 180 rf pulse, '1 is inverted to ÿ'1. The second pulse will produce a phase shift '2: '2 ˆ Gz2

…2†

where z2 is the spin position during the second pulse. The net transverse magnetization Mxy is: Mxy=M0 ˆ exp‰i…'2 ÿ '1 †Š ˆ exp‰i G…z2 ÿ z1 †Š

…3†

where all relaxation effects have been incorporated into M0.

90°

180°

Echo

G

G

d

d D

Figure 1 Stejskal and Tanner diffusion spin-echo sequence. Gradient pulses, G, are separated by a time interval
. Their duration is 

For References see p. 16

2 METHODS AND APPLICATIONS OF DIFFUSION MRI Obviously for `static' spins, z1=z2 and the bipolar gradient pair has no effect. For `moving' spins, however, there is a net dephasing that will depend on the spin history during the time interval
between the pulses. Indeed, we must now consider a population of spins, which may have different motion histories. This leads to an attenuation of the Mxy and the resulting signal. Assuming <<
(negligible displacement during ), these equations give a time-dependent part that represents the attenuation owing to diffusion, A: … … …4† A ˆ d z1 P…z1 † exp‰i G…z2 ÿ z1 †ŠP…z2 ; z1 ;
†d z2

the selected slice pro®le. In addition, the relaxation time T2 of the medium must not be too short to allow enough diffusion attenuation to occur during TE.

where P(z1) is the probability for a labeled spin to be at initial position z1 and P(z2, z1,
) is the diffusion propagator, i.e., P(z2, z1,
)dz2 is the conditional probability to ®nd a labeled spin initially at position z1 in position z2 after a time interval
. For a true, isotropic diffusion process:

A ˆ exp‰ÿ 2 G2 D2 …
ÿ =3†Š

@P…z2 ; z1 ; t†=@t ˆ Dr2 P…z2 ; z1 ; t†

…5†

and P is Gaussian, at t=
: …6†

where D is the diffusion coef®cient. Combining Equations (1) and (2) gives: A ˆ exp‰ÿ… G†2
DŠ  exp‰ÿ… G†2 hz2 i=2Š

…7†

where the root mean square diffusion displacement hz2i that occurred during the diffusion time
is easily interchangeable with D according to Einstein's equation, hz2i=2DTd. Diffusion coef®cients or root mean square diffusion distances are classically obtained by varying either G or  and by measuring the slope of the semilogarithmic plot of the signal intensity versus (G)2. However, it is the diffusion path that the NMR experiment is actually sensitive to, not the diffusion coef®cient. Therefore, in this simple bipolar gradient experiment, clear information on spin displacements can directly be obtained from the signal attenuation A, assuming P is Gaussian. This model is valid only for diffusion in an in®nite and homogeneous medium. If diffusion is restricted or anisotropic (see below), the signal attenuation must be calculated using a more general formalism based on diffusion tensors.12 2.1 Diffusion and the NMR Signal: Different Approaches Almost any NMR sequence can be designed to measure diffusion.

Bipolar Gradient Pulse Spin Echo Technique

The bipolar gradient pulse spin echo technique is, by far, the sequence most widely used for NMR diffusion measurements. In practice, as  is usually not negligible compared with
, movement during the application of the gradient pulses can no longer be neglected. Taking into account spin diffusion during , Equation (7) becomes:11 …9†

The diffusion time can be formally de®ned as (
ÿ/3), but its physical signi®cance is clear only when  is suf®ciently short compared with
. Tissues with short T2 times require short TE values, which may not allow suf®ciently long diffusion times to produce enough diffusion effects. 2.1.3

P…z2 ; z1 ;
† ˆ …4
D†ÿ1=2 exp‰ÿ…z2 ÿ z1 †2 =4
DŠ

2.1.1

2.1.2

Stimulated Echo Technique

A stimulated echo is generated from a sequence consisting of three rf pulses separated by time intervals  1 and  2. Gradient pulses must be inserted within the ®rst and the third periods of the stimulated echo sequence (Figure 2). The diffusion time now includes  2 and can be much longer than with the spin echo sequence without T2-related signal decay because only longitudinal relaxation occurs during this time interval. The Stejskal±Tanner relation [Equation (9)] still applies, provided that the period  2 is included in
.13 The longer diffusion time is useful for studying very slow diffusion rates or for compensating for the unavailability of large gradients.7 Unfortunately, the signal is reduced by 50% compared with the spin echo signal. 2.1.4

Gradient Echo Technique

The effect of diffusion on the amplitude of a gradient echo formed by a bipolar gradient pulse pair of reversed polarity does not differ from that of a spin echo sequence.14 However, if the gradient echo is part of a steady-state free precession (SSFP) sequence, where some degree of phase coherence is

90°

90°

Stimulated echo

90°

t1

t2

t1

G

G

d

d

Constant Field Gradient Spin Echo Method

In the presence of a simple constant linear gradient G0, the echo attenuation is:9,10 A…nTE† ˆ exp…ÿ

2

G20 DTE 3 =12n†

…8†

where TE is the echo time and n the echo number in a multiple echo experiment. This simple approach is dif®cult to combine with NMR imaging because the presence of a constant gradient during the rf pulses of the imaging sequence severely impairs For list of General Abbreviations see end-papers

D

Figure 2 Diffusion-stimulated echo sequence. The stimulated echo results from the spin excitation by three rf pulses separated by time intervals  1 and  2. During  2, relaxation is driven by T1 and not T2. Diffusion effects can be enhanced by placing gradient pulses within the  1 periods, where transverse magnetization is sensitive to ®eld inhomogeneities

METHODS AND APPLICATIONS OF DIFFUSION MRI

2.1.5

0 –0.1 –0.2 log (S/S0)

propagated throughout successive cycles, multiple echo paths with different diffusion times and different diffusion weighting must be considered.15 In MRI, the contrast-enhanced (CE-Fast) scheme has been proposed.16±18 Unfortunately, this sequence remains very sensitive to motion artifacts.19 Moreover, the effects of diffusion and relaxation are mingled so that diffusion measurements are always contaminated to some degree by relaxation effects.20

3

–0.3 –0.4 –0.5

Diffusion Measurements with B1 Field Gradients

Diffusion measurements can also be achieved by means of the rf (B1) ®eld produced by a NMR rf coil oriented perpendicularly to the main transmit/receive NMR coil.21±24 With rf gradients, extremely short switching times can be achieved since there are no eddy currents. Furthermore, substantial gradient strength may be produced from the rf transmitter, allowing measurements of very low diffusion coef®cients.

–0.6 –0.7 0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 b (¥103 s mm–2)

Figure 3 The relationship of the logarithm of the signal intensity in brain white matter with the gradient factor b is linear; the slope is equal to the diffusion coef®cient D=0.60 (‹0.01) 10ÿ3 mm2 sÿ1

2.2 Combining Diffusion NMR with MRI 2.2.1

Principles

Diffusion-weighted (DW) images are obtained by inserting gradient pulses into any NMR imaging sequence. For quanti®cation, it is necessary to determine the degree of diffusion weighting of the sequence. When inserting gradient pulses for diffusion, however, the combination of the imaging and the diffusion gradient pulses produces cross-terms. These cross-terms, which depend on the sequence design, may also lead to signi®cant diffusion-related attenuation effects,14,25 which must be taken into account; consequently the Stejskal±Tanner relation [Equation (9)] is incorrect in most cases. A more general formalism must then be developed to solve the Bloch±Torrey equation.26 Solutions may become analytically complex,27 and it has been suggested that all gradient effects should be combined in a term generally known as the `b factor':14,28 b ˆ 2

… TE … t 0

0

2 G…t0 †d t0 d t

…10†

The signal attenuation for, unrestricted, isotropic diffusion is then reduced to a simple, convenient expression: A ˆ exp…ÿbD†

…11†

This b factor characterizes the diffusion sensitivity of the sequence, as TE characterizes the degree of T2 weighting of a spin echo sequence. Without additional gradients, typical spin echo imaging sequences have intrinsically very low b values, typically less than 1 s mmÿ2; consequently diffusion effects are negligible (for pure water at room temperature the attenuation is less than 1%).14,29 As the raw signal depends not only on diffusion but also on other MR parameters, such as T1 or T2, a quantitative determination of the diffusion coef®cient requires at least two images acquired with different diffusion sensitivities (i.e., b factor values), but better accuracy is obtained when more than two images are used (see below). With the b factor values associated with each image, it is possible to compute diffusion images, i.e., images where the diffusion coef®cient is determined for each pixel. The computation of such diffusion

images is obtained by ®tting the signal intensity of each pixel obtained for different b factor values with Equation (11)4,6,7,14,28 using a regression analysis (Figure 3). On diffusion images, diffusion values are displayed using a gray scale where brightness corresponds to high, fast diffusion and darkness to low, slow diffusion (Figure 4). To save on acquisition and processing time, some investigators have proposed to limit the diffusion studies to the analysis of the raw images obtained with some degree of diffusion sensitivity or weighting. For this reason, these images are called DW images. Contrast in these images is opposite to that found in true diffusion images: regions with high diffusion have more pronounced MRI signal attenuation and appear dark, while regions with low diffusion appear bright. The DW image may be convenient to use, but its content is affected by many parameters other than diffusion, as it is also usually strongly T1 and T2 weighted. As these parameters may not have the same behavior as diffusion, variations in image intensity may be dif®cult to interpret. For instance, acute stroke lesions have decreased diffusion (see below) and appear bright on DW images, while subacute infarcted areas, which are associated with increased diffusion, may also appear bright on DW images because T2 has increased (T2 `shine-through' effect). DW images may be convenient to use in clinical practice, but, whenever possible, absolute diffusion images should be preferred. 2.2.2

Diffusion MRI Sequences

Using the spin echo sequence, it is possible to vary the strength or the duration of the diffusion-sensitizing gradients or their direction in order to enhance anisotropic diffusion effects14 (see below). The spin echo two-dimensional Fourier (2DFT) method is certainly the simplest to implement.6,8,28,29 Other sequences4 that have been investigated are stimulated echo sequences,7 line-integral projection reconstruction,30±32 variants of the SSFP technique; and `turbo' sequences such as HASTE,33 TurboFlash34 and fast spin echo.35 These schemes have been suggested to overcome, at least partially, some of the problems encountered with the spin echo 2DFT method. For References see p. 16

4 METHODS AND APPLICATIONS OF DIFFUSION MRI

Figure 4 Diffusion image (left) calculated from a set of 16 diffusion-weighted coronal echo-planar images of the brain of a normal volunteer. These are 64  64 pixel images with a ®eld of view of 16 cm and, therefore, an in-plane resolution of 2.5 mm  2.5 mm. The section thickness was 10 mm. The diffusion gradients varied from 0 to 38 mT mÿ1 in the z direction; the duration of each lobe was 20 ms and their separation was 6 ms (b=0 to 872 s mmÿ2). The acquisition time for each image was about 50 ms. Note that diffusion is the highest in ventricular cavities containing cerebrospinal ¯uid (free water). Because of diffusion anisotropy, diffusion along the x axis in corpus callosum is very low, while vertical frontal white matter tracts has the higher diffusion (see Table 1)

180°

,, ,,

rf

Slice select Readout

Phase encode

Figure 5 Diffusion echo planar imaging sequence. The spin echo is split into a series of gradient echoes by quickly switching the readout gradient. Sensitization to diffusion of an echo planar imaging sequence can be easily achieved by additional gradient pulses (dashed boxes)

For list of General Abbreviations see end-papers

With EPI, motion artifacts are virtually eliminated. The accuracy of assessments of diffusion achieved with EPI is generally extremely good, as many images differently sensitized to diffusion can be generated because of the very short acquisition time (typically less than 100 ms). The EPI technique is the method of choice for in vivo diffusion imaging, although it is very vulnerable to susceptibility artifacts, which are responsible for image distortion or signal dropout and to chemical shift artifacts, which require ef®cient fat suppression. Despite this, diffusion-EPI has become widely available on many clinical scanners and has been successfully used for measuring diffusion of water in the human brain in volunteers (Figure 4)37,41 and patients (see below). Recently, alternative fast acquisition schemes, such as BURST42,43 and spiral MRI,44 have been proposed to overcome some EPI artifacts, such as ghosting or susceptibility artifacts. 2.2.3

Echoes


, 
,,  ,

90°

,, ,,

However, echo planar imaging (EPI) remains generally the imaging method of choice.36±39 With EPI, the entire set of echoes needed to form an image is collected within a single acquisition period (single shot) of 25±100 ms. This is obtained by switching the echo signal formation in a train of gradient echoes by means of a large gradient in which polarity is very rapidly inverted as many times as is required to achieve the desired image resolution.40 EPI may easily be sensitized to diffusion (Figure 5).36±39 Sensitization consists of providing a pair of large compensated gradients for a period of time before rapid gradient switching and data acquisition. The refocusing may be achieved either by simply reversing the polarity of the gradient halfway through the period over which it is applied or by inserting a 180 rf refocusing pulse at the midpoint without reversing the gradient polarity.

Sequence Optimization

Although the b factor can be exactly determined for any pulse sequence,27,45 the diffusion time is impossible to de®ne accurately for a MR sequence where gradient pulses of ®nite duration are scattered all over the pulse sequence. A mixture of diffusion-related events with differently weighted contributions are obtained depending on their sequence in time with respect to the many pulse gradients. That is why it has been suggested that the parameter derived from the NMR signal attenuation using Equation (7) should be called an `apparent diffusion coef®cient' (ADC).28 MR sequences should be designed to minimize these problems, i.e., by placing gradient pulses to get as close as possible to the simple Stejskal±Tanner bipolar pulse sequence. However, the use of the ADC term remains justi®ed,

METHODS AND APPLICATIONS OF DIFFUSION MRI

as biological tissues usually differ signi®cantly from the ideal `free, unlimited, isotropic' medium. Another issue is to optimize the values and the number of b factors to be used. A simple calculation4 shows that the best accuracy for the diffusion coef®cient is obtained from a set of two acquisitions if the two b factors, b1 and b2 differ by about 1/D.46 In the brain, this translates to (b2ÿb1)~1000 to 1500 s mmÿ2. If more than two acquisitions are to be used, error propagation theory47 shows that it is better to accumulate n1 and n2 measurements at each of the small and large b factors, b1 and b2, than to use a range of b factors. The accuracy dD/D can be obtained from the raw image signal-to-noise ratio (SNR): dD=D ˆ ‰1=n1 ‡ exp‰2D…b1 ÿ b2 †Š=n2 Š1=2 =‰…SNR†D…b1 ÿ b2 †Š …12† Acquisition performed with linear sets of b value ranges remains, however, extremely useful to validate the sequence/ hardware quality and to assess the nature of the diffusion process or the presence of multiple diffusion compartments as it allows the exponential diffusion attenuation to be visualized according to Equation (11).

2.3 Experimental Considerations in Diffusion MRI 2.3.1

Gradient Hardware

Since the minimum length of molecular diffusion paths detectable with gradient-pulsed NMR is primarily determined by the intensity of the gradient pulses, hardware must be capable of providing stable gradients of the utmost intensity.48 This requirement may be extremely challenging when considering whole-body instruments designed for clinical studies. The lack of gradient power is usually compensated by using somewhat long gradient pulse widths; as a result, the classical condition <<
is not satis®ed. The case of such ®nite pulse widths is signi®cantly more dif®cult to treat and to interpret compared with the situation with -sharp pulses. If the duration of the pulses is ignored, signi®cant underestimation of the diffusion distances may result.49 For instance, to achieve at least a 20% signal attenuation and limit signal loss from transverse relaxation, the minimum diffusion coef®cient Dmin measurable is 3710ÿ3 mm2 sÿ1 using =
/10,
=T2=100 ms, and Gmax=10 mT mÿ1, which is more than 16 times the diffusion coef®cient of free water!14 Maximum sensitivity is achieved when =
(equivalent to a constant gradient) and
=T2/2. Then Dmin is 0.410ÿ3 mm2 sÿ1. There are partial solutions to this problem, such as using stimulated echoes to increase the effective diffusion time without penalizing the signal by T2 relaxation effects,13 or, better, by building high-performance gradient coils (e.g., up to 50, 100 or even 200 mT mÿ1). Safety should not be too much a concern, considering that slew rates will remain low, as there is no need to switch such gradient coils very rapidly (which also limits eddy currents). Hence, diffusion effects during the application of the gradient pulses can no longer be neglected, although their contribution to the overall signal attenuation does not scale the same way with time as

5

for diffusion occurring between pulses. The diffusion time then becomes more dif®cult to de®ne, although (
ÿ/3) is often taken instead of
as the ef®cient diffusion time.11 Another concern is that any mismatch between the diffusion-sensitizing gradient pulses may cause artifactual signal losses through an improper spin rephasing. Two major sources of problem are commonly seen. One is gradient instability, which may arise when gradient ampli®ers are driven hard for fast switching of large gradient intensities. Variations from shot to shot of the bipolar gradient balance result in widely distributed ghost artifacts. While such artifacts do not occur when a single-shot technique such as EPI is used, it is still essential to have high-quality gradient ampli®ers with some reserve capacity for accurate diffusion imaging. The second problem results from eddy currents generated mainly in the cryostat when switching large gradient pulses rapidly. Eddy currents may also be a major cause of image distortion and misregistration between images obtained with different b values, leading to ADC miscalculation. The best solution to this problem is undoubtedly to remove eddy currents at the source by using actively shielded gradient coils, which have no fringe ®elds and, therefore, do not generate eddy currents. The use of a gradient coil of small dimensions, which remains at a fair distance from the magnet core and with which it is easier to generate large gradient amplitudes, is certainly an attracting alternative.21 If residual eddy current effects persist, other approaches can be combined.50±52 In summary, there is clearly a need for very-high-performance gradient coils. With the hardware that is now becoming available,53,54 shorter diffusion times or larger b values can be reached while maintaining high SNR, giving access to slow diffusion components (e.g., intracellular water, metabolites).

2.3.2

Motion Artifacts

As the sequences used are deliberately sensitized to motion by the addition of large gradients, a major problem occurring with in vivo imaging of diffusion arises from motion of the object. Artifacts result from discontinuities that occur between successive cycles of an imaging sequence. Results of such temporal incoherence are commonly visible as `ghosts' along the phase-encoding direction. These ghosts are particularly intense in the presence of the diffusion gradients and render the diffusion measurements meaningless. Cardiac gating has been used to mitigate this problem, but even this motion is not strictly cyclic. It is dif®cult to compensate diffusion imaging sequences for motion, since the use of successive bipolar gradient pulses considerably reduces the value of the b factor.4,14 Chenevert et al. have suggested eliminating any phase encoding from the acquisition by sacri®cing one dimension of the image, which is then reduced to a single line.55,56 Ultimately, however, the best way to avoid motion artifacts is to use a single-shot technique, such as EPI, and to secure patients comfortably within the magnet using cushions or in¯ating devices. In spite of this, motion artifacts may persist, especially when multiple shots are used, as with diffusion spectroscopy (see below) or segmented EPI. Motion effects can be estimated by monitoring the phase of the signal and corrected.57,58 Navigator echoes have been used with some success by several investigators.59±62 For References see p. 16

6 METHODS AND APPLICATIONS OF DIFFUSION MRI 2.3.3

Background Inhomogeneities

In addition to the magnetic ®eld gradient pulses, residual ®eld inhomogeneities arising from the imperfect shimming of the magnet and from inhomogeneities in the sample must be considered. These inhomogeneities result from susceptibility variations, especially at high ®eld, and may not be negligible.63 By adding a constant, linear gradient G0, it is clear that two contributions of the residual gradients coexist.11 One is a unique contribution (term in G20 ); the other results from a cross-term between the residual and the applied gradients (GG0 term).14 While the ®rst contribution can easily be eliminated from a series study where only G is varied, the second term is more dif®cult to handle. There are however, diffusion pulse sequences that have been designed to reduce signi®cantly or to eliminate effects of background gradients.64±70 Most of them are based on multiple rf pulses. For instance, cross-terms can be completely eliminated by alternating the gradient pulses between the successive 180 rf pulses of a quadrupole spin echo sequence.71 If cross-terms with the background gradients are not eliminated, the logarithm of A is no longer linear with the b factor and the ADC overestimates the true diffusion coef®cient.25 The ADC will, furthermore, appear as a function of the diffusion time. The examination of the linearity of the relationship between log A and b is therefore, the ®rst step in interpreting any NMR diffusion study in biological tissues, although there are several possible causes that would explain a deviation from linearity. Notably, similar nonlinear behavior will be seen if the b factor is miscalculated, i.e., cross-terms coming from all applied (diffusion and imaging) gradient pulses are not appropriately considered in the calculation.27 The b factor must be carefully calculated before the presence of background gradients can be assessed. However, ®nding a linear plot for log A versus b does not exclude background gradients. First, the effect may be small. Second, the interpretation of the data may become very complicated when gradients are nonuniform or nonlinear and slowly varying in space, as in the case of capillaries ®lled with materials with a high susceptibility constant (e.g. contrast agent). Simulations and experiments have shown, in this case, that the ADC is decreased and that the parameter to consider is the variance of the internal gradients.72,73 The ultimate test to evaluate background gradients, however, is to vary TE while  and
are kept constant. The measured ADC should normally be independent of TE. Otherwise, the contribution of background gradients must be suspected. Furthermore, as self-induced gradients may depend on the orientation of the sample with respect to the direction of B0, they may cause isotropic diffusion to appear anisotropic.65 In this case also, the measured diffusion coef®cient depends on the diffusion time. Once it is established that background gradients are present, it may be desirable to estimate them from the attenuation behavior in order to obtain information on the structure of the medium. For instance, in the ideal case where the internal gradients result from inhomogeneities caused by the presence of closely packed particles, a rough estimation of the mean size R, of these particles can be obtained from:74 R  60 B0 =G0 For list of General Abbreviations see end-papers

…13†

where 0 is the magnetic susceptibility constant. Furthermore, these susceptibility effects may be of biological interest,75 as variations in the oxygenation level in brain capillaries and small vessels following brain activity have been shown to induce susceptibility effects measurable by T2 .76,77 A signi®cant drop of the measured ADC has also been observed during status epilepticus,78 cortical electroshocks79 and spreading depression.80±82 In all cases, large increases in oxygenated blood ¯ow could result in important reductions in local internal gradients and apparent variations to diffusion, although cytotoxic edema and cell swelling associated with energy failure has been mainly evoked as the reason for this diffusion drop. 2.4 2.4.1

Other Approaches to Diffusion Diffusion Tensor Imaging

Diffusion is a three-dimensional process. However, the molecular mobility may not be the same in all directions. This anisotropy may be a result of the physical arrangement of the medium (liquid crystal) or the presence of obstacles that limit diffusion in some directions. The result is that the ADC appears different when pulse gradients are imposed in different directions; this has been observed in muscle83 and, more recently, in brain white matter.37,56,84 The proper way to study anisotropic diffusion is to consider the diffusion tensor.1,11 Diffusion is no longer characterized by a single scalar coef®cient but by a tensor D, which fully describes molecular mobility along each axis and the correlation between these axes: Dxx D ˆ Dyx Dzx

Dxy Dyy Dzy

Dxz Dyz Dzz

…14†

In the reference frame [x', y', z'] that coincides with the principal or main directions of diffusivity, the off-diagonal terms do not exist and the tensor is reduced only to its diagonal terms Dx'x', Dy'y', Dz'z', which represent molecular mobility along axes x', y', and z', respectively. The echo attenuation then becomes: A ˆ exp…ÿbx0 x0 Dx0 x0 ÿ by0 y0 Dy0 y0 ÿ bz0 z0 Dz0 z0 †

…15†

where bii are the elements of the b matrix (which now replaces the b factor) expressed in the coordinates of this reference frame. In practice, measurements are made in the reference frame [x, y, z] of the gradients, which usually does not coincide with that of the tissue. Therefore, one must also consider the coupling of the nondiagonal elements bij of the b matrix with the nondiagonal terms Dij (i6ˆj ), of the diffusion tensor (now expressed in the gradient frame), which re¯ect correlation between molecular displacements in perpendicular directions:85 ! X X A ˆ exp ÿ bij Dij …16† iˆx;y;z jˆx;y;z

Calculation of the value of b may quickly become very complicated when many gradient pulses are used,45 however, the full

METHODS AND APPLICATIONS OF DIFFUSION MRI

determination of the diffusion tensor is necessary to assess anisotropic diffusion properly and fully. To determine the diffusion tensor, DW images are collected along several gradient directions. As the diffusion tensor must be symmetric, measurements along only six directions are mandatory (instead of nine), along with an image acquired without diffusion sensitivity (b=0). A typical set of gradient combinations that preserves uniform space sampling and similar b values along each direction is as follows (coef®cients for gradient pulses along the x, y, z axes, normalized to a given amplitude, G): p p p p p p ‰…1= 2; 0;1= 2†; …ÿ1= 2; 0;1= 2†; …0; 1= 2;1= 2†; p p p p p p …0; 1= 2; ÿ1= 2†; …1= 2;1= 2; 0†; …ÿ1= 2;1 2; 0†Š This minimal set of images may be repeated for averaging, as the SNR may be low12,85 In the case of `axial symmetry', only four directions are necessary (tetrahedral encoding),86 as suggested for the spine.87 The acquisition time and the number of images to process is then reduced, but efforts should be made to collect data along as many directions in space as possible to avoid sampling direction biases for applications such as the determination of ®ber orientation and gain in SNR (see below).88 A second step is to estimate the Dij values from the set of diffusion/orientation-weighted images, generally by multiple linear regression, using Equation (23) below. The ®nal step is to determine the main direction of diffusivities in each voxel and the diffusion values associated with these directions. This is equivalent to determine the reference frame [x', y', z'] where the off-diagonal terms of D are null. This `diagonalization' of the diffusion tensor provides `eigen-vectors' and `eigen-values', , which correspond, respectively, to the main diffusion directions and associated diffusivities,12,85 (the eigendiffusivities  are not to be confused with the tortuosity factors; see below). As it is dif®cult to display tensor data with images (multiple images would be necessary), the use of `diffusion ellipsoids' has been proposed.12,85 Ellipsoids are a three-dimensional representation of the diffusion distance covered in space by molecules in a given diffusion time (Td). These ellipsoids, which can be displayed for each voxel of the image, are easily calculated from the eigen-diffusivities 1, 2, and 3 (corresponding to Dx'x', Dy'y', and Dz'z'): x02 =…21 Td † ‡ y02 =…22 Td † ‡ z02 =…23 Td † ˆ 1

…17†

where x', y', and z' refer to the frame of the main diffusion direction of the tensor. These eigen-diffusivities represent the unidimensional diffusion coef®cients in the main directions of diffusivity of the medium. The main axis of the ellipsoid gives the main diffusion direction in the voxel (coinciding with the direction of the ®bers), while the eccentricity of the ellipsoid provides information about the degree of anisotropy and its symmetry (isotropic diffusion would be seen as a sphere). The length of the ellipsoids in any direction in space is given by the diffusion distance covered in this direction. In other words, the ellipsoid can also be seen as the three-dimensional surface of constant mean squared displacement of the diffusing molecules.

7

Recent work has demonstrated the feasibility of diffusion tensor imaging (DTI) in animal models, as well as in the human brain.41,86,89,90

2.4.2

The Diffusion Propagator and q-space Imaging

In the case where diffusion deviates from the Gaussian model, as with restricted diffusion (see below), it becomes extremely dif®cult to get any meaningful information on molecular displacements from the measured NMR signals. The distribution of the propagator is no longer Gaussian; consequently the value obtained for the diffusion coef®cient using Equation (11) may not necessarily properly re¯ect tissue structure of properties. Diffusion coef®cients may become, in these conditions, meaningless. The main dif®culty is that the medium structure is generally not known in detail and modeling is required. In particular, one must be cautious in the use of Fick's Law and derived relations. The probability distribution driven by the diffusion process (Equation (6)] may now signi®cantly deviate from a Gaussian distribution; as a result the relationship between A and b is no longer exponential as would be expected from Equation (7). A successful analysis should, therefore, start from a reformulation of the probability distribution, taking into account the medium structure and particular boundary or limit conditions. Hence, more useful information may be obtained by directly inferring molecular displacements from the MR signal attenuation without going through a diffusion coef®cient formulation. Molecular displacement pro®les, or more exactly their probability density distribution, can be obtained using q-space imaging.91,92 This does not differ, in principle, from diffusion imaging. Assuming P(z1) is homogeneous throughout the medium, Equation (4) can be rewritten by substituting the relative diffusion displacement z for (z2ÿz1) and by calling q the quantity ( G) as: … A…q;
† ˆ exp‰iqzŠP…z;
†d z …18† It now appears that P(z,
) is just the Fourier transform of the attenuation A(q,
). By inverting Equation (15), direct and unequivocal data on molecular displacements can be obtained. In practice, data are collected for a series of different gradient strengths and an inverse Fourier transform is applied to the corresponding set of signal attenuation values. The experiment can be repeated with different values of
in order to characterize the probability distribution for different diffusion times.93 With q-space imaging, a more comprehensive picture of the system under study is obtained than when using a unique ADC.92 It then becomes straightforward to verify whether P(z,
) is a Gaussian distribution and to determine the value of the diffusion time when the value of P(z,
) may start to deviate from a Gaussian behavior, suggesting that molecules have reached another structural order in the medium, or to obtain the molecular displacement distribution in different conditions. Localized q-space imaging has been used, for instance, to obtain water-displacement pro®les from normal and ischemic mouse brain, showing an increase of the fraction of molecules undergoing displacements of less than 10 m.94 Such ®ndings are very useful for examining the microstructural changes occuring during acute brain ischemia (see below). For References see p. 16

8 METHODS AND APPLICATIONS OF DIFFUSION MRI 2.4.3

Diffusion Spectroscopy

Water may not be the most suitable molecule for diffusion measurements in biological tissues because of its ubiquity and the permeability of most tissue interfaces, such as membranes, to water molecules. Consequently, diffusion measurements of larger molecules that are more speci®c to a tissue or compartment appear to be more promising for tissue characterization. However, such measurements are technically more challenging, because of the low concentration of such molecules and their relatively low diffusion coef®cients. Further, this requirement for signal averaging, and, therefore, long acquisition times, makes in vivo diffusion spectroscopy particularly sensitive to motion.57,95 Recent progress made in in vivo Fourier NMR spectroscopy allows the extension of diffusion measurements to molecules other than water. For a given species, the chemical shift can be used to determine independent diffusion coef®cients of compounds in complex mixtures.96 Diffusion of phosphocreatine, for instance can be studied by 31P spectroscopy.97.98 Phosphocreatine is a true intracellular space probe, in contrast to water, which diffuses across cell membranes; as a result it can be used to observe true restricted diffusion. Phosphocreatine may be used to provide unique information on the intracellular medium, such as viscosity or geometry. Similarly, diffusion of Nacetylaspartate (NAR) and myo-inositol may provide useful information on neuronal and glial cell populations, respectively. Monitoring of exchanges of metabolites or drugs through cell membranes could also bene®t from diffusion ®lter techniques designed to separate intracellular from extracellular compounds.99 Although diffusion coef®cients of metabolites are low because of their size, for nuclear species with spin >1/2 or for coupled-spin systems, multiple quantum experiments are less demanding on gradient hardware because the effective gradient amplitudes are increased by the power of the coherence order (n). With n=2, a fourfold increase in the diffusion effect would occur. Such spectral editing techniques have been used for selective measurement of diffusion properties of J-coupled compounds such as lactate.100,101 3 DIFFUSION IN BIOLOGICAL SYSTEMS: EFFECTS OF MICRODYNAMICS AND MICROSTRUCTURE Signi®cant technical progress has been achieved in biological systems to ensure that meaningful data can be acquired with minimal artifacts. However, even in the ideal case with perfect data, the interpretation of diffusion NMR experiments performed in biological tissues remains challenging. The diffusion coef®cient of water in tissues has been found to be two to ten times less than that of pure water (Table 1).14 This can be largely understood by considering that water molecules are obliged to divert tortuously around obstructions presented by ®bers, intracellular organelles, or macromolecules. Water molecules may be con®ned in bounded compartments or retained by attractive centers or surfaces. Biological systems, therefore, differ greatly from an `in®nitely large medium'. They are very heterogeneous and made of multiple subcompartments (microstructure). Depending on the permeability of the barriers that limit these compartments, exchanges and transport between them (microdynamics) may occur. A classic treatment of the NMR signal may not then re¯ect properly tissue structure or For list of General Abbreviations see end-papers

Table 1 Diffusion Coef®cients of Water in Human Brain Mean diffusivity Anisotropy (10ÿ3 mm sÿ1) (1ÿvolume ratio) Cerebrospinal ¯uid Gray matter (frontal cortex) Caudate nucleus White matter Pyramidal tract Corpus callosum (splenium) Internal capsule Centrum semiovale

3.19 ‹ 0.10 0.83 ‹ 0.05 0.67 ‹ 0.02

0.02 ‹ 0.01 0.08 ‹ 0.05 0.08 ‹ 0.03

0.71 ‹ 0.04 0.69 ‹ 0.05 0.64 ‹ 0.03 0.65 ‹ 0.02

0.93 ‹ 0.04 0.86 ‹ 0.05 0.70 ‹ 0.08 0.27 ‹ 0.03

Measurements (mean ‹ SD) were obtained in normal volunteers using diffusion tensor MRI. (With permission from Pierpaoli et al.41)

properties. Water molecules will `sense' all these obstacles only if the time over which diffusion is measured is made suf®ciently long that signi®cant interaction of the water molecules with cellular compartments may occur. The concept of `diffusion time' is, therefore, central to any diffusion study in biological tissues. While it may appear that using long diffusion times may reveal more tissue interactions, it is perhaps even more interesting to make the diffusion time as short as possible, up to the point where the diffusion coef®cient will reach its value in pure water. By doing so, and plotting measured diffusion coef®cients versus diffusion times, the many mechanisms contributing to in vivo diffusion may be revealed and identi®ed. This approach, which will bene®t from the newly developed gradient hardware allowing very large gradient strength, may be compared in some ways to that used in NMR dispersion studies, where multiple elementary relaxation mechanisms can be identi®ed and separated by studying dispersion curves.102 Diffusion coef®cients may become, in these conditions, meaningless if the measurement time scale or the measurement direction are not provided. The main dif®culty is that the medium structure is generally unknown in detail. The issue is, therefore, to infer meaningful information on tissue from the measured NMR signals. Initially, this problem could be reversed to consider how known tissue features relevant to tissue microstructure and dynamics affect diffusion NMR signals. Diffusion NMR has been used to study ¯uid-®lled porous media and derive information on microstructure (shape of the pore space) and ¯uid permeability (porosity).103±105 It is clear that many of the questions addressed in those studies are relevant to diffusion measurements in biological media.5 Differences in diffusion coef®cient and available diffusion space can be used to distinguish compartments and exchanges between them.99 3.1

Effect of Temperature

The ®rst obvious effect on diffusion is that of temperature, as diffusion directly results from the molecular thermal motion. The diffusion sensitivity to temperature is high, about 2.4% for 1 C change (Figure 6),106,107 consequently temperature must be carefully controlled in diffusion MRI experiments. Based on the strong and unique relationship that exists between temperature and molecular diffusion, diffusion imaging has been proposed for the real-time and noninvasive monitoring of tem-

METHODS AND APPLICATIONS OF DIFFUSION MRI

Temperature (MRI) (°C)

37

32

T1 Diffusion

27

22 22

27 32 Temperature (Luxtron) (°C)

37

Figure 6 Comparison of temperature measurements using T1 and diffusion MRI, and using probes (Luxtron) in a phantom. The accuracy was found to be 0.2 C with diffusion and 0.5 C with T1

perature. Noninvasive and nondestructive temperature imaging in biological systems may be particularly useful to monitor hyperthermia treatments in real-time, whether using rf electrical ®elds108 or focused ultrasound.109,110 It could also be used to study and control tissue interactions in surgical and medical laser procedures. However, in the context of this chapter, the diffusion±temperature relationship also provides some indication about diffusion mechanisms in tissues. This relationship has been established semi-empirically in liquids1 and veri®ed experimentally.106,111 D ˆ D0 exp…ÿEa =kT †

…19†

where k is the Boltzmann constant, T is the temperature, D0 is the diffusion that would be obtained at an in®nite temperature, and Ea is introduced as the translational diffusion activation energy, which is approximately equal to the energy required to break hydrogen bonds, both in pure water and in vivo (Ea=0.2 eV).107,111 This important result can be understood by considering that the mechanism at the molecular scale for water to move, and thus diffuse, involves continuous breaking and reforming of hydrogen bonds. The fact that Ea is identical for water molecules diffusing in vitro and in vivo is not really surprising as the laws of physics at this level should not be different. 3.2 Restricted Diffusion Diffusion is restricted when boundaries in the medium prevent molecules from moving freely.112,113 Restriction must be related to the experimental parameters. When measurement times are very short, most molecules do not have enough time to reach boundaries and so they behave as if diffusing freely. Once the diffusion time increases, an increasing fraction of molecules will strike the boundaries, and diffusion will deviate

9

from the free, Gaussian behavior. Hence, the usual way to check for restricted diffusion is to plot the diffusion distance, calculated as for free diffusion using Einstein's equation, as a function of the square root of the diffusion time. In the case of restricted diffusion, this plot shows a curvature and ®nally a leveling off when the diffusion distance reaches the size of the restricting compartment. The effects of restriction will, therefore, appear in the NMR signal for diffusion times such that molecular displacements are in the order of the size of the restricting volumes. These effects will depend on the type of restriction (impermeable or permeable barriers, attractive centers, etc.), the shape of the restricting volumes (spherical, cylindrical, parallel walls, etc.), and the type of NMR experiment (constant or pulsed gradients). As a result, there is not a unique analytical expression that could describe any con®guration. A simple example is represented by molecules diffusing between two impermeable parallel walls separated by a distance a.112 If the theoretical, free diffusion distance greatly exceeds a, the echo attenuation A in the case of the bipolar gradient pulse experiment signi®cantly deviates from an exponential decay and becomes independent of the diffusion time,112 implying that molecules are trapped in the direction of the applied gradient: A ˆ ‰sin… G a=2†=… G a=2†Š2

…20†

Another interesting, but somewhat more complicated, case is represented by diffusion restricted in a spherical cavity of radius R0. In the limit where the theoretical, free diffusion distance largely exceeds R0, the attenuation is again independent of the diffusion time112,114 and the measured ADC decreases when the diffusion time is increased: A ˆ exp‰ÿ… G†2 R20 =5Š

…21†

corresponding to an asymptotic ADC value, Dasymp of R20 /5. The factor 5 would be replaced by 3 if diffusion was con®ned to the surface of the sphere. Whatever the geometry of the restrictive medium, the deviation from linearity in the semi-log plot of the signal attenuation versus b is crucial to determine whether diffusion is restricted, although other causes may be responsible, such as diffusion in inhomogeneous systems or anisotropic diffusion. The ultimate test is to show that the measured diffusion coef®cient or the signal attenuation varies when the diffusion time is changed. Such studies can theoretically lead to the determination of the geometry and size of the restricting boundaries, as with the q-space concept. To avoid restricted diffusion effects, the diffusion time must be decreased to ensure that the diffusion distance during that period remains less than the size R of the restricted region. Unfortunately, the diffusion effect in these conditions becomes small unless considerable gradient intensities are used. A further complication is that in biological tissues walls may not be re¯ecting boundaries but rather occur as partially absorbing borders.115 3.3

Permeable Barriers

When the restrictive barriers become permeable to diffusing molecules, the restricted diffusion pattern changes. The mathematical treatment of diffusion in systems partitioned by For References see p. 16

10 METHODS AND APPLICATIONS OF DIFFUSION MRI permeable barriers is far from simple. An example was given by Tanner for equally spaced, plane barriers having a permeability constant .116 For short diffusion times, the ADC is D0. When the diffusion time increases, the ADC decreases, as expected for restricted diffusion, but saturates at Dasymp, which depends on the permeability constant: Dasymp ˆ D0 =…1 ‡ D0 =a†

…22†

where a is the barrier spacing. This spacing can be estimated by the equivalent free diffusion distance that would be obtained from Einstein's equation with D=(D0+Dasymp)/2 as the diffusion coef®cient and Td1=2 , the corresponding diffusion time.117 The plot of D versus Td then shows a typical sigmoid pattern and it becomes possible to estimate the barrier permeability  by ®tting the data with Equation (22).118 This approach is, however, very optimistic, as the geometrical arrangement of the medium is generally not known. In particular, this formalism does not apply to the case where the system consists of spherical cavities separated by permeable barriers.

3.4

Hindered Diffusion

True `restricted' diffusion, which occurs in bounded media and leads to a decrease of diffusion with the diffusion time should be distinguished from `hindered' diffusion. With hindered diffusion, diffusion is decreased by the presence of obstacles but there is no limit to the diffusion distance; consequently diffusion does not change with the diffusion time (once the hindered diffusion regime has been reached). Perhaps the most powerful concept associated with hindered diffusion is that of `tortuosity', a concept that has been widely used in solid porous media studies and more recently in brain diffusion experiments using external tracers.119±121 The idea is that because of the presence of obstacles such as ®bers, macromolecules, and organelles water molecules must travel longer paths to cover any given distance. In other words, molecules can no longer go straight from A to B, but must diffuse around structures that are impermeable to them (Figure 7). This situation results in a longer diffusion time to diffuse from A to B, or to an apparent decrease in the diffusion distance covered in a given diffusion time and in the measured ADC. This `hindered'

Figure 7 Restricted and hindered diffusion in white matter. Several models have been suggested to explain white matter diffusion anisotropy. First, diffusion may simply be totally restricted in axons (diameter, d). Plots of diffusion distance versus diffusion time (as in Figure 8, below) however, do not support this hypothesis. One may also take into account permeability of membranes and myelin to water. Going from A to B, water molecules would have to cross several interfaces. However, one may argue that those results may also be compatible with a `tortuosity' or `hindrance' model, where water molecules going from A to B would have to move around ®bers. As there is no limit to diffusion, this model is totally compatible with Figure 8 if extracellular water diffusion predominates

For list of General Abbreviations see end-papers

METHODS AND APPLICATIONS OF DIFFUSION MRI

11

14 Diffusion distance (mm)

12 10 8 6 Gray matter White matter, x White matter, z

4 2 0 0

2

4

6

8

10

Td (ms1/2)

Figure 8 Plot of the diffusion distance against diffusion time in gray and white matter. The diffusion distance was obtained from the diffusion coef®cient using Einstein's relation. There is no leveling off of the diffusion distance, suggesting that water diffusion is not completely restricted

diffusion effect is classically expressed quantitatively using a `tortuosity' coef®cient, , such that: ADC ˆ D=2

Figure 9 Diffusion image showing anisotropic diffusion in white matter. Excellent contrast is achieved, although T1 and T2 effects have been removed. Corpus callosum and temporal white matter ®bers, which are horizontal, are dark. Vertical corona radiata frontal ®bers and internal capsule ®bers are bright. In brainstem, vertical fast-conducting motor and somatosensory tracts are also bright

…23†

where D would be the diffusion coef®cient observed in the absence of obstacles. Furthermore, as there is no real barrier, molecules can, in principle, diffuse over very large distances compared with those seen in restricted diffusion. Therefore, no curvature would be seen in the plot of the diffusion distance versus the square root of the diffusion time. Similarly, the measured ADC would not depend on the diffusion time, unless the diffusion time is very short, and hindered diffusion paths would not differ signi®cantly from free diffusion paths (Figure 8). It would then be dif®cult to distinguish between hindered diffusion and diffusion restricted by permeable barriers. 3.5 Anisotropic Diffusion Diffusion in tissues may be different for different directions of motion; this anisotropy will give rise to variations in the measured diffusion coef®cient with the direction of measurement. Diffusion anisotropy has been observed in muscle,83 in brain white matter,37,56,84 (Figure 9) and, recently, in gray matter.122,123 It may result from restriction of diffusion inside ®bers (intracellular water) or from an increased tortuosity when diffusion occurs around ®bers (extracellular water). Diffusion can be both anisotropic and unrestricted. This behavior is well known in nematic liquid crystals124 and can be found in the water lamellar phase of amphiphilic lyotropic systems.125 Using DTI, it appears that diffusion data can be analyzed in three ways to provide information on tissue microstructure and architecture for each voxel or region of interest:126 the mean diffusivity, which characterizes the overall mean-squared displacement of molecules (average ellipsoid size) and the overall presence of obstacles to diffusion; the degree of anisotropy, which describes how much molecular displacements vary in space (ellipsoid eccentricity) and is related to the presence of

oriented structures; and the main direction of diffusivities (main ellipsoid axes), which is linked to the orientation in space of the structures. These three DTI `meta-parameters' can all be derived from complete knowledge of the diffusion tensor. However because of the complexity of data acquisition and processing for full DTI, and the sensitivity to noise of the determination of the diffusion tensor eigen values, simpli®ed approaches have been proposed. 3.5.1

Mean Diffusivity

To obtain an overall evaluation of the diffusion in a voxel or region, anisotropic diffusion effects must be avoided and the result limited to an `invariant', i.e., a quantity that is independent of the orientation of the reference frame.12,86 Among several combinations of the tensor elements, the trace of the diffusion tensor, Tr…D† ˆ Dxx ‡ Dyy ‡ Dzz

…24†

is such an invariant. The mean diffusivity is then given by Tr(D)/3. A slightly different de®nition of the trace has proved useful in assessing the diffusion drop in brain ischemia127 (see below). Unfortunately, the correct estimation of Tr(D) still requires the complete determination of the diffusion tensor. Diffusion coef®cients obtained by separately acquiring data with gradient pulses added along the x, y, and z axes cannot be used as these measured coef®cients usually do not coincide with Dxx, Dyy, and Dzz, respectively. The reason is that the diffusion attenuation that results, for instance from inserting gradients on the x axis, is A=exp[ÿbxxDxx+2bxyDxy+2bxzDxz] and not simply exp[ÿbxxDxx] unless diffusion is isotropic (no nondiagonal terms) or there are no gradient pulses at all on the other axes (y and z, here) (no localization) during the diffusion measurement time. For References see p. 16

12 METHODS AND APPLICATIONS OF DIFFUSION MRI To avoid this problem and to simplify the approach, several groups have designed sequences based on multiple echoes or acquisitions with tetrahedral gradient con®gurations to cancel nondiagonal term contributions to the MRI signal directly.86,128±130 3.5.2

Diffusion Anisotropy Indices

Several scalar indices have been proposed to characterize diffusion anisotropy. Initially, simple indices calculated from DW images84 or ADC values obtained in perpendicular directions were used, such as ADCx/ADCy and displayed using a color,131 scale. Other groups have devised indices mixing measurements along the x, y, and z directions, such as maximum [ADCx, ADCy, ADCz]/minimum [ADCx, ADCy, ADCz] or the standard deviation of ADCx, ADCy, and ADCz divided by their mean value.127 Unfortunately, none of these indices is really objective as they do not correspond to a single meaningful physical parameter; more importantly, they are clearly dependent on the choice of directions made for the measurements. The degree of anisotropy would then vary upon the respective orientation of the gradient hardware and the tissue frames of reference and would generally be underestimated. Here again invariant indices must be found to avoid such biases and provide objective, intrinsic structural information.132 Invariant indices are made of combinations of the terms of the diagonalized diffusion tensor, i.e., the eigen-values 1,2, and 3. The most commonly used invariant indices are the relative anisotropy, RA, the fractional anisotropy, FA, and the volume ratio, VR: q p …1 ÿ hi†2 ‡ …2 ÿ hi†2 ‡ …3 ÿ hi†2 = 3hi

RA ˆ

…25†

where hi ˆ …1 ‡ 2 ‡ 3 †=3 FA ˆ

q 3‰…1 ÿ hi†2 ‡ …2 ÿ hi†2 ‡ …3 ÿ hi†2 Š= q 2…21 ‡ 22 ‡ 23 †

VR ˆ 1 2 3 =hi3

…26† …27†

RA, a normalized standard deviation, also represents the ratio of the anisotropic part of D to its isotropic part. FA measures the fraction of the `magnitude' of D that can be ascribed to anisotropic diffusion. p FA and RA vary between 0 (isotropic diffusion) and 1 ( 2 for RA) (in®nite anisotropy). As for VR, which represents the ratio of the ellipsoid volume to the volume of a sphere of radius hi, its range is from 1 (isotropic diffusion) to 0; as a result some authors prefer to use (1ÿVR).133 Once these indices have been de®ned, it is possible to evaluate them directly from DW images, i.e., without the need to calculate the diffusion tensor.134 For instance, A which is very similar to RA, has been proposed as:135 A ˆ

s X p  …Dii ÿ hDi†2 ‡ …D2xy ‡ D2xz ‡ D2yz = 6hDi …28† iˆx;y;z

with hDi=

X

Dii/3.

iˆx;y;z

For list of General Abbreviations see end-papers

Also, images directly sensitive to anisotropy indices, or anisotropically weighted images, can be obtained.136 Finally, the concept of these intravoxel anisotropy indices can be extended to a family of intervoxel or `lattice' measures of diffusion anisotropy, which allows neighboring voxels to be considered together, in a region of interest, without losing anisotropy effects resulting from different ®ber orientations across voxels.133 Clinically relevant images of anisotropy indices have been obtained in the human brain (Table 1).41,137 3.5.3

Fiber Orientation Mapping

The last family of parameters that can be extracted from the DTI concept relates to the mapping of the orientation in space of tissue structure. The assumption is that the direction of the ®bers is colinear with the direction of the eigen-vector associated with the largest eigen-diffusivity. This approach opens a completely new way to obtain directly in vivo information on the organization in space of tissues such as muscle, myocardium, and brain or spine white matter, which is of considerable interest, clinically and functionally. Direction orientation can be derived from DTI directly from diffusion/orientationweighted images or through the calculation of the diffusion tensor. A ®rst issue is to display ®ber orientation on a voxelby-voxel basis. The use of color maps was ®rst suggested,131 followed by representation by ellipsoids,12,133 octahedra138 or vectors pointing in the ®ber direction.139,140 A second issue is to assess connectivity from DTI data in order to visualize anatomical connections between different parts of the brain on an individual basis. Studies of neuronal connectivity are tremendously important to the interpretation of functional MRI data and for establishing how activated foci are linked together through networks.141,142 This issue is dif®cult, as continuity of ®ber orientation from voxel to voxel has to be inferred. Fiber orientation may appear to be varying because of the occurrence of noise in the data. In a given voxel ®bers may be merging, branching or dividing. In addition, several fascicles may cross in a given voxel, which cannot be detected with the diffusiontensor approach in its present form. Structures that exhibit anisotropic diffusion at the molecular level can be isotropically oriented at the microscopic level, resulting in a `powder average' effect that is dif®cult to resolve.143 The semilog plot of the signal attenuation versus b may not be linear in this case.125 This deviation from linearity can be ascribed to anisotropy and not to restricted diffusion because the diffusion measurements are independent of the diffusion time. Several groups have recently approached the dif®cult problem of inferring connectivity from DTI data in the rat brain144 in vitro and in the living human brain145 (Figure 10). 3.6

Diffusion in Multiple Compartment Systems

Most diffusion measurements in biological tissues refer to an ADC and yet it is generally considered that diffusion in the measurement volume (voxel) has a unique diffusion coef®cient. This simpli®cation may not always be legitimate, since partial volume effects may occur and most tissues are made of multiple subcompartments (with at least intracellular and extracellular components). Assuming measurement times are short and diffusion is unrestricted in each subcompartment i, and that there is no exchange, the signal attenuation is:

METHODS AND APPLICATIONS OF DIFFUSION MRI

Aˆ

X

i exp…ÿbDi †

…29†

i

where i is the density of molecules diffusing in compartment i, and Di the associated diffusion coef®cient. In this case, the ADC that would be measured would depend on the range used for the b values and would not re¯ect properly the diffusion in the voxel. Measurements with low b values would then be

13

more sensitive to fast diffusion components. The ideal approach would be to separate all subcompartments by ®tting the data with a multiexponential decay. Unfortunately, the values for Di are often low and not very different from each other; consequently very large b values and very high SNR would be required. Relaxation effects must also be considered if compartments have different relaxation rates.146 If all spins do not have the same transverse relaxation during the pulse sequence, Equation

Figure 10 Diffusion tensor imaging and ®ber orientation mapping in the brain of a normal volunteer at 1.5 T. The direction of the white matter ®bers can be determined from the eigen vectors of the diffusion tensor for each voxel and superimposed on a 3D rendering of internal brain structures. (Contribution of J.F. Mangin, C. Clark, C. Poupon, D. Le Bihan)

For References see p. 16

14 METHODS AND APPLICATIONS OF DIFFUSION MRI (5) must be complemented with a decay term:115 @P…z2 ; z1 t†=@t ˆ Dr2 P…z2 ; z1 t† ÿ P…z2 ; z1 ; t†=T2 …30† For two compartments, assuming measurement times are short and, therefore, exchanges are small, the diffusion signal is given by: S ˆ F1 exp…ÿbD1 † ‡ F2 exp…ÿbD2 †

…31†

where b is the sequence diffusion factor, D1, D2 are the diffusion coef®cients and F1, F2 are relative weights to the signal of the two compartments. Estimates of F1, F2 and D1, D2 are obtained by ®tting a series of signals acquired with multiple b values. For a spin echo sequence with TR>>T1, F depends on the volume fraction f and the relaxation time T2 of each compartment: F ˆ f exp…ÿTE=T2 †

…32†

Two diffusion compartments have been found in the rat brain.147 Assignment of these compartments to extra- and intracellular water is not, however, straightforward, as the values found for F1 and F2 are apparently opposite to the known volume fractions for the intra- and extracellular spaces (82.5% and 17.5%, respectively). A ®rst explanation for this discrepancy is that T2 effects must be taken into account when converting F values into f values. A second factor is that the exchange rate may not be negligible. It has been shown that if the residence rate in each compartment is considered, the contribution of the fast diffusing component is overestimated.148 It is, therefore, not yet certain that the two compartments that have been observed do correspond to the intra- and extracellular compartments. The situation is even more complex when measurement times are longer. First, restricted diffusion may be seen in the smallest subcompartments. Second, molecular exchanges may occur between communicating compartments. As a result, the analytic treatment becomes dif®cult. Applying the central limit theorem for statistically distributed compartments in the case of long diffusion times, the use of a single apparent diffusion constant Da can be justi®ed: X i Di …33† Da ˆ i

This result is also consistent with NMR dispersion studies,102 which consider that cell membranes can be ignored on the NMR time scale. In intermediate situations, the geometrical arrangement and the diffusion coef®cients of each compartment must be considered as well as their rates of exchange.149 The comprehensive analysis of the diffusion attenuation curves obtained with different diffusion times may lead to an accurate description of the medium microstructure. It is clear that the ADC depends on the range used for the b values and would not re¯ect properly diffusion in the tissue. Measurements with low b values (less than 1000 s mmÿ2) would be more sensitive to fast diffusion components, such as those occurring in the extracellular compartment. In clinical studies and most animal experiments, especially when data are ®tted to a single exponential, diffusion patterns observed in tissues have to be explained by features of the extracellular For list of General Abbreviations see end-papers

space, although this compartment is physically small. Changes in the ADC calculated in these conditions should be interpreted in terms of changes in the diffusion coef®cient of the extracellular space (tortuosity) and in its fractional volume relative to the intracellular volume. This explains why diffusion has appeared as a sensitive marker of the changes in the extracellular/intracellular volume ratio, as observed in brain ischemia, spreading depression,80±82 status epilepticus,78 or extraphysiological manipulations of the cell size through osmotic agents.150 Using large b values, it becomes increasingly possible to assess separately the intra- and extracellular compartments and their relative volumes in vitro151 and in vivo, as attempted in the rat brain147 and the human brain.152

3.7

Metabolites

Data on magnitudes and even directional anisotropy of diffusion coef®cients of molecules such as choline, creatine/ phosphocreatine and NAA in animals,97,153,154 and human brain95 have been made available. Although diffusion rates of metabolites in pure water are lower than that of water because of the difference in molecular weight and hydration layers, the ADC values of these metabolites in brain are considerably smaller. Typical values from a series of 10 normal volunteers (18 cm3 voxels located in white matter, diffusion time 240 ms) are 0.1310ÿ3 mm2 sÿ1 for choline and creatine and 0.1810ÿ3 mm2 sÿ1 for NAA compared with 1.2410ÿ3 mm2 sÿ1 for choline and 0.8510ÿ3 mm2 sÿ1 for NAA at 35 C in vitro.155 The fact that metabolites, as well as molecules such as ¯uorodeoxyglucose 6-phosphate,156 with different molecular weights had very similar ADC values in vivo is striking and remains to be explained. The very low values of the ADC obtained with long diffusion times are, of course, compatible with restriction of these metabolites in compartments that were about the same size. For instance, recent experiments varying the diffusion times have shown that diffusion of NAA is considerably restricted in a manner compatible with two compartments, one of 7±8 m and one of approximately 1 m, possibly representing the cell bodies and the intra-axonal space.157 However, intracellular obstacles may also play an important role, as water molecules could diffuse easily in small spaces between small obstacles, such as mitochondria, organelles or macromolecules, while larger metabolites would `feel' a more obstructed medium. Tortuosity factors may, therefore, be larger for metabolites than for water and this would explain, at least partially, the diffusion decrease for intracellular metabolites that has been observed during ischemia.158,159

3.8

Summary: Application to Brain White Matter

The concepts of restriction, hindrance, tortuosity, and multiple compartments are particularly useful to understand diffusion ®ndings in brain white matter. Water diffusion is highly anisotropic in white matter37,56,84 and this anisotropy is observed even before ®bers are myelinated, though at a lesser degree.137,160±166 ADC values obtained by measurements made parallel and perpendicularly to the ®bers do not seem to depend on the diffusion time167,168 at least for diffusion times longer than 20 ms (see Figure 8).

METHODS AND APPLICATIONS OF DIFFUSION MRI

Initial reports suggested that the anisotropic water diffusion could be explained by restriction of the water molecules to the axons (anisotropically restricted diffusion) by the myelin sheath.169,170 However, although restricted diffusion has been seen for intra-axonal metabolites, such as NAA, or for truly intra-axonal water,171 it now appears that most studies were performed with relatively low b values and are mostly sensitive to the extracellular, interaxonal space. In this condition, diffusion anisotropy in white matter should be linked to the anisotropic tortuosity of the interstitial space between the ®bers: diffusion would be more impeded perpendicular to the ®bers because of their geometric arrangement (Figure 7).167 Considering a bundle where ®bers are organized in the most compact way, molecules would have to actually travel over a distance of d/2, where d is the ®ber diameter, for an apparent diffusion distance equal to the ®ber diameter. Therefore, on average, the ADC measured perpendicularly to the ®bers would be reduced, whatever the diffusion time, to: ADC ˆ …2=†2 D0  0:4D0

…34†

where the reference value, D0, is the diffusion value measured parallel to the ®bers. This ratio of about 0.4 ®ts reasonably very well with literature data,167 although much larger ratios have been reported.133 This rough model also implies that the tortuosity factor would be anisotropic in white matter with a perpendicular/parallel ratio of /2 (&1.15). Unfortunately no systematic measurements of  have yet been made in white matter using ionic extracellular tracers,2,119 although anisotropy has been seen.172 Another interesting point about this model is that it is compatible with the fact that no true restricted effects are observed when the diffusion time is increased, as there are no actual boundaries to diffusing molecules. Also, the parallel organization of the ®bers may be suf®cient to explain the presence of anisotropy before myelination. However, as the axonal membranes should be more permeable to water than the myelin sheaths, the degree of anisotropy should be less pronounced in the absence of myelin, as signi®cant exchanges with the axonal spaces should occur. Oriented ®laments within the axoplasm do not seem to play an important role.173 Fiber orientation mapping and connectivity studies derived from anisotropic diffusion in white matter will clearly bene®t from a better understanding of the respective contributions of intraaxonal and extra-axonal compartments to anisotropy mechanisms.174

4 CLINICAL APPLICATIONS Diffusion imaging is a truly quantitative method. The diffusion coef®cient is a physical parameter that directly re¯ects the physical properties of the tissues in terms of the random translational movement of the molecules under study (most often water molecules, but sometimes metabolites. The diffusion coef®cient does not depend on the ®eld strength of the magnet or the pulse sequence used, which is not the case for the other classical MRI parameters such as T1 or T2. Diffusion coef®cients obtained at different times in a given patient, or in different patients, or in different hospitals can be compared without any need for normalization.

4.1

15

Central Nervous System

The most clinically relevant ®eld of application of diffusion MRI is in the nervous system. Diffusion MRI has already appeared as a breakthrough in two areas: early brain ischemia and white matter diseases. During the acute stage of brain ischemia, water diffusion is decreased in the ischemic territory by as much as 50%, as shown in cat brain models.175 This diffusion slowdown is linked to the cytotoxic edema that results from the energetic failure of the cellular membrane Na+/K+ pumping system. The exact mechanism by which diffusion is reduced is still unclear (increase of the slow-diffusion intracellular volume fraction, changes in membrane permeability,176 or shrinkage of the extracellular space resulting in increased tortuosity for water molecules.177,178 have been suggested). Diffusion MRI has been extensively used in animal models to establish and test new therapeutic approaches. These results have been con®rmed in patients with stroke, offering the potential to highlight ischemic regions within the ®rst hours of the ischemic event, when brain tissue might still be salvageable,179,180 well before conventional MRI becomes abnormal (vasogenic edema). Combined with perfusion MRI, diffusion MRI is under clinical evaluation as a tool to help clinicians to optimize their therapeutic approach to individual patients,181 to monitor patient progress on an objective basis, and to predict clinical outcome.182±185 In white matter, DTI has already shown its potential in diseases such as multiple sclerosis186±188 leukoencephalopathy,189 Wallerian degeneration, Alzheimer disease,190,191 and Creutzfeld±Jacob disease.192±194 Mean diffusivity indices, such as the trace of the diffusion tensor, re¯ect overall water content, while anisotropy indices indicate myelin ®ber integrity. It has been shown that the degree of diffusion anisotropy in white matter increases during the myelination process,137,162,195 and diffusion MRI could be used to assess brain maturation in children,196 newborns, or premature babies.137,197 Abnormal connectivity in frontal white matter based on DTI data has also been reported in schizophrenic patients.198 The potential of diffusion MRI has also been studied in brain tumor grading,199±201 trauma,202 hypertensive hydrocephalus,203 AIDS,204 eclampsia,205 leukoaraiosis,206,207 and the spinal cord.87,185,208,209 4.2

Body

The use of diffusion imaging has been less successful in areas of the body apart from the brain because of the occurrence of strong respiratory motion artifacts and of the short T2 values of body tissues, which require shorter TE than in the brain and, therefore, leaves less room for the diffusion gradient pulses. These obstacles, however, can sometimes be overcome with ad hoc MR sequences and hardware. Potential for tissue characterization has been shown in the extremity muscles,210 the spine,211,212 the breast,54,213,214 the kidney, and the liver.215±218 Muscle ®ber orientation can be approached using DTI in organs such as the tongue219 or the heart. Myocardium DTI220,221 has tremendous potential for providing data on heart contractility, a very important parameter, but remains technically very challenging to perform in vivo because of heart motion. Other applications include temperature imaging For References see p. 16

16 METHODS AND APPLICATIONS OF DIFFUSION MRI through the sensitivity of diffusion coef®cients to temperature.107,108,210,222

5 CONCLUSION Many tissue features at the microscopic level may in¯uence NMR diffusion measurements. So far, many theoretical analyses of the effect of restriction, membrane permeability, hindrance, anisotropy, or tissue inhomogeneity have been published. These analyses underline how much care is necessary to conduct diffusion NMR studies properly and to interpret the results. Although the results of these analyses have been applied to characterize nonliving systems, such as porous media, much work remains to be done to produce accurate information on microstructure and microdynamics in vivo in biological systems. Powerful tools, such as diffusion spectroscopy of metabolites, DTI or q-space imaging, that are still under development are expected to provide such information.

6 RELATED ARTICLES Anisotropically Restricted Diffusion in MRI; Diffusion: Clinical Utility of MRI Studies; Ischemic Stroke; Male Pelvis Studies Using MRI.

7 REFERENCES 1. W. Jost, `Diffusion in Solids, Liquids and Gases', Academic Press, San Diego, 1960. 2. C. Nicholson, `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', ed. D. Le Bihan, Raven Press, New York, 1995, p. 127 3. O. B. Paulson, M. M. Hertz, T. G. Bolwig, and N. A. Lassen, Microvasc. Res., 1977, 13, 113. 4. D. Le Bihan (ed.), `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', Raven Press, New York, 1995. 5. P. T. Callaghan, `Principles of Nuclear Magnetic Resonance Microscopy', Oxford University Press, Oxford, 1991. 6. D. Le Bihan and E. Breton, C. R. Acad. Sci. Paris, Series II, 1985, T.301, 1109. 7. K. D. Merboldt, W. Hanicke, and J. Frahn, J. Magn. Reson., 1985, 64, 479. 8. D. G. Taylor and M. C. Bushell, Phys. Med. Biol., 1985, 30, 345. 9. E. L. Hahn, Phys. Rev., 1950, 80, 580. 10. H. Y. Carr and E. M. Purcell, Phys. Rev., 1954, 94, 630. 11. E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288. 12. P. J. Basser, J. Mattiello, and D. Le Bihan, Biophys. J., 1994, 66, 259. 13. J. E. Tanner, J. Chem. Phys., 1970, 52, 2523. 14. D. Le Bihan, Magn. Reson. Q., 1991, 7, 1. 15. R. Kaiser, E. Bartholdi, and R. R. Ernst, J. Chem. Phys., 1974, 60, 2966. 16. D. Le Bihan, Magn. Reson. Med., 1988, 7, 346. 17. D. Le Bihan, R. Turner, and J. R. MacFall, Magn. Reson. Med., 1989, 10, 324. 18. K. D. Merboldt, W. Hanicke, M. L. Gyngell, J. Frahm, and H. Bruhn, J. Magn. Reson., 1989, 82, 115.

For list of General Abbreviations see end-papers

19. K. D. Merboldt, H. Bruhn, J. Frahm, M. L. Gyngell, W. Hanicke, and M. Deimling, Magn. Reson. Med., 1989, 9, 423. 20. E. X. Wu and R. B. Buxton, J. Magn. Reson., 1990, 90, 243. 21. D. Canet, B. Diter, A. Belmajdoub, J. Braondeau, J. C. Boubel, and K. Elbayed, J. Magn. Reson., 1989, 81, 1. 22. G. S. Karczmar, D. B. Twieg, T. J. Lawry, and M. W. Weiner, Magn. Reson. Med., 1988, 7, 111. 23. F. Humbert, M. Valtier, A. Retournard, and D. Canet, J. Magn. Reson., 1998, 134, 245. 24. F. Humbert, B. Diter, and D. Canet, J. Magn. Reson., 1996, A123, 242. 25. M. Neeman, J. P. Freyer, and L. O. Sillerud, J. Magn. Reson., 1990, 90, 303. 26. H. C. Torrey, Phys. Rev., 1965, 104, 563. 27. J. Mattiello, P. J. Basser, and D. Le Bihan, J. Magn. Reson., 1994, 108, 131. 28. D. Le Bihan, E. Breton, D. Lallemand, P. Grenier, and E. A. Cabanis, Radiology, 1986, 161, 401. 29. G. A. Ojemann, J. Neurosci., 1991, 11, 2281. 30. R. V. Mulkern and R. G. S. Spencer, Magn. Reson. Imag., 1988, 6, 623. 31. H. Gubjartsson, S. E. Maier, and F. A. Jolesz, Magn. Reson. Med., 1997, 38, 101. 32. S. E. Maier, H. Gudbjartsson, S. Patz, L. Hsu, K. O. LoÈvblad, R. R. Edelman, S. Warach, and F. A. Jolesz, Am. J. Roentgenol., 1998, 171, 85. 33. K. O. LoÈvblad, P. M. Jakob, Q. Chen, A. E. Baird, G. Schlaug, S. Warach, and R. R. Edelman, Am. J. Neuroradiol., 1998, 19, 201. 34. D. L. Thomas, G. S. Pell, M. F. Lythgoe, D. G. Gadian, and R. J. Ordidge, Magn. Reson. Med., 1998, 39, 950. 35. F. Schick, Magn. Reson. Med., 1997, 38, 638. 36. R. Turner and D. Le Bihan, J. Magn. Reson., 1990, 86, 445. 37. R. Turner, D. Le Bihan, J. Maier, R. Vavrek, L. K. Hedges, and J. Pekar, Radiology, 1990, 177, 407. 38. R. C. McKinstry, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, K. K. Kwong, R. R. Rzedzian, T. J. Brady, and R. R. Rosen, Proc. IXth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 5. 39. H. R. Avram and L. E. Crooks, Proc. VIIth Annu. Mtg. Soc. Magn. Reson. Med., San Francisco, 1988, p. 80. 40. P. Mans®eld, J. Phys. C., 1977, 10, L55. 41. C. Pierpaoli, P. Jezzard, P. J. Basser, A. Barnett, and G. DiChiro, Radiology, 1996, 201, 637. 42. J. S. Doran, P. Jakob, and M. DeÂcorps, Magn. Reson. Med., 1996, 35, 547. 43. J. S. Doran and M. DeÂcorps, J. Magn. Reson., 1995, 117, 311. 44. L. Li, Magn. Reson. Med., 1999, 41, 143. 45. J. Mattiello, P. J. Basser, and D. Le Bihan, in `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', ed. D. Le Bihan, Raven Press, New York, 1995, p. 77. 46. J. H. Burdette, A. D. Elster, and P. E. Ricci, J. Comput. Assist. Tomogr., 1998, 22, 792. 47. D. Xing, N. G. Papadakis, C. L. Huang, V. M. Lee, T. A. Carpenter, and L. D. Hall, Magn. Reson. Imag., 1997, 15, 771. 48. R. M. Vavrek and J. R. MacFall, in `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', ed. D. Le Bihan, Raven Press, New York, 1995, p. 67. 49. P. P. Mitra and B. I. Halperin, J. Magn. Reson., 1995, 113, 94. 50. J. C. Haselgrove and J. R. Moore, Magn. Reson. Med., 1996, 36, 960. 51. P. Jezzard, A. S. Barnett, and C. Pierpaoli, Magn. Reson. Med., 1998, 39, 801. 52. F. Calamante, Magn. Reson. Med., 1999, 41, 95.

METHODS AND APPLICATIONS OF DIFFUSION MRI 53. M. R. Thompson, R. Venkatesan, K. Kuppusamy, A. Celik, W. L. Lin, D. K. Kido, and E. M. Haacke, Radiology, 1999, 210, 253. 54. C. F. Maier, H. N. Nikolov, K. C. Chu, B. A. Chronick, and B. K. Rutt, Magn. Reson. Med., 1998, 39, 392. 55. T. L. Chenevert and J. G. Pipe, Magn. Reson. Med., 1991, 19, 261. 56. T. L. Chenevert, J. A. Brunberg, and J. G. Pipe, Radiology, 1990, 177, 401. 57. S. Posse, C. A. Cuenod, and D. Le Bihan, J. Magn. Reson., 1993, 102, 222. 58. A. M. Ulug, P. B. Barker, and P. C. M. van Zijl, Magn. Reson. Med., 1995, 34, 476. 59. R. J. Ordidge, J. A. Helpern, Z. X. Qing, A. Knight, and V. Nagesh, Magn. Reson. Imag., 1994, 12, 455. 60. A. W. Anderson and J. C. Gore, Magn. Reson. Med., 1994, 32, 379. 61. A. J. de Crespigny, M. P. Marks, D. R. Enzmann, and M. E. Moseley, Magn. Reson. Med., 1995, 30, 720. 62. M. P. Marks, A. De Crespigny, D. Lentz, D. Enzmann, G. W. Albers, and M. E. Moseley, Radiology, 1996, 199, 403. 63. Z. H. Endre, B. E. Chapman, and P. W. Kuchel, Biochim. Biophys. Acta, 1984, 804, 137. 64. D. W. Williams, E. F. W. Seymour, and R. M. Cotts, J. Magn. Reson., 1978, 31, 271. 65. J. D. Trudeau, W. T. Dixon, and J. Hawkins, J. Magn. Reson., 1995, B108, 22. 66. R. F. Karlicek and I. J. Lowe, J. Magn. Reson., 1980, 37, 75. 67. W. D. Williams, E. F. W. Seymour, and R. M. Cotts, J. Magn. Reson., 1978, 31, 271. 68. M. I. Hrovat and C. G. Wade, J. Magn. Reson., 1981, 44, 62. 69. R. M. Cotts, M. J. R. Hoch, T. Sun, and J. T. Marker, J. Magn. Reson., 1989, 83, 252. 70. L. L. Latour, L. Li, and C. H. Sotak, J. Magn. Reson., 1993, 101, 72. 71. E. von Meerwall and M. Kamat, J. Magn. Reson., 1989, 83, 308. 72. R. P. Kennan, J. Zhong and J. C. Gore, Magn. Reson. Med., 1994, 31, 9. 73. R. P. Kennan, J. Zhong, and J. C. Gore, in `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', ed. D. Le Bihan, Raven Press, New York, 1995, p. 110. 74. J. S. Murdey and R. M. Cotts, J. Chem. Phys., 1968, 48, 4938. 75. S. Majumdar and J. C. Gore, J. Magn. Reson., 1988, 78, 41. 76. R. M. Weisskoff, C. S. Zuo, J. L. Boxerman, and B. R. Rosen, Magn. Reson. Med., 1994, 31, 601. 77. J. L. Boxerman, L. M. Hamberg, B. R. Rosen, and R. M. Weisskoff, Magn. Reson. Med., 1995, 34, 555. 78. J. Zhong, O. A. C. Petroff, J. W. Prichard, and J. C. Gore, Magn. Reson. Med., 1993, 30, 241. 79. J. Zhong, O. A. C. Petroff, L. A. Pleban, J. C. Gore, and J. W. Prichard, Magn. Reson. Med., 1997, 37, 1. 80. L. L. Latour, Y. Hasegawa, J. E. Formato, M. Fisher, and C. H. Sotak, Magn. Reson. Med., 1994, 32, 189. 81. Y. Hasegawa, L. L. Latour, J. E. Formato, C. H. Sotak, and M. Fisher, J. Cereb. Blood Flow Metab., 1995, 15, 179. 82. E. Busch, M. Hoehn-Berlage, M. Eis, M. L. Gyngell, and K.-A. Hossmann, NMR Biomed., 1995, 8, 59. 83. G. G. Cleveland, D. C. Chang, and C. F. Hazelwood, Biophys. J., 1976, 16, 1043. 84. M. E. Moseley, Y. Cohen, and J. Kucharczyk, Radiology, 1990, 176, 439. 85. P. J. Basser, J. Mattiello, and D. Le Bihan, J. Magn. Reson., 1994, 103, 247. 86. T. E. Contouro, R. C. McKinstry, E. Akbudakr, and B. H. Robinson, Magn. Reson. Med., 1995, 35, 399. 87. V. Gulani, G. A. Iwamoto, H. Jiang, J. S. Shimony, A. G. Webb, and P. C. Lauterbur, Magn. Reson. Med., 1997, 38, 868.

17

88. N. G. Papadakis, D. Xing, C. L. H. Huang, L. D. Hall, and T. A. Carpenter, J. Magn. Reson., 1999, 137, 67. 89. P. J. Basser, J. Mattiello, R. Turner, and D. Le Bihan, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 584. 90. T. L. Davis, V. J. Wedeen, R. M. Weisskoff, and B. R. Rose, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 289. 91. P. T. Callaghan, J. Magn. Reson., 1990, 88, 493. 92. D. G. Cory and A. N. Garroway, Magn. Reson. Med., 1990, 14, 435. 93. M. D. King, J. Houseman, S. A. Roussel, N. Van Bruggen, S. R. Williams, and D. G. Gadian, Magn. Reson. Med., 1994, 32, 707. 94. M. D. King, J. Houseman, D. G. Gadian, and A. Connelly, Magn. Reson. Med., 1997, 38, 930. 95. S. Posse, C. A. Cuenod, and D. Le Bihan, Radiology, 1993, 188, 719. 96. T. L. James and G. G McDonald, J. Magn. Reson., 1973, 11, 58. 97. C. T. W. Moonen, P. C. M. van Zijl, D. Le Bihan, and D. Despres, Magn. Reson. Med., 1990, 13, 467. 98. K. Yoshizaki, Y. Seo, H. Nishikawa, and T. Morimoto, Biophys. J., 1982, 38, 209. 99. P. C. M. van Zijl, C. T. W. Moonen, P. Faustino, J. Pekar, O. Kaplan, and J. S. Cohen, Proc. Natl. Acad. Sci., USA, 1991, 88, 3228. 100. L. E. Kay and J. H. Prestegard, J. Magn. Reson., 1986, 67, 103. 101. C. H. Sotak, J. Magn. Reson., 1990, 90, 198. 102. S. H. Koenig, R. D. Brown, M. Spiller, and N. Lundbom, Magn. Reson. Med., 1990, 14, 482. 103. M. D. Hurlimann, K. G. Helmer, L. L. Latour, and C. H. Sotak, J. Magn. Reson., 1994, 111, 169. 104. P. T. Callaghan, A. Coy, T. P. J. Halpin, D. MacGowan, K. J. Packer, and F. O. Zelaya, J. Chem. Phys., 1992, 97, 651. 105. D. J. Bergman and K. J. Dunn, Phys. Rev., 1995, 51, 3401. 106. R. Mills, J. Phys. C., 1973, 77, 685. 107. D. Le Bihan, J. Delannoy, and R. Levin, Radiology, 1989, 171, 853. 108. J. Delannoy, D. Le Bihan, D. I. Hoult, and R. L. Levin, Med. Phys., 1990, 17, 855. 109. K. Hynynen, A. Darkazanli, and E. Unger, Med. Phys., 1993, 20, 7. 110. H. E. Cline, K. Hynynen, R. D. Watkins, W. J. Adams, J. F. Schenck, R. H. Ettinger, W. R. Freund, J. P. Vetro, and F. A. Jolesz, Magn. Reson. Imag., 1995, 194, 731. 111. J. H. Simpson and H. Y. Carr, Phys. Rev., 1958, 111, 1201. 112. E. O. Stejskal, J. Chem. Phys., 1965, 43, 3597. 113. J. E. Tanner, Biophys. J., 1979, 28, 107. 114. C. H. Neuman, J. Chem. Phys., 1974, 60, 4508. 115. A. V. Barzykin, K. Hayamizu, W. S. Price, and M. Tachiya, J. Magn. Reson., 1995, 114, 39. 116. J. E. Tanner, J. Chem. Phys., 1978, 69, 1748. 117. R. L. Cooper, D. B. Chang, and A. C. Young, Biophys. J., 1974, 14, 161. 118. E. von Meerwall and R. D. Ferguson, J. Chem. Phys., 1981, 74, 6956. 119. C. Nicholson and J. M. Philipps, J. Physiol., 1981, 321, 225. 120. C. Nicholson and E. Sykova, Trends Neurosci., 1998, 21, 470. 121. D. A. Rusakov and D. M. Kullmann, Proc. Natl. Acad. Sci., USA, 1998, 95, 8975. 122. M. F. Lythgoe, A. L. Busza, F. Calamante, C. H. Sotak, M. D. King, A. C. Bingham, S. R. Williams, and D. G. Gadian, Magn. Reson. Med., 1997, 38, 662. 123. M. Hoehn Berlage, M. Eis, and B. Schmitz, NMR Biomed., 1999, 12, 45.

For References see p. 16

18 METHODS AND APPLICATIONS OF DIFFUSION MRI 124. M. E. Moseley, J. Chem. Phys., 1983, 87, 18. 125. P. T. Callaghan and O. Soderman, J. Chem. Phys., 1983, 87, 1737. 126. P. J. Basser, `Imaging Brain Structure and Function', New York Academy of Sciences, New York, 1997, p. 123. 127. P. van Gelderen, M. H. M. De Vleeschouwer, D. Des Pres, J. Pekar, P. C. M. Van Zijl, and C. T. W. Moonen, Magn. Reson. Med., 1994, 31, 154. 128. S. Mori and P. C. M. van Zijl, Magn. Reson. Med., 1995, 33, 41. 129. T. Chun, A. M. Ulug, and P. C. M. van Zijl, Magn. Reson. Med., 1998, 40, 622. 130. E. W. Hsu and S. Mori, Magn. Reson. Med., 1995, 34, 194. 131. P. Douek, R. Turner, J. Pekar, N. J. Patronas, and D. Le Bihan, J. Comput. Assist. Tomogr., 1991, 15, 923. 132. P. J. Basser and C. Pierpaoli, J. Magn. Reson., 1996, B111, 209. 133. C. Pierpaoli and P. J. Basser, Magn. Reson. Med., 1996, 36, 893. 134. P. J. Basser and C. Pierpaoli, Magn. Reson. Med., 1998, 39, 928. 135. T. E. Conturo, R. C. McKinstry, E. Akbudak, and B. H. Robinson, Magn. Reson. Med., 1996, 35, 399. 136. R. I. Shrager and P. J. Basser, Magn. Reson. Med., 1998, 40, 160. 137. J. J. Neil, S. I. Shiran, R. C. McKinstry, et al., Radiology, 1998, 209, 57. 138. R. J. Gilbert, T. G. Reese, S. J. Daftary, R. N. Smith, R. M. Weisskoff, and V. J. Wedeen, Am. J. Physiol., 1998, 38, G363. 139. N. Makris, A. J. Worth, A. G. Sorensen, et al., Ann. Neurol., 1997, 42, 951. 140. D. J. Werring, C. A. Clark, G. J. Barker, D. H. Miller, G. J. M. Parker, M. J. Brammer, E. T. Bullmore, V. P. Giampietro, and A. J. Thompson, J. Neurol. Neurosurg. Psychiatr., 1998, 65, 863. 141. D. J. Werring, C. A. Clark, G. J. M. Parker, D. H. Miller, A. I. Thompson, and G. J. Barker, Neuroimage, 1999, 9, 352. 142. J. W. Meyer, N. Makris, J. F. Bates, V. S. Caviness, and D. N. Kennedy, Neuroimage, 1999, 9, 1. 143. J. Karger and H. Pfeifer, Magn. Reson. Imag., 1994, 12, 235. 144. S. Mori, B. J. Crain, V. P. Chacko, and P. C. M. van Zijl, Ann. Neurol., 1999, 45, 265. 145. C. Poupon, J. F. Mangin, V. Frouin, J. ReÂgis, F. Poupon, M. Pachot-Clouard, D. Le Bihan, and T. Bloch, `Regularization of MR Diffusion Tensor Maps for Tracking Brain White Matter Bundles', Springer-Verlag, 1998, p. 489. 146. E. D. von Meerwall, J. Magn. Reson., 1982, 50, 409. 147. T. Niendorf, R. M. Dijkhuizen, D. G. Norris, M. van Lookeren Campagne, and K. Nicolay, Magn. Reson. Med., 1996, 36, 847. 148. J. G. Li, G. J. Stanisz, and R. M. Henkelman, Magn. Reson. Med., 1998, 40, 79. 149. G. P. Zientara and J. H. Freed, J. Chem. Phys., 1980, 72, 1285. 150. J. Pfeuffer, U. Flogel, and D. Leibfritz, NMR Biomed., 1998, 11, 11. 151. U. Pilatus, H. Shim, D. Artemov, D. Davis, P. C. van Zijl, and J. D. Glickson, Magn. Reson. Med., 1997, 37, 825. 152. R. V. Mulkern, H. Gudbjartsson, C. F. Weston, et al., NMR Biomed., 1999, 12, 51. 153. K. D. Merboldt, D. HoÈrstermann, W. HaÈnicke, H. Bruhn, and J. Frahm, Magn. Reson. Med., 1993, 29, 125. 154. S. Posse, C. A. Cuenod, and D. Le Bihan, in `Diffusion and Perfusion Magnetic Resonance Imaging. Applications to Functional MRI', ed. D. Le Bihan, Raven Press, New York, 1995, p. 57. 155. M. K. Nishijima, R. C. Koehler, and P. D. Hurn, Am. J. Physiol., 1989, 257, H1860. 156. T. Q. Duong, J. J. Ackerman, H. S. Ying, and J. J. Neil, Magn. Reson. Med., 1998, 40, 1. 157. Y. Assaf and Y. Cohen, J. Magn. Reson., 1998, 131, 69. 158. M. Wick, Y. Nagatoma, F. Prielmeier, and J. Frahm, Stroke, 1995, 26, 1930.

For list of General Abbreviations see end-papers

159. A. van der Toorn, R. M. Dijkhuizen, C. A. F. Tulleken, and K. Nicolay, Magn. Reson. Med., 1996, 36, 914. 160. M. A. Rutherford, F. M. Cowan, and A. Y. Manzur, J. Comput. Assist. Tomogr., 1991, 15, 188. 161. P. B. Toft, H. Leth, B. Peitersen, H. C. Lou, and C. Thomsen, J. Comput. Assist. Tomogr., 1996, 20, 1006. 162. C. Baratti, A. S. Barnett, and C. Pierpaoli, Radiology, 1999, 210, 133. 163. I. Vorisek and E. Sykov, J. Neurophysiol., 1997, 78, 912. 164. D. Prayer, T. Roberts, A. J. Barkovich, L. Prayer, J. Kucharczyk, M. Moseley, and A. Arieff, Neuroradiology, 1997, 39, 320. 165. K. Takeda, Y. Nomura, H. Sakuma, T. A. Tagami, Y. Okuda, and T. Nakagawa, J. Comput. Assist. Tomogr., 1997, 21, 1. 166. C. Beaulieu, F. R. Fenrich, and P. S. Allen, Magn. Reson. Imaging., 1998, 16, 1201. 167. D. Le Bihan, R. Turner, and P. Douek, NeuroReport, 1993, 4, 887. 168. C. T. W. Moonen, J. Pekar, M. H. M. De Vleeschouwer, P. Van Gelderen, P. C. M. Van Zijl, and D. Des Pres, Magn. Reson. Med., 1991, 19, 327. 169. J. V. Hajnal, M. Doran, and A. S. Hall, J. Comput. Assist. Tomogr., 1991, 15, 1. 170. M. A. Rutherford, F. M. Conan, and A. Y. Manzur, Radiology, 1991, 180, 229. 171. Y. Assaf and Y. Cohen, J. Magn. Reson., 1996, B112, 151. 172. C. Nicholson and E. Sykova, Trends Neurosci., 1998, 21, 207. 173. C. Beaulieu and P. S. Allen, Magn. Reson. Med., 1994, 32, 579. 174. G. J. Stanisz, A. Szafer, G. A. Wright, and R. M. Henkelman, Magn. Reson. Med., 1997, 37, 103. 175. M. E. Moseley, J. Kuchareczyk, J. Mintorovitch, Y. Cohen, J. Kurhanewicz, N. Derucin, H. Asgari, and D. Normou, Am. J. Neuroradiol., 1990, 11, 423. 176. L. L. Latour, K. Svoboda, P. P. Mitra, and C. H. Sotak, Proc. Natl. Acad. Sci., USA, 1994, 91, 1229. 177. D. G. Norris, T. Niendorf, and D. Leibfritz, NMR Biomed., 1994, 7, 304. 178. J. Pfeuffer, W. Dreher, E. Sykova and D. Leibfritz, Magn. Reson. Imag., 1998, 16, 1023. 179. A. G. Sorensen, F. S. Buonanno, R. G. Gonzalez, et al., Radiology, 1996, 199, 391. 180. S. Warach, D. Chien, W. Li, M. Ronthal, and R. R. Edelman, Neurology, 1992, 42, 1717. 181. S. Warach, M. Boska, and K. M. A. Welch, Stroke, 1997, 28, 481. 182. K. O. LoÈvblad, A. E. Baird, G. Schlaug, et al., Ann. Neurol., 1997, 42, 164. 183. R. G. Gonzalez, P. W. Schaefer, F. S. Buonanno, L. H. Schwamm, R. F. Budzik, G. Rordorf, B. Wang, A. G. Sorensen, and W. J. Koroshetz, Radiology, 1999, 210, 155. 184. S. Warach, M. Boska, and K. M. Welch, Stroke, 1997, 28, 481. 185. W. Dreher, B. KoÈhn, M. L. Gyngell, E. Busch, T. Niendorf, K. A. Hossmann, and D. Leibfritz, Magn. Reson. Med., 1998, 39, 878. 186. J. Ono, K. Harada, T. Mano, K. Sakurai, and S. Okada, Pediatr. Neurol., 1997, 16, 63. 187. T. Iwasawa, H. Matoba, A. Ogi, H. Kurihara, K. Saito, T. Yoshida, S. Matsubara, and A. Nozaki, Magn. Reson. Med., 1997, 38, 484. 188. M. A. Hors®eld, H. B. Larsson, D. K. Jones, and A. Gass, J. Neurol. Neurosurg. Psychiatry, 1998, 64(Suppl 1), S80. 189. H. Ay, F. S. Buonanno, P. W. Schaefer, D. A. Le, B. Wang, R. G. Gonzalez, and W. J. Koroshetz, Neurology, 1998, 51, 1369. 190. H. Hanyu, H. Shindo, D. Kakizaki, K. Abe, T. Iwamoto, and M. Takasaki, Gerontology, 1997, 43, 343.

METHODS AND APPLICATIONS OF DIFFUSION MRI 191. H. Hanyu, H. Sakurai, T. Iwamoto, M. Takasaki, H. Shindo, and K. Abe, J. Neurol. Sci., 1998, 156, 195. 192. M. M. Bahn, D. K. Kido, W. L. Lin, and A. L. Pearlman, Arch. Neurol., 1997, 54, 1411. 193. S. Kropp, M. Finkenstaedt, C. Laske, W. SchulzSchaeffer, I. Zerr, A. Kretschmar, and S. Poser, J. Neurol., 1999, 246, 66. 194. P. Demaerel, L. Heiner, W. Robberecht, R. Sciot, and G. Wilms, Neurology, 1999, 52, 205. 195. T. Q. Li, Z. G. Chen, and T. Hindmarsh, Acta Radiol., 1998, 39, 460. 196. R. A. Zimmerman, J. C. Haselgrove, Z. Y. Wang, J. V. Hunter, M. C. Morriss, A. Hoydu, and L. T. Bilaniuk, Brain Dev., 1998, 20, 275. 197. P. S. Huppi, S. E. Maier, S. Peled, G. P. Zientara, P. D. Barnes, F. A. Jolesz, and J. J. Volpe, Pediatr. Res., 1998, 44, 585. 198. M. S. Buchsbaum, C. Y. Tang, S. Peled, et al., Neuroreport, 1998, 9, 425. 199. D. Le Bihan, P. Douek, M. Argyropoulou, R, Turner, N. Patronas, and M. Fulham, Top. Magn. Reson. Imag., 1993, 5, 25. 200. K. Ikezaki, M. Takahashi, H. Koga, J. Kawai, Z. KovaÈcs, T. Inamura, and M. Fukui, Acta Neurochir. Suppl. (Wien), 1997, 70, 170. 201. K. Krabbe, P. Gideon, P. Wagn, U. Hansen, C. Thomsen, and F. Madsen, Neuroradiology, 1997, 39, 483. 202. P. Barzo, A. Marmarou, P. Fatouros, K. Hayasaki, and F. Corwin, J. Neurosurg., 1997, 87, 900. 203. R. B. Schwartz, R. V. Mulkern, H. Gudbjartsson, and F. Jolesz, Am. J. Neuroradiol., 1998, 19, 859. 204. L. Chang and T. Ernst, Neuroimaging Clin. North Am., 1997, 7, 409. 205. P. W. Schaefer, F. S. Buonanno, R. G. Gonzalez, and L. H. Schwamm, Stroke, 1997, 28, 1082. 206. K. Okada, L. H. Wu, and S. Kobayashi, Stroke, 1999, 30, 478. 207. D. K. Jones, D. Lythgoe, M. A. Hors®eld, A. Simmons, S. C. R. Williams, and H. S. Markus, Stroke, 1999, 30, 393. 208. J. C. Ford, D. B. Hackney, E. Lavi, M. Phillips, and U. Patel, JMRI, 1998, 8, 775.

19

209. B. A. Inglis, L. Yang, E. D. Wirth, D. Plant, and T. H. Mareci, Magn. Reson. Imag., 1997, 15, 441. 210. D. Morvan, Magn. Reson. Imag., 1995, 13, 193. 211. A. Baur, A. Stabler, R. Bruning, R. Bartl, A. Krodel, M. Reiser, and M. Deimling, Radiology, 1998, 207, 349. 212. G. Akansel, V. M. Haughton, R. A. Papke, S. Censky, Am. J. Neuroradiol., 1997, 318, 443. 213. S. A. Englander, A. M. Ulug, R. Brem, J. D. Glickson, and P. C. M. van Zijl, NMR Biomed., 1997, 10, 348. 214. C. F. Maier, Y. Paran, P. Bendel, B. K. Rutt, and H. Degani, Magn. Reson. Med., 1997, 37, 576. 215. T. Namimoto, Y. Yamashita, S. Sumi, Y. Tang, and M. Takahashi, Radiology, 1997, 204, 739. 216. T. Ichikawa, H. Haradome, J. Hachiya, T. Nitatori, and T. Araki, Am. J. Roentgenol., 1998, 170, 397. 217. Y. Yamashita, Y. Tang, and M. Takahashi, JMRI, 1998, 8, 367. 218. I. Yamada, W. Aung, Y. Himeno, T. Nakagawa, and H. Shibuya, Radiology, 1999, 210, 617. 219. R. J. Gilbert, T. G. Reese, S. J. Daftary, R. N. Smith, R. M. Weisskoff, and V. J. Wedeen, Am. J. Physiol., 1998, 275, G363. 220. D. F. Scollan, A. Holmes, R. Winslow, and J. Forder, Am. J. Physiol., 1998, 44, H2308. 221. E. W. Hus, A. L. Muzikant, S. A. Matulevicius, R. C. Penland, and C. S. Henriquez, Am. J. Physiol., 1998, 274, H1627. 222. Y. Zhang, T. V. Samulski, W. T. Joines, J. Mattiello, R. L. Lebin, and D. Le Bihan, Int. J. Hyperthermia, 1992, 8, 263.

Biographical Sketch Denis Le Bihan. b 1957. Ph.D., 1987, Physics, M.D., 1984, Radiology, University of Paris, France. Former Chief of the Radiology Research Section, Clinical Center, NIH, Bethesda, USA. Presently Research Director, French Atomic Energy Commission, Director, Anatomical and Functional Neuroimaging Laboratory, Orsay, France, and Clinical Professor of Radiology, Harvard University, Cambridge, USA. Corresponding Member, French Academy of Sciences. Approx. 400 articles, book chapters, abstracts and patents. Research interests: diffusion, perfusion and functional MRI.

For References see p. 16

BRAIN MRS OF HUMAN SUBJECTS

Brain MRS of Human Subjects James W. Prichard Yale University, New Haven, CT, USA

1 INTRODUCTION A biomedical revolution based on NMR technology is under way. Many of the NMR methods described in this Encyclopedia are already in use for medical research and diagnosis, and many others will ®nd such application soon. As the full impact of mature NMR methods becomes evident in the latter 1990s, it will be measured on the same scale with such things as microscopy and genetics. Two well-established characteristics of biomedical NMR technology ensure this outcome: it is remarkably noninvasive (see Health and Safety Aspects of Human MR Studies) and it is remarkably versatile, as attested in many other chapters. In each respect separately, it exceeds most other technologies available for the study of living tissue. By their combination, it stands alone. This chapter deals with human brain research done by the branch of biomedical NMR technology designated magnetic resonance spectroscopy (MRS) to distinguish it from magnetic resonance imaging (MRI), which makes pictures of anatomical structure from the water proton signal. MRS obtains chemically speci®c information from biological tissue by the measurement of much smaller signals from 1H, 31P, 13C, 17O, and other magnetic nuclei in a variety of compounds. It is extending the range of biomedically useful NMR measurements to practical applications that would have courted dismissal as visionary a decade ago. In 1994, signals from more than two dozen compounds can be detected routinely in the human brain, and many more are in prospect. All report on biochemical phenomena previously accessible only by surgical removal of tissue or at autopsy. As various forms of MRS supported by MRI reach technical maturity, their collective contributions to an understanding of normal and deranged biochemistry in the brain will amount to nothing less than a new neurobiological discipline comparable in scope to today's neurochemistry, neuroanatomy, and neuropathology combined. All stages of normal human brain development and disease processesÐincluding intrauterine stagesÐwill be open to inspection along the many axes MRS can measure. The MRS studies mentioned in this chapter are intended to help scientists and physicians interested in the brain to evaluate the above assertions, which may appear extravagant to readers who are exploring modern NMR for the ®rst time. Many more studies appropriate for that purpose have been published than can be cited. Selection among them was intended to favor efforts that illuminate each other or set new problems by failing to do so, but it cannot have escaped all of the author's biases.

2

1

H AND

1

31

P SPECTRA

Most human brain MRS research to date has used these two nuclei. Figures 1 and 2 illustrate the normal features and some disease-related abnormalities of brain 1H and 31P spectra with simulations that are free of the wide technical variation among spectra in original publications. Differences in acquisition parameters and data processing can cause spectra that are biologically equivalent to look quite different from each other. For readers new to either NMR or neuroscience, the variation can easily distract attention from features of biological importance. The simulated spectra in Figures 1 and 2 were generated by Mathematica# and annotated in Adobe Illustrator#. Within each ®gure, random noise is of the same intensity in all spectra. Resonance variables were adjusted to mimic the appearance typical of many published spectra, in which less intense resonances are not evident because of such factors as long spin echo times, J coupling, and processing methods. The resonances shown are usually identi®able in all spectra from normal adult brain. The disease-related variations are large ones that have been reported by more than one group. They are presented together to emphasize their potential for metabolic ®ngerprinting of different disease processes.

3

NORMAL BRAIN FUNCTION: LACTATE IN HUMAN SENSORY SYSTEMS

A signal from lactate is detectable in 1H spectra from the normal human brain.1 Stimulation of the human visual2±4 and auditory5 systems can cause increases in lactate observable by 1 H MRS in the primary cortical receiving areas. These ®ndings are consistent with positron emission tomography (PET) data showing that similar stimulation can increase glucose uptake more than oxygen extraction in the regions affected.6 The PET workers interpreted their observations as evidence for preferential activation of nonoxidative glycolysis. The MRS data strengthen that view. The notable aspect of this research area is the evidence from at least ®ve different PET and NMR groups showing that glycolysis can increase more than respiration in response to some kinds of activation within the normal range of brain function. The phenomenon is well known in skeletal muscle. Selective activation of muscle cells with differing capacities for glycolysis and respiration is a normal feature of muscle function, and heterogeneous distribution of glycolytic and respiratory enzymes has long been employed for histochemical characterization of individual muscle cells. Evidence for similarly heterogeneous enzyme distribution already exists at the level of cell groups in the animal brain.7 Like muscle cells, neurons and possibly glia may well have variable capacity for glycolysis and respiration related to their physiological functions. These PET and MRS data prompt the question of what adaptive utility lactate elevation resulting from brain activity might have. Two novel possibilities that can be tested by experiment are the following: (1) Lactate may be a local energy storage medium that is increased in anticipation of sudden high energy demand. Ammunition stacked beside cannons defending a fort is more useful than ammunition in a remote storehouse if the fort is

2 BRAIN MRS OF HUMAN SUBJECTS

N-Acetyls

NORMAL Cholines

Creatines

Simulations of 1H SPECTRA from the brain

Lactate

4

3

2

1

ppm PATHOLOGICAL – filled resonances are abnormal Stroke

Glioma

Epilepsy – during seizure

Epilepsy – mesial temporal sclerosis

potentials a second and surrounding cells to recycle transmitter at proportional rates. The immediate metabolic burden of this intense activity falls principally on synaptic terminals, which are quite small and in consequence have low ratios of cytoplasmic volume to the area of excitable surface membrane that they must support. In such a structure, the Krebs cycle can respond to sudden maximum energy demand more quickly if it is not limited by the time necessary for activation of glycolysis, which is long on the scale of intense neuronal discharge and transmitter release. Local lactate elevation by previous activity would prime synaptic terminals to deliver their maximum response to sudden demand while glycolytic rate is increasing. Existing techniques for the measurement of glucose and oxygen uptake and lactate concentration would not detect most instances of this process, because the ones that can be used in vivo, including MRS, report on overall changes in volumes of tissue much larger than the scattered locations at which such priming would ordinarily occur, and the others are bedeviled by the possibility of agonal artifact. Only when it occurs in many cells at the same timeÐas during sensory barrages and seizure dischargeÐcould such a process be detected by present methods.

PCr

Alzheimer disease

Multiple sclerosis plaque

PDE Simulations of 31P SPECTRA from the brain

PME

10

Figure 1 Simulated 1H spectra from brain, created in Mathematica# and annotated in Adobe Illustrator#. Alterations typically caused by six pathological conditions are contrasted with each other and to normal. Spectral variables were adjusted to mimic the appearance common in published spectra. Random noise and each un®lled resonance are of the same intensity in all spectra. The `normal' spectrum represents resonances from choline-containing compounds (cholines), creatine and phosphocreatine (creatines), N-acetylcontaining compounds, mostly N-acetylaspartate (N-acetyls), and lactate methyl protons (lactate). Glutamate and glutamine signals are represented as unresolved low-intensity resonances between 2 and 3 ppm. The `pathological' spectra were created by varying the intensities of signals reported to be affected by the disease states indicated by labels on each spectrum. The abnormalities shown are the ones most often associated with each condition, but they are not all present in every case. The elevated lactate signal in four spectra displays the characteristic 7 Hz splitting caused by J coupling to the C-2 proton

suddenly attacked. Suspicion of imminent attack would certainly cause the fort's defenders to move extra ammunition from the storehouse to the cannons, to shorten the response time and lengthen the duration of maximum ®ring. In the brain, neurons can suddenly be called upon to ®re hundreds of action

Pi

Normal

ATP Alpha Gamma

0

ppm

–10

Beta

–20

PATHOLOGICAL – filled resonances are abnormal

Stroke – moderate

Stroke – severe

Epilepsy – during seizure Glioma

Figure 2 Simulated 31P spectra from normal and diseased brain, created and presented by the same procedures used for Figure 1. The resonances represented are from phosphomonoesters (PME), inorganic phosphate (Pi), phosphodiesters (PDE), phosphocreatine (PCr) and the , , and phosphates of adenosine triphosphate (ATP). In the `stroke - severe' example, all of the signal present is Pi, indicating total loss of high-energy phosphate compounds

BRAIN MRS OF HUMAN SUBJECTS

(2) Lactate may have a signaling function in the brain. Lactate is well suited to be a neuromodulatorÐa substance which alters the excitability of local neural ensembles. It is produced at increased rates whenever glycolysis accelerates in response to neural activity. It accumulates during pathological activation caused by ischemia or lack of oxygen, and, as described above, during some kinds of activation in the physiological range. IfÐupon escape from the active cell that produced itÐit reduced the excitability of adjacent cells, it could both inhibit seizure discharge and `sharpen' transmission along multi®ber pathways, a process known to occur by other mechanisms. A ®nding consistent with this idea is available: in Torpedo electroplax, lactate in concentrations that can be produced by brain activation inhibited release of acetylcholine.8

4 STROKE Stroke is the commonest major brain disease and a principal medical burden on society through the lost productivity it causes and its consumption of resources by extended disability. It is a natural place to test the bene®ts of new biomedical technology. MRS observations on human stroke are far enough advanced to allow initial assessment of the role MRS will come to have in patient management. Lactate and N-acetyl (NA) resonances in the 1H spectrum and intracellular pH (pHi) derived from the 31P spectrum appear to have the best prospects for clinical utility.

4.1

1

H MRS in Stroke

Figure 1 illustrates the kind of change commonly observed in 1H spectra of human brain infarction: loss of the NA signal and elevation of the signal from lactate. Loss of NA signal should ensue from death of neurons, which are the only cells in mature brain thought to contain N-acetylaspartate (NAA), the principal source of the NA resonance. The NA signal was depressed both in the ®rst few days after stroke onset and several weeks later in six patients studied at least twice.9 In 10 patients studied within 60 h of stroke onset, it was reduced relative to the homologous region of the contralateral hemisphere in all but two; in seven of the same patients, repeat examination 1±2 weeks later showed additional decline at an overall rate calculated at ÿ29% ‹ 9% per week.10 Similar ®ndings have been reported by other groups.11±13 Decline of the NA signal in the ®rst days after a stroke surely re¯ects, in part, clearing of debris from neurons killed at the outset, but it may also be evidence of continuing neuronal loss after the acute period, possibly in the ischemic penumbra.14 Therapy which prevents delayed neuronal loss would presumably also reduce eventual ®xed de®cit. Monitoring of the NA signal may prove useful as a surrogate endpoint for evaluation of therapies intended to reduce continuing neuronal loss in the subacute period. Lactate elevation by acute infarction is readily detectable by 1H MRS (Figure 1). However, the ®rst report of increased lactate associated with stroke14 was on patients studied months after the stroke occurred. Many observers including

3

the present writer doubted that stroke-associated lactate elevation would persist so long, but subsequent observations by several groups have shown that it does. Serial study of individual patients demonstrated continuous lactate elevation for weeks after stroke in most and for several months in some.9 Again, similar observations have been made by other groups.11±13 Multiple mechanisms of lactate elevation must be at work over so long a period. A newly infarcted region of brain accumulates lactate to a concentration of 15±30 mM within minutes of losing its blood supply, as glucose and glycogen in unperfused tissue are metabolized in the absence of oxygen, which is exhausted in the ®rst few seconds. If the region is never reperfused, the lactate in it can dissipate only by diffusion, but the process would not take weeks to months in the presence of active tissue repair processes. Other possible sources of elevated lactate associated with infarction include adjacent regions of surviving but impaired tissue referred to as an `ischemic penumbra',15 in®ltrating cells involved in the tissue's reaction to injury, and altered metabolism of surviving brain. MRS has provided evidence for the second of these:16 the brain of a patient studied two weeks after a stroke and autopsied a week later had large numbers of macrophages in the regions where 1H MRS had shown elevated lactate. The role of altered metabolism is conjectural at present; the longterm metabolic response of brain tissue that survives ischemic insult is not well understood. Lactate that accumulates in the killed core of an infarct is metabolically inert, while that in an ischemic penumbra turns over, albeit probably at a rate different from that of normal brain tissue. The anatomical boundary between pools of lactate in these two states need not be sharp; the pools may even be anatomically interdigitated. Their relative sizes re¯ect the proportion of killed to surviving tissue early in the history of the lesion, and may therefore indicate how a particular lesion is likely to evolve. An advanced form of combined 1H/13C MRS that has been shown to be feasible in humans can make a pertinent measurement. Inert and metabolically active lactate associated with a human stroke can be distinguished from each other by determining how much of the total stroke-associated lactate pool can be labeled with 13C from blood glucose. Explanation of this remarkable possibility requires a brief description of the NMR properties of 13C.

4.2

13

C MRS

Carbon in Nature is nearly 99% 12C, which is not magnetic and hence gives no NMR signal. The stable magnetic isotope 13 C is 1.1% naturally abundant. Its potential for biological studies was appreciated early in the second decade of empirical NMR work. In 1958, the following passage was written by P. C. Lauterbur:17 `Most practical applications of 13C spectra must await some improvement in the signal-to-noise ratio. One application that might be made immediately is in the use of 13C as an isotopic tracer in reactions. Various chemical forms of carbon could be identi®ed by their chemical shifts and ®ne structures, even in complex mixtures, after the introduction of a compound enriched in 13C. Even some biological systems might be studied by this technique.' (Emphases added.)

4 BRAIN MRS OF HUMAN SUBJECTS These words are the conceptual origin of biomedical 13C MRS. In 1957, their author had published the ®rst report of 13 C chemical shifts,18 which was followed shortly by another.19 Years became decades while NMR technology evolved to a stage at which Lauterbur's idea could be explored by 13C labeling studies, ®rst on enzyme systems and functioning cells,20 later on intact animal liver,21 animal brain,22 and human brain.23,24 The potential adumbrated by these studies is so great that even the most advanced of them belong to what later generations of biomedical scientists will regard as the early history of 13C MRS. Lauterbur was remarkably prescient. In living subjects including humans, natural abundance 13C MRS has much still unexploited potential for characterization of normal and pathological variation among tissues, but the most inviting opportunity that 13C MRS presents to biomedical workers is 13C enrichment of molecules observable in vivo. As demonstrated more than two decades ago in Candida utilis,25 feeding a 13C-enriched nutrient to a living organism provides the simultaneous advantages of increased signal-to-noise ratio for observation of molecules that receive the 13C through metabolic processes and a nondestructive means of measuring the rates of those processes. The adult brain normally derives nearly all of its energy from glucose, most of which it converts to two molecules of lactate; if the glucose was enriched in 13C at the C-1 position, one of the lactates receives 13C glucose in its methyl position. Further metabolism in the Krebs cycle creates several other 13C-labeled molecules observable in vivo, the most prominent of which is 4-13C-glutamate. After animal experiments demonstrated that 1-13C-glucose, 3-13C-lactate, and 13 C-labeled amino acids could be detected in living brain by 13 C MRS,22 similar observations were made in the human.24 4.3 Detection of

13

C by 1H MRS

Direct 13C MRS has the disadvantage for in vivo work of long acquisition times due to the low sensitivity of 13C compared to 1H. In samples containing enough 1H±13C bonds, the presence of the 13C can be detected in 1H spectra by the characteristic way in which it splits the signal from protons bonded to it. Exploitation of this phenomenon confers proton sensitivity on NMR measurement of 13C in some positions of intact, functioning molecules. The principle was used in studies of organic molecular structure as early as the 1960s.20 Later, technological advances and diligent effort allowed its successful adaptation to observation of 13C-labeled compounds in the brains of living animals26 and humans.23 By the use of appropriate models, cerebral metabolic rates can be calculated from the time course of 13C accumulation in observable metabolic pools.27 These and related NMR methods are still in the early stages of development; in their fully mature forms, they are likely to be the preeminent means of measuring metabolic rates in the living human brain. 4.4 Labeling of Stroke-Elevated Lactate with

13

C

In stroke, 13C labeling can make the important distinction between lactate which is trapped in a nonmetabolizing compartment and lactate that is persistently elevated in the presence of competent metabolic machinery. This strategy was used to show that shock-elevated lactate in rabbit brain28 is all metabo-

lically active,29 and it is equally applicable to assessment of stroke-elevated lactate in the human, as has been demonstrated.30 However, the prospect for the routine use of this very sophisticated technique in emergency stroke evaluation depends on other things. Carbon-13 labeling of the kind described can certainly provide information about how much of a fresh stroke is an ischemic penumbra,14 but rapid evolution of MRI may provide equivalent information from simpler measurements. The best candidates are diffusion weighted imaging (DWI)31 (see Diffusion: Clinical Utility of MRI Studies) and magnetization transfer contrast imaging (see Magnetization Transfer Contrast: Clinical Applications). The point here is that if 13C labeling of stroke-elevated lactate proves to be a suf®ciently novel predictor of a fresh infarct's later course to justify its relative complexity, its minor risk, and a few hundred dollars for each patient, it can be widely implemented. 4.5

31

P MRS in Stroke

Figure 2 illustrates two degrees of loss of energy stores and elevation of inorganic phosphate that can be caused by stroke. Intracellular acidosis is measurable from the resonant frequency of the inorganic phosphate signal, but the frequency difference is too small to be evident in the ®gure. The ®rst 31P MRS study of human stroke found normal metabolite ratios with reduced total phosphorus signal in chronic stroke, consistent with the replacement of infarcted tissue by cerebrospinal and interstitial ¯uid.32 After developing the capability to study acutely ill patients in a magnetÐa decidedly nontrivial acheivementÐa group at Henry Ford Hospital in Detroit used it to monitor 31P changes from the acute to the subacute period.33 Phosphocreatine and ATP were reduced in the acute period, and inorganic phosphate was elevated, as shown in Figure 2. Intracellular pH was acidotic. All of these changes are consistent with the traditional understanding of stroke pathophysiology, but they did not correlate well with clinical measures. Alkalosis replaced acidosis within a few days. The authors suggested that effective therapy might have to be instituted during the period of acidosis. An estimate of tissue Mg2+ concentration can be made from information in 31P spectra.34 In stroke, Mg2+ was elevated in the acidotic period; it might be a pathophysiological factor or a marker of cellular injury.35 1 H and 31P spectra can both be obtained in the same session. Combined 1H/31P observations are a powerful way of analyzing relationships among a wide range of brain metabolites, as was ®rst demonstrated in animal work on hypoglycemia.36 Despite rather considerable technical obstacles to doing this in human stroke patients, two groups have accomplished it,37,38 and the results are illuminating. Both groups found that stroke-associated lactate and pH were usually not inversely correlated. In combined 1H/31P observations ranging from the acute to the chronic period, the more common association was of elevated lactate with alkalosis, rather than acidosis or normal pH. Because 31P and 1H spectra come from tissue volumes that are not the same size and are both large compared with the dimensions of any of the several metabolic compartments they contain, MRS-observable metabolites should not be expected to behave as though they were in a well-stirred test tube. Dissociation of lactate from pH has been observed in ex-

BRAIN MRS OF HUMAN SUBJECTS

perimental status epilepticus by NMR39 and, by biochemical techniques, in tumors and one day after global ischemia.40

5 EPILEPSY 5.1

31

P MRS in Epilepsy

Several animal studies on seizure phenomena in the early years of in vivo MRS demonstrated that the PCr/Pi (Figure 2) decline well established by traditional biochemical research could be observed in the the living brain.41 Due to the limited bore size of available spectrometers, the ®rst MRS observations on the epileptic human brain were 31P spectra from infants.42 Findings during seizures were as expected: the PCr/Pi ratio was decreased about 50%, as shown in Figure 2, and it returned to normal after seizure discharge ceased. Infants who had the lowest ratios during seizures developed long-term neurological sequelae. The development of spectrometers with bores large enough for adults was quickly followed by studies of chronic temporal lobe epilepsy, which is a major problem in modern epilepsy management. Two groups have published data on enough patients to allow comparison of results, which are somewhat different. One found alkaline pHi associated with the seizure focus in eight patients, seven of which also had increased Pi and decreased PME.43,44 The other group reported low PCr/Pi ratios without other changes.45 Further work will be necessary to determine whether the discrepancy re¯ects technical factors of differences between patient populations. 5.2

1

H MRS in Epilepsy

The principal component of the NA signal in 1H spectra of normal brain is from NAA, which occurs mainly if not exclusively in neurons. Reduction of the NA signal implying loss of neurons is a common feature of chronically epileptogenic brain tissue, having been documented in two papers,46,47 and in six preliminary reports by these groups and others at the Annual Meeting of The Society of Magnetic Resonance in Medicine in 1993. The phenomenon is illustrated in Figure 1. These data have the important implication that chemically speci®c 1H MRS abnormalities may be detectable in vivo before structural changes in patients with chronic temporal lobe epilepsy, which is commonly associated with a neuropathological state known as mesial temporal sclerosis and can often be relieved by surgery. MRI techniques that appear to be especially sensitive to this pathology have been reported,48,49 but even if they become routine, noninvasive preoperative detection of chemical abnormality is likely to improve accurate selection of the tissue to be resected for relief of complex partial epilepsy. MRS will be especially important for that purpose if the NA signal proves to be the ®rst NMR quantity to change as mesial temporal sclerosis develops. Noninvasive electroencephalographic and NMR techniques together may soon eliminate the need for implantation of intracranial electrodes to determine which patients with this kind of intractable epilepsy can bene®t from surgery. Observation by 1H MRS of two patients with a form of chronic localized epileptogenic encephalitis known as Rasmus-

5

sen's syndrome produced the ®rst report of elevated lactate associated with seizure discharge in human brain.46 Lactate elevation caused by seizure discharge is illustrated in Figure 1. Monitoring of signals from -aminobutyric acid and glutamine in the human brain by newly developed 1H MRS methods50 has allowed direct observation of the effects of vigabatrin, a new antiepileptic drug.51 Such observation of drug effects directly in the living target organ opens a new era in neuropharmacology.

6

DEMENTIA

Primary dementiaÐdecline of mental function not secondary to tumors, trauma, drugs, or other obvious causesÐis about one-half dementia of the Alzheimer type (hereinafter `Alzheimer disease'). Vascular, infectious, and other dementias are much less common. Alzheimer disease is a condition of unknown etiology that is de®ned by its characteristic neuropathology, although the diagnosis can usually be made correctly in life from its distinctive constellation of clinical and laboratory abnormalities. The protracted disability that it causes places a large burden on society, motivating intense research effort which now includes MRS. 6.1

31

P Studies of Alzheimer Disease

Published 31P data on Alzheimer patients do not agree with each other. Con¯icting claims about what changes, if any, are present have persisted for several years. A series of studies recently summarized52 reported that PME and Pi were above normal in the brains of Alzheimer patients, while another group found no clear 31P changes associated with the disease.53,54 No obvious difference in patient selection or characterization accounts for the difference. Technical factors might; the two studies used somewhat different 31P acquisition methods. 6.2

1

H Studies of Alzheimer Disease

In contrast to the 31P ®ndings, 1H MRS done by several groups has produced general agreement that a reduced NA signal is characteristic of Alzheimer disease (see Figure 1). Preliminary reports of nine studies of living Alzheimer patients all describe decreased NA signals (Proc. 12th Ann Mtg. (Int) Soc. Magn. Reson. Med., New York, 1993). These very consistent data urge the conclusion that NA is characteristically reduced in Alzheimer disease, apparently in proportion to the severity of neuron loss.

7

BRAIN TUMORS

The cellular heterogeneity of brain tumors is a substantial impediment to progress in understanding neoplasia in the nervous system by in vivo techniques, as PET workers have known for years. The heterogeneity is far below the anatomical resolution of current MRS techniques, and technical improvements that appear feasible offer no hope that the gap can be closed. Its effects were evident in an early 31P MRS study of

6 BRAIN MRS OF HUMAN SUBJECTS brain tumors, which commented on the `striking diversity' of the metabolic patterns observed.55 Later studies associated decreased PCr, alkaline pH, increased PME, and altered PDE signals with heightened aggressivity of gliomas.56,57 One constellation of 31P changes that can occur in gliomas is illustrated in Figure 2. More detailed observation of metabolic properties of brain tumors is possible by 1H MRS due to its ®ner anatomical resolution. In the form of chemical shift imaging (CSI), it allows variations within single lesions to be detected. A study that combined 1H CSI with PET found that regions of high lactate tended to coincide with regions of high glucose uptake.58 Later work with improved techniques by the same group revealed a more complicated situation: high lactate was also associated with loculations of extracellular ¯uid.59 Another group able to study patients with both 1H CSI and PET reported similar variability and metabolic ®ndings.60,61 Variable reductions in NA and increased choline signals were observed by both groups. These changes and the elevated lactate seen in some brain tumors are illustrated in Figure 1. Metabolic maps made from 31P have lower anatomical resolution than 1H maps, due to the lower sensitivity of 31P, but they are capable of showing metabolite distributions across major brain structures and defects caused by large lesions (also see the related Chemical Shift Imaging).62,63 The information about energy state, pH, and phospholipids available in 31P spectra is so valuable for the understanding of many disease states that continued vigorous efforts to re®ne 31P CSI are certain.

8 MULTIPLE SCLEROSIS Demonstration of the extensive, clinically silent pathology of cerebral white matter in multiple sclerosis was among the ®rst major new ®ndings of MRI in the nervous system. MRS of the disease was not practical until several years later, for the usual reason that MRS signals are of much lower intensity. By 1993, NMR technology had advanced to the point that nine preliminary reports on 1H MRS studies of multiple sclerosis appeared in that year's Proc. 12th Ann Mtg. (Int) Soc. Magn. Reson. Med. Reduced NA and increased choline signals associated with plaques were the most common ®ndings. Elevated lactate was also observed; it may re¯ect the presence of highly glycolytic white cells, as in subacute cerebral infarction.16 All three changes are illustrated in Figure 1. These metabolic abnormalities were followed over a period of months in a single multiple sclerosis patient with an unusually large cerebral plaque.64 Metabolic aspects of plaque evolution have not previously been accessible for study in human patients. The combination of MRI and chemically speci®c MRS will produce new understanding of the pathophysiology of multiple sclerosis, as they are used together to obtain new information on the natural history and therapeutic responsiveness of the disease in individual patients. All of the data reported by Arnold and colleagues64 are new information bearing on the underlying cellular and molecular biology of episodic demyelination. The opening of so wide a window on a pathophysiological process is certain to improve understanding of it.

9

THE FUTURE OF HUMAN BRAIN MRS

The chemical speci®city of MRS guarantees a major role for it in neurobiology. That property, together with the noninvasiveness that it shares with all NMR methods, offers scope for the investigation of the normal human brain that has no close precedent in the history of any earlier technology. Detailed biochemical characterization of the living human brain at all its stages of development and decline is coming within the reach of noninvasive, chemically speci®c MRS. Investigation of how the human brain works when it is normal and when it is diseased will move more rapidly, and in new directions, with abundant bene®t to both science and medicine. 9.1

Science

Neurochemistry is a dif®cult discipline, because nervous tissue is well protected and highly intolerant of the kind of disruption required for study by standard chemical techniques. MRS can reduce that barrier considerably by providing abundant neurochemical data from the living organ, free of agonal artifact and remeasurable as often as necessary in the same individual. The MRS studies of normal function and metabolic rates mentioned above are early examples of work that will grow into a new dimension of neurochemistry touching nearly every aspect of human brain biology. While MRS can measure only a small fraction of the compounds present in living brain, information about that fraction is unique because previously it was not available at all. Neurobiologists can now look through the window of several dozen MRS-measurable compounds at the biochemical milieu of which they are part. As the number of compounds observable in vivo grows, MRS will become an increasingly powerful complement to cellular and molecular methods in neurobiological research. 9.2

Medicine

In the late 1990s, no routine clinical application of MRS is yet standard practice of the kind that every hospital must provide, like X-ray equipment and electrocardiographs, but MRI has not reached that point either. Both will. Diagnostic MRI is so much more versatile and ef®cient than earlier technologies that its emergence as the premier medical imaging method of the latter 1990s is certain. Implementation of MRS (small signal capability) on standard clinical MRI machines is no longer a large step in either technique or money. The widespread availability of MRI machines needed for ef®cient medical diagnosis will facilitate the introduction of MRS into routine clinical practice as rapidly as MRS research demonstrates useful applications. The most important prospect that MRS offers clinical medicine is chemically speci®c characterization of disease processes at all of their stages. Noninvasive longitudinal MRS data that provide new understanding of how pathophysiological processes evolve are unique. They will affect medicine no less than the data from microscopic and chemical study of removed brain tissue that are much of the basis for modern conceptions of disease. MRS-de®ned biochemical pro®les that distinguish ischemic, neoplastic, in¯ammatory, degenerative, and other pathophysiological categories from each other in vivo will emerge, as will pro®les that identify speci®c diseases. Many

BRAIN MRS OF HUMAN SUBJECTS

MRS measurements will be useful in monitoring the effects of therapy. As this body of knowledge grows, the use of speci®c MRS measurements in the management of individual patients will become routine. 10

RECENT PROGRESS

This article was ®rst written in 1994. In the four intervening years, MRS of the human brain has advanced rapidly, as have nearly all biomedical applications of NMR technology. Most of the advances in MRS have been along paths predictable from the work described in the article, which continues to provide useful orientation to the origins of a ®eld that is both revolutionary and still young. However, in several areas recent progress is either novel or extensive to a degree that the article does not adequately indicate. The following recent citations, which are mostly reviews, will help the interested reader ®nd relevant literature. 10.1 Spectroscopic Imaging This procedure, also known as chemical shift imaging, allows mapping of metabolites in two dimensions. The major technical problems that it presents have been the object of intense development efforts in recent years,65,66 and it has been used by several groups in clinical research studies on brain tumors,67 multiple sclerosis,68 and various aspects of brain metabolism.69,70 10.2 High-®eld Magnets Spectrometers suitable for human studies at ®elds as high as 4.1 T have been used in research for several years, and instruments with ®elds up to 8 T are under development. Notable examples of the neurobiologic opportunities opened by work at 4 T include improved observation of glucose71 and amino acid69 resonances in the human brain. 10.3 Measurement of Brain pH in 1H Spectra Human 1H MRS at 2.1 T has shown that a titrating signal from homocarnosine can provide a measure of cytosolic pH in neurons with high concentrations of that compound, probably a subset specialized for synaptic release of -aminobutyric acid.72 A general review of brain pH measurements by MRS has appeared.73 11

RELATED ARTICLES

Anisotropically Restricted Diffusion in MRI; Brain Infection and Degenerative Disease Studied by Proton MRS; Brain MRS of Infants and Children; Brain Neoplasms Studied by MRI; Chemical Shift Imaging; Diffusion: Clinical Utility of MRI Studies; Echo-Planar Imaging; Health and Safety Aspects of Human MR Studies; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Sodium-23 Magnetic Resonance of Human Subjects; Structural and Functional MR in Epilepsy; Systemically

7

Induced Encephalopathies: Newer Clinical Applications of MRS; Whole Body Studies: Impact of MRS.

12

REFERENCES

1. C. C. Hanstock, D. L. Rothman, J. W. Prichard, T. and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1988, 85, 1821. 2. J. Prichard, D. Rothman, E. Novotny, O. Petroff, T. Kuwabara, M. Avison, A. Howseman, C. Hanstock, and R. Shulman, Proc. Natl. Acad. Sci. USA, 1991, 88, 5829. 3. D. Sappey-Marinier, G. Calabrese, G. Fein, J. Hugg, C. Biggins, and M. Weiner, J. Cereb. Blood Flow Metab., 1992, 12, 584. 4. B. G. Jenkins, J. W. Belliveau, and B. R. Rosen, Proc. 11th Ann Mtg. (Int) Soc. Magn. Reson. Med., Berlin, 1992, 2145. 5. M. Singh, IEEE Trans. Nucl. Sci., 1992, 39, 1161. 6. P. T. Fox, M. E. Raichle, M. A. Mintun, and C. Dence, Science, 1988, 241, 462. 7. I. W. Borowsky and R. C. Collins, J. Comp. Neurol., 1989, 288, 401. 8. Y. M. Gaudry-Talarmain, Eur. J. Pharmacol., 1986, 129, 235. 9. G. D. Graham, A. M. Blamire, A. M. Howseman, D. L. Rothman, P. B. Fayad, L. M. Brass, O. A. Petroff, R. G. Shulman, and J. W. Prichard, Stroke, 1992, 23, 333. 10. G. D. Graham, A. M. Blamire, D. L. Rothman, L. M. Brass, P. B. Fayad, O. A. C. Petroff, and J. W. Prichard, Stroke, 1993, 24, 1891. 11. J. H. Duijn, G. B. Matson, A. A. Maudsley, J. W. Hugg, and M. W. Weiner, Radiology, 1992, 183, 711. 12. C. C. Ford, R. H. Griffey, N. A. Matwiyoff, and G. A. Rosenberg, Neurology, 1992, 42, 1408. 13. P. Gideon, B. Sperling, P. Arlien-Soborg, T. S. Olsen, and O. Henriksen, Stroke, 1994, 25, 967. 14. J. W. Berkelbach van der Sprenkel, P. R. Luyten, P. C. van Rijen, C. A. Tulleken, and J. A. den Hollander, Stroke, 1988, 19, 1556. 15. J. W. Prichard, in `Molecular and Cellular Approaches to the Treatment of Brain Disease', ed. S. G. Waxman, Raven Press, New York, 1993, pp. 153±174. 16. O. A. Petroff, G. D. Graham, A. M. Blamire, R. M. al, D. L. Rothman, P. B. Fayad, L. M. Brass, R. G. Shulman, and J. W. Prichard, Neurology, 1992, 42, 1349. 17. P. C. Lauterbur, Ann. N.Y. Acad. Sci., 1958, 70, 841. 18. P. C. Lauterbur, J. Chem. Phys., 1957, 26, 217. 19. C. H. Holm, J. Chem. Phys., 1957, 26, 707. 20. N. A. Matwiyoff, and D. G. Ott, Science, 1973, 181, 1125. 21. J. R. Alger, L. O. Sillerud, K. L. Behar, R. J. Gillies, R. G. Shulman, R. E. Gordon, D. Shae, and P. E. Hanley, Science, 1981, 214, 660. 22. K. L. Behar, O. A. C. Petroff, J. W. Prichard, J. R. Alger, and R. G. Shulman, Magn. Reson. Med., 1986, 3, 911. 23. D. L. Rothman, E. J. Novotny, G. I. Shulman, A. M. Howseman, O. A. C. Petroff, G. Mason, T. Nixon, C. C. Hanstock, J. W. Prichard, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1992, 89, 9603. 24. R. Gruetter, E. J. Novotny, S. D. Boulware, D. L. Rothman, G. F. Mason, G. I. Shulman, R. G. Shulman, and W. V. Tamborlane, Proc. Natl. Acad. Sci. USA, 1992, 89, 1109. 25. R. T. Eakin, L. O. Morgan, C. T. Gregg, and N. A. Matwiyoff, FEBS Lett., 1972, 28, 259. 26. D. L. Rothman, K. L. Behar, H. P. Hetherington, J. A. den Hollander, M. R. Bendall, O. A. C. Petroff, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1985, 82, 1633. 27. G. F. Mason, D. L. Rothman, K. L. Behar, and R. G. Shulman, J. Cereb. Blood Flow Metab., 1992, 12, 434. 28. J. W. Prichard, O. A. Petroff, T. Ogino, and R. G. Shulman, Ann. N.Y. Acad. Sci., 1987, 508, 54.

8 BRAIN MRS OF HUMAN SUBJECTS 29. O. A. C. Petroff, E. J. Novotny, M. Avison, D. L. Rothman, J. R. Alger, T. Ogino, G. I. Shulman, and J. W. Prichard, J. Cereb. Blood Flow Metab., 1992, 12, 1022. 30. D. L. Rothman, A. M. Howseman, G. D. Graham, O. A. C. Petroff, G. Lantos, P. B. Fayad, L. M. Brass, G. I. Shulman, R. G. Shulman, and J. W. Prichard, Magn. Reson. Med., 1991, 21, 302. 31. S. Warach, D. Chien, W. Li, M. B. Ronthal, and R. R. Edelman, Neurology, 1992, 42, 1717. 32. P. A. Bottomley, B. P. Drayer, and L. S. Smith, Radiology, 1986, 160, 763. 33. S. R. Levine, J. A. Helpern, K. M. Welch, A. M. Vande Linde, K. L. Sawaya, E. E. Brown, N. M. Ramadan, R. K. Deveshwar, and R. J. Ordidge, Radiology, 1992, 185, 537. 34. H. R. Halvorson, A. M. Vande Linde, J. A. Helpern, and K. M. Welch, NMR Biomed., 1992, 5, 53. 35. J. A. Helpern, A. Vande Linde, K. M. Welch, S. R. Levine, L. R. Schultz, R. J. Ordidge, H. R. Halvorson, and J. W. Hugg, Neurology, 1993, 43, 1577. 36. K. L. Behar, J. A. den Hollander, O. A. C. Petroff, H. Hetherington, J. W. Prichard, and R. G. Shulman, J. Neurochem., 1985, 44, 1045. 37. J. W. Hugg, J. H. Duijn, G. B. Matson, A. A. Maudsley, J. S. Tsuruda, D. F. Gelinas, and M. W. Weiner, J. Cereb. Blood Flow Metab., 1992, 12, 734. 38. P. Gideon, B. Sperling, O. Henriksen, N. A. Lassen, T. Skyhùj, T. S. Olsen, P. Sidenius, and P. Arlien-Soborg, Proc. 12th Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, 1484. 39. O. A. C. Petroff, J. W. Prichard, T. Ogino, M. J. Avison, J. R. Alger, and R. G. Shulman, Ann. Neurol., 1986, 20, 185. 40. W. Paschen, B. Djuricic, G. Mies, R. Schmidt-Kastner, and F. Linn, J. Neurochem., 1987, 48, 154. 41. J. W. Prichard, Epilepsia, 1994, 35(Suppl 6), S14. 42. D. P. Younkin, P. M. Delivoria, J. Maris, E. Donlon, R. Clancy, and B. Chance, Ann. Neurol., 1986, 20, 513. 43. J. W. Hugg, G. B. Matson, D. B. Tweig, A. A. Maudsley, D. Sappey-Marinier, and M. W. Weiner, Magn. Reson. Imag., 1992, 10, 227. 44. K. D. Laxer, B. Hubesch, M. D. Sappey-Marinier, and M. W. Weiner, Epilepsia, 1992, 33, 618. 45. R. Kuzniecky, G. A. Elgavish, H. P. Hetherington, W. T. Evanochko, and G. M. Pohost, Neurology, 1992, 42, 1586. 46. P. M. Matthews, F. Andermann, and D. L. Arnold, Neurology, 1990, 40, 985. 47. G. Layer, F. Traber, U. Muller-Lisse, J. Bunke, C. E. Elger, and M. Reiser, Radiologe, 1993, 33, 178. 48. G. D. Jackson, S. F. Berkovic, J. S. Duncan, and A. Connelly, Am. J. Neuroradiol., 1993, 14, 753. 49. G. D. Jackson, A. Connelly, J. S. Duncan, R. A. Grunewald, and D. G. Gadian, Neurology, 1993, 43, 1793. 50. D. L. Rothman, O. A. Petroff, K. L. Behar, and R. H. Mattson, Proc. Natl. Acad. Sci. USA, 1993, 90, 5662. 51. O. Petroff, D. Rothman, K. Behar, and R. Mattson, Proc. 12th Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, 434. 52. J. N. Kanfer, J. W. Pettegrew, J. Moossy, and D. G. McCartney, Neurochem. Res., 1993, 18, 331. 53. P. A. Bottomley, J. P. Cousins, D. L. Pendrey, W. A. Wagle, C. J. Hardy, F. A. Eames, R. J. McCaffrey, and D. A. Thompson, Radiology, 1992, 183, 695. 54. D. G. Murphy, P. A. Bottomley, J. A. Salerno, C. DeCarli, M. J. Mentis, C. L. Grady, D. Teichberg, K. R. Giacometti, J. M. Rosenberg, C. J. Hardy, M. B. Schapiro, S. I. Rapoport, J. R. Alger, and B. Horwitz, Arch. Gen. Psychiatr., 1993, 50, 341.

55. R. D. Oberhaensli, J. D. Hilton, P. J. Bore, L. J. Hands, R. P. Rampling, and G. K. Radda, Lancet, 1986, 2, 8. 56. J. Jeske, K. Herholz, W. Heindel, and W. D. Heiss, Onkologie, 1989, 1, 42. 57. W. D. Heiss, W. Heindel, K. Herholz, J. Rudolf, J. Bunke, J. Jeske, and G. Friedmann, J. Nucl. Med., 1990, 31, 302. 58. P. R. Luyten, A. J. Marien, W. van G. P. Heindel, K. den H. J. Herholz, G. Friedmann, and W. D. Heiss, Radiology, 1990, 176, 791. 59. K. Herholz, W. Heindel, P. R. Luyten, J. A. denHollander, U. Pietrzyk, J. Voges, H. Kugel, G. Friedmann, and W. D. Heiss, Ann. Neurol., 1992, 31, 319. 60. J. R. Alger, J. A. Frank, A. Bizzi, M. J. Fulham, B. X. DeSouza, M. O. Duhaney, S. W. Inscoe, J. L. Black, P. C. Van Zijl, C. T. Moonen, and G. Di Chiro, Radiology, 1990, 177, 633. 61. M. J. Fulham, A. Bizzi, M. J. Dietz, H. H. Shih, R. Raman, G. S. Sobering, J. A. Frank, A. J. Dwyer J. R. Alger, and G. Di Chiro, Radiology, 1992, 185, 675. 62. T. R. Brown, NMR Biomed., 1992, 5, 238. 63. B. J. Murphy, R. Stoyanova, R. Srinivasan, T. Willard, D. Vigneron, S. Nelson, J. S. Taylor, and T. R. Brown, NMR Biomed., 1993, 6, 173. 64. D. L. Arnold, M. D. Matthews, G. S. O. C. J. Francis, and J. P. Antel, Ann. Neurol., 1992, 31, 235. 65. S. J. Nelson, D. B. Vigneron, J. Star-Lack, and J. Kurhanewicz, NMR Biomed., 1997, 10, 411. 66. R. V. Mulkern, H. Chao, J. L. Bowers, and D. Holtzman, Ann. N. Y. Acad. Sci., 1997, 820, 97. 67. M. C. Preul, Z. Caramanos, R. Leblanc, J. G. Villemure, and D. L. Arnold, NMR Biomed., 1998, 11, 192. 68. D. L. Arnold, J. S. Wolinsky, P. M. Matthews, and A. Falini, J. Neurol. Neurosurg. Psychiat., 1998, 64(Suppl. 1), S94. 69. H. P. Hetherington, J. W. Pan, W. J. Chu, G. F. Mason, and B. R. Newcomer, NMR Biomed., 1997, 10, 360. 70. P. C. van Zijl and P. B. Barker, Ann. N. Y. Acad. Sci., 1997, 820, 75. 71. R. Gruetter, M. Garwood, K. Ugurbil, and E. R. Seaquist, Magn. Reson. Med., 1996, 36, 1. 72. D. L. Rothman, K. L. Behar, J. W. Prichard, and O. A. Petroff, Magn. Reson. Med., 1997, 38, 924. 73. J. W. Prichard, D. L. Rothman, and O. A. C. Petroff, in `pH and brain function', eds K. Kaila and B. R. Ransom, John Wiley, New York, 1998, p. 149.

Acknowledgements The author's own work was supported by USPHS Grants NS 27883, DK 34576, and NS 21708.

Biographical Sketch J. W. Prichard. b 1934. A.B., 1955, Philosophy, Washington University, St. Louis, M.D., 1959 Harvard Medical School, Boston. Clinical and postdoctoral training at Bellevue Hospital (New York), The National Hospital for Nervous Diseases (Queen Square, London), Yale (New Haven), and The National Institutes of Health (Bethesda). Research career in neurology electrophysiology prior to 1981. Entered NMR research then through collaboration with Yale spectroscopists in development and validation of NMR methods for study of the brain in vivo. Approx. 110 publications, half in NMR. Principal research interests: application of NMR methods to clinical neurology and their relationship to cerebral electrophysiology.

Brain: Sensory Activation Monitored by Induced Hemodynamic Changes with Echo Planar MRI Peter A. Bandettini, Jeffrey R. Binder, Edgar E. DeYoe and James S. Hyde Medical College of Wisconsin, Milwaukee, WI, USA

1 2 3 4 5 6 7 8

Introduction Postprocessing Methodology Pulse Sequence and Hardware Auditory Stimuli Visual Stimuli Conclusions Related Articles References

1

INTRODUCTION

1 2 2 2 2 3 4 4

In recent years, it has been demonstrated that MRI is capable of detecting changes in cerebral blood volume,1 flow,2 and oxygenation,2 – 5 that accompany an increase in neuronal activity. MRI methods that observe these activation-induced changes have been termed functional MRI (FMRI). The most widely used FMRI method for the noninvasive mapping of human brain activity is based on blood oxygenation level dependent (BOLD) contrast.6 A localized signal increase in activated cortical regions is generally observed using a time-course collection of T 2 *-weighted images. A working model of this phenomenon is that an increase in neuronal activity causes local vasodilatation, which in turn causes blood flow to increase in such a manner that the amount of paramagnetic deoxyhemoglobin in the local vasculature is reduced. This reduction causes an increase in spin coherence and, therefore, an increase in signal when using T 2 - or T 2 *weighted pulse sequences. Support for the hypothesized BOLD contrast mechanism, as it related to FMRI, comes from several studies. Local cerebral blood oxygenation has been observed to increase with neuronal activity.7 – 10 Brain tissue T 2 , T 2 *, and T 2
have been shown to be decreased and increased by cerebral blood oxygenation decreases and increases, respectively.6,11 – 15 In addition, brain activation has been shown to increase T 2 , T 2 *, and T 2
,16 – 20 in proportions that show general agreement with mathematical simulations based on simplified models of the cerebral vasculature.21,22 The applicability of FMRI depends on the degree to which the induced signal enhancement magnitude, location, and timing can be correlated with underlying neuronal activation. The signal enhancement magnitude is not only affected by MRI parameters, but is also dependent on hemodynamic factors (i.e. blood volume, oxygenation, vessel orientation, and radii), which vary significantly from voxel to voxel.

Correlation between BOLD signal enhancement magnitude and cerebral blood flow has, nevertheless, been observed. Visual cortex activation studies have shown that BOLD signal enhancement has essentially the same flicker frequency dependency2 as that of cerebral blood flow changes observed using positron emission tomography (PET).23 The FMRI signal enhancement location and distribution is also an issue. The upper limit of spatial resolution of FMRI may be influenced by the degree to which an activation induced modulation of the velocity or oxygenation of rapidly inflowing spins or large collecting veins may contribute to the signal change. Generally, the larger the collecting vessel, the more spatially removed the signal change is from the area of neuronal activation. Such vessel size weighting might depend strongly on the pulse sequence, field strength, and resolution used.16 – 26 Nevertheless, studies have shown that FMRI can reveal activity localized to patches of cortex smaller than 1.5 mm.27 The temporal resolution of FMRI depends on the latency and consistency of the activation-induced signal change. Rise latency of 8–9 s from stimulus onset to maximal signal change and a fall latency of 9–10 s from stimulus cessation to baseline signal have been reported.2,28,29 The hemodynamic response time sets upper limits on the functional temporal resolution. A significant increase from baseline is generally observed to take place within 2–3 s after stimulus onset.2,28,29 Activation durations of less than 1 s are detectable30,31 and relative differences (in the rise time from adjacent regions or from different experiments) in the onset of signal enhancement are discriminable to within a second.32 Platforms on which FMRI are performed vary considerably. Primary differences are in field strength, pulse sequence, gradient and radiofrequency coil hardware, and postprocessing methods. The many trade-offs that exist between platform types have not been completely characterized, but it is clear that FMRI based on BOLD contrast benefits from high field strength, high system stability, high signal-tonoise, (S/N) ratio, and minimal pulsatile motion sensitivity. From reported results, it appears that these criteria are most readily met using echo planar imaging. Nevertheless, other techniques may be suitable. Initial successes in the performance of FMRI using BOLD contrast were published by groups using either EPI at 1.5 T,2,3 and 4 T,16 or fast multishot gradient-recalled imaging techniques at 4 T,4 and at 2 T.5 Results have since been reported using fast multishot gradient-recalled imaging techniques at 1.5 T.33 – 35 Pulsatile brain, cerebrospinal fluid, and blood motion apparently cause nonrepeatable ghosting patterns in sequential images obtained by conventional multishot techniques,36,37 thus adding significantly to the image signal variation in time. Multishot spiral-scan techniques have lower sensitivity to these contaminating pulsatile effects,36,37 and have been used successfully to perform high resolution FMRI at 1.5 T.27 The use of a time course of images in conjunction with control over activation timing allows for the application of postprocessing methods which include the use of z maps,38 and other statistical techniques.35 In addition, other methods, including Fourier analysis,39 temporal cross correlation,39 and time–frequency analysis40 have been successfully applied. While significant signal changes are easily observed using the most conservative statistical tests, a standardized method by which functional images are created has not been established.

2 BRAIN: SENSORY ACTIVATION MONITORED BY INDUCED HEMODYNAMIC CHANGES WITH ECHO PLANAR MRI Using EPI, studies of the fluctuations in the susceptibilityweighted signal from a quiescent brain have been carried out,41 not only to determine the nature of the noise for application to statistical tests, but also to obtain potentially useful physiological information, and to determine the actual relationship between functional contrast to noise and field strength.42 Because of the ease of use and accessibility of FMRI, many areas of the brain and many different tasks have already been studied. Studies that have been carried out include those of primary cortical regions including visual cortex,2,4,5,16,29,35,43 motor cortex,2,3,26,33,34,39,44 – 46 and auditory cortex.47 Studies of higher cognitive function, including word generation,48,49 higher visual processing,35,43 visual recall,50 complex motor control,46 and single word semantic processing,51 have also been performed. In this article, applications of FMRI to the mapping of cortex activated by sensory stimulation are summarized using time-course collection of gradient recalled echo planar images at 1.5 T.3,39 and temporal crosscorrelation postprocessing techniques.39 Specifically, auditory cortex regions were differentially activated by noise and speech sounds, and visual cortex regions were selectively activated by visual stimuli of different eccentricity.

2

POSTPROCESSING METHODOLOGY

Stimuli are presented in a repetitive on/off fashion for several cycles throughout each time course. Foci of brain activation are identified by crosscorrelation of the time course of each voxel with a reference vector resembling the expected activation-induced response.39 In general, a reference vector may be obtained by: (a) choosing the voxel containing what appears to be the ‘best’ temporal response; (b) averaging, in time, the activation cycles in a ‘best temporal response’ voxel and then duplicating the time-averaged cycle for the length of the time course; (c) averaging in space several of the ‘best temporal response’ voxels; (d) synthesizing a reference vector; or (e) choosing a vector obtained from principal component analysis of the time course. Pixels having a temporal correlation coefficient below a given threshold (typically 0.4–0.6) are removed. After thresholding, the vector product of the reference vector with each of the surviving time courses is calculated to yield an index of change in the signal magnitude. These ‘activation’ images are then colorized and superimposed on high resolution anatomical scans of the same slice obtained in the same imaging session.

3

PULSE SEQUENCE AND HARDWARE

All studies presented in this article were performed using single-shot 64 × 64 gradient-recalled EPI (TE = 40 ms) on a clinical 1.5-T GE Signa scanner. To perform EPI without additional stress to the standard gradient amplifiers, we used an insertable balanced torque three-axis head gradient coil designed for rapid gradient switching.52 To obtain high-quality images throughout the entire brain volume, a shielded quadrature elliptical endcapped transmit/receive birdcage radiofrequency coil was used.53 Typically, single-

or multi-slice time-course series of 64–1024 images were obtained with a TR of 0.5–3 s, flip angle of 65–90◦ , field of view (FOV) of 24 cm, and slice thickness of 4–10 mm.

4 AUDITORY STIMULI

This study47 presents findings using FMRI of brain regions involved in auditory speech perception. Specifically, regions activated by speech sounds (words and pseudowords) and nonspeech sounds (noise) were compared. Five right-handed subjects were tested. Symmetric lateral sagittal slices (slice thickness 10 mm) of the left and right hemispheres were obtained, centered at positions 8 mm medial to the most lateral point of the temporal lobe on each side. In each time-course series, 64 sequential images were collected (TR = 3 s), during which activation alternated with baseline every 9 s (6 images per cycle, 18 s per cycle, 10 cycles). During baseline periods, subjects heard only the ambient scanner noise. During activation periods, prepared digitized auditory stimuli were delivered to the subject via air conduction through a semi-rigid 1-cm-bore plastic tube. The tube conducted the sound stimulus approximately 20 feet from control room wall to the subject, at which point a Y-connector split the tube for binaural stimulation through a tightly fitting headset with occlusive earplugs to reduce scanner noise exposure. Subjects passively listened to stimuli; no response was required. Stimuli differed in both semantic and acoustic (frequency modulation) content. White noise presentation was compared to presentation of nouns (e.g. ‘barn’), and pseudowords (e.g. ‘narb’) with stimuli matched for duration, presentation rate, average sound pressure level, and spectral range. All voxels having a temporal correlation coefficient to a sinusoidal reference vector below 0.5 were removed. The activation images were superimposed on high resolution scans of the same slices. Figure 1 illustrates typical results from two of the subjects studied with white noise, pseudowords, and words, respectively. In these and all other subjects studied, the area activated by white noise was considerably smaller than that activated by speech sounds, and was restricted to the dorsal aspect of the superior temporal gyrus. In most instances, this region coincided with or included the transverse temporal (Heschl’s) gyrus. Presentation of speech sounds activated a larger region, including more anterior and posterior areas of the dorsal superior temporal gyrus, as well as cortex in or near the superior temporal sulcus bilaterally. Both words and pseudowords differed significantly from noise bilaterally, while no significant differences were seen between word and pseudoword conditions. Activity occurred symmetrically in the left and right temporal lobes. Unlike white noise, therefore, processing of speech sounds appears to elicit extensive participation of auditory association areas, even when the subject is not engaged in any ‘active’ task.

5 VISUAL STIMULI

Much of the utility of FMRI depends on its ability to depict spatial patterns of neural activity. To test this capacity in the visual system, FMRI was used for retinotopic mapping of primary visual cortex activation.43

BRAIN: SENSORY ACTIVATION MONITORED BY INDUCED HEMODYNAMIC CHANGES WITH ECHO PLANAR MRI

3

Figure 1 Sagittal images of the left and right temporal lobes of two subjects. Demonstrated are regions activated during passive listening to (A) white noise, (B) pseudowords, and (C) words. The area activated by white noise was smaller than the area activated by either pseudowords or words. The area activated by pseudowords was generally the same size and shape as the area activated by words. Reproduced with permission of the American Neurological Association, 1994)

Dynamic, computer graphics-based visual images were directly projected onto the subjects’ retinae. The image generator was a modified Sharp XG2000U video projector driven by Cambridge Instruments VSG video graphics board installed in a personal computer. The image plane was then viewed through a custom optical system that included a wide field, magnifying eyepiece, a 45◦ prism, and additional objective lenses for adjusting magnification and minimizing chromatic aberration. Two sets of imaging optics were combined to provide full binocular viewing.54 To map the retinotopic organization of the visual cortex, three highly trained subjects viewed a small white fixation dot on a uniform black or gray field subtending 60◦ of visual angle. A black-and-white checkered annulus surrounding the fixation point was presented for 5 on/off cycles of 10 s on and 10 s off. Time-course series (TR = 2 s) of 100 images (slice thickness 8 mm) were used. When on, the checkered pattern was either counter-phase modulated or flickered at 6–8 Hz. Six successively larger annuli were tested. The width of each annulus as well as the check size were scaled in proportion to eccentricity. Only voxels having a correlation coefficient above 0.55 with respect to a chosen reference vector were used in the creation of functional images. All functional images were then superimposed on high resolution anatomical scans. Figure 2 illustrates the relative sizes of the annuli and shows the corresponding brain activation images. In these experiments, the subject passively viewed the stimuli and was not required to respond to them. A small checkerboard annulus presented at the fixation point elicited activation in striate cortex only at the occipital poles bilaterally. Annuli presented at increasing eccentricities activated successively more anterior

regions of the calcarine sulcus. The most eccentric stimulus activated only the anterior calcarine cortex while sparing the occipital poles. Detailed examination of the sequence of activity foci in Figure 2, shows a progression that closely follows the folded cortical mantle within the calcarine sulcus. While such a precise progression is not always observed, these data do show that, under optimal conditions, a detailed mapping of the visual field representation is possible with FMRI. Resolution is not limited by the coarseness of the distribution of large blood vessels, even though such vessels may sometimes introduce artifacts.

6 CONCLUSIONS

MRI of human brain activation using BOLD contrast is a relatively new functional brain imaging method. Accompanying the novelty of the technique are many unknowns regarding the upper limits of spatial and temporal resolution as well as an unclear understanding of physiological and the biophysical mechanisms that regulate hemodynamic changes. In addition, the ways in which the hemodynamic changes affect the magnetic resonance signal are incompletely understood. Nevertheless, the applications described here and elsewhere empirically establish the utility of this approach. These studies demonstrate observation by FMRI of human brain activation by sensory stimulation. Regions activated in temporal lobes by various speech and nonspeech auditory stimuli were observed. In addition, retinotopic organization of the

4 BRAIN: SENSORY ACTIVATION MONITORED BY INDUCED HEMODYNAMIC CHANGES WITH ECHO PLANAR MRI 7 RELATED ARTICLES

Diffusion and Perfusion in MRI; Echo-Planar Imaging; Functional MRI: Theory and Practice; Functional Neuroimaging Artifacts; Hemoglobin; Susceptibility Effects in Whole Body Experiments

8 REFERENCES

Figure 2 Axial brain activation images created by passive viewing of visual stimuli with six different eccentricities while fixating at the center. The active foci traveled in an anterior direction along the calcarine fissure as the stimulus became more peripheral

primary visual cortex was observed using stimuli of varying visual field eccentricity. Additionally, in our laboratory, preliminary studies involving activation by tactile, aromatic, and taste stimuli are in progress. FMRI is a new technique that holds great promise in uncovering unique and useful information about brain function and physiology.

1. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen, Science, 1991, 254, 716. 2. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H. M. Cheng, T. J. Brady, and B. R. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 3. P. A. Bandettini, E. C. Wong, R. S. Hinks, R. S. Tikofsky, and J. S. Hyde, Magn. Reson. Med., 1992, 25, 390. 4. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellermann, S.-G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1992, 89, 5951. 5. J. Frahm, H. Bruhn, K.-D. Merboldt, and W. Hanicke, J. Magn. Reson. Imaging, 1992, 2, 501. 6. S. Ogawa, T. M. Lee, A. R. Kay, and D. W. Tank, Proc. Natl. Acad. Sci. USA, 1990, 87, 9868. 7. A. Grinvald, R. D. Frostig, R. M. Siegel, and E. Bratfeld, Proc. Natl. Acad. Sci. USA, 1991, 88, 11559. 8. P. T. Fox and M. E. Raichle, Proc. Natl. Acad. Sci. USA, 1986, 83, 1140. 9. R. D. Frostig, E. E. Lieke, D. Y. Ts’o, and A. Grinvald, Proc. Natl. Acad. Sci. USA, 1990, 87, 6082. 10. A. Villringer, J. Planck, C. Hock, L. Schleinkofer, and U. Dirnagl, Neurosci. Lett., 1993, 154, 101. 11. R. Turner, D. Le Bihan, C. T. Moonen, D. Despres, and J. Frank, Magn. Reson. Med., 1991, 22, 159. 12. B. E. Hoppel, R. M. Weisskoff, K. R. Thulborn, J. B. Moore, K. K. Kwong, and B. R. Rosen, Magn. Reson. Med., 1993, 30, 715. 13. M. K. Stehling, F. Schmitt, and R. Ladebeck, J. Magn. Reson. Imaging., 1993, 3, 471. 14. A. J. DeCrespigny, M. F. Wendland, N. Derugin, E. Kozniewska, and M. E. Moseley, Magn. Reson. Med., 1992, 27, 391. 15. P. Jezzard, F. Heineman, J. Taylor, D. Taylor, H. Wen, R. S. Balaban, and R. Turner, NMR Biomed., 1994, 7, 35. 16. R. Turner, P. Jezzard, H. Wen, K. K. Kwong, D. Le Bihan, T. Zeffiro, and R. S. Balaban, Magn. Reson. Med., 1993, 29, 277. 17. R. S. Menon, S. Ogawa, D. W. Tank, and K. Ugurbil, Magn. Reson. Med., 1993, 30, 380. 18. B. E. Hoppel, J. R. Baker, R. M. Weisskoff, and B. R. Rosen, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 1384. 19. P. A. Bandettini, E. C. Wong, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, NMR Biomed., 1994, 7, 12. 20. P. A. Bandettini, E. C. Wong, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 169. 21. S. Ogawa, R. S. Menon, D. W. Tank, S.-G. Kim, H. Merkle, J. M. Ellermann, and K. Ugurbil, Biophys. J., 1993, 64, 803. 22. R. P. Kennan, J. Zhong, and J. C. Gore, Magn. Reson. Med., 1994, 31, 9. 23. P. T. Fox and M. E. Raichle, J. Neurophysiol., 1984, 51, 1109. 24. J. Frahm, K.-D. Merboldt, and W. Hanicke, Proceedings, 12th Annual Scientific Meeting, Society of Magnetic Resonance Medicine, New York, 1993, p. 1427.

BRAIN: SENSORY ACTIVATION MONITORED BY INDUCED HEMODYNAMIC CHANGES WITH ECHO PLANAR MRI 25. J. H. Duyn, C. T. W. Moonen, R. W. de Boer, G. M. van Yperen, and P. R. Luyten, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 168. 26. S. Lai, A. L. Hopkins, E. M. Haacke, D. Li, B. A. Wasserman, P. Buckley, L. Friedman, H. Meltzer, P. Hedera, and R. Friedland, Magn. Reson. Med., 1993, 30, 387. 27. S. A. Engel, D. E. Rumelhart, B. A. Wandell, A. T. Lee, G. H. Glover, E. J. Chichilnisky, and M. N. Shadlen, Nature (Lond.), 1994, 369, 525. 28. E. A. DeYoe, J. Neitz, P. A. Bandettini, E. C. Wong, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1991 Meet., p. 1824. 29. A. M. Blamire, S. Ogawa, K. Ugurbil, D. Rothman, G. McCarthy, J. M. Ellermann, F. Hyder, Z. Rattner, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1992, 89, 11069. 30. P. A. Bandettini, E. C. Wong, E. A. DeYoe, J. R. Binder, S. M. Rao, D. Birzer, L. D. Estkowski, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 1382. 31. J. W. Belliveau, J. R. Baker, K. K. Kwong B. R. Rosen, J. S. George, C. J. Aine, J. D. Lewine, J. A. Sanders, G. V. Simpson, and J. Foxe, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 6. 32. J. R. Binder, A. Jesmanowicz, S. M. Rao, P. A. Bandettini, T. A. Hammeke, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 1383. 33. Y. Cao, V. L. Towle, D. N. Levin, and J. M. Balter, J. Magn. Reson. Imaging, 1993, 3, 869. 34. A. Connelly, G. D. Jackson, R. S. Frackowiak, J. W. Belliveau, F. Vargha-Khadem, and D. G. Gadian, Radiology, 1993, 188, 125. 35. W. Schneider, D. C. Noll, and J. D. Cohen, Nature, 1993, 365, 150. 36. G. H. Glover and J. M. Pauly, Magn. Reson. Med., 1992, 28, 275. 37. G. H. Glover, A. T. Lee, and C. H. Meyers, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 197. 38. D. Le Bihan, P. Jezzard, R. Turner, C. A. Cuenod, L. Pannier, and A. Prinster, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 11. 39. P. A. Bandettini, A. Jesmanowicz, E. C. Wong, and J. S. Hyde, Magn. Reson. Med., 1993, 30, 161. 40. B. Biswal, P. A. Bandettini, A. Jesmanowicz, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 722. 41. R. M. Weisskoff, J. Baker, J. Belliveau, T. L. Davis, K. K. Kwong, M. S. Cohen, and B. R. Rosen, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 7. 42. P. Jezzard, D. Le Bihan, C. Cuenod, L. Pannier, A. Prinster, and R. Turner, Proc. Soc. Magn. Reson. Med. 1993 Meet., p. 1392. 43. E. A. DeYoe, P. A. Bandettini, J. Neitz, D. L. Miller, and P. Winans, J. Neurosci. Meth., 1994, 54, 171. 44. S.-G. Kim, J. Ashe, A. P. Georgopoulos, H. Merkle, J. M. Ellermann, R. S. Menon, S. Ogawa, and K. Ugurbil, J. Neurophysiol., 1993, 69, 297. 45. S.-G. Kim, J. Ashe, K. Hendrich, J. M. Ellermann, H. Merkle, K. Ugurbil, and A. P. Georgopoulos, Science, 1993, 261, 615. 46. S. M. Rao, J. R. Binder, P. A. Bandettini, T. A. Hammeke, F. Z. Yetkin, A. Jesmanowicz, L. M. Lisk, G. L. Morris. W. M. Meuller, L. D. Estkowski, E. C. Wong, V. M. Haughton, and J. S. Hyde, Neurology, 1993, 43, 2311. 47. J. R. Binder, S. M. Rao, T. A. Hammeke, F. Z. Yetkin, A. Jesmanowicz, P. A. Bandettini, E. C. Wong, L. D. Estkowski, M. D. Goldstein, V. M. Haughton, and J. S. Hyde, Ann. Neurol., in press. 48. G. McCarthy, A. M. Blamire, D. L. Rothman, R. Gruetter, and R. S. Shulman, Proc. Natl. Acad. Sci. USA, 1993, 90, 4952. 49. R. M. Hinke, X. Hu, A. E. Stillman, S.-G. Kim, H. Merkle, R. Salmi, and K. Ugurbil, Neuroreport, 1993, 4, 675.

5

50. D. Le Bihan, R. Turner, T. Zeffiro, C. A. Cuenod, P. Jezzard, and V. Bonnerot, Proc. Natl. Acad. Sci. USA, 1993, 90, 11802. 51. J. R. Binder, S. M. Rao, T. A. Hammeke, J. A. Frost, P. A. Bandettini, A. Jesmanowicz, and J. S. Hyde, Arch. Neurol., 1995, 52, 593. 52. E. C. Wong, P. A. Bandettini, and J. S. Hyde. Proc. Soc. Magn. Reson. Med. 1991 Meet., p. 105. 53. E. C. Wong, E. Boskamp, and J. S. Hyde, Proc. Soc. Magn. Reson. Med. 1992 Meet., 3 vol.3 , p. 4015. 54. E. A. DeYoe, J. Neitz, D. Miller, and J. Wieser, Proc. Soc. Magn. Reson. Med. 1993 Meet., p, 1394.

Acknowledgments This work was supported in part, by grants CA41464 and RR01008 from the National Institute of Health. P.A.B. thanks GE Medical Systems for financial support. In adition, the support of R. Scott Hinks at GE Medical Systems and of Donald Dickerson, Lloyd D. Estkowski, Andrew S. Greene, Thomas A. Hammeke, Victor M. Haughton, Andre Jesmanowicz, George L. Morris, Wade M. Meuller, Joel B. Myklebust, David Miller, Jay Neitz, Steven M. Rao, Elliot Stein, Eric C. Wong, F. Zerrin Yetkin, and Jeffrey R. Zigun, at the Medical College of Wisconsin, are gratefully appreciated.

Biographical Sketches Peter A. Bandettini. b 1966. B.S., 1989, Ph.D., 1994, Biophysics, Medical College of Wisconsin, USA (supervisors James S. Hyde and R. Scott Hinks). Postdoctoral fellow, Massachusetts General Hospital NMR Center, 1994–95. Approx. 15 publications. Research specialties: functional MRI contrast mechanisms, postprocessing techniques, and applications. Jeffrey R. Binder. b 1958. B.M., 1980, M.D., 1986, University of Nebraska, USA. Neurology Residency, 1987–90,. Neurological Institute of New York. Fellow in cerebrovascular disease, 1990–92. Neurological Institute (with J. P. Mohr). Currently Assistant Professor of Neurology, Medical College of Wisconsin, USA. Approx. 15 publications. Research specialties: behavioral neurology, neuroscience applications of functional MRI, cerebrovascular disease. Edgar A. DeYoe III. b 1950. B.S. Electrical Engineering, 1972; Ph.D. 1983, Experimental Psychology; Ph.D. 1983, Neuroscience, University of Rochester, USA (Advisor: Robert W. Doty). Postdoctoral fellowship at California Institute of Technology (Supervisor: David Van Essen). Currently assistant professor in Department of Cellular Biology and Anatomy with adjunct appointment to Department of Biophysics at the Medical College of Wisconsin, Milwaukee, WI. Approx. 21 publications. Research interests: Neural mechanisms of visual perception—anatomy and physiology; Functional magnetic resonance imaging (FMRI) of the human visual system; development of systems for testing normal vision and visual system; development of systems for testing normal vision and visual dysfunction during FMRI. James S. Hyde. b 1932. B.S., 1954, Ph.D., 1959, Physics, M.I.T, Cambridge, MA, USA. Member of the scientific staff, Varian Associates, 1959–75, working under the direction of W. A. Anderson and M. E. Packard. Professor of Biophysics, Medical College of Wisconsin, 1975–present. Approx. 275 publications, 27 patents. Research specialties: ESR spectroscopy, ENDOR, muscloskeletal MRI, surface coils, functional MRI, electron and nuclear spin physics, and magnetic resonance instrumentation.

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA

Fluorine-19 MRS: General Overview and Anesthesia David K. Menon University of Cambridge, UK

1 INTRODUCTION Fluorine-19 has 100% natural abundance and possesses a spin of 12 and an NMR sensitivity of 80±85% relative to protons. Consequently, ¯uorine-containing compounds produce NMR signals that are nearly as easily detected as proton signals. The range of chemical shifts for 19F compounds is about 1000 ppm, far greater than that observed for 1H spectroscopy, leading to much greater spectral dispersion. These attributes, coupled with the fact that 19F magnetic resonance (MR) does not require solvent suppression, would be expected to make it an ideal nucleus for study in biological systems. Unfortunately, no naturally occurring molecules of biological consequence contain adequate concentrations of MR-visible 19F (bone and teeth contain a substantial amount of 19F, but this is contained in compounds that possess extremely short T2 values). However, several pharmacologically active compounds contain ¯uorine, and 19F magnetic resonance spectroscopy (MRS) has been used for the study of the biodistribution, pharmacokinetics, and metabolism of several drugs.1 Other groups have also used 19F-containing compounds as tracers for blood,2 either to image tissue blood ¯ow (perfusion) or intravascular volume (angiography). In addition, 19F chemical shifts and relaxation parameters can be greatly in¯uenced by the physical and chemical environment, and 19F containing molecules in biological systems exhibit signi®cant and variable 19F{1H} nuclear Overhauser effects (NOEs) which depend on their chemical environment and mobility.3 These properties allow 19 F MR to be used to probe molecular interactions in biological systems. Section 2 of this article provides an overview of the applications of in vivo 19F MR to biological systems, while Section 3 covers the application of 19F MR to research in anesthesia.

2 APPLICATIONS OF

19

F MR: AN OVERVIEW

In the following discussion illustrative examples are described, the aim being to outline the principles of applications of 19F MR. The discussion will focus on experiments that have involved whole body in vivo MR either in experimental animals or in humans, but occasional reference will be made to in vitro studies of tissue preparations or cell cultures to illustrate a potential in vivo application. 2.1 Biodistribution and Pharmacokinetics 19

F MRS has been used to study the biodistribution of volatile anesthetic agents and antimitotic compounds, and studies

1

addressing this area are discussed in some detail in later sections of this article. In addition, the technique has also been used to study the metabolism of other drugs, two examples of which are discussed here. Karson et al.4 studied subjects being treated with the antidepressant agent ¯uoxitine, and found brain concentrations of 1.3±5.7 g mLÿ1 on a daily dose of 40 mg dayÿ1 in volunteers. They found acquisition of signals from patients more dif®cult, but suggested a correlation between dose and measured brain concentrations. In a more recent paper,5 the same group describe the use of a quadrature coil for localized 19 F MRS of ¯uoxitine in patients and reported on T1 and T2 values for the ¯uoxitine resonance, which was predominantly intracerebral in location. Ex vivo 19F MRS of brain obtained from patients at autopsy showed that at least part of the in vivo resonance was due to the active metabolite nor¯uoxitine, and in vitro analysis of lipid extracts of brain showed far higher concentrations (12.3±18.6 g mLÿ1) than were observed in vivo, suggesting that some of the compound may be NMR invisible in vivo. These studies clearly demonstrate that, while 19 F MRS is technically suitable for clinical pharmacokinetic research, in vivo MRS needs to be underpinned in its initial phases by ex vivo and in vitro data. Relevant clinical studies might investigate the relationship of brain concentrations to drug response, and the behavior of brain concentrations in relation to changes in dosage schedules. Lee et al.6 used 19F MRS to study the rate of absorption of the nonsteroidal antiin¯ammatory agent ¯ubiprofen from a topical preparation in an animal model. This technique may be easily transferred to human studies, and may provide substantial advantages over traditional methods (which either involve excised human skin or require the use of radioisotope labeling) for studying transcutaneous drug absorption. 2.2

Metabolism

Most work reported in the literature has concentrated on the metabolism of drugs, and several examples are discussed in later sections in the context of volatile anesthetic agents and antimitotics. However, 19F MRS may also be used to study the metabolism of modi®ed physiological substrate molecules. Some recent studies have investigated the use of ¯uorinated glucose analogs, such as 2-¯uoro-2-deoxy-D-glucose (2-FDG), in imaging glucose utilization in intact animals.7±9 Such studies provide a regional map of glucose utilization in a manner analogous to 18FDG positron emission tomography (PET), but without the need to expose subjects to radiation.7 In addition, 19 F MRS also provides the ability to interrogate individual metabolic pathways by detecting and quantifying downstream metabolites. Thus, 3-¯uoro-3-deoxy-D-glucose (3-FDG) has been used to study glucose metabolism to sorbitol via the aldose reductase pathway,9 a process that is thought to be intimately involved in the pathogenesis of diabetic cataract and peripheral neuropathy. This is of particular relevance because of the need to evaluate new aldose reductase inhibitors which may be of use in postponing or avoiding the long-term complications of diabetes mellitus. Unfortunately, while many animal studies are underway, the large doses of FDG required, coupled with the potential toxicity of the molecule, dictate that clinical studies are not likely to be possible in the immediate future.

2 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA 2.3 Insights into Molecular Mechanisms of Drug Action and Metabolism The interaction of 19F-containing compounds with other molecules at their site of action may be elucidated by studying changes in NMR properties, including T1, T2, and NOE values. Such information may provide insight into the mechanisms of action or metabolism of drugs. For example, the demonstration of short T2 environments for volatile anesthetics in the brain in some studies implies that the agents may be immobilized at their site of residence (see later), and that this environment is substantially different from that in adipose tissue, where these agents possess long T2 values. This distinction may be of considerable importance in validating putative mechanisms of action of volatile anesthetic agents. The T1 relaxation of ¯uorinated compounds in biological systems is dominated by 1H±19F interactions,10 and the resulting NOEs can provide useful information regarding these interactions.3 In small molecules, irradiation of the proton frequency often results in an enhancement of signal intensity (a positive NOE). In other systems, where the 19F nucleus is bound to a large molecule (e.g. a protein), the NOE is typically negative (i.e. there is a reduction in signal intensity with irradiation of the proton frequency). In situations where the 19F nucleus under study is part of a small molecule which attaches reversibly to a macromolecule, the observed NOE depends on the mole fraction of the small molecule that is bound and its rate of dissociation from the complex. In many biological systems the observed effects are dominated by macromolecular binding, even when the bound fraction is small and the rate of dissociation slow. Jacobson et al.3 have used these effects to investigate the interaction of a cytochrome P450 inducer and a novel herbicide with macromolecules in the liver using in vivo 19 F MRS, and the technique may be applicable to many more experimental situations. Direct transfer to human research must be preceded by studies that estimate energy deposition produced by 1H irradiation, but this is unlikely to be major problem, since similar issues have been addressed in the context of 13C and 31P MRS.

2.4 Information Regarding Physicochemical Environment The T1 of per¯uorocarbons varies with, and may thus provide information regarding, their chemical and physical environment. The paramagnetic O2 molecule reduces their T1 from 1±4 to 0.3±0.5 s in linear proportion to the local partial pressure of oxygen.11 When administered intravenously, these compounds are sequestered in tissues, where their T1 value provides a measure of pO2. Alternatively, the compound may be directly introduced into the site of interest (e.g. the vitreous humor). This technique has been used to estimate tissue pO2 in experimental tumors, vitreous humor in animals and humans, and in animal myocardium. Unfortunately, the T1 of ¯uorocarbons also varies with temperature, making estimation of pO2 dif®cult in experimental situations such as brain ischemia where there may be simultaneous changes in temperature. Berkowitz et al.12 suggest the use of the compound per¯uorotributylamine (FTBA) in this situation. While the T1 value of FTBA changes with pO2 and temperature, the chemical shift of its 19F resonance is indepen-

dent of pO2, but shows a linear change with temperature. 19F MRS of this compound can thus provide an independent measure of the tissue temperature from chemical shift data, and the resulting information can be used to correct T1 data to provide independent and valid measures of pO2. The chemical shift of other 19F compounds is sensitive to pH, and 19F MRS has been used to measure pH in biological systems. For example, Beech and Iles13 used the chemical shift of exogenously administered F-Quene 1 to estimate intracellular pH (pHi) in rat liver in vivo. While the technique was practicable, it did not work consistently, and they found small differences when comparisons were made to estimates of pHi from 31P MRS (using the chemical shift of Pi). Finally, when 19F is covalently attached to various chelators (e.g. 5,5'-di¯uoro-1,2-bis(o-aminophenoxy)ethane-N,N,N',N'tetraacetic acid (5-FBAPTA)), the 19F chemical shift of the resultant compound is sensitive to binding by cations, and the local concentration of many cations (of which calcium is the most biologically important) can be estimated by measuring the relative concentration of the bound species. 5-FBAPTA and other ¯uorinated chelates have been used to study intracellular Ca2+ concentrations in cell preparations and tissue slices,14 and more recent studies have used the techniques to quantify other cations.15 However, no in vivo study has been reported.

2.5

Imaging of Large Vessel Flow and Tissue Perfusion

The per¯uorocarbons have been used as `contrast agents' for 19F angiography, with varying degrees of success.2,16 While their distribution in tissues provides a measure of local tissue blood ¯ow, they do not give a true estimate of perfusion since they are retained in the intravascular compartment, and are not freely diffusible tracers. However, 19F-containing gases and vapors are freely diffusible and can be used as tracers for estimating brain perfusion. Rudin and Sauter17 used the washout of halothane to estimate cerebral blood ¯ow (CBF) in rats. However, halothane is known to be a potent cerebral vasodilator and increases CBF in the doses used in this study. Consequently, its use to determine CBF is inappropriate. However, Pekar et al.18 administered the inert diffusible gas tri¯uoromethane via the inhalational route in cats, and estimated CBF from wash-in and wash-out data. Their results suggest that the measurement of rCBF may be possible using this technique, with a spatial resolution of 0.4 mL, at least in experimental animals. 3

3.1

19

F MRS STUDIES OF FLUORINATED ANESTHETIC AGENTS Clinical and Pharmacological Context

Recent 19F MRS studies of ¯uorinated anesthetics have been the source of some controversy, as results from some groups challenge generally held notions of anesthetic action. Two main issues are pertinent: (a) theories regarding the site of action of anesthetic agents; and (b) the duration of residence of modern anesthetic agents in the brain. Conventional viewpoints on these issues are brie¯y described below, in order to put the following discussion into some sort of context.

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA F

1.0

Br

Desflurane

F N N O

F Cl Halothane

F O

Cl

F

F F

F O

F F

Cl

F Enflurane F F

0.8

F F

Isoflurane Enflurane

0.6

Halothane 0.4

Isoflurane

F

0.2 O

F

FAx/FIx

Nitrous Oxide

3

F

F

Sevoflurane

F

F F

F O F

F F

Desflurane

Figure 1 Chemical structures of general anesthetic agents in common clinical use

General anesthetic agents vary widely in chemical structure (Figure 1), but still continue to have surprisingly similar behavioral effects. However, we have little information on how and where they act in the brain. Two opposing theories claim very different sites of anesthetic action.19 One school of thought attributes general anesthesia to action at a nonspeci®c hydrophobic site in the lipid bilayer of cell membranes which is `disordered' by the entry of the anesthetic molecule.20 The opposing school of thought suggests that these agents may act at speci®c stereoselective hydrophilic sites on membrane proteins.21 Data that suggest molecular speci®city of the interaction between anesthetic agents and the brain would provide support to the second theory.21 The ¯uorinated volatile agents constitute the most widely used class of general anesthetics in the world, and individual agents that are presently in use (Figure 1) include halothane, en¯urane, iso¯urane, sevo¯urane, and des¯urane.22 These compounds are characterized, on clinical grounds, by an apparently rapid onset and offset of action. Conventionally, the rapidity of wash-in and wash-out of an anesthetic agent is thought to vary inversely with its solubility in blood and tissue.22 On this basis, the agents mentioned above rank as follows: des¯urane > sevo¯urane > iso¯urane > en¯urane > halothane, where halothane has the highest blood gas solubility, and hence slowest wash-in and wash-out characteristics22 (Figure 2). Nevertheless, even with halothane, the offset of clinical anesthesia is relatively rapid, with patients awakening within a few minutes of ceasing administration of the agent. However, patients are reported to continue to have subtle psychomotor impairments for several hours to days after general anesthesia.23 3.2 Pharmacokinetics of Fluorinated Anesthetics: Is Anesthetic Residence Prolonged in the Brain? Over the last 10 years there have been numerous papers from several groups that have used 19F MRS to study anesthetic action in the brain. These address two main issues; the ®rst of which is discussed in this section; the second issue is discussed in Section 3.3. The ®rst in vivo surface coil study of ¯uorinated anesthetics demonstrated the feasibility of such studies, and suggested that

0

10

20

30

Time (min)

Figure 2 Increase in the alveolar fractional concentration (FAx) toward that of inspired fractional concentration (FIx) with time, compared for different ¯uorinated volatile agents. Note that the less soluble agents (iso¯urane and des¯urane) show a more rapid rise

the compounds could be detected in the brain for substantially longer than had been expected.24 Further studies from Wyrwicz et al.25 conducted with 2.8 cm surface coil on the scalp, suggested that, following a 2-h exposure to 1% halothane, halothane was washed out of rabbit brain with a biexponential temporal pro®le (Figure 3). The initial rapid decay had a time constant of approximately 25 min, while the later slower washout phase had a time constant of 320 min. These ®ndings were con®rmed by in vitro 19F MRS of extracts of excised rabbit brain at different intervals after the cessation of anesthesia (Figure 4). Although ®gures for washout half-times were not presented for the in vitro data, they were reported to show an `elimination pro®le similar to that observed in intact animals'. These in vitro studies provided better spectral resolution than in vivo studies, and showed that, starting at 90 min after cessation of halothane anesthesia, brain halothane could be resolved into two resonances: a doublet with a 5.6 Hz proton J coupling that was attributed to the tri¯uoromethyl resonance of halothane, and a second peak 0.7 ppm down®eld whose proportion increased with time, such that it represented 40% of the residual 19F signal by 6 h. These two resonances were also differentiated by their relaxation properties. The halothane resonance showed a T1 of 1.3 s and a T2 of 3.8 ms, while the new singlet resonance had a T1 of 2.8 s and a T2 of 10.6 ms, suggesting that the two resonances were in different chemical environments. Further subcellular fractionation showed that the singlet resonance was con®ned to the cytosol, while the halothane tri¯uoromethyl resonance was widely present, and was loosely bound, being dissociated from cellular components by washing with 0.32 M sucrose. The authors concluded that the new singlet resonance represented a nonvolatile metabolite of halothane, possibly tri¯uoroacetate. These ®ndings were surprising, since they implied prolonged halothane residence in the brain, a concept that ¯ew in the face of conventional perfusion-limited models of anesthetic elimination.26 The possibility that a nonvolatile metabolite might be responsible for the prolonged 19F signal provides some explanation for these ®ndings, but did not account for the fact that even 6 h after anesthesia, 60% of the observed 19F signal in excised brain appeared to arise from the tri¯uoro-

4 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA

% 19F NMR signal observed

100

50

30

20 40

80

120

160 200 240 Time (min)

280

320

360

where volatile anesthetics tend to be retained, owning to lower perfusion. Accordingly, they used a spatially selective depth pulse in an effort to obtain a selective signal from deeper brain tissue during 19F MRS studies of halothane elimination in rats.34 They found that the 19F NMR signal from halothane decreased to 40% of its initial value by 34 min (in contrast to about 240 min as reported by Wyrwicz et al.25). Several factors may have explained this discrepancy. First, halothane elimination may be signi®cantly faster in rats when compared with rabbits. Second, the depth pulse imposed an acquisition delay of about 0.5 ms, and may have resulted in signi®cant loss of signal from the short T2 component. In a separate paper,35 the same group reported data on iso¯urane elimination in rabbits with two protocols, both obtained with pulse-and-collect sequences (Figure 6). In the ®rst set of experiments they used

Figure 3 A representative time course of decay of the 19F signal from halothane during recovery from anesthesia. Spectra were obtained during a pulse-and-collect experiment with a surface coil placed over the head of a rabbit after a 120-min exposure to 1% halothane. All signal intensities are expressed as a percentage of the initial signal obtained. (Reproduced with permission from A. M. Wyrwicz, C. B. Conboy, B. G. Nichols, K. R. Ryback, and P. Eisle, Biochim. Biophys. Acta, 1987, 929, 271)

methyl resonance of unmetabolized halothane.25 Furthermore, this theory could not explain the fact that in a similar experiment Wyrwicz et al.27 observed similar two-compartment elimination for iso¯urane (Figure 5), a compound that is only minimally (<1%) metabolized.28 The initial decay phase for both halothane and iso¯urane were similar (t1/2 of 25 and 26 min, respectively), but the later elimination of halothane was signi®cantly slower than iso¯urane (t1/2 of 320 and 174 min, respectively). The authors suggested that the slow, late elimination of halothane could be attributed to its greater tissue solubility and the presence of nonvolatile metabolites. While the two pharmacokinetic compartments might represent brain tissue with varying perfusion (e.g. gray and white matter), the t1/2 values were at odds with data obtained from non-NMR techniques. Cohen et al.29 administered labeled halothane intravenously, and found minimal (<10%) retention of radiolabel after 20 min. Similarly, Wolff30 found that 97% of a subanesthetic dose of halothane was eliminated within 3 h. However, it has been argued by Topham and Longshaw31 that the pharmacokinetics of halothane may be signi®cantly dependent on the route of administration and the dose administered. Consequently, the preceeding two experiments may be irrelevant to the process of clinical anesthesia. Indeed, Divakaran et al.32 used gas chromatography to measure halothane levels in excised brain, and found elimination rates that correlate well with the ®gures obtained by Wyrwicz et al. However, Strum et al.33 used gas chromatography to measure residual concentrations of iso¯urane in rabbit tissue following 90 min of 1.3% iso¯urane administration, and showed that brain concentrations of iso¯urane were reduced to about 10% by 90 min. In an effort to address these discrepancies, the research group at the University of California studied elimination of halothane and iso¯urane in animal models. They hypothesized that much of the delayed elimination in published in vivo studies arose from contaminating signals in extracranial tissue,

360 min

180 min

90 min

45 min

0 min

4

2

0

–2

–4 ppm

Figure 4 High-resolution 19F MRS experiments on excised rabbit brain removed after varying periods of recovery (between 0 and 6 h) following halothane anesthesia. Chemical shifts are reported relative to an external 5 mM tri¯uoroacetic acid standard. Note the appearance of a new resonance in the 90-min spectrum, attributed to tri¯uoroacetic acid. (Reproduced with permission from A. M. Wyrwicz, C. B. Conboy, B. G. Nichols, K. R. Ryback, and P. Eisle, Biochim. Biophys. Acta, 1987, 929, 271)

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA 100

% isoflurane

50

10 5 40

80

120 160 200 240 280 320 360 400 440 480 Time (min)

Figure 5 A representative time course of decay of the 19F signal from iso¯urane during recovery from anesthesia. Spectra were obtained during a pulse-and-collect experiment with a surface coil placed over the head of a rabbit after a 90-min exposure to 1.5% iso¯urane. All signal intensities are expressed as a percentage of the initial signal obtained. (Reproduced with permission from A. M. Wyrwicz, C. B. Conboy, K. R. Ryback, B. G. Nichols, and P. Eisele, Biochim. Biophys. Acta, 1987, 927, 86)

% maximum isoflurane intensity

a 3 cm surface coil over the scalp, similar to that employed by Wyrwicz et al.,27 and found similar kinetics for iso¯urane elimination. In another experiment they placed a 1 cm surface coil directly over the exposed dura in rabbits. This study revealed a monoexponential decay of iso¯urane concentrations in the brain, with a reduction to 15% of initial values by 90 min. These data agreed well with ®ndings of Strum et al.,33 who used non-NMR methods and showed a reduction to 10% of initial values by 90 min. The University of California group concluded that the later prolonged elimination kinetics for iso-

100 3 cm coil over scalp

50 1 cm coil on brain 10 5.6 3.1 0

50

100 150 200 250 300 350 400 Minutes after inspired isoflurane discontinued

450

Figure 6 Time course of decay of the 19F signal from iso¯urane during recovery from anesthesia. Spectra were obtained during a pulseand-collect experiment with a surface coil placed over the head of a rabbit after a 90-min exposure to 1.5% iso¯urane. All signal intensities are expressed, on a log scale, as a percentage of the initial signal obtained. (*) Signal obtained from a 1.0 cm diameter surface coil placed directly over the dura in craniectomized animals, ®tted to a single exponential decay. (*) Signal obtained from a similar experiment, acquired using a 3.0 cm diameter surface coil placed over the intact scalp of the rabbit; the line represents the optimum leastsquares ®t for a biexponential decay. (Reproduced with permission from P. Mills, D. I. Sessler, M. Moseley, W. Chew, B. Pereora, T. L. James, and L. Litt, Anesthesiology, 1987, 67, 169)

5

¯urane observed by Wyrwicz et al. arose from signal contamination by iso¯urane in extracerebral tissues. The papers discussed above show that the issue of the spatial localization of the 19F signal from anesthetic agents is an important one. Attempts have been made to clarify this point using 19F MRI studies. While both spin echo36 and gradient echo37 sequences have been used to image halothane and en¯urane in anesthetized animals, the echo times for these imaging studies have been 9 and 10 ms, respectively. As the T2 of volatile anesthetics in the brain is short (typically 3±4 ms),25 it is not surprising that these studies were unable to demonstrate any anesthetic in the brain, despite being conducted while the animal was still anesthetized. On the other hand, they clearly demonstrate fairly high concentrations of ¯uorinated agent in extracerebral tissues. When these distributions were compared with either 1H lipid imaging36 or 13C imaging,37 it appeared that the anesthetics localized in lipid rich areas in adipose tissue stores, glands, and muscle. Localized spectroscopy of volatile anesthetic agents therefore demands techniques that do not involve appreciable delays in data acquisition. Wyrwicz and Conboy38 used rotating frame zeumatography (RFZ) to examine the distribution of halothane in the rat head. Localized spectra were obtained from animals that were allowed to recover for varying periods after 90 min of anesthesia with 1% halothane (Figure 7). The localization of signals was necessarily imperfect since RFZ localizes signals along a single axis only. However, the 19F spectra obtained immediately after anesthesia demonstrated very little variation in chemical shift across most of the rat head, but a heterogeneous distribution of 19F signal across the animal's head with apparently lower peak heights in the `brain' slices [Figures 7(a) and (b)]. The authors did not report on variations in T2 or linewidths across the blocks of spectra, so estimates of signal intensity are impossible to obtain. RFZ experiments after 4.5 and 6.5 h of recovery from anesthesia showed signals from both halothane and halothane metabolite in all areas of the head [Figures 7(c) and (d)]. With time, the proportion of 19F signal arising from halothane decreased, but this reduction followed a heterogeneous pattern across the head with more rapid reductions being seen in the `brain' slices. Nevertheless, spectra obtained at 6.5 h (390 min) continued to show clearly detectable peaks from halothane in `brain' slices. As the metabolite signal contribution increased, its distribution also changed, with a larger proportion of the 19F signal arising from the brain. By 10 h after anesthesia, virtually all the 19F signal arose from the metabolite resonance, and was mainly con®ned to `brain' slices. No information regarding relative changes in signal intensity over time in different regions was provided, so even crude estimates of regional elimination hal¯ives are impossible. In a more recent study, Lockhart et al.39 measured the cerebral uptake and elimination of halothane, iso¯urane, and des¯urane from rabbit brain, using a 1 cm surface coil positioned over the exposed dura. They found that rates of change of cerebral concentrations of anesthetic paralleled alveolar levels, and both cerebral uptake and elimination of des¯urane were 3 times as fast as for halothane and 1.7 times as fast as for iso¯urane (Figure 8). These ®ndings would ®t with standard clinical and pharmacokinetic premises, since des¯urane has a rapid onset and offset of action owing to its lower blood and tissue solubility. The data for the elimination of iso¯urane were similar to those

6 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA

4

(a)

2 5 3

Intensity

3

1

1

16

32 Distance (block no.)

48

64

2 Coil plane

(b) 1

4

64

3 3

48 bl

oc

kn o.)

5

Di

sta

nc

e(

32

16

1 Chemical shift Metabolite

Halothane

(c)

Coil plane 64

e(

Di

sta

nc

32

bl oc

kn

o.)

48

16

obtained from the previous University of California study, with a monoexponential decay and only 10% of initial levels retained in the brain 90 min after discontinuation of iso¯urane. The data also clearly demonstrate a hysteresis between alveolar and cerebral levels of all three volatile agents; alveolar concentrations were higher during the uptake phase, while cerebral concentrations were higher during the elimination phase (Figure 9). This is consistent with conventional pharmacokinetic modeling. However, it was found that the degree of hysteresis was identical for all three agents, suggesting that the rates of transfer between brain and blood are governed by factors other than blood and tissue solubility. Chen et al.40 studied elimination of iso¯urane from the rat brain as a function of age using 19F MRS with a surface coil. They found a biphasic elimination of iso¯urane in both age groups with t1/2 values of 7±9 and 100±115 min for the fast and slow components, respectively. The t1/2 estimates for the fast compartment are consistent with data for elimination of iso¯urane from the brain obtained using other techniques. Consequently, the authors attributed the slow component to washout of iso¯urane from intracranial fatty tissue. They also found that older rats showed a slower elimination of iso¯urane than younger rats, a difference that the authors attributed to impaired cardiovascular function. What inferences can we draw from this review of 19F MRS studies of the cerebral pharmacokinetics of anesthetic agents? Clearly, 19F MRS provides a convenient technique for investigating these questions, but several issues need to be addressed. These include ef®cient spatial localization without acquisition delays, which will result in the loss of signal from short T2 components. While many of the studies that reported prolonged anesthetic residence in the brain may have been ¯awed due to contamination by signal from extracerebral tissues, three points cannot be ignored. First, many MR studies tend to suggest slower anesthetic elimination from the brain when compared to nonNMR techniques. However, two more recent studies39,40 suggest brain iso¯urane kinetics that are similar to those obtained from non-NMR techniques. Second, the RFZ studies strongly suggest the presence of residual unmetabolized anesthetic in the brain as long as 6.5 h after anesthesia.38 Finally, prolonged residence of at least some proportion of halothane is supported by the in vitro studies of brain extracts per-

1 Chemical shift

(d) Coil plane 64

16

1 Chemical shift

Di

sta

nc

bl

e(

32

oc

k

no

.)

48

Figure 7 Results of localized 19F MRS using rotating frame zeumatography performed following administration of 1.5% halothane for 90 min. (a) One-dimensional projection across the rabbit head (from above downwards) showing the location of different blocks of spectra: 1, soft tissue (skin and muscle); 2, brain, muscle, and calvarium; 3, brain; 4, brain, muscle, and calvarium; 5, tissue below the cranium. (b) Stacked, spatially resolved spectra obtained immediately after discontinuing anesthesia, blocks are numbered as in (a). (c) Stacked, spatially resolved spectra obtained 4.5 h after cessation of anesthesia. Note the appearance of a new resonance to the left of the halothane resonance, which is attributed to tri¯uoroacetic acid. (d) Stacked, spatially resolved spectra obtained 6.5 h after cessation of anesthesia. Note the decrease in intensity of the halothane resonance, and the increase in the intensity of the metabolite resonance. (Reproduced with permission from A. M. Wyrwicz and C. B. Conboy, Magn. Reson. Med., 1989, 9, 219)

7

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA

0.1 Halothane Isoflurane 0.01

Desflurane

0.001

Initial brain concentration (CC/FAO) (%)

0

20

40 60 80 Wash-out time (min)

100

Proportion of maximal cerebral concentration

(a)

0.8

0.6

Halothane Desflurane

0.4

Isoflurane

0.2

120

0.0 0.0

1

(b)

0.1 Halothane

0.2 0.4 0.6 0.8 Proportion of maximal alveolar concentration

1.0

Figure 9 Plot of the relationship between cerebral and alveolar concentrations of volatile anesthetics during induction of and recovery from anesthesia (lower and upper halves of ellipsoid, respectively). Both parts of the curve deviate from the line of identity (straight line), suggesting a signi®cant hysteresis between alveolar and brain levels of anesthesia. Note the unexpected result showing that all three anesthetic agents studied seem to ®t to the same curve. (Reproduced with permission from S. H. Lockhart, Y. Cohen, N. Yasuda, B. Freire, S. Taheri, L. Litt, and E. I. Eger II, Anesthesiology, 1991, 74, 575)

Isoflurane Desflurane

0.01 0

20

40 60 80 Wash-out time (min)

100

120

Figure 8 (a) Alveolar washout obtained from gas chromatography, and (b) cerebral washout obtained from in vivo 19F MRS of anesthetic agent following anesthesia with des¯urane (&), iso¯urane (*), and halothane (~). (Reproduced with permission from S. M. Lockhart, Y. Cohen, N. Yasuda, B. Freire, S. Taheri, L. Litt, and E. I. Eger II, Anesthesiology, 1991, 74, 575)

formed by Wyrwicz et al.25 However, as several of the later studies have shown, this may be an extremely small proportion of the initial concentrations. These small quantities of residual cerebral anesthetic are important for two reasons. First, they may provide an explanation for the minor but prolonged psychomotor de®cits that are recognized to persist for up to days after general anesthesia.23 Second, elucidation of the mechanisms that result in prolonged anesthetic retention may provide clues regarding the mechanisms of anesthetic action. 3.3 Sites of Anesthetic Residence in the Brain: Physical Environment and its Signi®cance, Saturability Many authors have reported that volatile anesthetics in the brain exhibit shorter T2 values than when they are dissolved in lipid solvents or adipose tissue. These ®ndings have been interpreted as implying immobilization of the ¯uorinated anesthetic molecules, and have led to speculation that the short T2 en-

vironment may be intimately related to the process of anesthesia. These latter speculations were further fuelled by the results of a study by Evers et al.41 in which T2 values of four ¯uorinated anesthetic agents (methoxy¯urane, en¯urane, iso¯urane, and ¯uoroxene) and one nonanesthetic ¯uorinated hydrocarbon (hexa¯uoroethane) in the brain were compared. The T2 values of the four anesthetic agents were signi®cantly shorter than in adipose tissue (0.5±4.5 versus 200±400 ms, respectively) and correlated linearly with the anesthetic potency of these agents (Figure 10). The nonanesthetic ¯uorocarbon exhibited a substantially higher T2 (18.5 ms) in brain tissue. 5 Methoxyflurane Isoflurane Enflurane Fluroxene

4

T2 (ms)

Initial alveolar concentration (FA/FAO) (%)

1.0 1

3 2 1 0 0

1

2 ED50 (vol%)

3

4

Figure 10 Relationship of anesthetic potency (measured as ED50) to 19 F spin±spin relaxation times of ¯uorinated volatile anesthetics in the brain. (Reproduced with permission from A. S. Evers, J. C. Haycock, and D. A. d'Avignon, Biochem. Biophys. Res. Commun., 1988, 151, 1039)

8 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA 1.8 6

75 1.0 50 0.6 25

5 3% ¥ 1 h ln (signal density)

Brain halothane (mM)

1.4

Area [19F ]halothane resonance (arb.units)

100

4

3

2 3% ¥ 2 min 1

0.2 12

0.0 0

1

2 3 Inspired halothane (vol%)

6

0

0

0

4

Figure 11 Brain halothane concentrations in rats as a function of inspired halothane concentrations, measured using either in vivo 19F MRS (*) or ex vivo 19F MRS (*) of excised brain. The horizontal dashed line indicates the half-maximal concentration of brain halothane concentration achieved in this experiment. This value (about 1.2%) corresponds closely to the ED50 for halothane in rats. The inset shows a representative 19F MR spectrum from the ex vivo experiments. The resonance at 0 ppm arises from methoxy¯urane in an external standard, while the peak at 10.2 ppm corresponds to halothane in the brain. (Reproduced with permission from A. S. Evers, B. A. Berkowitz, and D. A. d'Avignon, Nature, 1987, 328, 157)

These data were interpreted as supporting the premise that the short T2 environment represented a molecular site of anesthetic action. In a later study, Evers et al.42a studied spontaneously breathing rats, anesthetized with halothane, using high resolution in vitro 19F MRS of excised brain at different anesthetic concentrations and in vivo MRS of whole animals at 4.7 T. These experiments showed increasing cerebral levels of anesthetic with increasing inspired concentrations of halothane up to 2.5%. However, little or no further increases in cerebral halothane concentrations were observed when inspired halothane concentrations were increased from 2.5% to 4% (Figure 11). These data were interpreted as evidence for abundant saturable binding sites for halothane, a concept that did not support a nonspeci®c hydrophobic site of action of volatile anesthetic agents. Evers et al. also measured the T2 relaxation parameters for the brain halothane resonance, and found that there were at least two components, with one T2 value of about 3.6 ms (short T2) and a second one of about 43 ms (long T2). They also showed that the short T2 site represented 85% of the halothane molecules present in the brain, and was preferentially occupied during the initial period of induction of anesthesia (Figure 12). Continuing anesthetic administration beyond this point seemed to occupy the long T2 site. These ®ndings were thought to provide further proof for the importance of the short T2 site in the process of anesthesia. Unfortunately, these conclusions had to be revised owing to further experimental results.

0

8

16

24 t (ms)

32

40

Figure 12 Decay of the 19F MR signal from brain halothane in rats as a function of echo time (expressed as ) in a Carr±Purcell± Meiboom±Gill sequence from rats exposed to 3% halothane for 2 min (*) or 1 h (*). Note the biexponential T2 decay of the signal from rats exposed to halothane for 1 h, and the minimal increase in the amount of signal with short T2 characteristics with longer exposure. (Reproduced with permission from A. S. Evers, B. A. Berkowitz, and D. A. d'Avignon, Nature, 1987, 328, 157)

First, both Evers et al.42b and Litt et al.43 reported that the apparent saturability of brain halothane concentrations was a physiological artifact, being the consequence of respiratory depression at high inspired concentrations of halothane in spontaneously breathing animals. When the studies were repeated with arti®cially ventilated animals, the cerebral anesthetic concentrations scaled linearly with inspired concentrations of halothane (Figure 13). Furthermore, the short T2 environment was not speci®c to brain,44 but was also found in other tissues, including red blood cells, liver, and kidney, while the long T2 behavior was exhibited by halothane in serum. These more recent results have prompted a thorough review of the original paper by Evers et al.,41 and have led to a reappraisal of the pharmacological relevance of the T2 behavior of volatile anesthetic agents in the brain. The area clearly requires further study. 3.4

Metabolism of Fluorinated Agents

Many of the previous 19F MRS studies of halothane provided some data on the production of metabolites of the agent. Other groups have used the technique to study directly the metabolism of volatile anesthetics in the liver.46±48 This issue is of considerable importance, since reactive metabolites of halothane and other agents are thought to act as haptens and lead to rare but fatal immunologically mediated liver failure.49 For example, Preece et al.50 studied the metabolism of en¯urane in the liver of rats anesthetized with thiobarbitone. They demonstrated an elimination t1/2 of 76 min for en¯urane in the rat liver, and showed that this ®gure could be reduced to 39

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA (a) Inspired concentrations

Normalized peak area (% of initial total)

100 6.4% 4.8%

5.6% 4.0% 3.2% 2.4% 1.6% 0.8%

50

(b) Inspired concentrations

9

40

t1/2 = 76 min

t1/2 = 39 min 10

t1/2 = 74 min t1/2 = 29 min 1 0

30 ppm

30

60 90 Time (min)

120

150

Figure 14 Decrease in the normalized signal intensities (as a percentage of the initial total) of en¯urane in the liver with time. The plots show the decay of the major and minor resonances in control (&, ^) and isoniazid pretreated rats (*, ~). Points and bars represent the mean ‹ SEM of four rats; the calculated t1/2 values for signal decay are also shown. (Reproduced with permission from N. E. Preece, J. Challands, and S. C. Williams, NMR Biomed., 1992, 5, 101)

9.3% 7.0% 6.0% 4.5%

1.5%

60

50

40

30

20 ppm

Figure 13 Superimposed 19F MR spectra of brain iso¯urane obtained at 4.7 T at different inspired concentrations of the agent, in spontaneously breathing (a) and arti®cially ventilated (b) animals. The apparent saturability of brain iso¯urane at high inspired concentrations seen in (a) is a re¯ection of respiratory depression, since the spectra in (b) show a linear increase in brain halothane levels with inspired concentration. (Reproduced with permission from S. H. Lockhart, Y. Cohen, N. Yasuda, F. Kim, L. Litt, E. I. Eger II, L.-H. Chang, and T. James, Anesthesiology, 1990, 73, 455)

min by pretreatment with isoniazid, which is known to enhance hepatic metabolism of en¯urane (Figure 14). Similar studies may be useful in delineating pharmacogenetic differences in metabolism between individuals, or in elucidating the effects of drugs that enhance or inhibit the metabolism of volatile anesthetic agents. Data obtained from such studies may be quantitatively different, as in the studies by Preece et al.,50 or show qualitative differences in metabolic pathways in different individuals or induced by different agents. Such processes could result in an increase or decrease in the production of toxic metabolites. 3.5 Clinical Studies Clinical 19F MRS studies of volatile anesthetics may have substantial advantages, since they enable the assessment of

subtle cognitive dysfunction produced by trace residual anesthetic concentrations, and may be technically easier since the human head is larger, and the improved signal-to-noise ratio may enable more effective and speci®c volume selection. However, such studies also present considerable logistic and ethical problems. Menon et al.51 reported the use of 19F MRS to study halothane in the brain of eight patients recovering from halothane anesthesia of short duration. A 6 cm surface receiver coil was used, and resonances attributable to halothane were observed in these patients up to 90 min after withdrawal of the anesthetic agent (Figure 15). Localized spectroscopy with two-

400 Halothane signal (arb.units)

3.0%

350 300 250 200 150 100 50 0 0

10

20 30 40 50 60 70 Time from end of anaesthetic (min)

80

Figure 15 Decay of the 19F signal intensity (arbitrary units) of halothane resonance with time after anesthesia in a patient where a single 19F resonance was observed, and localized to the brain using a two-dimensional CSI sequence. (Reproduced with permission from D. K. Menon, G. G. Lockwood, C. J. Peden, I. J. Cox, J. Sargentoni, and J. D. Bell, Magn. Reson. Med., 1993, 30, 680)

10 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA 1

(a)

2

0 5 ppm/division (b)

3

2

ratio for an unlocalized spectrum was typically 20 with data collection times of 2 min. In seven patients a single resonance was seen with a mean (‹ SD) chemical shift of +43.3 (‹ 1.8) ppm, referenced to NaF at 0 ppm [Figure 16(a)]. This chemical shift is comparable to that observed in animal studies of halothane anesthesia. This resonance exhibited a T1 value of 0.5±1 s, and a T2* (estimated from the linewidth of the resonance) value of 3.5±10 ms. In one patient two resonances were observed with chemical shifts of +38 and +41 ppm [Figure 16(b)]. These two resonances may represent halothane in two different chemical environments within the brain, since phantom studies showed that the chemical shift of halothane in different environments (such as water, olive oil, methanol, and lecithin) could vary to an extent that accounted for the two resonances seen. The two resonances observed in this patient had different T1 values and linewidths (and, by implication, different T2* values). Intriguingly, the two-dimensional CSI localization in this patient suggested that the broader resonance at +38 ppm was con®ned to the brain, while the narrow resonance at +41 ppm arose from overlying scalp (Figure 17). Although susceptibility effects could not be excluded, it is at least possible that these two resonances represent halothane in two different tissues. While the results discussed above do not resolve the problems discussed in earlier sections, they do demonstrate the feasibility of in vivo 19F MRS studies of ¯uroinated volatile agents in humans, and con®rm several ®ndings observed in animal studies. In further studies, Lockwood et al.52,53 have used 19F MRS to study the pharmacokinetics of iso¯urane in volunteers. They found two-compartment kinetics within the a b c

(a) Brain

(b) Brain and scalp 0 5 ppm/division

Figure 16 Spectra acquired from the whole sensitive volume of a surface receiver coil in two patients recovering from halothane anesthesia. In both spectra the large resonance at 0 ppm arises from an external NaF standard. (a) The single resonance at +43 ppm is from halothane. (b) The two resonances at +38 and +41 ppm are clearly seen in the inset, where the peaks are enlarged. (Reproduced with permission from D. K. Menon, G. G. Lockwood, C. J. Peden, I. J. Cox, J. Sargentoni, and J. D. Bell, Magn. Reson. Med., 1993, 30, 680)

dimensional chemical shift imaging (CSI) showed that, although some of the signal arose from extracranial tissues, a substantial proportion arose from brain. The signal-to-noise

(c) Scalp

5 ppm/division

Figure 17 Spectra from three contiguous planes of the head acquired during the course of a two-dimensional CSI sequence, the location of which is shown on the line diagram of a transverse image of the head. Note that the resonances from super®cial and deep slices have different chemical shifts (+38 and +41 ppm, respectively) and linewidths. (Reproduced with permission from D. K. Menon, G. G. Lockwood, C. J. Peden, I. J. Cox, J. Sergentoni, and J. D. Bell, Magn. Reson. Med., 1993, 30, 680)

FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA

Isoflurane partial pressure (kPa)

(a) 1.2

4

RELATED ARTICLES

Brain MRS of Human Subjects; Cation Movements across Cell Walls of Intact Tissues Using MRS; pH Measurement In Vivo in Whole Body Systems; Temperature Measurement Using In Vivo NMR; Whole Body Studies: Impact of MRS.

1.0 0.8 0.6 0.4

5

0.2 0

(b) 200

MRS peak height

11

150

100

50

0 0

10

20

30 40 50 Time (min)

60

70

80

Figure 18 Iso¯urane anesthesia in a volunteer during wash-in and wash-out.53 The subject was unresponsive to verbal stimuli during the period between the two vertical dotted lines. (a) Partial pressure of iso¯urane measured in inspired gas (upper continuous line), expired gas (lower continuous line) and arterial blood (solid circles). (b) Actual 19 F MRS data in arbitrary units (solid circles) and the summed intensities from the two-compartment model calculated for the data (continuous line). The fast (broken line) and slow (dotted line) components of the model are calculated from the MR data. In (a) the calculated fast-compartment trace is largely identical with the experimental data and is not visible. (Reproduced by kind permission of Oxford University Press from G. G. Lockwood, D. P. Dob, D. J. Bryant, J. A. Wilson, J. Sargentoni, S. M. Sapsed-Byrne, D. N. Harris and D. K. Menon, Magnetic resonance spectroscopy of iso¯urane kinetics in humans, Part II: Functional localisation. British Journal of Anaesthesia, 79, 586±9 (1999).)

head with equilibrium half-times of 3.5 min and approximately 1 h with respect to expired iso¯urane concentrations.52 Using critical fusion ¯icker frequency as an objective measure of the cerebral effect of iso¯urane, they found evidence to identify the fast component as in brain tissue. Responsiveness to command was lost at a brain partial pressure of 0.3% iso¯urane.52 It was concluded that these measurements exactly matched the predictions of the classical perfusion-limited model. These compartments showed decay half-times of 9.5 and 130 min, but the signal was too weak to localize the compartments spatially. If the fast compartment is assumed to be the brain then these results match the predictions of the classical perfusionlimited pharmacokinetic model of inhalation anesthesia (Figure 18).53

REFERENCES

1. M. C. Malet-Marino and R. Martino, Biochimie, 1992, 74, 785. 2. P. M. Joseph, J. E. Fishman, B. Mukherji, and H. A. Sloviter, J. Comput. Assist. Tomogr., 1985, 9, 1012. 3. A. R. Jacobsons, L. A. Sylvia, D. L. Veenstra, and J. T. Gerig, J. Magn. Reson., 1992, 96, 387. 4. C. N. Karson, J. E. Newton, P. Mohanakrishnan, J. Sprigg, and R. A. Komoroski, Psychiatr. Res., 1992, 45, 95. 5. R. A. Komoroski, J. E. O. Newton, D. Cardwell, J. Sprigg, J. Pearce, and C. N. Karson, Magn. Reson. Med., 1994, 31, 204. 6. D. J. Lee, C. T. Burt, and R. L. Koch, J. Invest. Dermatol., 1992, 99, 431. 7. T. Nakada, I. L. Kwee, B. V. Griffey, and R. H. Griffey, Magn. Reson. Imag., 1988, 6, 633. 8. T. Nakada, I. L. Kwee, B. V. Griffey, and R. H. Griffey, Radiology, 1988, 168, 823. 9. T. Nakada, I. L. Kwee, P. J. Card, N. A. Matwiyoff, B. V. Griffey, and R. H. Griffey, Magn. Reson. Med., 1988, 6, 307. 10. J. T. Grieg, Methods Enzymol., 1989, 177, 1. 11. L. C. Clark Jr, J. L. Ackerman, S. R. Thomas, R. W. Millard, R. E. Hoffman, R. G. Pratt, H. RagleCole, R. A. Kinsey, and R. Janakiraman, Adv. Exp. Med. Biol., 1984, 180, 835. 12. B. A. Berkowitz, J. T. Handa, and C. A. Wilson, NMR Biomed., 1992, 5, 65. 13. J. S. Beech and R. A. Iles, Magn. Reson. Med., 1991, 19, 386. 14. H. S. Bachelard, R. S. Badar-Goffer, K. J. Brooks, S. J. Dolin, and P. G. Morris, J. Neurochem., 1988, 51, 1311. 15. R. Badar-Goffer, P. Morris, N. Thatcher, and H. Bachelard, J. Neurochem., 1994, 62, 2488. 16. P. M. Joseph, Y. Yuas, H. L. Kundel, B. Mukherji, and H. A. Sloviter, Invest. Radiol., 1985, 20, 504. 17. M. Rudin and A. Sauter, NMR Biomed., 1989, 2, 98. 18. J. Pekar, L. Ligeti, T. Sinnwell, C. T. Moonen, J. A. Frank, and A. C. McLaughlin, J. Cereb. Blood Flow Metab., 1994, 14, 656. 19. J. K. Ali®moff and K. W. Miller, `Principles and Practice of Anesthesiology', M. C. Rogers, J. H. Tinker, B. G. Covino, and D. E. Longnecker ed., Mosby, St Louis, 1993, pp. 1035±1052. 20. K. W. Miller, Int. Rev. Neurobiol., 1985, 27, 1. 21. N. P. Franks and W. R. Lieb, Nature, 1994, 367, 607. 22. D. E. Longnecker and F. L. Miller, `Principles and Practice of Anesthesiology', ed. M. C. Rogers, J. H. Tinker, B. G. Covino, and D. E. Longnecker, Mosby, St Louis, 1993, pp. 1053±1086. 23. M. Herbert, T. E. J. Healy, J. B. Bourke, I. R. Fletcher, and J. M. Rose, Br. Med. J., 1983, 286, 1539. 24. A. Wyrwicz, M. H. Pszenny, J. C. Scho®eld, P. C. Tillman, R. E. Gordon, and P. A. Martin, Science, 1983, 222, 428. 25. A. M. Wyrwicz, C. B. Conboy, B. G. Nichols, K. R. Ryback, and P. Eisle, Biochim. Biophys. Acta, 1987, 929, 271. 26. E. I. Eger II, `Anaesthetic Uptake and Action', Williams & Wilkins, Baltimore, 1974, p. 230. 27. A. M. Wyrwicz, C. B. Conboy, K. R. Ryback, B. G. Nichols, and P. Eisele, Biochim. Biophys. Acta, 1987, 927, 86. 28. J. H. Lewis, H. J. Zimmerman, K. G. Ishak, and F. G. Mulluk, Ann. Int. Med., 1983, 98, 984. 29. E. N. Cohen, K. L. Chow, and L. M. Mathers, Anesthesiology, 1972, 37, 324.

12 FLUORINE-19 MRS: GENERAL OVERVIEW AND ANESTHESIA 30. M. S. Wolff, J. Toxicol. Environ. Health, 1977, 2, 1079. 31. J. C. Topham and S. Longshaw, Anesthesiology, 1972, 37, 311. 32. P. Divakaran, F. Joiner, B. M. Rigor, and R. C. Wiggins, J. Neurochem., 1980, 34, 1543. 33. D. P. Strum, B. H. Johnson, and E. I. Eger II, Science, 1986, 234, 1586. 34. L. Litt, R. Gonzalez-Mendez, T. L. James, D. I. Sessler, P. Mills, W. Chew, M. Moseley, B. Pereira, J. W. Severinghaus, and W. K. Hamilton, Anesthesiology, 1987, 67, 161. 35. P. Mills, D. I. Sessler, M. Moseley, W. Chew, B. Pereora, T. L. James, and L. Litt, Anesthesiology, 1987, 67, 169. 36. T. L. James, L.-H. Chang, W. Chew, R. Gonzalez-Mendez, L. Litt, P. Mills, M. Moseley, B. Pereira, D. I. Sessler, and P. R. Weinstein, Ann. NY. Acad. Sci., 1987, 508, 64. 37. T. Hashimoto, H. Ikehira, H. Fukuda, Y. Ueshima, and Y. Tateno, Magn. Reson. Imag., 1991, 9, 577. 38. A. M. Wyrwicz and C. B. Conboy, Magn. Reson. Med., 1989, 9, 219. 39. S. H. Lockhart, Y. Cohen, N. Yasuda, B. Freire, S. Taheri, L. Litt, and E. I. Eger II, Anesthesiology, 1991, 74, 575. 40. M. Chen, J. I. Olsen, J. A. Stolk, M. P. Schweizer, M. Sha, and I. Ueda, NMR Biomed., 1992, 5, 121. 41. A. S. Evers, J. C. Haycock, and D. A. d'Avignon, Biochem. Biophys. Res. Commun., 1988, 151, 1039. 42. (a) A. S. Evers, B. A. Berkowitz, and D. A. d'Avignon, Nature, 1987, 328, 157; (b) correction: 1989, 341, 766. 43. S. H. Lockhart, Y. Cohen, N. Yasuda, F. Kim, L. Litt, E. I. Eger II, L.-H Chang, and T. James, Anesthesiology, 1990, 73, 455. 44. H. J. C. Yeh, E. J. Moody, and P. Skolnick, Nature, 1990, 346, 227. 45. H. Zimmerman, Hepatology, 1991, 13, 1251.

46. D. S. Selinsky, M. Thompson, and R. E. London, Biochem. Pharmacol., 1987, 36, 413. 47. B. S. Selinsky, M. E. Perlman, and R. E. London, Mol. Pharmacol., 1988, 33, 559. 48. B. S. Selinsky, M. E. Perlman, and R. E. London, Mol. Pharmacol., 1988, 33, 567. 49. H. Zimmerman, Hepatology, 1991, 13, 1251. 50. N. E. Preece, J. Challands, and S. C. R. Williams, NMR Biomed. 1992, 5, 101. 51. D. K. Menon, G. G. Lockwood, C. J. Peden, I. J. Cox, J. Sargentoni, G. A. Coutts, and J. G. Whitesam, Magn. Reson Med., 1993, 30, 680. 52. G. G. Lockwood, D. P. Dob, D. J. Bryant, J. A. Wilson, J. Sargentoni, S. M. Sapsed-Byrne, D. N. Harris, and D. K. Menon, Br. J. Anaesth., 1997, 79, 581. 53. G. G. Lockwood, D. P. Dob, D. J. Bryant, J. A. Wilson, J. Sargentoni, S. M. Sapsed-Byrne, D. W. Harris, and D. K. Menon, Br. J. Anaesth., 1997, 79, 586.

Biographical Sketch David Krishna Menon. b 1956. M.B.B.S. 1977, M.D. 1982, M.R.C.P. 1984, F.R.C.A. 1988, Ph.D. 1995, F. Ac. Med. Sci. UK 1998, F.R.C.P. 1999; internal medicine residency, Jawaharlal Institute, Pondicherry, India 1978±1983; Medical Registrar, Professorial Medical Unit, Leeds General In®rmary 1985±86; Registrar in Anaesthesia, Royal Free Hospital, London 1987±1988; MRC Research Fellow, NMR Unit, Hammersmith Hospital 1989±91. Presently: lecturer in anaesthesia, Cambridge University, and Director, Neurosciences Critical Care Unit, Addenbrooke's Hospital, Cambridge, UK. Current research interests: metabolic imaging of acute brain injury and brain in¯ammation following ischemia and trauma.

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Functional MRI at High Fields: Practice and Utility Kamil Ugurbil, Wei Chen, Xiaoping Hu, Seong-Gi Kim, Xiao-Hung Zhu Center for Magnetic Resonance Research, University of Minnesota, MN, USA

and Seiji Ogawa Bell Laboratories, Lucent Technologies, Murray Hill, NJ, USA

1 INTRODUCTION In the middle of the 19th century, a central debate about brain function revolved about the issue of whether the human brain processed all of its task as a single entity or function was compartmentalized. At about that time, an argument for the existence of regional specialization of human brain function was presented by Pierre Paul Broca.1 Broca examined a patient who was unable to speak as a result of a stroke but was otherwise normal. Based on an autopsy performed subsequent to the patient's death, Broca concluded that the seat of the damage was an egg-sized lesion located in the inferior frontal gyrus of the frontal lobe in the left hemisphere; this general area is now commonly referred to as Broca's area, although its precise topographical extent remains ambiguous. This type of study and, later, intraoperative mapping efforts with electrodes were, until recently, the primary source of our information on functional compartmentation in the human brain. Recent techniques using NMR permitted the acquisition of such information much more rapidly and with greater spatial accuracy, fueling explosive developments in the investigation of human brain function. For example, the language area ®rst identi®ed by Broca can now be visualized with unprecedented spatial resolution using functional MRI (fMRI), in data collection times that last only a few minutes. Figure 1 displays the three-dimensional result of such a study.2 The functional map generated is superimposed on the anatomical image during a covert language task. In these ®gures, the gray scale anatomic images are either opaque, permitting the visualization of activation only on the outer cortical surface, or rendered partially transparent so that activation in the interior of the brain and within its numerous folds (sulci) are visible, albeit with diminished intensity. These images are based on BOLD (blood oxygen leveldependent contrast), ®rst described by Ogawa in rat brain studies3±5 and subsequently applied to generate functional images in human brain.6±8 Today, images like those shown in Figure 1 can be generated by the 1.5 T scanners that are often found in hospitals, provided that the scanner is equipped with appropriate hardware to perform fast imaging. These are relatively `low' resolution images, in the several millimeter spatial domain, obtained with averaging over many executions of the

1

same task by a single individual. However, there have been studies beyond this type of imaging using signi®cantly higher magnetic ®elds. Shortly before the introduction of BOLD-based fMRI, efforts were initiated in three laboratories, Universities of Minnesota and Alabama, and the National Institutes of Health, to explore the possibility of using high magnetic ®elds (4 T) for human MRI. However, these high-®eld studies were initially met with skepticism, and the possibility of human imaging at magnetic ®elds much higher than 1.5 T was seriously questioned. This skepticism was not based on the existence of any experimental evidence; rather, it followed from concepts and theoretical considerations regarding the interaction of high-frequency electromagnetic waves with the conductive human body. These considerations had led prominent investigators in the ®eld of MRI research, as early as 1979, to suggest that human imaging would not be possible beyond 10 MHz (~0.24 T).9 Of course, even clinical imaging is now performed at magnetic ®elds much higher than 0.24 T (up to 1.5 T), and recent work at 3 and 4 T has demonstrated that exquisite anatomical and functional imaging of the human head is achievable at these high ®elds. The efforts toward the development of fMRI at our site was performed only at 4 T, coinciding with the effort to introduce and explore the use of these high magnetic ®elds. It is generally stated that the contrast-to-noise ratio (CNR) for the BOLD effect and hence for detection of activated areas increases at high ®elds, hence higher ®elds are better for fMRI. This may indeed be the case for some ®eld strengths. A clear demonstration of this is illustrated in Figure 2, where mapping of the various functionally distinct regions within the visual cortex at 1.5 and 3 T depict larger areas of activation at the higher ®eld strength. These are images of the ¯attened cortex. Similar functional maps have been published by Tootell and colleagues previously, mainly at 1.5 T.10±15 However, the ®eld dependence of the BOLD-based fMRI is rather complex. At high ®elds, for example, the fractional signal change coupled to alterations in neuronal activity may actually become smaller than that observed at lower magnetic ®elds such as 1.5 T; however, the speci®city of signal changes detected by MR in relation to the actual site of neuronal activation may be substantially improved. Such trade-offs between CNR and speci®city will also depend on the details of data acquisition. Therefore, consideration of the BOLD mechanism is imperative in understanding the underlying ®eld dependence of fMRI. This chapter starts with such a discussion and subsequently gives a few selected examples of accomplishments that remain unique to high-®eld fMRI.

2

2.1

MECHANISTIC CONSIDERATIONS RELEVANT TO FIELD DEPENDENCE Extravascular Blood Oxygen Level-dependent Effects

Modeling of the effect of susceptibility gradients across vascular boundaries on MR signals have been considered by several groups, with similar conclusions.16±24 If one considers an in®nite cylinder as an approximation for a blood vessel with magnetic susceptibility difference
, then the magnetic ®eld expressed in angular frequency at any point in space will be perturbed from the applied magnetic ®eld !0.16 Inside the

2 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 1 Three-dimensional, whole brain images of activation during a language task based on word generation from a phoneme. Subjects were presented phonemes and were asked to think of as many English words as they could that contained the phoneme until the presentation of the next phoneme. The gray scale picture is the anatomical image. In color is the superimposed functional map. Brain images shown are (a) left hemispheres (b) right hemispheres, and (c) at an angle where the left hemisphere and top of the brain are both partially seen. The extensive frontal cortex activity depicted in (a) largely de®nes the area of Broca and extends to area 46 as well. In (a)±(c), the anatomical images are opaque; consequently only activated regions that lie predominantly on the cortical surface are seen. Image (d) is identical to (c) except that the anatomical image was rendered partially transparent in order to `look through' and see the activated regions that would normally be blocked from the view by overlapping cortex (e.g., regions within sulci). In these particular views, activity in the medial part of the brain and other areas are also apparent, albeit with diminished intensity, in addition to the extensive activity in the left hemisphere frontal cortex. (With permission from Erhard et al.2)

cylinder, the perturbation,
!B, will be given by: 2
!in B ˆ 2
0 …1 ÿ Y †!0 fcos …† ÿ 1=3g

…1†

At any point outside the cylinder, the magnetic ®eld will vary, depending on the distance and orientation relative to the blood vessel and the external magnetic ®eld direction, according to the equation: 2 2
!out B ˆ 2
0 …1 ÿ Y †!0 frb =rg sin …†cos…2†

…2†

In these equations,
0 is the maximum susceptibility difference expected in the presence of fully deoxygenated blood, Y is the fraction of oxygenated blood present, rb is the cylinder radius, and r is the distance from the point of interest to the

center of the cylinder in the plane normal to the cylinder (Figure 3). Note that outside the cylinder the magnetic ®eld changes rapidly over a distance comparable to two or three times the cylinder radius; at a distance equal to the diameter of the cylinder from the cylinder center,
!out B is already down to 25% of its value at the cylinder boundary. If such a blood vessel is present in a given voxel, the magnetic ®eld within this voxel will be inhomogeneous. The effect of this inhomogeneity in tissues can be understood in terms of dynamic and static averaging, the former arising as a result of the diffusive motion of the water molecules. First, let us ignore the blood in the intravascular space (i.e., inside the cylinder) and focus on the extravascular space only. In BOLD-based fMRI, data are collected after excitation and an echo delay time TE in a gradient or a spin echo sequence.

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

3

Figure 2 Mapping eccentricity and motion versus stationary stimulus in the visual cortex. Images are presented as a ¯attened cortex. (a), (b) Data from an experimental mapping eccentricity; (c), (d) low-contrast moving stimulus versus stationary stimulus that activates areas MT and V3a. (a), (c) Data at 3 T, with statistical signi®cances ranging from 10ÿ5 to 10ÿ15. (b), (d) Data at 1.5 T analyzed with the same statistical criteria as in (a) and (c). Comparison of (a) and (c) and of (b) and (d) show the effects of magnetic ®eld strength at the same statistical signi®cance. (Figure supplied by N. Hadjikhani and R. Tootell)

Typical TE values used depend on the ®eld strength and the speci®cs of the pulse sequence but, in general, range from ~30 to ~100 ms. If the typical diffusion distances during the delay TE are comparable to the distances spanned by the magnetic ®eld gradients, then during this delay the magnetic ®eld inhomogeneities will be dynamically time-averaged. Therefore, blood vessel size compared with the diffusion distances in this 30±100 ms time domain becomes a critical parameter in the BOLD effect (Figure 4). In this time scale, small blood vessels (e.g., capillaries) that contain deoxyhemoglobin will contribute

to the dynamic averaging and result in a signal decay that will be characterized with a change in T2 time.17,18,25 In a spin echo experiment with a single refocusing pulse in the middle of the delay period, the dynamic averaging that has taken place during the ®rst half of the echo will not be recovered. Of course, applying many refocusing pulses, as in a Carr±Purcell pulse train, or applying a large B1 ®eld (relative to the magnitude of the magnetic ®eld inhomogeneity) for spin-locking during this delay will reduce or even eliminate this signal loss through dynamic averaging. In a gradient echo measurement,

4 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY capillaries exceeds 500 ms,27±29 signi®cantly longer than the typical T2 and T2 values in the brain tissue and longer than the period TE typically employed in fMRI studies. For larger blood vessels, complete dynamic averaging for the entire voxel will not be possible. Instead, there will be `local' or `partial' dynamic averaging over a subsection of the volume spanned by the magnetic ®eld gradients generated by the blood vessel. However, there will be signal loss from the voxel through static averaging if refocusing pulses are not used or asymmetric spin echoes are employed. Following the excitation and rotation onto the plane transverse to the external magnetic ®eld, a water molecule at a given point in space relative to the blood vessel will see a `locally' time-averaged magnetic ®eld that will vary with proximity to the large blood vessel. Therefore, the signal in the voxel will then be described by: X ck eÿTE=T2k …eÿi!k TE † …3† S…t† ˆ Figure 3 A cylinder representing a blood vessel and the parameters that determine magnetic ®eld at a point outside of the cylinder when the susceptibilities inside and outside the cylinder are not the same

dynamic averaging will occur during the entire delay TE. If the imaging voxel contains only small blood vessels at a density such that one half the average distance between them is comparable to diffusion distances (as is the case in the brain where capillaries are separated on the average by 25 m26), then the entire signal from the voxel will be affected by dynamic averaging. In considering the movement of water molecules around blood vessels, we need not be concerned with the exchange that ultimately takes place between intra- and extravascular water across capillary walls. Typical lifetime of the water in

Figure 4 Dynamic and static averaging regimes based on diffusion distances relative to the size of a compartment that differs in magnetic susceptibility from the surrounding tissue. The magnetic ®eld gradients are most prominent in the vicinity of the compartments with different susceptibility, i.e., red blood cells, capillaries, and large blood vessels. For large blood vessels, diffusion distances are not large compared with vessel radius, and hence they do not lead to dynamic averaging

k

where the summation is performed over the parameter k, which designates small volume elements within the voxel, and ck is a constant; the time-averaged magnetic ®eld experienced within  k in angular frequency units. these small volume elements is ! The summation over k, therefore, covers the entire voxel.  k TE varies across the voxel, the signal will be Because !

Figure 5 The susceptibility-induced R2 (1/T2 ) in the presence of water diffusion plotted as a function of cylinder radius at three different values of frequency shifts. TE is 40 ms and the fractional `blood volume' (i.e., volume within cylinder relative to voxel volume) is 0.02. (With permission from Ogawa et al.18)

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 6 The fractional signal loss (
S/S) for spin echo versus a gradient echo image acquisition as a function of cylinder radius TE, 40 ms; frequency shift owing to susceptibility difference, 40 Hz; fractional `blood volume' (i.e., volume within cylinder relative to voxel volume), 0.02. (With permission from Ogawa et al.18)

`dephased' and diminish with increasing TE. This signal loss k occurs from static averaging. In this domain, if the variation ! over the voxel is relatively large, signal decay can be approximated with a single exponential time constant T2 . Figure 5 illustrates R2 (i.e., 1/ T2 ) as a function of the radius of the cylinder from modeling studies by Ogawa et al.18 In this calculation, all other relaxation mechanisms that can contribute to transverse relaxation of water protons are ignored; only the effect of the susceptibility difference between the cylinders and their surrounding is considered. The imaging voxel was divided into many cubes (>106) with an edge dimension L, each of which contained a single cylinder of length L. The volume of the cylinder (i.e., `blood volume' bv) relative to the volume of the cube is given by (rb 2L)/L3. Figure 5 displays the results as a function of cylinder radius and for a frequency shift (
0(1ÿY)!0) equal to 32, 48, and 64 Hz, bv=0.02 (simulating the capillary blood volume to total tissue volume ratio in the brain,26 and TE of 40 ms (often used in studies at 4 T). Given the maximal susceptibility difference between fully oxygenated and fully deoxygenated blood5 and taking Y to be 0.6, typical of venous blood in the brain, the frequency difference
0(1ÿY)!0 is calculated to be 43 Hz at 4 T. The data in Figure 5 demonstrate that at a radius less than ~5 to 10 m, depending on the magnitude of the frequency shift, R2 decreases because of dynamic averaging; the cylinder radius where this R2 decrease becomes apparent is smaller at larger frequency shifts as expected. Above ~10 m, R2 is approximately independent of the radius as the static regime dominates for all
0(1ÿY)!0 values considered in these calculations. These modeling efforts also suggest that deoxyhemoglobincontaining microvessels in the brain [i.e., capillaries (5 m mean diameter) and venules (which can be 2±20 m in diameter30)] would be in the `dynamic' averaging regime. Note that the frequency difference between the intra- and extra-cylinder

5

compartments depends both on the applied external magnetic ®eld B0 and on the magnetic susceptibility difference
 between the cylinder and its surrounding. Consequently, either at very high magnetic ®elds or at very large values of
, the radius at which a transition occurs from static to dynamic averaging will shift to values smaller than the capillary size. Such large values of
 can be achieved even at 1.5 T with bolus injections of contrast agents. Figure 6 illustrates the dynamic versus static averaging regimes and their dependence on cylinder radius (i.e., blood vessel radius) in a different way. In a spin echo study, static averaging will not come into play because the refocusing pulse will undo the `dephasing' and `rephase' the spins. Therefore, spin echoes will only be sensitive to small cylinder radii. For gradient recalled echoes, both dynamic and static averaging are, in principle, operative. However, for small cylinder radii, only the dynamic averaging regime will apply. Therefore, a plot of the ratio of fractional signal loss during TE (i.e.
S/S) for a spin echo versus a gradient recalled echo will have a curve approaching unity as the cylinder radius approaches zero. At the other extreme, this ratio will diminish with increasing cylinder radius as dynamic averaging disappears and a static averaging regime dominates. The net result of these calculations (again so far not considering the intravascular effects) yields the following terms for contributions to R2 : R2 ˆ f
0 !0 …1 ÿ Y †gbvl R2 ˆ f
0 !0 …1 ÿ Y †g2 bvs p

…large vessels† …small vessels†

…4† …5†

where  and  are constants, !0 is the external magnetic ®eld in frequency units (rad/s) (i.e., !0= B0),
0!0(1ÿY ) is the frequency shift owing to the susceptibility difference between the cylinder simulating the deoxyhemoglobin-containing blood vessel, bvl is the blood volume for large blood vessels (veins and venules with a radius greater than ~5 m for 4 T) and bvs is the small vessel blood volume (capillaries and small venules, less than ~5 m in radius, that permit dynamic averaging), and p is the fraction of active small vessels (i.e., ®lled with deoxyhemoglobin-containing red blood cells). An important feature of Equation (5) is the fact that it varies as the square of the external magnetic ®eld for small vessels where the effect is dominated by dynamic averaging. In contrast, the dependence on the external magnetic ®eld is linear for large blood vessels because they are in the static averaging domain. Note, however, that even for the capillaries the quadratic dependence of the extravascular BOLD effect on !0 will not persist forever with increasing external magnetic ®eld. At some very high external magnetic ®eld strength, the frequency shift across the luminal boundaries of the blood vessel will be suf®ciently large to displace even the capillaries into the static averaging domain, and hence to linear dependence on the applied magnetic ®eld. At the present time, there is no experimental evidence as to what that ®eld strength is, although results from our laboratory demonstrate that at 9.4 T there still exists an extravascular BOLD effect owing to dynamic averaging during activation (presented below). Another important point that can be surmised from Equations (4) and (5) is that the BOLD effect will be proportional to three physiologic parameters: regional cerebral

6 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY blood ¯ow (CBF) and oxygen consumption rate (CMRO2) [since (1ÿY)=CMRO2/CBF] and regional blood volume. Neuronal activity is coupled to all of these physiologic parameters. It has been suggested that regional CBF increases while CMRO2 in the same area is not elevated commensurably,31±33 resulting in decreased extraction fraction and lower deoxyhemoglobin content per unit volume of brain tissue. Another important prediction of the modeling studies is that there are large and small vessel extravascular BOLD effects. This has implications with respect to the speci®city of functional images generated by the BOLD effect. While capillaries are uniformly distributed in tissue and are suf®ciently high in density, large venous vessels are not; consequently BOLD effects associated with large vessels will not be as closely correlated with the actual site of neuronal activity. This issue will be discussed in greater detail below. 2.2 Intravascular Blood Oxygen Level-dependent Effects In the blood, hemoglobin is also compartmentalized within red blood cells and when the deoxy form is present, there are ®eld gradients around the red cells. However, because the dimensions of red cells are very small compared with diffusion distances, the effect is dynamically averaged and becomes a T2 effect only. The dynamic averaging in this case also involves exchange across the red blood cell membrane, which is highly permeable to water. Consequently, in the presence of deoxyhemoglobin-containing red blood cells, the T2 value of blood decreases. This effect was noted by Thulborn and was shown to increase quadratically with ®eld strength as expected from dynamic averaging owing to diffusion in the presence of ®eld gradients.34,35 Therefore, even when we neglect the extravascular effect described above, the T2 time of blood itself will change when the deoxyhemoglobin content is altered by elevated neuronal activity, and this will lead to a signal change in a T2- or T2 -weighted image. This effect will be present wherever the deoxyhemoglobin content has changed, potentially both in large and small blood vessels. As discussed previously, an increase in cerebral blood volume (CBV) results in signal loss in the extravascular BOLD effect. If the extravascular BOLD effect is neglected and only the intravascular contribution is considered, the increase in CBV with elevated neuronal activity can be more complex. Essentially, the ratio of blood to tissue volume will increase in a given voxel when CBV is elevated during increased neuronal activity. If the blood T2 time is longer than that of tissue, then the CBV increase will lead to a signal increase rather than the decrease that is always expected from the extravascular BOLD effect. At 1.5 T, blood T2 can be calculated from:36 …1=T2 † ˆ 4:02 ‡ 41:5…1 ÿ Y †2

…6†

which yields ~250 ms for arterial blood (Y&1), and 94, 129, and 176 ms for venous blood with Y=0.6, 0.7, and 0.8, respectively. Experimental determinations give human arterial and venous blood T2 times of 254‹26 ms and 181‹23 ms, respectively, at 1.5 T.37 Compared with these blood T2 values, cerebral tissues display T2 values that range from ~70 to 90 ms at 1.5 T,38 where the longest value is associated with cortical gray matter. These investigators measured T2 values at 1.4 T. These values are expected to be also valid for 1.5 T given the

relatively slow variation of T2 with magnetic ®eld strength for water in the brain. For example T2 of cortical gray matter is 63‹6.2 ms at 4 T39 as opposed to 87‹2 ms reported for 1.4 T.38 At 1.5 T, the value of T2 of both arterial and venous blood is longer than that of gray matter where the blood volume and alterations in blood volume coupled to neuronal activity are most signi®cant. Taking into account that the coef®cient 41.5 in Equation (6) should be proportional to the square of the static ®eld (except at very high magnetic ®elds), the T2 time for venous blood can be calculated at other ®eld strengths. At the same Y values, the predicted blood values for T2 are approximately 4, 7, and 14 ms, respectively, for 9.4 T, in agreement with the experimentally measured values of ~7 to 8 ms in rat venous blood for a Y value in the 0.7 to 0.8 range (unpublished data from our laboratory). This rapid decrease in venous blood T2 with increasing magnetic ®elds has signi®cant implications for high®eld fMRI studies and will be discussed below. At 4 T, where high-®eld human fMRI studies have so far been performed, blood T2 is estimated to be 20, 31 and 63 ms for Y=0.6, 0.7 and 0.8, respectively, using Equation (6); these values are comparable to or signi®cantly less than the gray matter T2 time of 63‹6.2 ms.39 Blood contribution comes into the BOLD phenomenon in a second special way when blood occupies a large fraction of the volume of the voxel, in other words when a large blood vessel occurs in the voxel. When deoxyhemoglobin is present in the blood, the blood water will result in dynamic averaging in the gradients surrounding the red blood cells and it will behave as if it encounters the uniform magnetic ®eld given by Equation (1). This will differ from the magnetic ®eld experienced by the rest of the voxel. In the immediate vicinity of the blood vessel, the magnetic ®eld will vary and it will approach a constant value in tissue distant from the blood vessel. For simplicity, we can neglect the gradients near the blood vessel and consider the voxel to be composed of two large bulk magnetic moments, one associated with blood and the rest with the extravascular volume. These magnetic moments will precess at slightly different frequencies, the difference in frequency given by Equation (1); therefore, the signal from the voxel will decrease with time as the two moments lose phase coherence. In this scenario, the signal can even oscillate as the phase between the two magnetic moments increases and then decreases. We will refer to this as a type 2 blood effect in fMRI. When a voxel only contains capillaries, the blood volume is ~2%;26 hence, the type 2 blood effect cannot exist for such a voxel. However, when a large blood vessel or vessels are present in the voxel, blood volume can signi®cantly increase and become comparable to or even exceed the tissue volume in the voxel. If, of course, the voxel is smaller than the blood vessel dimensions, and the entire voxel is occupied by blood, then here too the type 2 effect does not come into play. Note that the type 2 blood effect and the extravascular BOLD effect in the static averaging regime are similar in nature because they both involve signal modulation owing to dephasing of magnetization within a voxel. The main difference is the presence of a large blood component and the lack of suf®cient variation in resonance frequency within the voxel to approximate the signal modulation as an exponential signal decay in the type 2 blood effect. Similar to the extravascular BOLD effect in the static averaging domain, the type 2 blood

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

effect will be refocused in a symmetric spin echo and thus nulli®ed. Consequently, it will not be present in purely T2weighted BOLD and functional images derived from it. Another important point is that this type 2 effect will diminish and even disappear at high magnetic ®elds because the T2 value of blood gets very short and blood signal contribution becomes negligible. 2.3 Blood Oxygen Level Effects Detected in T2 and T2 by fMRI The origin of the signal intensity changes that are detected in T2 versus T2 -based BOLD fMRI images, and hence the dependence on magnetic ®eld of these effects, differ signi®cantly, as can be surmised from the above discussion. These differences are summarized here. The T2 -based BOLD signal can arise from both intravascular (blood) and extravascular effects originating from both large and small blood vessels. The relative contributions of these effects will depend on the magnetic ®eld strength. In a T2based BOLD fMRI map, the signal changes come from intravascular effects (i.e., blood T2 changes), hence from both large and small blood vessels, and from an extravascular effect associated only with microvessels such as capillaries and small venules. The major difference is that the extravascular BOLD effect in a T2 image can only arise from the microvasculature, whereas in T2 images, it can originate from blood vessels of all sizes. It is often suggested that T2-based fMRI avoids a large vessel contribution. This claim is not correct because it ignores the intravascular blood-associated BOLD effect. At ®eld strengths of 1.5 T, or even 4 T, there clearly exists a signi®cant blood contribution to the BOLD effect during increased neuronal activity. At 1.5 T, T2-based fMRI maps are predominantly if not exclusively af®liated with blood T2 changes, hence with intravascular space, as discussed in the next section. Such blood contribution can originate from both large and small blood vessels. Only at very high ®elds (e.g., 9.4 Tesla) is T2based BOLD fMRI largely associated with the capillaries because intravascular blood contributions are suppressed by the very short T2 value of blood (discussed in greater detail in the next section). 2.4 Experimental Studies Early in the history of fMRI, it was demonstrated experimentally that the BOLD effect in human brain functional maps contained contributions from macroscopic venous blood vessels (i.e., vessels of ~0.5 mm or larger that can be visualized in MRI images) as well as from regions where no such macroscopic vessels were identi®ed.40 This has profound implications for the effective spatial speci®city and spatial resolution that can be achieved with fMRI since large blood vessels do not exist at high density in the brain, where capillaries are common. Therefore, functional maps based on the macrovasculature can be signi®cantly distant from the actual site of increased neuronal activity and thus misleading if high-resolution mapping in the millimeter or smaller scale is desired. The effect of the blood vessel size on BOLD will depend on magnetic ®eld strength and the relative contributions of intra- and extravascular effects. These relationships can be experimentally evaluated.

7

The issue of extra- and intravascular BOLD effects has been experimentally examined using a Stejskal and Tanner41 gradient pair of either the same or opposite polarity, depending on whether gradient or spin echoes are used, respectively. Such gradients were ®rst used to examine molecular diffusion; therefore, their use to alter the image signal intensity is often referred to as `diffusion weighting' even though there are additional perturbations that arise from the use of such gradients. In experiments employing the Stejskal±Tanner gradients, the important parameters are the magnitude and the duration of the gradient pulses and the time separation between them. Frequently, the results are evaluated in terms of a parameter b, which is equal to ( G)2 (
ÿ/3) where is the gyromagnetic ratio (rad sÿ1 Gÿ1), G is the magnetic ®eld gradient magnitude (G cmÿ1),  is the duration of the gradient pulse, and
is the separation in time of the onset of the two gradient pulses (where 1 G=10ÿ4 mT). In simple isotropic diffusion, the MR signal in the presence of Stejskal±Tanner gradient, decays according to exp(ÿbD) where D is the diffusion constant. For ¯owing spins, b does not have such an immediately obvious physical meaning. If there is ¯ow, the spins will acquire a phase that will depend on their velocity along the direction of the gradient and the gradient magnitude. Within vasculature, however, blood velocities are not uniform especially for vessels of large diameter. Furthermore, the blood vessels may change directions within a voxel, and there may be several different blood vessels with different ¯ow rates and/or different orientations relative to the gradient direction. Since the blood signal detected from the voxel will be the sum of all these, the net result can be nulling of signal from ¯owing spins through dephasing. The ability to nullify the intravascular signals by use of Stejskal±Tanner pulsed gradients provides the means to distinguish between intra- and extravascular BOLD effects in functional images. Such experiments have been performed at 1.5 and 4 T on humans. The studies at 1.5 T have indicated that most of the BOLD-based signal increase during elevated neuronal activity is eliminated by Stejskal±Tanner gradients, leading to the conclusion that most of the fMRI signal at 1.5 T arises from intravascular effects.42,43 This intravascular BOLD effect may be associated with macroscopic blood vessels since it is debatable whether the gradient pulses used can suppress intravascular signals from microscopic blood vessels such as capillaries and small venules.44 As discussed above, the intravascular BOLD effect in functional brain imaging can arise from the change in blood T2 time or from what we described as the type 2 blood effect. The latter is refocused by symmetric spin echoes and, therefore, can only be present in gradient recalled echo or asymmetric spin echo measurements. Given the fact that symmetric spin echo fMRI experiments at 1.5 T yield very weak effects compared with gradient recalled echo studies, the data acquired with suppressing the blood component suggests that most of the fMRI signal at 1.5 T arises from type 2 blood effects. The same conclusion was reached in high-resolution two-, or three-dimensional gradient recalled echo studies of motor cortex activation.45 The effect of the Stejskal±Tanner gradients on brain tissue signal intensity at 4 T is illustrated in Figure 7a for a spin echo sequence. When activation studies were performed with such gradients at this ®eld strength, ~60% of the `activated' pixels disappeared at small b values but the remaining pixels persisted

8 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 7 (a) The effect on signal intensity of variations in the value of b in the human brain under diffusion weighting in a spin echo sequence. (b) Number of activated pixels in the human visual cortex during visual stimulation at 4 T before and during application of bipolar gradients with increasing b value. The sequence was a spin echo sequence yielding T2-weighted BOLD images. Residual signal intensity that persists at high b values can come only from the microvasculature. Data were from four subjects (*) or one subject (*). (With permission from Menon et al.46)

as the gradient strength was increased to very large b values (Figure 7b). This suggests that at 4 T extravascular and/or capillary-level intravascular BOLD effects exist during activation as well as a signi®cant intravascular contribution associated with the macrovasculature. Note that we have grouped extravascular and capillary-level intravascular effects together because it is not clear whether these gradients are suf®cient to nullify the intravascular signal from capillaries. At 9.4 T, the effects of the Stejskal±Tanner gradients become even more interesting. In a T2-weighted fMRI study conducted in the rat brain (forepaw stimulation, symmetric spin echo with one 180 pulse), activation does not alter at all with changes from very small to very high b values (Figure 8).47 The T2-based BOLD effects can only come from the blood through a change in the blood T2 or from extravascular effects associated with capillaries and comparably sized venules. The gradient pair will suppress the blood effect, except possibly in capillaries and postcapillary small venules. Therefore, it can be concluded that at this very high magnetic ®eld there exists a strong and dominant BOLD effect originating from microscopic vessels. Intravascular effects associated with a blood T2 change is a priori not expected to be signi®cant at this ®eld strength because the T2 value of venous blood is very short at 9.4 T (~5 ms),47 as discussed above. Even arterial blood has a short T2 time at this high magnetic ®eld strength (~30 ms).47 3 FUNCTIONAL IMAGING BASED ON CEREBRAL BLOOD FLOW BOLD contrast relies on the interplay between CBF and CMRO2 as well as blood volume, and, as such, it represents a complex response controlled by several parameters.18,21,23±25,48,49 Recent MR techniques, however, can also generate images based on quantitative measures of changes in CBF coupled

with neuronal activity.7,50±54 These CBF techniques rely on tagging the blood spins differentially within and outside of a well-de®ned volume. For example, in the FAIR technique51±53 frequency-selective inversion pulses are used to invert the longitudinal magnetization within a `slab' along one direction (typically axial); in the absence of blood ¯ow, the spins relax back to thermal equilibrium only by spin-lattice relaxation mechanisms characterized with the time constant T1. If ¯ow is present, however, the relaxation becomes effectively faster as unperturbed spins outside the inverted slab ¯ow in and replenish the net magnetization within the slab. Consequently, the effective spin-lattice relaxation in FAIR as well as other, similar ¯ow-sensitive techniques48,51,53±58 becomes characterized by a shorter time constant, T1 , which is related to blood ¯ow. It also follows naturally that if the inversion pulse in FAIR does not de®ne a slab but inverts everything in the whole body (i.e., it is nonselective), blood ¯ow does not enter into the problem. Therefore, in the FAIR technique, two images are acquired consecutively, each after a ®xed delay period subsequent to the inversion pulse; in one, the inversion pulse is slab selective and in the other it is nonselective. The difference image generated from this pair is a ¯ow-sensitive image. One of the unique aspects of CBF-based functional maps is that macrovascular ¯ow components can be selectively suppressed while sensitivity to microvascular ¯ow and tissue perfusion changes is enhanced. In an experimental approach such as FAIR, this selective microvascular sensitivity is accomplished by changing the delay time after the initial inversion pulse and the subsequent signal excitation before image acquisition. In CBF-based functional imaging, longer delays (>1 s) emphasize microvascular ¯ow and perfusion whereas shorter delays yield predominantly large vessel images.51,59 The former, of course, is highly desirable for generating functional maps. Herein lies the magnetic ®eld dependence of CBF-based functional images. The T1 time of tissue water gets signi®cantly longer at high magnetic ®elds. Consequently, it is easier

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

9

Figure 8 Diffusion-weighted spin echo fMRI maps at 9.4 T with b values of 6.1 (a) and 438 s mmÿ2 (b) overlaid on one of the original consecutively acquired echo planar images (BOLD and diffusion weighted) collected during the functional imaging study. Coronal single-slice singleshot spin echo planar images of rat brain were acquired with a matrix size of 64  32, a Field of View of 3.0 cm  1.5 cm, a slice thickness of 2 mm, and TE of 30 ms. Somatosensory stimulation was used. The color bar indicates a maximum cross-correlation value from 0.7 to 0.9. Signal intensity (shown in background) was signi®cantly reduced by bipolar gradients, as expected owing to diffusion. Localized activation is observed at the somatosensory cortex in the contralateral side of a stimulated forepaw. Foci of activation site (color) agree very well in both fMRI maps. (c) A Turbo FLASH image shows the region of interest. (d) Time courses of diffusion-weighted images within the region of interest. If the macrovascular contribution was signi®cant, relative BOLD signal changes would decrease when a higher b value was used. However, relative signal changes remained the same in both images, suggesting that extravascular and microvascular components predominantly contribute to spin echo BOLD at 9.4 T

to detect the `microvascular' ¯ow and perfusion and suppress the macrovascular component (see Tsekos et al.59). In view of the macrovascular problems that can deleteriously affect BOLD images, the question arises as to why CBF-based images are not the preferred approach in generating functional maps. The answer is that the CNR is higher in BOLD images, and, in particular, rapid acquisition techniques covering the whole or a large subsection of the brain are yet to be developed with CBF-based methods. Unlike the CBF-based techniques, however, BOLD images can be acquired rapidly over the whole brain. Consequently, this approach remains the main method employed in fMRI applications.

4 SPATIAL SPECIFITY In fMRI studies, there are two reasons for concern regarding spatial speci®city; one is the sensitivity to different size blood vessels and the presence of a macrovascular contribution, and the second is the spatial speci®city of the physiologic and metabolic events that ultimately yield the functional images.

4.1

Early Blood Oxygen Level-dependent Responses

Optical measurements in animal experiments have demonstrated that the onset of task-related activation ®rst results in signal changes interpreted as caused by an increase in deoxyhemoglobin content.60,61 This deoxyhemoglobin increase peaks at ~3 s after task onset and is subsequently reversed, ultimately resulting in a relatively large decrease in overall deoxyhemoglobin content. If CMRO2 is elevated because of energy requirements of increased neuronal transmission, oxygen extraction and consequently the deoxyhemoglobin level will also be elevated provided the blood ¯ow remains constant. In contrast, a CBF increase alone without any alterations in CMRO2 will cause only a decrease in deoxyhemoglobin; if there is a difference in the metabolic and hemodynamic response times of these two processes, with the latter lagging behind, the time dependence of deoxyhemoglobin content in the `activated' region will resemble the biphasic curve illustrated in Figure 9. Presence of early deoxygenation has been recently reported based on direct measurements of blood partial pressure of oxygen,62 indicating that there indeed exists an

10 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 9 Hypothetical time courses of cerebral blood ¯ow (CBF) and oxygen consumption rate (CMRO2) that would predict an early negative BOLD response

early increase in CMRO2 that precedes the onset of enhancement in blood ¯ow. The early deoxyhemoglobin increase can also arise from a rapid elevation in blood volume before the onset of blood ¯ow augmentation; such a volume increase would actually result in elevated oxy- and deoxyhemoglobin contents. A subsequent rush of oxygenated blood through enhanced blood ¯ow would then decrease the deoxyhemoglobin level, provided oxygen utilization does not increase commensurately with CBF. The possible occurrence of such a rapid volume increase has been suggested by recent studies in the rat brain.63 Malonek and Grinvald have argued that the early response is spatially more speci®c and de®nes better the columnar structure in the cat visual cortex examined in that study.61 In contrast, they suggest that the CBF increase is not as speci®c, ¯ooding not only the active but also inactive columns and surpassing in spatial extent the actual area of activation by several millimeters. These claims suggest that the BOLD effect associated with the hyperoxygenated, high-¯ow phase during increased neuronal activity will also be spatially nonspeci®c over several millimeters, re¯ecting the spatial distribution of the CBF response. The question arises whether the early response associated with increased deoxyhemoglobin can also be detected as a negative signal change in BOLD-weighted MRI and whether this response can be used as a means of obtaining functional images with the MR approach. An early negative response to activation was ®rst reported by Hennig and colleagues using a MR spectroscopy method, monitoring signal from a relatively large voxel but with high temporal resolution.64 Subsequently, ®rst using averaging of data from a number of subjects65 and later with single subjects, imaging studies documented the presence of a small but detectable early negative signal change, a `dip', at high magnetic ®elds.66±70 Recently, the presence of this initial response was also observed in fMRI studies conducted in monkeys at 4.7 T;71 the magnitude and the time course was very similar to the human data obtained at 4 T. Figure 10 illustrates the time dependence of BOLD contrast fMRI signal changes observed in the human visual cortex during and subsequent to a brief visual stimulation that lasted

2.4, 3.6, and 4.8 s.67 In this study, T2 -weighted, gradientrecalled echo planar images were acquired rapidly covering only a few slices in the visual cortex, thus sacri®cing spatial extent of coverage and spatial resolution in favor of time resolution. Furthermore, the brief stimuli were repeated several times (four to ten) and images were collected in synchrony with the stimulus presentation so that they could be averaged. The data revealed that during and following the brief visual stimulation period the BOLD-based fMRI signal initially decreased; this decrease was reversed at about 3 s, resulting subsequently in a large signal intensity increase (Figure 10). For the longer stimulation period, a signal intensity decrease below baseline was also seen towards the end of the time course; this `late-phase', poststimulation decrease has been observed before, even in the very ®rst fMRI papers, and may re¯ect a difference in the post-task temporal responses of blood volume72 and/or CMRO2 values73 returning to basal levels much more slowly than did the blood ¯ow. Figure 11 displays visual stimulation activation maps in the sagittal plane constructed from these data with the negative signal changes color coded in blue/purple colors and the positive response in red/yellow colors. Examining the images constructed from data collected during the early (negative) and the late (positive) response, we see that the early response is restricted to the anatomically well-de®ned visual area V1 (primary visual cortex) along the calcarine ®ssure while the later `positive' BOLD image displays apparent `activation' in areas distant from this region; in this study, these distant areas represent artifacts associated with the MR methodology, namely the macrovascular in¯ow effect, which has been already discussed. The macrovascular in¯ow effect was not suppressed in these images because they were rapidly acquired and were obtained using a surface coil both for signal detection and excitation; consequently, it was practically impossible to achieve the condition of `full relaxation' between images that would have eliminated these artifacts without signi®cantly sacri®cing signal-to-noise ratio (SNR). The macrovascular in¯ow effect does not appear in the early images because it is absent owing to the slower hemodynamic response time. If the macrovascular effects are ignored and the functional maps around the calcarine ®ssure are compared, they are similar for the early negative and the later positive BOLD images. This observation has implications with respect to the spatial speci®city of the functional maps generated by fMRI. It suggests that, provided the macrovascular `in¯ow' artifacts are eliminated, the spatial speci®city of the early negative BOLD-based maps and the later
CBF-dominated positive BOLD images are very similar at the resolution of this study (~3 to 4 mm isotropic). However, this study does not yet answer the question of spatial speci®city raised by Malonek and Grinvald61 because the study is not conducted at a suf®ciently high spatial resolution. It is likely that higher magnetic ®elds will be necessary to achieve greater spatial resolution with this early response because it is a very weak effect. The fact that it is detectable by BOLDbased fMRI, however, is signi®cant both from a mechanistic point of view and for future developments in fMRI. The early negative response has been examined further74 and demonstrated to be linearly dependent on TE. The percentage signal change varied from individual to individual in the nine subjects studied, ranging between 0.38 and 0.95% at 21 ms TE and 0.93 and 3.2% at 45 ms TE at 4 T; however, in

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

11

Figure 10 Signal intensity time course for activated areas in the primary visual cortex (V1) for three different periods of brief visual stimulation. Initially, the signal decreased to a valley, which is the early negative response or dip; subsequently, the signal increased leading to a peak that corresponded to the positive BOLD effect employed in functional images. This positive BOLD effect is much larger in magnitude than the early negative BOLD effect. Note that the poststimulation undershoot is seen prominently only when the visual stimulation duration is 3.6 ms or larger. (With permission from Hu et al.67)

each case, a linear dependence on TE was strongly evident. Figure 12 illustrates this linear dependence in a way that accounts for the intersubject variability. The presence of this linear response with TE is a priori expected if the initial negative change arises from a BOLD effect re¯ecting an increase in regional cerebral deoxyhemoglobin content. If the early response arises from an elevated CMRO2 prior to the onset of an increase in blood ¯ow, then initial stages of this early response will be associated with the capillary bed. At later times, this deoxyhemoglobin alteration will show up in the venules and veins as a result of blood ¯ow. Capillary BOLD effects scale as the square of the magnetic ®eld, which would explain why this early negative response has not been detected or has been reported to be very small at 1.5 T. A recent study in our group was able to identify a small early response at 1.5 T in the visual cortex; comparison with the 4 T data suggests that the ratio of the peaks corresponding to the early negative and subsequent positive responses increased linearly with ®eld strength. Given the large macrovascular contribution at both 1.5 and 4 T to the late hyperoxygenated state, only a linear dependence on B0 is expected for this positive BOLD response. This suggests that the early response must increase quadratically with the magnetic ®eld.

4.2

High-Resolution Imaging

The issue of spatial speci®city and resolution of fMRI can also be addressed using speci®c experiments to map functionally distinct structures with well-de®ned organization and topography in the human brain. Early experiments introducing the fMRI methodology employed such a strategy and examined the hemispheric lateralization in brain function.6±8,75 For example, simple motor tasks are expected to be lateralized ipsilaterally in the cerebellum;75 in other words, a simple motor task with the left or the right hand should predominantly activate the left or the right cerebellar hemisphere, respectively. This is in direct contrast to what is expected and observed in V1. However, detection of functional specialization with respect to hemispheric laterality only reveals the existence of a very coarse level of spatial speci®city in a spatial domain that can be characterized as several centimeters; certainly, this level of speci®city was demonstrated in the very ®rst studies introducing fMRI.6±8 On a much ®ner spatial scale (e.g., millimeter and submillimeter), it is possible to examine activation of small subcortical nuclei and even that of lower-level neuronal functional organizations such as the ocular dominance columns (ODCs).

12 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 11 Functional images of visual stimulation constructed from (a) the early negative response (color coded in blue) and (b) the peak of the subsequent positive response (color coded in yellow/red). (With permission from Hu et al.67)

The thalamus provides an excellent case for evaluating the question whether structures that are only a few millimeters in size can be accurately mapped by fMRI methodology. The thalamus contains several distinct, anatomically well-de®ned regions or nuclei. These nuclei serve as relay points for a remarkably large number of pathways. For example, retinal output projects mainly to the lateral geniculate nucleus (LGN), a small, subcentimeter nucleus located posteriorly and ventrally within the thalamus. In turn, the LGN activates V1.76±80 Detection of LGN activation by fMRI has previously been reported at 1.5 and 4 T ®eld strength.81±83 The LGN is functionally and spatially segregated and compartmentalized. Each of six LGN layers receives inputs from the speci®c visual ®eld via the retina and then retinotopically signals to V1.84 The speci®city and resolution of fMRI can be examined by testing the feasibility of mapping the retinotopic organization within this small nucleus. We have recently conducted such a study using a checkerboard visual stimulus covering different sections of the visual ®eld.85 In particular, either upper or lower hemi®eld stimulation against a dark control period was utilized. This is expected to activate LGN bilaterally but not uniformly. Spatial differentiation re¯ecting the retinotopical relationship between the upper and lower visual ®elds was detected and distinguished within the LGN. This is illustrated as composite maps for four individual subjects in Figure 13. The LGN activation induced by the upper visual ®eld stimulation (green and red pixels) was more inferior in location (closer to the hippocampal formation) compared with that induced by the lower visual ®eld stimulation (yellow

Figure 12 Echo time dependence of the early negative response (dip) to activation. For each of nine individuals studied, an average percentage change was calculated for the three echo delays used. Subsequently, the percentage change for the dip at each echo time was divided by this average for that individual. Then, the `normalized' percentage change at each echo time was averaged across the nine subjects studied. The data points represent the mean and standard deviation obtained from this procedure. The line is a linear ®t to the data. (With permission from Yacoub et al.74)

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

13

Figure 13 High-resolution fMRI mapping of lateral geniculate nucleus activation during the upper-®eld and lower-®eld red/black checkerboard visual stimulations (illustrated at the top of the ®gure). Images are in the coronal orientation. The activations in color represent composite fMRI maps obtained by combining the activation maps generated by upper-visual ®eld versus a dark control period and lower visual stimulation versus a dark control period. The red pixels represent the overlap between these two activation maps. Yellow and green identify pixels activated only by the lower- and upper-®eld stimulation, respectively. (Adapted from Chen et al.85)

and red pixels). This relationship is consistent with data obtained in the non-human primate visual system, which has been extensively studied using microelectrode recording86 and selective lesions.87,88 Our results also illustrate that the LGN in humans behaves similarly to V1.10 However, the upper and lower visual ®eld representations in V1 are anatomically separated by the calcarine ®ssure and are distinguishable without overlap. In contrast, they are continuous in LGN layers. This contributes to the partial overlap of LGN activation between the upper and lower visual ®elds (Figure 13). From the LGN, geniculostriate projections to V1 continue to carry left or right eye input separately and terminate in layer IVC of V1, where they are arranged in a system of roughly parallel alternating stripes known as ODCs. In non-human primates and other vertebrates (e.g., cats), the organization of these columns has been studied by histological stains, autoradiography, and microelectrode recordings80,89,90 and by optical imaging of intrinsic signals.60,61,91±93 In humans, the ODCs have been demonstrated at autopsy in the striate cortex by histochemical staining for cytochrome oxidase;94,95 however, a noninvasive technique for examining human striate cortex organization on the scale of cortical functional subunits has not been available.

The hemodynamic-response mechanism that allows visualization of orientation columns and ODCs in awake monkeys by optical imaging of intrinsic signals demonstrates that corticovascular responses to visual stimuli can be localized to the columnar level in several mammalian species. In particular, the optical data demonstrate that, while the CBF response may not be speci®c at the ODC level,61 a deoxyhemoglobin difference across the active and inactive columns is generated presumably because of the enhanced CMRO2 in the active but not the inactive column (Figure 14). This deoxyhemoglobin difference between the two columns should, in principle, be detectable by BOLD-based fMRI provided the technique has suf®cient speci®city as well as sensitivity (SNR) to achieve the required high spatial resolution. For example, if large vessel contributions dominate the BOLD contrast observed, speci®city will be inadequate to detect ODCs. For fMRI studies using clinically available hardware, ~3 mm in-plane resolution and ~5 mm slices are typical because of the limited SNR available without extensive data averaging. Using the same high-resolution fMRI pulse sequence with imaging hardware and parameters optimized at three different ®eld strengths, we have found that the SNR at 4 T is at least four times higher than at the much more commonly available 1.5 T

14 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 14 Deoxy- and oxyhemoglobin response as a function of time during a brief period of visual stimulation in active and inactive columns in the cat visual cortex. The ®gure represents a summary of ®ndings by Maloneck and Grinvald.61

®eld strength.96,97 This increase is suf®ciently large to attempt imaging of ODCs in human V1, which are approximately 0.8± 1 mm on a side for a column and 5±10 mm long.94,95 Using a simple visual paradigm in combination with an optimized rf coil, head restraints, subvoxel image registration, and the enhanced SNR provided by 4 T ®eld strength, it has been possible to demonstrate adjacent image pixels in human V1 that respond primarily to left or right eye photic input.98±101 Figure 15a demonstrates a magni®ed picture of cortical ribbon along a sulcus in V1. The image plane is along the calcarine ®ssure and columns appear as rectangles of approximately 1 mm1 mm separated by ~1 mm in the cortical gray matter. The color map represents pixels that had higher signal intensity during left eye monocular photic stimulation than during right eye simulation. The dimensions of the `activated' pixels are approximately 1 mm1 mm or slightly smaller and are reproducible.98,99 The technique in this study was to ®rst use a binocular stimulation and then to use alternating left and right monocular stimulation. The data were, however, analyzed by looking for statistically signi®cant differences in pixel intensities for only the monocular stimulation period, ignoring the binocular stimulation completely. The pixels identi®ed as activated during either the left or right eye monocular stimulation also showed activation for the binocular period, as they should

if they indeed represent ODCs as opposed to random statistical correlations. Figure 15b demonstrates ODCs in the human brain from a more recent study at 4 T by Menon and colleagues.102 These images are obtained in the sagittal plane adjacent to the interhemispheric ®ssure and the ODCs are visualized as they appear on the cortical surface in this region of the visual cortex. The blue and red colors illustrate the columns associated with the two eyes. The columns are now seen along their long axis rather than in cross-section. These data demonstrate for the ®rst time that mapping of functional subunits in humans is possible in a noninvasive manner. The fMRI time courses show that the hemodynamic response at the `hyperoxygenation phase' can be used at 4 T as a direct indicator of neuronal activity in cortical columns. This opens up the possibility of mapping specialized populations of neurons in humans that are not accessible to electrophysiological or other methods of invasive mapping. However, they do not yet resolve the issue of whether CBF increase coupled to increased neuronal activity and consequently BOLD response is spatially speci®c and selective at the column level. The column images illustrated in Figure 15 were obtained using alternating monocular stimulation. Therefore, columns would be detectable as long as there was a difference in the deoxyhe-

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

15

Figure 15 Detection of human ocular dominance columns (ODCs) during alternating monocular stimulation at 4 T. (a) Data obtained on a plane parallel to the calcarine ®ssure that intersects the ODCs perpendicular to their long axis. In this view, the ODCs should appear approximately as squares of 11 mm cross-section, separated by ~1 mm. There is a curving sulcus lined by the ODCs. (With permission from Menon et al.99) (b) A more recent study at 4 T showing a sagittal plane adjacent to the interhemispheric ®ssure; the ODCs are visualized as they appear on the cortical surface. The blue and red colors illustrate the columns associated with different eyes. (With permission from Menon and Kim101 and Goodyear and Menon102)

moglobin content of the inactive and active columns, and hence a difference in the BOLD effect, even if the deoxyhemoglobin content changed for both active and inactive columns, as illustrated in Malonek and Grinvald (Figure 14).61 The results summarized above demonstrate that, despite the presence of several potential problems, fMRI at 4 T has the speci®city and sensitivity to map organizations in the millimeter or slightly smaller scale with the use of appropriate paradigms. This is unprecedented in human brain studies. We must emphasize the high-®eld aspect of all of the afore-mentioned high-resolution studies; to date, the ODCs have not been detected at lower ®eld strengths. 5 SINGLE-TRIAL FUNCTIONAL MRI Most fMRI studies utilize a `block' design where periods of a control state are interleaved with periods of task performance and/or sensory stimulation. The control period itself may also require the subject to perform a task and/or be subjected to sensory stimulation. The images are generated by examining the difference between the control and the tasking periods. Each of these periods are relatively long, typically a minute or more, and the subject executes the task many times. The picture that emerges from such a study is a time average that blurs important information regarding the temporal evolution of the neuronal activity in different parts of the brain. Equally im-

portant, cognitive effects such as learning, alteration in strategy, etc. that evolve with repeated executions are also averaged into the ®nal image generated. fMRI is actually a realtime measurement. Single-slice fMRI images can be acquired in tens of milliseconds, adequate to monitor neuronal responses. Unfortunately, the temporal response of fMRI signals is dictated by the response of the vascular system, which is characterized with a time constant of seconds. This was demonstrated in one of the early fMRI studies103 and subsequently con®rmed in numerous other reports.67,104±108 However, even within this temporal regime, useful information on brain function can be obtained since numerous tasks and processes exist that necessarily engage the human brain for prolonged periods. In this temporal domain, signal changes in the fMRI images track the temporal evolution of stimulation or mental task performance very well, albeit with a shift in time or a delay that lasts several seconds. This capability permits the acquisition of fMRI data gated to a particular time point in stimulus onset or the instruction±execution sequence. In either case, this type of fMRI data collection is referred to as event-related fMRI. In event-related fMRI, two distinct types of experiments have been performed. Buckner et al. acquired images gated to the onset of a task in a paradigm so that the temporal evolution of the fMRI signal during and following the execution of the task could be temporally co-registered and averaged following repeated executions of the same task.109 However, such aver-

16 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY aging loses unique information associated with each execution of the task; subjects do not perform the same way each time because both brain function and performance is modulated by effects such as learning, alterations in strategy, errors, and habituation. In addition, when averages of single trials are performed, it is not possible to conclude much about the temporal differences that may be observed among different regions of the brain; such a difference was reported in the study by Buckner et al., where activation in left prefrontal areas involved in language were delayed ~2 s relative to extrastriate areas during a word generation task.109 Of course, the human brain performs this task much faster than 2 s, and this time difference cannot really be attributed to differences in brain activation. Rather, differences must exist in the hemodynamic response of the fMRI signal in different regions of the brain. This distinction, however, cannot be made from the data alone in this study, and the interpretation of the results would be ambiguous in less obvious cases when this approach is utilized. The second type of experiment was a true single-trial fMRI study achieved with the use of high magnetic ®elds (4 T).105,106,110±113 Intentionally and speci®cally, it was demonstrated that (at least at 4 T) there exists enough sensitivity to monitor fMRI signal evolution in a single execution of a task without averaging over many trials. This point is important. With this capability, it is then possible to perform many such single trial executions of a task and not average them but rather to store them separately and subsequently analyze and correlate the fMRI data with differences in aspect of the subject's performance (e.g., response time, errors, etc.). In this way, the hemodynamic response time differences can also be factored out and distinguished from temporal behavior of neuronal activity. An alternative approach is to average such single trials but using performance or response criteria to pool together only those response that are similar. An example of a true single-trial study conducted at 4 T monitored the evolution of fMRI signals in the human brain before, during, and subsequent to an instruction and execution of a motor task (Figure 16). Both the single-trial fMRI signal intensity time courses in motor areas and the electromyograph (EMG) changes detected in the muscle are shown. The EMG showed no movements during the motor preparation period between Instruction and GO. Activation of the primary motor area and of the supplementary and premotor areas were observed during motor execution and preparation. These data are virtually identical to electrode recordings taken from the corresponding areas in a monkey cortex during execution of the same task, except that the fMRI data are displaced in time by seconds relative to the actual time of neuronal activity. The data presented in Figure 16 also illustrate the sensitivity of the high-®eld fMRI method. Like the electrode recordings from primates, the fMRI data demonstrate that V1 is active during the motor preparation period. This, however, has been a controversial issue with humans because position emission tomographic studies did not reveal activation during such motor preparation.114 In the high-®eld fMRI study, this activation is detectable in a single execution of the task. This type of true single-trial experiment was also utilized to correlate aspects of activation with task performance.106,113 The paradigm employed was the `mental rotation' task of Shepard and Metzler:115 The subjects were presented with drawings of

Figure 16 A single-subject, single-trial fMRI (without averaging) study showing time courses in three regions of the brain together with the electromyograph (EMG) recording of muscle movement during a visually instructed delayed cued four-®nger movement task. The presentation lasted 2.1 s and the GO signal was given at 9.1 s. M1, contralateral primary motor area; PM, bilateral premotor area; SMA, bilateral supplementary motor area. (With permission from Richter et al.111)

three-dimensional objects, examples of which are illustrated in Figure 17. In each task, a pair of objects were presented; they were either identical or mirror images and they were rotated relative to each other through varying degrees. The subject had to identify whether the pair was identical or a mirror image and report it by pressing one of two buttons. In this task, subject's response time depended on the angle through which the two objects were rotated relative to each other. The experiment started with the subject looking at similar but identical twodimensional objects. When ready, the subject commenced the scanning process by pressing a button. The three-dimensional objects were shown and the subject made a decision; after a suitable delay to allow for the hemodynamic response to return to basal levels, the process was repeated. Figure 18 displays signal intensity curves from the parietal lobe from two trials where the response time of the subject was different. The width of the response was evaluated for correct response only with respect to the response time. In each individual, linear correlation was found. However, the intercept corresponding to a response time of zero was signi®cantly different for the different individuals, presumably re¯ecting fundamental differences in the hemodynamic response to elevated neuronal activity among individuals. When this subject-dependent variable was subtracted, each single-trial, single-subject data point for all subjects yielded an excellent correlation with response time (Figure 19). One potential confounding problem with such single-trial fMRI studies is the presence of various types of spatiotemporal patterns in the fMRI signals even under basal `resting' conditions. The T2 -weighted MRI signals ¯uctuate with heart beat and respiration, although ¯uctuations can be removed from the fMRI data.116±121 When cardiac and respiratory ¯uctuations are suppressed, fMRI signal from resting human brain still exhibits low-frequency oscillations at about 0.1 Hz.121±124 Near-infrared

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY

Figure 17

17

The paradigm in a true single-trial `mental rotation' study. (With permission from Richter et al.107)

optical studies also show this slow oscillation. Therefore, techniques that can separate the fMRI data into its various components (e.g., Mitra et al.121) and in the process signi®cantly improve effective SNR for detection of function will play a crucial role in mapping brain function, particularly in single-trial fMRI studies with temporal resolution.

6 CONCLUSION Since its introduction, fMRI has rapidly evolved to become the most signi®cant method for investigating human brain function, and unique accomplishments have been realized at high magnetic ®elds. However, it must be realized that high magnetic ®eld MRI instruments are not optimized and re®ned machines as clinical scanners. Presence of a high-®eld magnet does not guarantee superior functional imaging results; instead, Figure 19 The width of the fMRI response (in Figure 18) versus performance data from all subjects subsequent to removal of the time zero intercept. Each point represents a true single trial in a single subject (no averaging). (With permission from Richter et al.107)

advantages that are inherent in the high ®eld for functional mapping can easily be lost through instrumentation imperfections. Therefore, improvement in high-®eld instrumentation are essential for future expansion of these fMRI applications. Additional re®nements in data collection schemes, motion correction, and statistical methods for data analysis, which are areas that are being actively pursued, will undoubtedly improve the already impressive capabilities of this methodology.

7 Figure 18 The T2 -weighted BOLD response in the parietal lobe for two different single trials (no averaging) for the paradigm outlined in Figure 17. The arrows indicate the time at which the subject responded. (With permission from Richter et al.106)

RELATED ARTICLES

Hemodynamic Changes owing to Sensory Activation of the Brain Monitored by Echo-Planar Imaging; Image Processing of Functional MRI Data.

18 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY 8 REFERENCES

1. P. Broca and C. E. Brown-Sequard, `Proprietes et fonctions de la moelle epiniere: Rapport quelques experiences de M. Borwn-Sequard'. (Lu a loa Soceite de biologies le 21 Juillet). Bonaventure et Ducessois. 1855. 2. P. Erhard, P. Strick, J. Strupp, A. K. Helms Tillery, S. NaeveVelguth, and K. Ugurbil, unpublished results, 1998. 3. S. Ogawa, T.-M. Lee, A. R. Kay, and D. W. Tank, Proc. Natl. Acad. Sci., USA, 1990, 87, 9868. 4. S. Ogawa, T.-M. Lee, A. S. Nayak, and P. Glynn, Magn. Reson. Med., 1990, 14, 68. 5. S. Ogawa and T. M. Lee, Magn. Reson. Med., 1990, 16, 9. 6. P. A. Bandettini, E. C. Wong, R. S. Hinks, R. S. Tikofsky, and J. S. Hyde, Magn. Reson. Med., 1992, 25, 390. 7. K. K. Kwong, J. W. Belliveau, D. A. Chesler et al., Proc. Natl. Acad. Sci., USA, 1992, 89, 5675. 8. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellermann, S. G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci., USA, 1992, 89, 5951. 9. D. I. Hoult and P. C. Lauterbur, J. Magn. Reson., 1979, 34, 425. 10. M. I. Sereno, A. M. Dale, J. B. Reppas, K. K. Kwong, J. W. Belliveau, T. J. Brady, B. R. Rosen, and R. B. H. Tootell, Science, 1995, 268, 889. 11. R. B. H. Tootell, J. B. Reppas, A. M. Dale, R. B. Look, M. I. Sereno, R. Malach, T. J. Brady, and B. R. Rosen, Nature, 1995, 375, 139. 12. R. B. H. Tootell, J. B. Reppas, K. K. Kwong, R. Malach, R. T. Born, T. J. Brady, B. R. Rosen, and J. W. Belliveau, J. Neurosci., 1995, 15, 3215. 13. R. G. H. Tootell, A. M. Dale, M. I. Sereno, and R. Malach, Trends Neurosci., 1996, 19, 481. 14. J. B. Reppas, S. Niyogi, A. M. Dale, M. I. Sereno, and R. B. Tootell, Nature, 1997, 388, 175. 15. R. B. Tootell, J. D. Mendola, N. K. Hadjikhani, P. J. Ledden, A. K. Liu, J. B. Reppas, M. I. Sereno, and A. M. Dale, J. Neurosci., 1997, 17, 7060. 16. C. S. Springer and Y. Xu, New Devel. Contrast Agent Res., 1991, 13. 17. J. L. Boxerman, R. M. Weisskoff, B. E. Hoppel, and B. R. Rosen, in Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, 389. 18. S. Ogawa, R. S. Menon, D. W. Tank, S.-G. Kim, H. Merkle, J. M. Ellermann, and K. Ugurbil, Biophys. J., 1993, 64, 800. 19. R. M. Weisskoff, J. L. Boxerman, C. S. Zuo, and B. R. Rosen, `Functional MRI of the Brain', Arlington, Virginia, 1993. 20. E. C. Wong and P. A. Bandettini, in Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, 10. 21. R. P. Kennan, J. Zhong, and J. C. Gore, Magn. Reson. Med., 1994, 31, 9. 22. D. A. Yablonskiy and E. M. Haacke, Magn. Reson. Med., 1994, 32, 749. 23. J. L. Boxerman, L. M. Hamberg, B. R. Rosen, and R. M. Weisskoff, Magn. Reson. Med., 1995, 34, 555. 24. S. Ogawa, R. S. Menon, S. G. Kim, and K. Ugurbil, Annu. Rev. Biophys. Biomol. Struct., 1998, 27, 447. 25. R. M. Weisskoff, C. S. Zuo, J. L. Boxerman, and B. R. Rosen, Magn. Reson. Med., 1994, 31, 601. 26. G. Pawlik, A. Rackl, and R. J. Bing, Brain Res., 1981, 208, 35. 27. J. O. Eichling, M. E. Raichle, R. L. Grubb, and M. M. TerPogossian, Circ. Res., 1974, 35, 358. 28. O. B. Paulson, M. M. Hertz, T. G. Bolwig, and N. A. Lassen, Acta Neurol. Scand., 1977, 64(Suppl.), 492.

29. O. B. Paulson, M. M. Hertz, T. G. Bolwig, and N. A. Lassen, Microvasc. Res., 1977, 13, 113. 30. A. C. Burton, Fed. Proc., 1966, 25, 1753. 31. P. T. Fox and M. E. Raichle, Proc. Natl. Acad. Sci., USA, 1986, 83, 1140. 32. M. E. Raichle, `Handbook of PhysiologyÐThe Nervous System', 1987, 643. 33. P. T. Fox, M. E. Raichle, M. A. Mintun, and C. Dence, Science, 1988, 241, 462. 34. K. R. Thulborn, J. C. Waterton, P. M. Mathews, and G. K. Radda, Biochim. Biophys. Acta, 1982, 714, 265. 35. K. R. Thulborn, J. C. Waterton, and P. M. Matthews, Biochim. Biophys. Acta, 1992, 714, 265. 36. G. A. Wright, B. S. Hu, and A. Macovski, JMRI, 1991, 1, 275. 37. M. Barth and E. Moser, Cell. Mol. Biol., 1997, 43, 783. 38. F. W. Wehrli, J. R. MacFall, D. Schutts, R. Breger, and R. J. Herfkens, J. Comput. Assist. Tomogr., 1984, 8, 369. 39. P. Jezzard, S. Duewell, and R. S. Balaban, Radiology, 1996, 199, 773. 40. R. S. Menon, S. Ogawa, D. W. Tank, and K. Ugurbil, Magn. Reson. Med., 1993, 30, 380. 41. E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288. 42. J. L. Boxerman, P. A. Bandettini, K. K. Kwong, J. R. Baker, T. L. Davis, B. R. Rosen, and R. M. Weisskoff, Magn. Reson. Med., 1995, 34, 4. 43. A. W. Song, E. C. Wong, S. G. Tan, and J. S. Hyde, Magn. Reson. Med., 1996, 35, 155. 44. R. M. Henkelman, J. J. Neil, and Q. S. Xiang, Magn. Reson. Med., 1994, 32, 464. 45. E. M. Haacke, A. Hopkins, and S. Lai et al., NMR Biomed., 1994, 7, 54. 46. R. S. Menon, X. Hu, G. Adriany, P. Andersen, S. Ogawa, and K. Ugurbil, in Proc. IInd Annu. Mtg (Int.) Soc. Magn. Reson., San Francisco, 1994, 2, 622. 47. S.-P. Lee, A. C. Silva, K. Ugurbil, and S.-G. Kim, Magn. Reson. Med., 1999, 42, 919. 48. R. B. Buxton and L. R. Frank, J. Cereb. Blood Flow Metab., 1997, 17, 64. 49. P. C. van Zijl, S. M. Eleff, J. A. Ulatowski, J. M. Oja, A. M. Ulug, R. J. Traystman, and R. A. Kauppinen, Nature Med., 1998, 4, 159. 50. R. E. Edelman, B. Siewer, D. G. Darby, V. Thangaraj, A. C. Nobre, M. M. Mesulam, and S. Warach, Radiology, 1994, 192, 513. 51. S.-G. Kim, Magn. Reson. Med., 1995, 34, 293. 52. S.-G. Kim and N. V. Tsekos, Magn. Reson. Med., 1997, 37, 425. 53. S. G. Kim and K. Ugurbil, Magn. Reson. Med., 1997, 38, 59. 54. E. C. Wong, R. B. Buxton, and L. R. Frank, Magn. Reson. Med., 1998, 39, 702. 55. J. A. Detre, J. S. Leigh, D. S. Williams, and A. P. Koretsky, Magn. Reson. Med., 1992, 23, 37. 56. J. A. Detre, W. Zhang, D. A. Roberts, A. C. Silva, D. S. Williams, D. J. Grandis, A. P. Koretsky, and J. S. Leigh, NMR Biomed., 1994, 7, 75. 57. D. A. Roberts, J. A. Detre, L. Bolinger, E. K. Insko, and J. S. Leigh, Proc. Natl. Acad. Sci., USA, 1994, 91, 33. 58. K. K. Kwong, D. A. Chesler, R. M. Weisskoff, K. M. Donahue, T. L. Davis, L. Ostergaard, T. A. Campbell, and B. R. Rosen, Magn. Reson. Med., 1995, 34, 878. 59. N. V. Tsekos, F. Zhang, H. Merkle, M. Nagayama, C. Iadecola, and S. G. Kim, Magn. Reson. Med., 1998, 39, 564. 60. R. D. Frostig, E. E. Lieke, D. Y. Ts'o, and A. Grinvald, Proc. Natl. Acad. Sci., USA, 1990, 87, 6082. 61. D. Malonek and A. Grinvald, Science, 1996, 272, 551.

FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY 62. I. Venzetta and A. Grinvald, Neurosci. Lett., 1998 Suppl. 51, S42. 63. J. B. Mandeville, J. J. A. Marota, C. Ayata, M. Moskowitz, B. R. Rosen, and R. M. Weisskoff, in Proc. VIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Sydney, 1998, p. 1422. 64. T. Ernst and J. Hennig, Magn. Reson. Med., 1994, 32, 146. 65. R. S. Menon, S. Ogawa, X. Hu, J. P. Strupp, P. Anderson, and K. Ugurbil, Magn. Reson. Med., 1995, 33, 453. 66. T. H. Le and X. Hu, in Proc. IVth Annu. Mtg (Int.) Soc. Magn. Reson. Med., New York, 1996, p. 285. 67. X. Hu, T. H. Le, and K. Ugurbil, Magn. Reson. Med., 1997, 37, 877. 68. J. McIntosh, Y. Zhang, S. Kidambi, T. Harshbarger, G. Mason, G. M. Pohost, and D. Twieg, in Proc. IVth Annu. Mtg (Int.) Soc. Magn. Reson. Med., New York, 1996, 1, p. 284. 69. G. G. Moore, Y. T. Zhang, and D. B. Twieg, in Proc. Vth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, 1, p. 377. 70. D. B. Twieg, G. G. Moore, and Y. T. Zhang, in Proc. Vth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, 1, p. 1647. 71. N. K. Logothetis, H. Gugenberger, S. Peled, and J. Pauls, Nature Neurosci., 1999, 2, 555. 72. J. B. Mandeville, J. J. Marota, B. E. Kosofsky, J. R. Keltner, R. Weissleder, B. R. Rosen, and R. M. Weisskoff, Magn. Reson. Med., 1998, 39, 615. 73. J. Frahm, K. D. Kruger, K. D. Merboldt, and A. Kleinschmidt, Magn. Reson. Med., 1996, 35, 143. 74. E. Yacoub, T.H. Le, K. Ugurbil, and X. Hu, Magn. Reson. Med., 1999, 41, 436. 75. J. M. Ellermann, D. Flament, S.-G. Kim, Q.-G. Fu, H. Merkle, T. J. Ebner, and K. Ugurbil, NMR Biomed., 1994, 7, 63. 76. D. H. Hubel and T. N. Wiesel, J. Physiol. (Lond.), 1968, 195, 215. 77. D. H. Hubel and T. N. Wiesel, J. Comp. Neurol., 1972, 146, 421. 78. D. H. Hubel, N. Wiesel, and D. M. K. Lam, Brain Res., 1974, 79, 273. 79. P. Rakic, Nature, 1976, 261, 467. 80. D. H. Hubel and T. N. Wiesel, in Proc. R. Soc. Lond. Ser. B, 1977, 198, 1. 81. A. Kleinschmidt, K. D. Merboldt, W. Hanicke, H. Steinmetz, and J. Frahm, J. Cereb. Blood Flow Metab., 1994, 14, 952. 82. C. Buchel, R. Turner, and K. Friston, Magn. Reson. Med., 1997, 38, 691. 83. W. Chen, T. Kato, X. H. Zhu, J. Strupp, S. Ogawa, and K. Ugurbil, Magn. Reson. Med., 1998, 39, 89. 84. S. Zeki, `A Vision of the Brain', Oxford, Blackwell Scienti®c, 1993. 85. W. Chen, X. H. Zhu, K. R. Thulborn, and K. Ugurbil, Proc. Natl. Acad. Sci., USA, 1999, 96, 2430. 86. J. G. Malpeli and F. H. Baker, J. Comp. Neurol., 1975, 161, 569. 87. P. H. Schiller, N. K. Logothetis, and E. R. Charles, Vis. Neurosci., 1990, 5, 321. 88. P. H. Schiller, N. K. Logothetis, and E. R. Charles, Nature, 1990, 343, 68. 89. C. Kennedy, M. H. Des Rosiers, and O. Sakurada, Proc. Natl. Acad. Sci., USA, 1976, 73, 4230. 90. S. LeVay, M. Connolly, and J. Houde, J. Neurosci., 1985, 5, 486. 91. A. Grinvald, E. Lieke, R. D. Frostig, C. D. Gilbert, and T. N. Wiesel, Nature, 1986, 324, 361. 92. D. Y. Ts'o, R. D. Frostig, E. E. Lieke, and A. Grinvald, Science, 1990, 249, 417.

19

93. A. Grinvald, R. D. Frostig, R. M. Siegel, and E. Bartfeld, Proc. Natl. Acad. Sci., USA, 1991, 88, 11559. 94. J. C. Horton and E. T. Hedley-White, Phil. Trans. R. Soc. Lond. Ser. B, 1984, 304, 255. 95. J. C. Horton, L. R. Dagi, and E. P. McCrane, Arch. Opthalmol., 1990, 108, 1025. 96. J. S. Gati, R. S. Menon, K. Ugurbil, and B. K. Rutt, in Proc. IIIrd Annu. Mtg (Int.) Soc. Magn. Reson., Nice, 1995, p. 772. 97. J. S. Gati, R. S. Menon, K. Ugurbil, and B. K. Rutt, Magn. Reson. Med., 1997, 38, 296. 98. R. S. Menon, S. Ogawa, and K. Ugurbil, Neuroimage, 1996, 3, S357. 99. R. S. Menon, S. Ogawa, J. P. Strupp, and K. Ugurbil, J. Neurophysiol., 1997, 77, 2780. 100. R. S. Menon and B. G. Goodyear, Magn. Reson. Med., 1999, 41, 230. 101. R. S. Menon and S.-G. Kim, Trends Cognit. Neurosci., 1999, 3, 207. 102. B. G. Goodyear and R. S. Menon, Submitted, 1999. 103. A. M. Blamire, S. Ogawa, K. Ugurbil, D. Rothman, G. McCarthy, J. M. Ellermann, F. Hyder, Z. Rattner, and R. G. Shulman, Proc. Natl. Acad. Sci., USA, 1992, 89, 11069. 104. R. L. Savoy, P. A. Bandettini, K. M. O'Craven, K. K. Kwong, T. L. Davis, J. R. Baker, J.W. Belliveau, R. M. Weisskoff, and B. R. Rosen, Proc. IIIrd Annu. Mtg (Int.) Soc. Magn. Reson. Med., Nice, 1995, p. 450. 105. S.-G. Kim, W. Richter, and K. Ugurbil, Magn. Reson. Med., 1997, 37, 631. 106. W. Richter, P. M. Andersen, A. P. Georgopoulos, and S.-G. Kim, Neuroreport, 1997, 8, 1257. 107. W. Richter, K. Ugurbil, A. P. Georgopoulos, and S.-G. Kim, Neuroreport, 1997, 8, 3697. 108. R. S. Menon, D. C. Luknowsky, and J. S. Gati, Proc. Natl. Acad. Sci., USA, 1998, 95, 10902. 109. R. L. Buckner, P. A. Bandettini, K. M. O'Craven, R. L. Savoy, S. E. Petersen, M. E. Raichle, and B. R. Rosen, Proc. Natl. Acad. Sci. USA, 1996, 93, 14878. 110. S.-G. Kim and W. Richter, Proc. IVth Annu. Mtg (Int.) Soc. Magn. Reson. Med., New York, 1996, p. 289. 111. W. Richter and S.-G. Kim, Proc. Soc. Neurosci., 1996, 1, 22. 112. W. Richter, K. Ugurbil, and S.-G. Kim, Neuroimage, 1996, 3, S38. 113. W. Richter, A. P. Georgopoulos, K. Ugurbil, and S.-G. Kim, Proc. Vth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, p. 357. 114. J. Decety, R. Kawashima, B. Gulyas, and P. E. Roland, Neuroreport, 1992, 3, 761. 115. R. N. Shepard and J. Metzler, Science, 1971, 171, 701. 116. X. Hu and S.-G. Kim, Magn. Reson. Med., 1994, 31, 495. 117. J. S. Hyde, B. Biswal, A. W. Song, and S. G. Tan, Proceedings of the 2nd Annual Midwest Functional MRI Workshop, Madison, WI, 1994, p. 73. 118. X. Hu, T. H. Le, T. Parrish, and P. Erhard, Magn. Reson. Med., 1995, 34, 201. 119. P. P. Mitra, D. J. Thompson, S. Ogawa, X. Hu, and K. Ugurbil, Proc. IIIrd Annu. Mtg (Int.) Soc. Magn. Reson. Med., Nice, 1995, p. 817. 120. B. Biswal, E. A. DeYoe, and J. S. Hyde, Magn. Reson. Med., 1996, 35, 117. 121. P. P. Mitra, S. Ogawa, X. Hu, and K. Ugurbil, Magn. Reson. Med., 1997, 37, 511. 122. B. Biswal, F. Z. Yetkin, V. M. Haughton, and J. S. Hyde, Magn. Reson. Med., 1995, 34, 537.

20 FUNCTIONAL MRI AT HIGH FIELDS: PRACTICE AND UTILITY 123. S. Ogawa, P. P. Mitra, X. Hu, and K. Ugurbil, `Recent Advances in MEG and functional MRI', (EEG J. Suppl. 47), eds. I. Hashimoto, Y. C. Okada, and S. Ogawa, International Federation of EEG Societies, Limerick, 1996, p. 5. 124. B. Biswal, A. Hudetz, F. Yetkin, V. M. Haughton, and J. S. Hyde, J. Cereb. Blood Flow Metab., 1997, 17, 301.

Acknowledgements Work reported here from the Center for Magnetic Resonance University of Minnesota was supported by NIH P41 RR08079, an NIH Regional Resource grant, and NIH NS32191.

Biographical Sketches Kamil Ugurbil. b 1949. A.B. Physics, 1971, Ph.D. Chemical Physics, Columbia University, New York, 1976. Postdoctoral Fellowship, Bell Laboratories, 1976±79, Assistant Professor Biochemistry, Columbia University, New York, 1979±82, Associate Professor Biochemistry, University of Minnesota, Minneapolis, 1982±85, Professor in Biochemistry, Radiology, Medicine, and Neurosciences, University of Minnesota, 1985±present, Director Center for Magnetic Resonance Research (CMRR), University of Minnesota, 1991±present, and holder of Margaret & H.O. Peterson Chair of Neuroradiology, Department of Radiology, University of Minnesota, 1996±present. Approx. 160 publications. Current research interests: high ®eld magnetic resonance imaging and spectroscopy with particular focus on functional imaging in the brain and myocardial bioenergetics. Wei Chen. b 1959. B.S. Physical Chemistry, Fudan University, Shanghai, China, 1981, Ph.D. Physical Chemistry, Washington University, St. Louis, 1991. Postdoctoral Associate, Yale University, New Haven, 1991±94. Assistant Professor, Radiology, University of Minnesota, 1994±99, Associate Professor of Radiology, 1989±present. Approx. 50 publications. Current research interests: function and metabolism

study of brain and heart using magnetic resonance imaging/spectroscopy. Xiaoping Hu. b 1961. B.S. Physics, University of Science of Technology of China, Hefei, China, 1982, Ph.D. Medical Physics, University of Chicago, Chicago, 1988. Postdoctoral Associate, University of Chicago, 1988±90, Assistant Professor Radiology 1990±94, Associate Professor Radiology 1994±98, Professor Radiology 1998±present, University of Minnesota Minneapolis. Approx. 80 publications. Current research interests: methodological development, mechanistic studies, and applications of functional magnetic resonance imaging at high magnetic ®elds. Seong-Gi Kim. b 1958. B.E. Applied Chemistry, Kyungpook National University, Korea, 1980, Ph.D. Chemistry, Washington University, St. Louis, 1988. Postdoctoral Associate, Washington University, St. Louis and University of Washington, Seattle, 1988±91, Research Associate, Radiology, University of Minnesota, 1991±94, Assistant Professor, Radiology, University of Minnesota, 1994±98, Associate Professor of Radiology, 1988±present. Approx. 70 publications. Current research interests: physiological and functional MRI studies of animal and human brain. Seiji Ogawa. b 1934. B.S. Applied Physics, University of Tokyo, 1957, Ph.D. Chemical Physics, Stanford University, 1967, Postdoctoral Fellowship, Stanford University, 1967±68, Postdoctoral Fellowship, AT&T Bell Laboratories, 1968±69, Member of Technical Staff, AT&T Bell Laboratories, 1969±83, Distinguished Member of Technical Staff, AT&T Bell Laboratories, 1984±96, Distinguished Member of Technical Staff, Bell Laboratories, Lucent Technologies, 1996±present. Approx. 70 publications. Current research interests: functional magnetic resonance imaging of the brain. Xiao-Hong Zhu. b 1964. B.S. Chemistry, Fudan University, Shanghai, China, 1985, Ph.D. Chemistry, University of Missouri-St. Louis, St. Louis, 1991. Postdoctoral trainee, Yale University, Connecticut, 1991± 94. Postdoctoral associate, University of Minnesota, 1994±98, Research Associate, University of Minnesota, 1998±present. Approx. 30 publications. Current research interests: magnetic resonance spectroscopy and functional imaging of brain at high ®eld.

Functional MRI: Theory and Practice Robert Turner Institute of Neurology, London, UK

and David G. Gadian Institute of Child Health, London, UK

1 2 3 4 5 6 7 8 9

Introduction Contrast Mechanisms in Functional MRI Imaging Methods Sources of Artifact Image Analysis Brain Functions Studied Discussion Related Articles References

1

INTRODUCTION

1 1 3 3 4 4 6 6 6

Since 1990, interest has grown rapidly in MRI methods for mapping human brain function. These methods have good spatial and temporal resolution, and are completely noninvasive. As a result, repeated single-subject studies are feasible, allowing new types of experiments in cognitive neuroscience to be formulated. In addition, there are many potential clinical applications. Functional brain mapping involves the visualization of local physiological changes within the brain that are associated with activation of the visual, motor, or other brain systems. The technique that has so far contributed most to functional brain mapping is positron emission tomography (PET).1 These PET studies rely on the detection of changes in local hemodynamics that are associated with cerebral activation; in particular, there are changes in regional cerebral blood flow that can be mapped with PET through the use of 15 O-labeled water. The MRI studies of functional activation that have recently been described also rely on hemodynamic changes, and on the detection of signals from water. However, in contrast to PET, the signal is from the protons of the water and the effects are observed without any need for radioisotope labeling. Instead, the detection of activation relies on perturbations of the water magnetization that are associated with the hemodynamic changes. The first brain activation study using functional MRI involved the administration of the paramagnetic contrast agent Gd-DTPA to act as a marker of cerebral blood volume.2 This investigation of the human visual cortex used a visual stimulus paradigm that, on the basis of PET studies, produces changes in regional cerebral blood flow of about 30–50%. Echo planar MR images were collected at intervals of 750 ms,

before, during and after injection of Gd-DTPA, which was administered as a bolus into the antecubital vein. As the Gd-DTPA passed through the brain, it produced a transient reduction in water signal intensity because of its magnetic susceptibility effects. From the time course of the signal changes, it was possible to generate images that reflected relative cerebral blood volume. The imaging procedure was carried out under control conditions, and also during visual stimulation produced by goggles that generated 8 Hz patterned flashes. Subtraction of the control from the activated blood volume image revealed those areas in which there was a change in blood volume associated with the task activation. These areas corresponded to the primary visual cortex. As it happened, this initial activation study was very rapidly superseded by an alternative MRI approach which had the major advantage that no extrinsic contrast agent was required. It had been known for some time that there were several mechanisms whereby changes in blood flow, volume, and oxygenation can influence signal intensities without any requirement for exogenous agents, and it soon became apparent that activated regions of the human brain could be visualized using such intrinsic contrast mechanisms. In these studies, particular interest has focused on T 2 *weighted signal intensity changes that have been attributed primarily to changes in local venous blood oxygenation, in particular to the effects generated by the paramagnetic iron centers of deoxyhemoglobin. There is a long history to NMR studies of this compound, and some of the salient points are described briefly below.

2 CONTRAST MECHANISMS IN FUNCTIONAL MRI 2.1 Blood Oxygenation Level Dependent (BOLD) Effects

It has long been known that the magnetic state of hemoglobin in red cells is dependent on oxygen saturation; with deoxygenation of hemoglobin, the electron spin of the heme Fe2+ changes from a diamagnetic state to a paramagnetic state. As a result, deoxygenated blood is more paramagnetic than oxygenated blood, the difference being about 0.2 ppm. There are magnetic susceptibility effects associated with this paramagnetism, and therefore local field gradients are generated in and around the blood vessels. These cause signal attenuation in T 2 *-weighted images, which is not confined just to water within the vessels; it can extend significantly into the surrounding tissue. As a result, there may be substantial signal changes associated with changes in oxygenation state, even if the blood volume is relatively small. In the early 1990s, a number of animal imaging studies were published, in which it was demonstrated that these magnetic susceptibility effects could be exploited as a means of detecting oxygenation changes in the brain3 – 5 (see Figure 1). In particular, T 2 *-weighted images showed a reduction in signal intensity under conditions of reduced oxygenation, i.e. under conditions where there are increased susceptibility effects associated with the presence of deoxyhemoglobin. Before very long, studies were carried out in humans, using T 2 *-weighted imaging sequences6 – 8 that showed focal changes in signal intensity in response to visual tasks. In the work of Kwong et al.,6 subjects experienced visual stimulation for periods of 30 s,

2 FUNCTIONAL MRI: THEORY AND PRACTICE

Figure 1 Sequential coronal difference images of a cat brain, at 3-intervals. The cat was subjected to respiratory anoxia from images 20–40. Image 20 was subtracted from each of the others to give this demonstration of contrast changes. The development of hypoxemia is well seen as a darkening, predominantly in gray matter, and the overshoot on recovery is also clearly visible. The first image shows enhancement of cerebrospinal fluid due to the completely unsaturated condition of the spins at the outset of scanning. Images were obtained using a 2.0 T GE Omega, 45 cm bore, MR scanner. (Reproduced by permission from Turner et al.5 )

interspersed by equal periods of darkness, while echo planar gradient echo images of a section passing through the primary visual cortex were obtained at 3 s intervals. When a control image was subtracted from each of the time-series of images, the resulting difference images showed unequivocal increases in intensity in the visual cortex, correlating very well with the visual stimulus. A lag time of 6–9 s was noted between onset of stimulation and rise to maximum of the difference signal, reflecting the timescale of the hemodynamic changes associated with activation. It is significant that a rise in signal can be observed during visual stimulation, indicating a relative decrease in the concentration of paramagnetic deoxyhemoglobin. This confirms earlier work using PET9 in which the measured rise in oxygen uptake rate during visual stimulation was found to be much smaller than the rise in blood flow. Earlier observations during neurosurgery (see, for example, Penfield10 ) had also demonstrated that the blood leaving active cortical regions is brighter red, i.e. more oxygenated, than normal, as a result of this mismatch between demand and supply. The contrast in T 2 *-weighted images is generated because gradient echoes do not refocus the dephasing effects of local field homogeneities associated with the presence of deoxyhemoglobin. If spin echoes are used instead of gradient

echoes, then such dephasing effects will be refocused, provided that there is no significant diffusion of the water molecules through the field gradients. This explains why T 2 *-weighted images are more sensitive than T 2 -weighted images to the effects of brain activation. However, if the water molecules diffuse through local field gradients during the echo time TE , then this will result in imperfect refocusing and hence to signal loss in T 2 -weighted spin echo sequences11 (see also, earlier 1 H NMR studies of red cell metabolism12 ). Because water molecules can only diffuse a small distance in an echo time that may be of the order of 60 ms, this diffusiondependent mechanism of signal loss should be more effective when the distances that characterize the gradients are small. Within the blood vessels themselves, these distances are indeed small, because the localization of deoxyhemoglobin to red cells results in local field gradients in the periphery of each individual cell. In the adjacent tissue, however, the dimensions characterizing these local gradients should be determined by the dimensions of the vessels. Therefore it can be anticipated that, in comparison with large draining vessels, small vessels such as capillaries should produce relatively large diffusiondependent T 2 effects in the adjacent tissue. This has been explored as a means of distinguishing between the effects of large and small vessels (see below). However, it needs to

FUNCTIONAL MRI: THEORY AND PRACTICE

be borne in mind that changes in oxygenation in vessels of size comparable to a voxel will result in large signal intensity changes even in T 2 -weighted images. 2.2

Flow Effects

In theory, the water signal intensity can be affected by many hemodynamic effects, including changes in blood arterial oxygenation, blood volume, blood flow, hematocrit, tissue oxygen uptake, and possibly blood velocity. Indeed, the initial study of Kwong et al.6 highlighted not only T 2 *-dependent effects, but also T 1 -dependent effects that could be attributed to the direct influence of changes in local blood flow that were associated with activation. There is considerable interest in the influence of blood flow on signal changes associated with task activation, and numerous investigations have been concerned with distinguishing between flow and oxygenation effects (see, for example, Frahm et al.13 ). One approach to visualizing flow effects involves the use of arterial spin tagging, using saturation or inversion of inflowing spins as a magnetic label. For example, if spins in the neck region are selectively inverted, then arterial flow will carry these labeled spins into the brain tissue and hence influence the observed signal from brain water. Comparison of the data obtained with and without this labeling procedure permits measurements of regional perfusion to be made. The signal change resulting from the magnetic labeling depends upon a number of factors, including the perfusion rate, the T 1 value of the tissue water, and any relaxation of the labeled spins during the time that they take to travel from the labeling plane to the detection volume. This approach has been used for the quantitative measurement of cerebral perfusion, initially in small animals, and more recently in man.14 Although the method has a number of well-recognized problems, it does offer a very promising approach to the noninvasive investigation of cerebral perfusion and of changes in perfusion associated with task activation and with disease processes. A related method, termed EPISTAR, has recently been used to obtain qualitative maps of cerebral blood flow, with spatial resolution of 2 × 2 × 2 mm.15 This method is an echo planar imaging modification of a time-of-flight angiography technique. It involves the alternate acquisition of two echo planar images with and without a radiofrequency inversion pulse applied to inflowing arterial spins. Image subtraction provides a picture of large proximal vessels when short inflow times are used, while progressively more distal portions of the tagged vessels are seen as the inflow time is lengthened. At these longer inflow times, the images represent a qualitative map of cerebral blood flow. The acquisition of such images also provides a means of visualizing activated regions of the brain.

3

IMAGING METHODS

Functional MRI studies generally involve the acquisition of data during a series of interleaved periods of rest and stimulation, each of these periods typically lasting about 30 s. Rapid imaging methods are therefore required. These divide into two categories: methods that use a series of low flipangle excitation pulses (e.g. FLASH, Turbo-FLASH etc.),16

3

and methods which use a single excitation pulse (echo planar imaging).17 While both have proved successful, there is an increased tendency toward the use of echo planar imaging, which permits imaging of a complete slice in a time of 60–100 ms. With an imaging time of 100 ms, a complete multislice data set covering the whole brain can be obtained very rapidly (data from 30 slices could be acquired in 3 s), and it seems likely that multislice echo planar imaging will become the preferred method for the majority of functional MRI studies of the brain.

4 SOURCES OF ARTIFACT

In any magnetic resonance study, it is important to appreciate the various ways in which artifacts can influence the appearance of spectra or images. This is particularly true in the case of functional brain mapping methods, which suffer from certain problems which can give false positive results, or obscure the effect altogether. It must be understood that the signal changes of interest are usually not much larger than the thermal and physiological noise that inevitably accompany MR images of living tissue, so that equipment stability is of great importance. But two other issues have been the subject of controversy in the literature. These relate to the precision of localization of neural activity, and to artifactual image differences associated with subject motion. 4.1 Localization

Changes in signal intensity may be observed as a result of hemodynamic changes that are associated not only with capillaries, but also with larger draining vessels (see Lai et al.18 ). The draining vessels may be situated a long way ‘downstream’ of the activated tissue, and on task activation they may give rise to changes in signal intensity in regions that do not correspond directly to the regions that are actually activated, particularly with MRI sequences that are sensitive to flow velocity. For this reason, it is essential to establish, by one means or another, that the location of the observed signal changes does indeed reflect the focally activated regions. There are several ways in which this might be achieved. One method is to combine functional MRI with angiographic techniques, so that the regions of increase in signal intensity can be compared directly with the anatomical location of the draining vessels. Alternatively, the signal from these vessels may be suppressed by using imaging sequences that give reduced signal from regions of high flow. In addition, multislice techniques that effectively give a three-dimensional activation pattern should help to establish the source of the signal increase. Finally, it has been pointed out that the combination of T 2 -weighted and T 2 *-weighted imaging should help to distinguish between large and small vessels (see above), T 2 -weighted and T 2 *weighted effects becoming more similar as vessels become smaller.19,20 In practice, there is a lot to be said for visualizing both the activated regions and the hemodynamic changes in the draining vessels, and the best approaches will no doubt preserve both types of information. Further information about the possible importance of downstream effects, and also about the extent to which delivery of blood to brain tissue is spatially matched to

4 FUNCTIONAL MRI: THEORY AND PRACTICE the increased demand, is available from other techniques. In particular, optical imaging methods, using a cranial window in animal models, or intraoperatively in human subjects, can provide information of direct relevance to the interpretation of functional MRI data.21,22 4.2

Subject Motion

To visualize task-related regions of increased signal, the earliest functional MRI studies relied on single or averaged image differences between experimental and control conditions. While this method has the merit of explicitness and simplicity, it takes no account of spatial variations in image noise, and it is extremely vulnerable to head motion temporally correlated with the experimental task.23 Normalization of the image difference by division of the difference at each pixel by the standard error of the mean in that pixel, to give a z score, accounts for artifacts caused by large intensity fluctuations related to cardiac or respiratory cycle. Misregistration artifact, however, may continue to confound the results. Clearly, either the brain must be rigidly fixed in place during each experiment, or a robust means of reregistration of images must be employed. Rigid head fixation may entail considerable discomfort, though a number of workers have successfully used bite bars. Image reregistration can be performed either by noting fiducial changes in head position during a study, and applying the required rotations and translations to the images afterwards, or by using internal image features and least-squares search procedures to determine post hoc the corrections needed. The iterative algorithm of Woods et al.24 has been successfully used by a number of workers (e.g. Tyszka et al.25 ), and very recently a faster, more robust, and more accurate algorithm has been developed.26 Any such algorithm requires the acquisition of multislice image data to specify fully the three translations and three rotations characteristic of solid body displacement. This encourages the use of echo planar imaging (EPI) methods for functional MRI which, as mentioned above, can obtain images of the complete brain within 3 s.

5

to become less popular. Of greater importance is the increase in statistical power obtainable by including the spatial correlation between activated pixels,29 which has not yet been fully explored in the context of functional MRI.

6 BRAIN FUNCTIONS STUDIED

Since the discovery of BOLD contrast in 1989, MR studies of human brain function have expanded very rapidly. As an index of this growth, at the Annual Meeting of the Society for Magnetic Resonance in Medicine in 1991 there was one presentation mentioning (and introducing) functional MRI. In 1992, at the next meeting, there were 21 presentations dealing with functional MRI, and 94 in 1993. At the corresponding meeting in 1994 there were more than 120 such presentations. Much of this early research has been published only in the form of long refereed abstracts for these meetings. Similar growth rates have occurred in submissions to the Society for

IMAGE ANALYSIS

Once images have been coregistered, there are a variety of methods for extracting significant activated regions. Maps of F score, z score or Student’s t score can be used, but only for images acquired with a repeat time of more than about 10 s, since the slowness of the hemodynamic response (about 6–10 s) introduces temporal correlations into the data, which reduces the number of degrees of freedom in each image. When images are acquired more rapidly, it is preferable to calculate the cross correlation between the time series data in each pixel with a suitably smoothed representation of the time variation of the task, the so-called contrast function.27,28 As a rule of thumb, a correlation coefficient r of >0.5 can be considered highly significant. A number of workers have explored the use of nonparametric measures of signal change significance. Use of Kolmogorov–Smirnov statistics has attracted attention, but as the character of the image intensity distributions in functional MRI becomes more fully investigated, this approach is likely

Figure 2 Areas V5/MT of occipital cortex specialized for motion perception. Protocol consisted of alternate presentations of still and moving visual grids, separated by rest periods of darkness. Multislice functional MRI was performed using echo planar imaging (TR 3 s, TE 40 ms) at 1.5 T, with a surface radiofrequency receiver coil at the back of the head to improve signal-to-noise ratio. (a) Map of echo planar imaging functional data, processed to give a Kolmogorov–Smirnov statistic, superimposed on a high-resolution structural MR oblique axial scan. Areas responding specifically to the moving grid are indicated by arrows. (b) Plots of time course data of regions of interest in V1 and in V5, showing the lack of response in V5 to a static object. (Figure provided by R. Tootell)

FUNCTIONAL MRI: THEORY AND PRACTICE

Figure 3 Brain activations in axial slices obtained with echo planar imaging at 1.5 T. The simple task consisted of flexing and extending all four digits together, while the complex task involved tapping each finger to the thumb sequentially in a repeating sequence. The imagined task required subjects mentally to rehearse this complex task without actual finger motion. Active areas were determined using a correlation technique, and were superimposed in color on a gray-scale highresolution structural scan. (Figure provided by Rao et al.34 summarizing their results) (SMA denotes supplementary motor area; M1 denotes primary motor cortex; S1 denotes primary somatosensory cortex)

Neuroscience annual meetings. In view of the current relative crudity of tools for registration and statistical analysis of functional MRI data, as already noted, the findings in many of these studies must be regarded at this stage as provisional but highly suggestive. For these reasons, we provide just a brief summary of the findings so far. The study of the primary visual and motor cortices provides an appropriate means of establishing and validating this new method of functional neuroimaging, partly because these systems are relatively well characterized. Numerous studies have now been reported, including, for example, investigations of hemispheric asymmetry and handedness.30 Another advantage of investigating these cortical areas is that the hemodynamic changes (and hence signal intensity changes) associated with the visual and motor tasks are likely to be considerably greater than those associated with higher cognitive functions. One of the main requirements is to establish the extent to which such higher functions are accessible to investigation by MRI. Initial studies, for example observations involving word generation,31 – 33 mental imagery (e.g imagining of motor and visual tasks34,35 ) and other cognitive tasks (see, for example, Kim et al.36 ) appear encouraging, as illustrated by the studies shown in Figures 2–5. However, a great deal more work still needs to be carried out in order to establish the sensitivity of MRI to the wide range of cognitive tasks that are of interest to the neuroscience community. The demonstration that standard 1.5 tesla systems can be used for at least some applications of functional imaging, including simple motor and visual task activation studies (see, for example, Connelly et al.38 ) has greatly facilitated the use of functional MRI for clinical as well as the more neuroscientific applications. One clinical area of interest relates to neurosurgical removal of lesions that are close to the primary sensory or motor cortex. It can often be difficult

5

Figure 4 Functional maps of dentate activation during visually guided finger movements (a) and a cognitive puzzle task (b). The color-coded functional map is superimposed on a T 2 *-weighted image in gray scale. The dentate nuclei in the cerebellum are dark crescent-shaped regions with low background signal due to iron deposit. C, dentate contralateral to the moving limb; I, dentate ipsilateral to the moving limb; D, dorsal; V, ventral. For tasks, a small lightweight pegboard with nine holes was used. Subjects simply moved pegs for visually guided movements and tried to solve a pegboard puzzle for the cognitive task. All seven subjects examined showed a large bilateral activation during the cognitive task. The activated area was 3–4 times greater than that seen during the visually guided movements. These support the concept that the computational power of the cerebellum is applied not only to the control of movements, but also to cognitive functions. Data were obtained using a FLASH-type functional MRI sequence at 4.0 T. (Reproduced with permission from Kim et al.36 )

to predict the exact location of these primary cortical areas, due to normal biological variation and the distorting effects of the lesion itself. In order to permit maximal resection of such lesions, cortical mapping is performed. Until now, this has required direct electrical stimulation of the brain, often during an awake craniotomy. Using functional MRI, it is now possible to carry out noninvasive presurgical mapping of the sensory and motor cortex.39 The functional MRI maps can be validated by comparison with the standard intraoperative cortical mapping procedures, and once the MRI method has been extensively validated in this way, then it could have major implications with respect to preoperative planning and to counseling of patients with lesions in these areas of the brain. Functional MRI could also be of value in the investigation of patients with epilepsy, for preliminary studies have shown that the technique can be used to map the cortical activation that occurs during focal seizures. In these studies of a boy suffering from frequent partial motor seizures, functional MRI revealed sequential cortical activation in structurally abnormal regions of the brain.40 The activation was observed with

6 FUNCTIONAL MRI: THEORY AND PRACTICE

Figure 5 Effect of hand training on area of activated motor cortex.37 Two finger-tapping tasks were performed during the same functional MRI study, one unrehearsed and the other practised for 10 min per day. After 2–4 weeks the trained task activated a greater number of pixels, as shown. Echo planar imaging data (TR 2 s, TE 25 ms) at 4 T are projected as a z map in color on a high-resolution structural scan. Also shown are time course data from an area of cortex that activated much more for the trained task than for the untrained task, and from an area where the amplitudes of activation were more equal

each of five consecutive clinical seizures, and was also seen during a period that was not associated with a detectable clinical seizure. This latter finding, together with the time dependence of the signal changes that were observed, opens up the possibility of using functional MRI to detect not only the seizures themselves, but also subclinical or interictal events. The technique may therefore find relatively widespread application in the investigation of patients with intractable seizure disorders.

7

DISCUSSION

The scope of magnetic resonance functional neuroimaging is clearly very wide. Variations in the cerebral division of labor, both between subjects and over the course of time within a single subject, can be studied noninvasively with a hitherto unparalleled combination of temporal and spatial resolution. However, the small size of the signal changes results in vulnerability to misregistration artifact, and good reregistration software is essential. Vast quantities of image data can be generated rapidly using echo planar imaging methods (of the order of two gigabytes per day) so that fast-throughput computing facilities and major archiving capabilities are needed.

Be this as it may, a new era in human brain mapping is beginning. Research into the numerous areas of neuroscience and neurology will be accelerated by the relative ease, noninvasiveness, and good spatial and temporal resolution of these techniques.

8 RELATED ARTICLES

Diffusion and Perfusion in MRI; Echo-Planar Imaging; Structural and Functional MR in Epilepsy; Whole Body Magnetic Resonance Angiography; Functional Neuroimaging Artifacts; Head and Neck Investigations by MRI.

9 REFERENCES 1. D. J. Chadwick and J. Whelan, Ciba Found. Symp., 1991, 163, 1. 2. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen, Science, 1991, 254, 716. 3. S. Ogawa, T-M. Lee, A. S. Nayak, and P. Glynn, Magn. Reson. Med., 1990, 14, 68. 4. S. Ogawa and T.-M. Lee, Magn. Reson. Med., 1990, 16, 9.

FUNCTIONAL MRI: THEORY AND PRACTICE 5. R. Turner, D. LeBihan, C. T. W. Moonen, D. Despres, and J. Frank, Magn. Reson. Med., 1991, 22, 159. 6. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H.-M. Cheng, T. J. Brady, and B. R. Rosen, Proc. Natl. Acad. Sci. U. S. A., 1992, 89, 5675. 7. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellermann, S.-G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. U. S. A., 1992, 89, 5952. 8. P. A. Bandettini, E. C. Wong, R. S. Hinks, R. S. Tikofsky, and J. S. Hyde, Magn. Reson. Med., 1992, 25, 390. 9. P. T. Fox and M. E. Raichle, Proc. Natl. Acad. Sci. U. S. A., 1986, 83, 1140. 10. W. Penfield, Ann. Intern. Med., 1933, 7, 303. 11. K. R. Thulborn, J. C. Waterton, P. M. Matthews, and G. K. Radda, Biochim. Biophys. Acta, 1982, 714, 265. 12. K. M. Brindle, F. F. Brown, I. D. Campbell, C. Grathwohl, and P. W. Kuchel, Biochem. J., 1979, 180, 37. 13. J. Frahm, K.-D. Merboldt, W. H¨anicke, A. Kleinschmidt, and H. Boecker, NMR Biomed., 1994, 7, 45. 14. D. Roberts, J. A. Detre, L. Bolinger, E. K. Insko, and J. S. Leigh, Jr., Proc. Natl. Acad. Sci. U. S. A., 1994, 91, 33. 15. R. R. Edelman, B. Siewert, D. G. Darby, V. Thangaraj, A. C. Nobre, M.-M. Mesulam, and S. Warach, Radiology, 1994, 192, 513. 16. J. Frahm, M. L. Gyngell, and W. Hanicke, in Magnetic Resonance Imaging eds. D. D. Stark and W. G. Bradley, Jr., Mosby Year Book, St. Louis, 1992, pp. 165. 17. M. K. Stehling, R. Turner, and P. Mansfield, Science, 1991, 254, 43. 18. S. Lai, A. L. Hopkins, E. M. Haacke, D. Li, B. A. Wasserman, P. Buckley, L. Friedman, H. Meltzer, P. Hedera, and R. Friedland, Magn. Reson. Med., 1993, 30, 387. 19. S. Ogawa, R. S. Menon, D. W. Tank, S.-G. Kim, H. Merkle, J. M. Ellermann, and K. Ugurbil, Biophys. J., 1993, 64, 803. 20. R. M. Weisskoff, C. S. Zuo, J. L. Boxerman, and B. R. Rosen, Magn. Reson. Med., 1994, 31, 601. 21. R. D. Frostig, E. E. Lieke, D. Y. Ts’o, and A. Grinvald, Proc. Natl. Acad. Sci. U. S. A., 1990, 87, 6082. 22. R. Turner and A. Grinvald, 2nd Mtg., Proc. Soc. Magn. Reson., San Francisco, 1994 , 430. 23. J. V. Hajnal, R. Myers, A. Oatridge, J. E. Schwieso, I. R. Young, and G. M. Bydder, Magn. Reson. Med., 1994, 31, 283. 24. R. P. Woods, S. R. Cherry, and J. C. Mazziotta, J. Comput. Asst. Tomogr., 1992, 16(4), 620. 25. J. M. Tyszka, S. T. Grafton, W. Chew, R. P. Woods, and P. M. Colletti, Ann. Neurol., 1994, 35, 746. 26. K. J. Friston, J. Ashburner, C. D. Frith, J.-B. Poline, J. D. Heather, and R. S. J. Frackowiak, Hum. Brain Mapping, in press. 27. P. A. Bandettini, A. Jesmanowicz, E. C. Wong, and J. S. Hyde, Magn. Reson. Med., 1993, 30, 161. 28. K. F. Friston, P. Jezzard, and R. Turner, Hum. Brain Mapping, 1994, 1, 153.

7

29. W. Schneider, D. C. Noll, and J. D. Cohen, Nature (London), 1993, 365, 150. 30. S.-G. Kim, J. Ashe, K. Hendrich, J. M. Ellermann, H. Merkle, K. Ugurbil, and A. P. Georgopolis, Science, 261, 615. 31. G. McCarthy, A. M. Blamire, D. L. Rothman, R. Gruetter, and R. G. Shulman, Proc. Natl. Acad. Sci. U. S. A., 1993, 90, 4952. 32. R. M. Hinke, X. Hu, A. E. Stillman, S.-G. Kim, H. Merkle, R. Salmi, and K. Ugurbil, Neuroreport, 1993, 4, 675. 33. L. Rueckert, I. Appollonio, J. Grafman, P. Jezzard, R. Johnson, D. Le Bihan, and R. Turner, J. Neuroimaging, 1994, 4, 67. 34. S. M. Rao, J. R. Binder, P. A. Bandettini, T. A. Hammeke, F. Z. Yetkin, G. L. Morris, W. M. Mueller, and L. D. Estkowski, Neurology, 1993, 43, 2311. 35. D. Le Bihan, R. Turner, T. Zeffiro, C.-A. Cuenod, P. Jezzard, and V. Bonnerot, Proc. Natl. Acad. Sci. U. S. A., 1993, 90, 11 802. 36. S.-G. Kim, K. Ugurbil, and P. L. Strick, Science, 1994, 265, 949. 37. P. Jezzard, A. Karni, G. Meyer, M. Adams, A. Prinster, L. Ungerleider, and R. Turner, 2nd Mtg. Proc. Am. Soc. Magn. Reson. San Francisco, CA, 1994 , p. 330. 38. A. Connelly, G. D. Jackson, R. S. J. Frackowiak, J. W. Belliveau, F. Vargha-Khadem, and D. G. Gadian, Radiology, 1993, 188, 125. 39. C. R. Jack, R. M. Thompson, F. W. Sharbrough, P. J. Kelly, D. P. Hanson, R. K. Butts, N. J. Hangiandreou, S. J. Riederer, R. L. Ehman, and G. D. Cascino, Radiology, 1994, 190, 1. 40. G. D. Jackson, A. Connelly, J. H. Cross, I. Gordon, and D. G. Gadian, Neurology, 1994, 44, 850.

Acknowledgments We thank the Wellcome Trust for their support.

Biographical Sketches Robert Turner. b 1946. B.A., 1968, Cornell University, Ph.D., 1972, Simon Fraser University, B.C., Canada. Postdoctoral work in NMR of liquid crystals, University of Cambridge, UK, 1973–75; Lecturer in Physics, University of Nottingham, UK (working with Peter Mansfield on MRI hardware and software), 1975–88 visiting scientist, NIH, Bethesda, MD, 1988–93; Institute of Neurology, London, UK, 1993–present. Approx. 70 publications. Current research interests include gradient coil design, applications of echo-planar MRI, and functional imaging of human brain. David G. Gadian. b 1950. B.A., 1971, Physics, D. Phil., 1975, University of Oxford, UK (supervisor Rex Richards). Postdoctoral research with Rex Richards and George Radda in Oxford. Moved in 1983 to the Royal College of Surgeons in London. Currently Rank Professor of Biophysics and Head of the RCS Unit of Biophysics and of the Radiology and Physics Unit at the Institute of Child Health, London. Approx. 140 publications. Research interests: development and application of magnetic resonance techniques for noninvasive investigation of brain metabolism and physiology.

Functional Neuroimaging Artifacts Joseph V. Hajnal Hammersmith Hospital, London, UK

1 2 3 4 5 6 7 8 9

Introduction Recognition of Activation-Induced Signal Changes The Problem of Subject Motion Image Registration Evidence for Stimulus Correlated Motion Correction of Misregistration Artifacts Conclusion Related Articles References

1

INTRODUCTION

1 1 2 3 3 5 9 9 9

Functional neuroimaging with MRI (fMRI) may open up enormous possibilities for the study of human brain function in health and disease.1,2 It offers the combination of being noninvasive with the potential for high spatial and moderate temporal resolution. Most functional MRI studies involve the acquisition of a series of images over time while the subject performs a task or experiences a stimulus. Images acquired with the subject in different states are then compared in order to detect signal changes associated with brain activation. The origin of such changes is widely thought to be associated with a localized hemodynamic response to increased neuronal activity. This response is thought to produce signal changes by two principal mechanisms: (a) changes in blood oxygenation state resulting in susceptibility mediated signal changes [the blood oxygen level dependent (BOLD) effect];3 and (b) signal changes associated with increased blood throughput.4 The signal changes attributed to these effects range from 1–2% up to 20–30%, depending on the MRI technique employed and the strength of the B0 field.5,6 Thus the task in functional MRI studies is to detect very small localized signal changes in a series of images of nominally the same anatomical structures. A key feature in this endeavor is the suppression of the wealth of anatomical detail that has hitherto been the hallmark of MRI in order to reveal functionally mediated signal variations. This change of emphasis has introduced a class of artifacts that are quite distinct from the imperfections usually seen on conventional in vivo magnetic resonance (MR) images. This article is concerned with the artifacts that may occur in fMRI studies including ways in which they may be identified or reduced. 2

RECOGNITION OF ACTIVATION-INDUCED SIGNAL CHANGES

Detection of activation-induced signal changes generally relies on correlating changes in the signals from individual

image pixels or groups of pixels with the temporal structure of the activation protocol under study. The basic methodology employed owes much to studies of brain activity using positron emission tomography (PET),7 a technique with which functional MRI shows some similarities as well as some essential differences. The majority of PET activation studies rely on detecting blood flow changes made visible by the injection of radioactive water. The data obtained typically display both a low effect-tonoise ratio and a large effect-to-resting-signal ratio. Thus the methodology employed is designed to detect a small coherent signal that may be buried deep in random noise. Statistical methods and spatial averaging are employed to discriminate against the noise,8 and signals that correlate with the activation protocol are taken to reflect activation-induced physiological changes. In contrast, functional MRI is characterized by a small effect-to-resting-signal ratio with a high overall signal-to-noise ratio. Thus, although it is necessary to distinguish the desired effect from random noise, for which statistical methods akin to those employed in PET may be appropriate,9,10 it is also essential to differentiate signals induced by activation from all other sources of change in the resting anatomical signal. This latter task requires a different approach because the competing signal changes may also be temporally coherent with the activation paradigm. In addition, the resting signals also have pronounced spatial structure, so that spatial averaging (smoothing) may not preferentially reduce them as would be the case for random noise. Coherent signal changes that may confound functional MRI studies can arise from a variety of sources. Both respiration and cardiac pulsations can cause signal changes in brain images.11 Changes in the subject’s state of arousal during activation studies thus may lead to signal changes that correlate with the task protocol. Large effects may be produced simply by changes in the subject’s position during the course of a study. The structural content of MR images may result in steep intensity gradients at interfaces between tissues, so that a slight head movement between the baseline and the activated images may result in a substantial signal variation. If subject motion is correlated with the task paradigm, the signal change may become task locked. As a result, it is insufficient to rely solely on coherence with the task under study as a sine qua non of a genuine activation. Recognition of artifacts is hampered by the suppression of anatomical details which removes many of the signs that radiologists conventionally rely on to detect them. Processing procedures that remove these landmarks may lead to confusion. For example, the discovery that detected signals are actually from the scalp or ventricles provides an obvious indication that something is amiss. It is important not simply to disregard these contributions to the overall activation map, but to recognize that whatever mechanism has produced them may also be contributing to less obviously erroneous signals. Even when activation signals are apparently appropriately located, detailed study of the local anatomy is often revealing. For example, direct visualization of the local cortical structure can be invaluable for checking if ‘cortical activation’ signals cross tissue boundaries or do not in fact follow the line of the cortex at all. Figure 1 illustrates this for a hemifield visual stimulation experiment designed to detect susceptibility changes associated with the BOLD effect. Long echo time field

2 FUNCTIONAL NEUROIMAGING ARTIFACTS echo images acquired during the activation study [Figure 1(a)] do not provide clear differentiation of the cortex, so that the difference image (illumination of half the visual field minus darkness) appears to show a plausible activation signal in one occipital cortex [Figure 1(b)]. However, acquisition of images at the same resolution, but designed to display the cortex alone [Figure 1(c)], reveals that the ‘active’ region follows the outline of the brain and not the highly convoluted cortex. The difference image shown in Figure 1(b) is thus most unlikely to represent cortical activation, and further analysis showed it to be artifactual. The process of recognizing genuine evidence for activation and rejecting other signal changes is still evolving. A cautious approach is indicated at present in which potential sources of error are identified and their effects at least quantified if they cannot be reduced or eliminated.

3 THE PROBLEM OF SUBJECT MOTION

Subject motion affects most MRI techniques to a greater or lesser extent. In most cases problems arise because of changes in subject position during the acquisition of image data. This may lead to gross motion artifacts or simply to an apparent reduction in the signal-to-noise ratio (see also Whole Body Magnetic Resonance Artifacts). Many techniques have been developed to minimize these effects, and some are now being applied to improve the quality of the individual images obtained in functional MRI studies.12 The long TE field echo sequences used to detect BOLD changes are particularly vulnerable to instabilities of various kinds. Fast imaging techniques such as EPI are, in general, less vulnerable to image artifacts produced by motion of the subject during the scan. Another kind of motion artifact is of particular importance in functional MRI. If the subject changes position between scans the anatomical content will change from one image to the next. This may result in artifactual signal changes even though the individual images are themselves relatively free from motion artifacts. It has been shown that subjects generally move to some degree during functional MRI studies and that this motion frequently has components that are correlated with the stimulus protocol.13 The result of such stimuluscorrelated displacement is that the intensity in some pixels varies synchronously with the stimulus. Analyses designed to detect correlated signal changes extract these motion-induced effects along with any genuine activation effects. A simple calculation illustrates the magnitude of the likely effects of changes in position. The signal intensity change
I caused by a small spatial displacement
x in a region of an image where there is a pixel intensity gradient (dI /dx ) is approximately Figure 1 A visual stimulation study in which (a) an oblique transverse slice was repeatedly imaged with a field echo sequence (TR 150, TE 100, flip angle 65◦ , 18 cm FOV, 64 × 256 matrix) while the subject experienced periods of hemifield visual stimulation and darkness. (b) A difference image reveals apparent activation in only one hemisphere as expected. However, (c) use of a double inversion recovery sequence (TR 6080, TI A 2600, TI B 245, TE 16) to image selectively the cortex at the same resolution reveals that the “activation” signals in (b) follow the outline of the brain as visualised in (a) rather than the more complex cortical structure (c)


I 

dI
x dx

(1)

If the intensity difference between adjacent pixels is 20% and the subject has moved by 1/10 of a pixel, a signal change of 2% will result. This is the same magnitude as the expected signal change from brain activation, so that clearly only a much smaller shift could be tolerated without some form of correction being necessary. With typical imaging parameters (16–22 cm field of view, 64–128 phase encode steps), 1/10

FUNCTIONAL NEUROIMAGING ARTIFACTS

pixel represents about 300 µm or less. The restraint necessary to prevent motions of this size cannot be reliably achieved by noninvasive head-holding techniques.

4

IMAGE REGISTRATION

Fortunately, the wealth of structural (anatomical) information contained in MR images allows subject movements to be detected retrospectively and corrected. Image registration algorithms that can match MR images to a tiny fraction of a pixel are now available. Hitherto, most techniques have been designed for cross-modality image matching [MRI to computerized tomography (CT), PET to MRI, etc.],14,15 and frequently operator intervention has been required to select landmarks. However, Woods et al.16,17 have developed a simple, fully automatic procedure that can be used for intramodality matching, and their approach is now being refined specifically for MRI to MRI registration.

5

EVIDENCE FOR STIMULUS CORRELATED MOTION

Changes in subject position during brain activation studies may result in gross misregistration artifacts which can easily be recognized and may have little to do with the task under study. However, smaller more subtle movements are commonplace and their effects may be more difficult to recognize by image inspection alone. The nature of subject motion during functional MRI examinations has been studied for both motor and visual activation paradigms.13 In these studies subjects were imaged using long TE rapid gradient echo sequences with a surface coil for signal reception. Repeated single-slice images of appropriate anatomical regions were acquired at a rate of one every 20 s. The stimulus condition was changed every five images. For the motor studies, the subjects performed finger to thumb touching continuously with one hand for the duration of five images and then changed hands for the next five images, and so on. At the end of the experiment all the images acquired with the left hand exercising were subtracted from all those with the right hand exercising and the results summed to form a cumulative difference image. Figure 2 shows an example of the results obtained. The anatomical slice which was acquired in the coronal plane is shown in Figure 2(a) and the cumulative difference image is shown in Figure 2(b). The highlighted region in Figure 2(b) (arrow) shows where there was a localized signal change that correlated with the stimulus protocol; it increased when the right hand was active and fell when the left hand was active. The ‘active’ region is in an appropriate cortical location for the task being performed. To examine the subject’s motion during the study, an image registration technique16 was used to find the rigid body translations and rotations that were required to match the head position in each of the images obtained with that in a single base image. This resulted in the time series data shown in Figure 2(d). The top graph shows left–right shifts from image to image, the middle graph shows head–foot displacements, and the bottom graph shows in-plane rotations of the head.

Figure 2

3

4 FUNCTIONAL NEUROIMAGING ARTIFACTS Stimulation cycle Right

Left 0

5

15

10

0.4 mm

20

25

30

35

40

25

30

35

40

25

30

35

40

25

30

35

40

Left–right position

0

–0.2 0

5

15

10

Head–foot position

0.5 mm

20

0

–0.5 0

5

15

10

20 In-plane rotation

deg.

0.2 0

–0.4 0

5

15

10

20

Left–right spectrum 0.06 0.04 0.02 0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

3.5

4

4.5

5

3.5

4

4.5

5

Head–foot spectrum 0.2

0 0

0.5

1

1.5

2

2.5

3

In-plane rotation spectrum 0.03 0.02 0.01 0 0

0.5

1

1.5

2

2.5

3

Figure 2 (a) Anatomical image of the coronal slice chosen for a motor stimulation study. (b) The cumulative difference image, right hand action minus left hand action, reveals a focal high signal region in an appropriate anatomical location (arrow). However, (c) a strikingly similar pattern of focal signal change can be demonstrated to have been produced purely as a result of small changes in position of the subject during the study. (d) The position changes, as determined by an image registration technique, are plotted against image number (time). They reveal that small displacements tend to occur as the hand being exercised is changed (i.e. at multiples of five images). (e) A power spectral analysis of the movement time series in (d) reveals dominant peaks associated with the periodicity of the stimulus protocol

FUNCTIONAL NEUROIMAGING ARTIFACTS

The magnitude of these displacements was less than 0.5 mm for translations and 0.3◦ for rotation. Although the motion is small, there is evidence that the positional changes occur in concert with the task [Figure 2(d) note the left–right shifts at images 20,25,30,35]. To investigate this more systematically a power spectrum analysis was performed [Figure 2(e)],18 to reveal the dominant periodic components of the motion. In this analysis the unit of frequency was normalized so that unity corresponds to one motion cycle per stimulus cycle of 10 images. There is a particularly pronounced peak at 1 for the in-plane rotation, indicating that the subject had a tendency to nod their head from side to side in time with each change of hand. The peak at 2 in the left–right displacement spectrum shows that the subject was moving in a way that correlated with changes of hand, irrespective of which hand was being exercised. To investigate the effect of the stimulus-correlated motion on individual pixel intensities in the difference image [Figure 2(b)], a synthetic data set was created. This was done by taking a single base image from the original functional series, replicating it, and then using the data obtained by coregistration to reposition the replicated images to match the position of each of the other images in the series. The resulting set of images matched the original experimental data by retaining the subject’s anatomy and history of their motion, but specifically excluded other sources of signal change such as changes in blood flow or oxygenation state. The calculated cumulative difference image from the synthetic data set is shown in Figure 2(c) and replicates the highlighted region in Figure 2(b) that appeared to be due to brain activation. Figure 2 thus illustrates that subject motion can correlate with the stimulus and cause signal changes that may easily be mistaken for evidence of brain activation. Similar analyses of four motor stimulation studies and four hemifield visual stimulation studies showed that stimulus-correlated motion was present in each case.13

6

CORRECTION OF MISREGISTRATION ARTIFACTS

The presence of signals that arise from changes in subject position during the course of a functional MRI study may not in itself be a problem, provided that the artifactual contribution can be recognized and dealt with in an appropriate manner. Estimating the maximum displacement that may have occurred allows the magnitude of potential displacement artifacts to be quantified. Using the study shown in Figure 2 as an example, if the subject was known to have moved by less than
x of a pixel from left to right,
y from head to foot and to have tilted the head by less than
θ radians in plane about a point (x 0 , y 0 ), then the maximum displacement-induced signal (
I ) for a pixel at (x , y) can be estimated as:
   2  2  ∂I (x, y) ∂I (x, y)
I (x, y) = 
x +
y ∂x ∂y

 2  ∂I (x, y) ∂I (x, y)
θ + −(y − y0 ) + (x − x0 ) ∂x ∂y (2)

5

Figure 3 Vulnerability maps produced from Figure 2(a) which reveal the spatial distribution of all potential signal changes that could occur as a result of displacements of
x =
y = 1/10 pixel and
θ = 0.02 rad (a) and
x =
y = 0 and
θ = 0.02 rad (b)

Applying this formula to the image shown in Figure 2(a) with
x =
y = 1/10 pixel and
θ = 0.002 rad, and dividing
I by the resting pixel intensity, results in the image shown in Figure 3(a). Note that most of the cortex is vulnerable to motion-induced changes, with typical signal changes of 10–20%, even for the small displacements considered. The image shown in Figure 3(a), which we refer to as a ‘vulnerability map’, indicates the potential combined effect of all possible movements up to the chosen maximum displacement in x , y, and θ. Figure 3(b) shows another vulnerability map, but this time calculated with
x =
y = 0 so that it shows the distribution of signal change induced by pure rotations of the head. Note that the most prominent features are the same as those that are highlighted in Figure 2(b) and (c). Vulnerability maps may be used as a thresholding device to subtract out potential motion effects. A signal change in a given pixel that was larger than the vulnerability for that pixel could then confidently be assigned to a cause other than motion. This approach is rather crude since, as can be seen from Figure 3, even modest displacements would lead to threshold criteria that would be unlikely to be exceeded by the expected genuine activation signals. A more rigorous approach is to quantify the exact displacements and correct for them, leaving a data set that can more confidently be analysed for evidence of genuine brain activation.

6 FUNCTIONAL NEUROIMAGING ARTIFACTS The image registration techniques already described provide a means of doing this; however, special care is required to ensure that the process of registration does not itself introduce artifacts. Since subpixel displacements are required, it is necessary to interpolate the image data to generate new interstitial pixel values.

6.1

Image Interpolation

Image matching is frequently performed using linear interpolation19 (see, for example, Woods et al.16 ). The effect that this has on MRI data can be determined with a simple experiment. An MR image, for example the T 1 -weighted image shown in Figure 4(a), is rotated using a computer, and new interpolated pixel values calculated to produce the image shown in Figure 4(b). The rotated image is then rotated back to its original orientation and the difference between the

original image and the doubly interpolated image is calculated. Figure 4(c) shows the result for linear interpolation. Substantial intensity errors have been introduced because the interpolation function does not match the structure of the image data. Much more faithful interpolation can be achieved by making use of the fact that MR data are strictly band limited and, therefore, their in-plane point spread function is the sinc function.20 Using sinc interpolation21 to rotate the images shown in Figures 4(a) and (b) produces much smaller differences [Figure 4(d)], with the maximum error now being less than 1% of the maximum signal.

6.2 In-Plane Registration

Sinc interpolation can form the basis of an in-plane misregistration correction scheme in which images are positionally matched with subvoxel precision and then compared in order

Figure 4 Displacement of image data by a fraction of a pixel requires image interpolation, which may introduce errors. These errors can be detected (a) by rotating a sample image by 10◦ to produce a second image (b) and then rotating (b) back to the original position. The final image, which has been interpolated twice, can then be subtracted from (a) to reveal any errors. Linear interpolation introduces substantial errors particularly at edges (c), whereas sinc interpolation is much more faithful to the original data structure (d)

FUNCTIONAL NEUROIMAGING ARTIFACTS

to detect changes that are independent of the transformations employed. A least-squares minimization procedure can be used to determine the required positional transformations. We employ the Levenberg–Marquart algorithm22 to minimize

χ = 2

(IA − IB )2 pixels N

(3)

where I A and I B are the intensities of corresponding pixels in the two images being matched, and N is the number of pixels included in the calculation of χ 2 . Minimization is achieved by rigid body rotation and translation of image A using sinc interpolation to generate new values for I A . Figure 5 illustrates the image registration procedure. A phantom [Figure 5(a)] was imaged twice with an in-plane displacement of approximately 2.5 pixels between the images. A difference image showing the effect of the displacement is shown in Figure 5(b). After χ 2 minimization, shifting one

7

image to match the other using sinc interpolation results in the difference image shown in Figure 5(c). The procedure has succeeded in reducing displacement-induced differences to the level of the background noise. For in vivo studies, a further refinement is generally required because subject movement can introduce image differences caused by the differential displacement of soft tissues. It is therefore advisable to restrict the registration to those parts of the image containing the tissue of interest. This can readily be achieved by segmenting23 one of the images.

6.3 Full Three-Dimensional Registration

The example illustrated in Figure 5 was restricted to inplane displacements. In a living subject, we must expect full three-dimensional movement to occur, and thus singleslice data do not provide sufficient information. Conventional multislice techniques can nominally provide the required

Figure 5 Image registration allows subject displacement to be corrected. (a) A phantom imaged twice with a differential offset of 2.5 mm results in substantial misregistration signals when the two images are subtracted (b). (c) Using an automatic image registration algorithm to determine the spatial shift required and sinc interpolation to reposition one image at the location of the other, reduces differences to the level of the background noise

8 FUNCTIONAL NEUROIMAGING ARTIFACTS spatial coverage, but problems can arise with uncertainties of data content at the boundaries between slices. The fidelity with which multislice data can be interpolated in the throughslice direction depends on the details of the slice profile. An idealized profile with uniform sensitivity in the selected slice and zero outside is uninterpolable because each slice is represented by independent isolated data. True three-dimensional volume data acquisitions (see also Image Formation Methods), in which phase encoding is used in the slice direction, are band limited in all directions and therefore can be faithfully interpolated using the sinc function. This type of acquisition is thus ideally suited to correction of arbitrary displacements. It has the drawback of being rather slower than the single-slice methods more commonly employed in functional MRI.

Figure 6 illustrates a motor stimulation study designed to detect changes in blood flow in draining veins, using three-dimensional acquisitions to image 40 slices. The subject performed hand exercises with one hand for the duration of a scan, changed hands for the next scan, and so on, for a total of six scans. Figure 6(a) shows one slice from one of the 40 slice sets. A cumulative difference image of the same slice [Figure 6(b)] shows obvious misregistration effects. In Figure 6(c) the images were registered before subtraction. Gross difference signals have now been eliminated from the brain, but some vessels can still be seen (arrow) as well as the scalp. Segmenting the image so that only the brain was used to calculate the transformations required for positional matching, but then transforming the original complete images, results in precise cancellation of signals from the brain and cerebral

Figure 6 A motor stimulation study in which each hand was exercised in turn as images were acquired repeatedly using an inflow sensitive 3D sequence (TR 40, TE 8, flip angle 20◦ , 192 × 256 matrix, 21 cm FOV, 40 slices of 1 mm) and a surface coil placed adjacent to the dominant hemisphere of the brain. (a) A single slice from the series and (b) cumulative difference image (right hand activation minus left hand activation). In (c) the images have been registered before subtraction and some vessels can still be seen (arrows) as well as the scalp. (d) Segmenting the image so that only the brain is used for detecting how the subject moved, but retaining all the information in the positionally matched images results in complete cancellation of brain and cortical vessel signals but leaves larger residual signals from the scalp

FUNCTIONAL NEUROIMAGING ARTIFACTS

vessels but leaves residual signals in the scalp [Figure 6(d)]. This indicates that the relative positions of the brain and scalp have altered. However, although a task was being performed, no significant signal changes were detected in the brain or its draining vessels when these were precisely matched for position.

7

CONCLUSION

Functional neuroimaging using MRI is a new and challenging area of research. The detection of signal changes that correlate with the applied stimulus protocol is not in itself sufficient evidence for localized brain activation as there are other sources of such signals. Most notably, subject movement during the course of an examination can result in task locked signal changes that may not be discriminated against by correlation and statistical analyses. Single-slice studies are particularly vulnerable because of the lack of information about tissues adjacent to the slice, which makes it difficult to detect the effects of through-slice movement. Image registration provides a means of correcting for the effect of interimage displacements, but great care is required to ensure that the correction procedure itself does not introduce spurious results.

8

RELATED ARTICLES

Brain: Sensory Activation Monitored by Induced Hemodynamic Changes with Echo Planar MRI; Image Formation Methods; Whole Body Magnetic Resonance Artifacts.

9

REFERENCES 1. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. A. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H.-M. Cheng, T. J. Brady, and B. R. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 2. S.-G. Kim, J. Ashe, K. Hendrich, J. M. Ellermann, H. Merkle, K. Ugurbil, and A. P. Georgopoulo, Science, 1993, 261, 615. 3. S. Ogawa, T. M. Lee, A. R. Kay, and D. W. Tank, Proc. Natl. Acad. Sci. USA, 1990, 87, 9868. 4. S. Lai, A. L. Hopkins, E. M. Haacke, D. Li, B. A. Wasserman, P. Buckley, L. Friedman, H. Meltzer, P. Hedera, and R. Friedland, Magn. Reson. Med., 1993, 30, 387. 5. R. Turner, P. Jezzard, H. Wen, K. K. Kwong, D. Le Bihan, T. Zeffiro, and B. Balaban, Magn. Reson. Med., 1992, 25, 390. 6. A. Connelly, G. D. Jackson, R. J. Frackowiak, J. W. Belliveau, F. Vargha-Khadem, and D. G. Gadian, Radiology, 1993, 188, 125. 7. R. Myers, T. J. Spinks, and D. J. Brooks, in Quantitative Methods in Neuroanatomy, ed. M. G. Steward, Wiley, Chichester, 1992, p. 117.

9

8. K. J. Friston and R. S. J. Frackowiack, in Brain Work and Mental Activity. Quantitative Studies with Radioactive Tracers, ed. N. A. Lassen, D. H. Ingvar, M. E. Raichle, and L. Friberg, Munskgaard, Copenhagen, 1991, p. 267. 9. P. A. Bandettini, A. Jesmanowicz, E. C. Wong, and J. Hyde, Magn. Reson. Med., 1993, 30, 161. 10. R. T. Constable, G. McCarthy, T. Allison, A. W. Anderson, and J. C. Gore, Magn. Reson. Imag., 1993, 11, 451. 11. R. M. Weisskoff, J. Baker, J. Belliveau, T. L. Davis, K. K. Kwong, M. S. Cohen, and B. R. Rosen, in Proceedings of the 12th Annual Meeting, SMRM, New York , Society of Magnetic Resonance in Medicine, Berkeley, CA, 1993, p. 7. 12. X. Hu and S.-G. Kim, Magn. Reson. Med., 1994, 31, 495. 13. J. V. Hajnal, R. Myers, A. Oatridge, J. E. Schwieso, I. R. Young, and G. M. Bydder, Magn. Reson. Med., 1994, 31, 283. 14. D. L. G. Hill, D. J. Hawkes, J. E. Crossman, M. J. Gleeson, T. C. Cox, E. E. Bracey, A. J. Strong, and P. Graves, Br. J. Radiol., 1991, 64, 1030. 15. C. A. Pelizzari, G. T. Y. Chen, D. R. Spelbring, R. R. Weichselbaum, and C. T. Chen, J. Comput. Assist. Tomogr., 1988, 13, 20. 16. R. P. Woods, S. R. Cherry, and J. C. Mazziotta, J. Comput. Assist. Tomogr., 1992, 16, 620. 17. R. P. Woods, J. C. Mazziotta, and S. R. Cherry, J. Comput. Assist. Tomogr., 1993, 17, 536. 18. P. Bloomfield, Fourier Analysis of Time Series—An Introduction, Wiley, Chichester, 1976. 19. E. Kreysig, Advanced Engineering Mathematics, 4th edn, Wiley, Chichester, 1979, p. 774. 20. A. K. Jain, Fundamentals of Digital Image Processing, PrenticeHall, Englewood Cliffs, NJ, 1992, p. 84. 21. A. K. Jain, Fundamentals of Digital Image Processing, PrenticeHall, Englewood Cliffs, NJ, 1992, p. 89. 22. W. H. Press, S. A. Tenkolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C , 2nd edn, Cambridge University Press, Cambridge, 1992, p. 683. 23. D. H. Ballard and C. M. Brown, Computer Vision, Prentice-Hall, Englewood Cliffs, NJ, 1982.

Biographical Sketch Joseph V. Hajnal. b 1958. B.Sc., Ph.D., 1984, University of Bristol, UK. Postdoctoral research fellow, Department of Physics, University of Bristol, 1984–86. Postdoctoral research fellow, School of Physics, University of Melbourne, Australia, 1986–89. Research scientist, GEC Hirst Research Centre, and Honorary Senior Lecturer in Diagnostic Radiology, Royal Postgraduate Medical School, Hammersmith Hospital, 1990–present. Approx. 80 publications. Current research speciality: in vivo MRI techniques and applications.

HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN

Hemodynamic Changes Owing to Sensory Activation of the Brain Monitored by Echo-Planar Imaging Peter A. Bandettini, Jeffrey R. Binder, Edgar E. DeYoe, and James S. Hyde Medical College of Wisconsin, Milwaukee, WI, USA

1 INTRODUCTION In recent years, it has been demonstrated that MRI is capable of detecting changes in cerebral blood volume,1 ¯ow,2 and oxygenation,2±5 that accompany an increase in neuronal activity. MRI methods that observe these activation-induced changes have been termed functional MRI (fMRI). The most widely used fMRI method for the noninvasive mapping of human brain activity is based on blood oxygenation level dependent (BOLD) contrast.6 A localized signal increase in activated cortical regions is generally observed using a time-course collection of T2*-weighted images. A working model of this phenomenon is that an increase in neuronal activity causes local vasodilatation, which in turn causes blood ¯ow to increase in such a manner that the amount of paramagnetic deoxyhemoglobin in the local vasculature is reduced. This reduction causes an increase in spin coherence and, therefore, an increase in signal when using T2- and T2*weighted pulse sequences. Support for the hypothesized BOLD contrast mechanism, as it relates to fMRI, comes from several studies. Local cerebral blood oxygenation has been observed to increase with neuronal activity.7±10 Brain tissue T2, T2*, and T2' have been shown to be decreased and increased by cerebral blood oxygenation decreases and increases, respectively.6,11±15 In addition, brain activation has been shown to increase T2, T2*, and T2',16±20 in proportions that show general agreement with mathematical simulations based on simpli®ed models of the cerebral vasculature.21,22 The applicability of fMRI depends on the degree to which the induced signal enhancement magnitude, location, and timing can be correlated with underlying neuronal activation. The signal enhancement magnitude is not only affected by MRI parameters, but is also dependent on hemodynamic factors (i.e. blood volume, oxygenation, vessel orientation, and radii), which vary signi®cantly from voxel to voxel. Correlation between BOLD signal enhancement magnitude and cerebral blood ¯ow has, nevertheless, been observed. Visual cortex activation studies have shown that BOLD signal enhancement has essentially the same ¯icker frequency dependency2 as that of cerebral blood ¯ow changes observed using positron emission tomography (PET).23

1

The fMRI signal enhancement location and distribution is also an issue. The upper limit of spatial resolution of fMRI may be in¯uenced by the degree to which an activation-induced modulation of the velocity or oxygenation of rapidly in¯owing spins or large collecting veins may contribute to the signal change. Generally, the larger the collecting vessel, the more spatially removed the signal change is from the area of neuronal activation. Such vessel size weighting might depend strongly on the pulse sequence, ®eld strength, and resolution used.16±26 Nevertheless, studies have shown that fMRI can reveal activity localized to patches of cortex smaller than 1.5 mm.27 The temporal resolution of fMRI depends on the latency and consistency of the activation-induced signal change. Rise latency of 8±9 s from stimulus onset to maximal signal change and a fall latency of 9±10 s from stimulus cessation to baseline signal have been reported.2,28,29 The hemodynamic response time sets upper limits on the functional temporal resolution. A signi®cant increase from baseline is generally observed to take place within 2±3 s after stimulus onset.2,28,29 Activation durations of less than 1 s are detectable30,31 and relative differences (in the rise time from adjacent regions or from different experiments) in the onset of signal enhancement may be discriminated to within a second.32 Platforms on which fMRI are performed vary considerably. Primary differences are in ®eld strength, pulse sequence, gradient and radiofrequency coil hardware, and postprocessing methods. The many trade-offs that exist between platform types have not been completely characterized, but it is clear that fMRI based on BOLD contrast bene®ts from high ®eld strength, high system stability, high signal-to-noise ratio (SNR), and minimal pulsatile motion sensitivity. From reported results, it appears that these criteria are most readily met using echo planar imaging. Nevertheless, other techniques may be suitable. Initial successes in the performance of fMRI using BOLD contrast were published by groups using either EPI at 1.5 T,2,3 and 4 T,16 or fast multishot gradient-recalled imaging techniques at 4 T,4 and at 2 T.5 Results have since been reported using fast multishot gradient-recalled imaging techniques at 1.5 T.33±35 Pulsatile brain, cerebrospinal ¯uid, and blood motion apparently cause nonrepeatable ghosting patterns in sequential images obtained by conventional multishot techniques,36,37 thus adding signi®cantly to the image signal variation in time. Multishot spiral-scan techniques have lower sensitivity to these contaminating pulsatile effects,36,37 and have been used successfully to perform high resolution fMRI at 1.5 T.27 The use of a time course of images in conjunction with control over activation timing allows for the application of postprocessing methods which include the use of z maps,38 and other statistical techniques.35 In addition, other methods, including Fourier analysis,39 temporal cross correlation,39 and time±frequency analysis40 have been successfully applied. While signi®cant signal changes are easily observed using the most conservative statistical tests, a standardized method by which functional images are created has not been established. Using EPI, studies of the ¯uctuations in the susceptibilityweighted signal from a quiescent brain have been carried out,41 not only to determine the nature of the noise for application to statistical tests, but also to obtain potentially useful physiological information, and to determine the actual relationship between functional contrast to noise and ®eld strength.42

2 HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN Because of the ease of use and accessibility of fMRI, many areas of the brain and many different tasks have already been studied. Studies that have been carried out include those of primary cortical regions including visual cortex,2,4,5,16,29,35,43 motor cortex,2,3,26,33,34,39,44±46 and auditory cortex.47 Studies of higher cognitive function, including word generation,48,49 higher visual processing,35,43 visual recall,50 complex motor control,46 and single word semantic processing,51 have also been performed. In this article, applications of fMRI to the mapping of cortex activated by sensory stimulation are summarized using time-course collection of gradient-recalled echo planar images at 1.5 T.3,39 and temporal cross correlation postprocessing techniques.39 Speci®cally, auditory cortex regions were differentially activated by noise and speech sounds, and visual cortex regions were selectively activated by visual stimuli of different eccentricity.

2 POSTPROCESSING METHODOLOGY Stimuli are presented in a repetitive on/off fashion for several cycles throughout each time course. Foci of brain activation are identi®ed by cross correlation of the time course of each voxel with a reference vector resembling the expected activation-induced response.39 In general, a reference vector may be obtained by: (a) choosing the voxel containing what appears to be the `best' temporal response; (b) averaging, in time, the activation cycles in a `best temporal response' voxel and then duplicating the time-averaged cycle for the length of the time course; (c) averaging in space several of the `best temporal response' voxels; (d) synthesizing a reference vector; or (e) choosing a vector obtained from principal component analysis of the time course. Pixels having a temporal correlation coef®cient below a given threshold (typically 0.4±0.6) are removed. After thresholding, the vector product of the reference vector with each of the surviving time courses is calculated to yield an index of change in the signal magnitude. These `activation' images are then colorized and superimposed on high resolution anatomical scans of the same slice obtained in the same imaging session.

3 PULSE SEQUENCE AND HARDWARE All studies presented in this article were performed using single-shot 64  64 gradient-recalled EPI (TE = 40 ms) on a clinical 1.5-T GE Signa scanner. To perform EPI without additional stress to the standard gradient ampli®ers, we used an insertable balanced torque three-axis head gradient coil designed for rapid gradient switching.52 To obtain high-quality images throughout the entire brain volume, a shielded quadrature elliptical endcapped transmit/receive birdcage radiofrequency coil was used.53 Typically, single- or multi-slice time-course series of 64±1024 images were obtained with a TR of 0.5±3 s, ¯ip angle of 65±90 , ®eld of view (FOV) of 24 cm, and slice thickness of 4±10 mm.

4

AUDITORY STIMULI

This study47 presents ®ndings using fMRI of brain regions involved in auditory speech perception. Speci®cally, regions activated by speech sounds (words and pseudowords) and nonspeech sounds (noise) were compared. Five right-handed subjects were tested. Symmetric lateral sagittal slices (slice thickness 10 mm) of the left and right hemispheres were obtained, centered at positions 8 mm medial to the most lateral point of the temporal lobe on each side. In each time-course series, 64 sequential images were collected (TR = 3 s), during which activation alternated with baseline every 9 s (6 images per cycle, 18 s per cycle, 10 cycles). During baseline periods, subjects heard only the ambient scanner noise. During activation periods, prepared digitized auditory stimuli were delivered to the subject via air conduction through a semi-rigid 1-cm-bore plastic tube. The tube conducted the sound stimulus approximately 20 feet (6 m) from control room wall to the subject, at which point a Y-connector split the tube for binaural stimulation through a tightly ®tting headset with occlusive earplugs to reduce scanner noise exposure. Subjects passively listened to stimuli; no response was required. Stimuli differed in both semantic and acoustic (frequency modulation) content. White noise presentation was compared to presentation of nouns (e.g. `barn'), and pseudowords (e.g. `narb') with stimuli matched for duration, presentation rate, average sound pressure level, and spectral range. All voxels having a temporal correlation coef®cient to a sinusoidal reference vector below 0.5 were removed. The activation images were superimposed on high resolution scans of the same slices. Figure 1 illustrates typical results from two of the subjects studied with white noise, pseudowords, and words, respectively. In these and all other subjects studied, the area activated by white noise was considerably smaller than that activated by speech sounds, and was restricted to the dorsal aspect of the superior temporal gyrus. In most instances, this region coincided with or included the transverse temporal (Heschl's) gyrus. Presentation of speech sounds activated a larger region, including more anterior and posterior areas of the dorsal superior temporal gyrus, as well as cortex in or near the superior temporal sulcus bilaterally. Both words and pseudowords differed signi®cantly from noise bilaterally, while no signi®cant differences were seen between word and pseudoword conditions. Activity occurred symmetrically in the left and right temporal lobes. Unlike white noise, therefore, processing of speech sounds appears to elicit extensive participation of auditory association areas, even when the subject is not engaged in any `active' task.

5

VISUAL STIMULI

Much of the utility of fMRI depends on its ability to depict spatial patterns of neural activity. To test this capacity in the visual system, fMRI was used for retinotopic mapping of primary visual cortex activation.43 Dynamic, computer graphics-based visual images were directly projected onto the subjects' retinae. The image generator was a modi®ed Sharp XG2000U video projector driven by

HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN

3

Figure 1 Sagittal images of the left and right temporal lobes of two subjects. Demonstrated are regions activated during passive listening to (A) white noise, (B) pseudowords, and (C) words. The area activated by white noise was smaller than the area activated by either pseudowords or words. The area activated by pseudowords was generally the same size and shape as the area activated by words. (Reproduced with permission of the American Neurological Association, 1994)

Cambridge Instruments VSG video graphics board installed in a personal computer. The image plane was then viewed through a custom optical system that included a wide ®eld, magnifying eyepiece, a 45 prism, and additional objective lenses for adjusting magni®cation and minimizing chromatic aberration. Two sets of imaging optics were combined to provide full binocular viewing.54 To map the retinotopic organization of the visual cortex, three highly trained subjects viewed a small white ®xation dot on a uniform black or gray ®eld subtending 60 of visual angle. A black-and-white checkered annulus surrounding the ®xation point was presented for 5 on/off cycles of 10 s on and 10 s off. Time-course series (TR = 2 s) of 100 images (slice thickness 8 mm) were used. When on, the checkered pattern was either counter-phase modulated or ¯ickered at 6±8 Hz. Six successively larger annuli were tested. The width of each annulus as well as the check size were scaled in proportion to eccentricity. Only voxels having a correlation coef®cient above 0.55 with respect to a chosen reference vector were used in the creation of functional images. All functional images were then superimposed on high resolution anatomical scans. Figure 2 illustrates the relative sizes of the annuli and shows the corresponding brain activation images. In these experiments, the subject passively viewed the stimuli and was not required to respond to them. A small checkerboard annulus presented at the ®xation point elicited activation in striate cor-

tex only at the occipital poles bilaterally. Annuli presented at increasing eccentricities activated successively more anterior regions of the calcarine sulcus. The most eccentric stimulus activated only the anterior calcarine cortex while sparing the occipital poles. Detailed examination of the sequence of activity foci in Figure 2, shows a progression that closely follows the folded cortical mantle within the calcarine sulcus. While such a precise progression is not always observed, these data do show that, under optimal conditions, a detailed mapping of the visual ®eld representation is possible with fMRI. Resolution is not limited by the coarseness of the distribution of large blood vessels, even though such vessels may sometimes introduce artifacts.

6

CONCLUSIONS

MRI of human brain activation using BOLD contrast is a relatively new functional brain imaging method. Accompanying the novelty of the technique are many unknowns regarding the upper limits of spatial and temporal resolution as well as an unclear understanding of physiological and the biophysical mechanisms that regulate hemodynamic changes. In addition, the ways in which the hemodynamic changes affect the magnetic resonance signal are incompletely understood. Never-

4 HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN theless, the applications described here and elsewhere empirically establish the utility of this approach. These studies demonstrate observation by fMRI of human brain activation by sensory stimulation. Regions activated in temporal lobes by various speech and nonspeech auditory stimuli were observed. In addition, retinotopic organization of the primary visual cortex was observed using stimuli of varying visual ®eld eccentricity. Additionally, in our laboratory, preliminary studies involving activation by tactile, aromatic, and taste stimuli are in progress. fMRI is a technique that holds great promise in uncovering unique and useful information about brain function and physiology. 7

RELATED ARTICLES

Echo-Planar Imaging; Methods and Applications of Diffusion MRI; MR-Guided Therapy in the Brain; Susceptibility Effects in Whole Body Experiments. 8

Figure 2 Axial brain activation images created by passive viewing of visual stimuli with six different eccentricities while ®xating at the center. The active foci traveled in an anterior direction along the calcarine ®ssure as the stimulus became more peripheral

REFERENCES

1. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen, Science, 1991, 254, 716. 2. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H. M. Cheng, T. J. Brady, and B. R. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 3. P. A. Bandettini, E. C. Wong, R. S. Hinks, R. S. Tikofsky, and J. S. Hyde, Magn. Reson. Med., 1992, 25, 390. 4. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellermann, S.-G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1992, 89, 5951. 5. J. Frahm, H. Bruhn, K.-D. Merboldt, and W. Hanicke, JMRI, 1992, 2, 501. 6. S. Ogawa, T. M. Lee, A. R. Kay, and D. W. Tank, Proc. Natl. Acad. Sci. USA, 1990, 87, 9868. 7. A. Grinvald, R. D. Frostig, R. M. Siegel, and E. Bratfeld, Proc. Natl. Acad. Sci. USA, 1991, 88, 11559. 8. P. T. Fox and M. E. Raichle, Proc. Natl. Acad. Sci. USA, 1986, 83, 1140. 9. R. D. Frostig, E. E. Lieke, D. Y. Ts'o, and A. Grinvald, Proc. Natl. Acad. Sci. USA, 1990, 87, 6082. 10. A. Villringer, J. Planck, C. Hock, L. Schleinkofer, and U. Dirnagl, Neurosci. Lett., 1993, 154, 101. 11. R. Turner, D. Le Bihan, C. T. Moonen, D. Despres, and J. Frank, Magn. Reson. Med., 1991, 22, 159. 12. B. E. Hoppel, R. M. Weisskoff, K. R. Thulborn, J. B. Moore, K. K. Kwong, and B. R. Rosen, Magn. Reson. Med., 1993, 30, 715. 13. M. K. Stehling, F. Schmitt, and R. Ladebeck, JMRI, 1993, 3, 471. 14. A. J. DeCrespigny, M. F. Wendland, N. Derugin, E. Kozniewska, and M. E. Moseley, Magn. Reson. Med., 1992, 27, 391. 15. P. Jezzard, F. Heineman, J. Taylor, D. Taylor, H. Wen, R. S. Balaban, and R. Turner, NMR Biomed., 1994, 7, 35. 16. R. Turner, P. Jezzard, H. Wen, K. K. Kwong, D. Le Bihan, T. Zef®ro, and R. S. Balaban, Magn. Reson. Med., 1993, 29, 277. 17. R. S. Menon, S. Ogawa, D. W. Tank, and K. Ugurbil, Magn. Reson. Med., 1993, 30, 380. 18. B. E. Hoppel, J. R. Baker, R. M. Weisskoff, and B. R. Rosen, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1384.

HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN 19. P. A. Bandettini, E. C. Wong, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, NMR Biomed., 1994, 7, 12. 20. P. A. Bandettini, E. C. Wong, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 169. 21. S. Ogawa, R. S. Menon, D. W. Tank, S.-G. Kim, H. Merkle, J. M. Ellermann, and K. Ugurbil, Biophys. J., 1993, 64, 803. 22. R. P. Kennan, J. Zhong, and J. C. Gore, Magn. Reson. Med., 1994, 31, 9. 23. P. T. Fox and M. E. Raichle, J. Neurophysiol., 1984, 51, 1109. 24. J. Frahm, K.-D. Merboldt, and W. Hanicke, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1427. 25. J. H. Duyn, C. T. W. Moonen, R. W. de Boer, G. M. van Yperen, and P. R. Luyten, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 168. 26. S. Lai, A. L. Hopkins, E. M. Haacke, D. Li, B. A. Wasserman, P. Buckley, L. Friedman, H. Meltzer, P. Hedera, and R. Friedland, Magn. Reson. Med., 1993, 30, 387. 27. S. A. Engel, D. E. Rumelhart, B. A. Wandell, A. T. Lee, G. H. Glover, E. J. Chichilnisky, and M. N. Shadlen, Nature, 1994, 369, 525. 28. E. A. DeYoe, J. Neitz, P. A. Bandettini, E. C. Wong, and J. S. Hyde, Proc. Xth Annu. Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 1824. 29. A. M. Blamire, S. Ogawa, K. Ugurbil, D. Rothman, G. McCarthy, J. M. Ellermann, F. Hyder, Z. Rattner, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1992, 89, 11069. 30. P. A. Bandettini, E. C. Wong, E. A. DeYoe, J. R. Binder, S. M. Rao, D. Birzer, L. D. Estkowski, A. Jesmanowicz, R. S. Hinks, and J. S. Hyde, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1382. 31. J. W. Belliveau, J. R. Baker, K. K. Kwong B. R. Rosen, J. S. George, C. J. Aine, J. D. Lewine, J. A. Sanders, G. V. Simpson, and J. Foxe, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 6. 32. J. R. Binder, A. Jesmanowicz, S. M. Rao, P. A. Bandettini, T. A. Hammeke, and J. S. Hyde, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1383. 33. Y. Cao, V. L. Towle, D. N. Levin, and J. M. Balter, JMRI, 1993, 3, 869. 34. A. Connelly, G. D. Jackson, R. S. Frackowiak, J. W. Belliveau, F. Vargha-Khadem, and D. G. Gadian, Radiology, 1993, 188, 125. 35. W. Schneider, D. C. Noll, and J. D. Cohen, Nature, 1993, 365, 150. 36. G. H. Glover and J. M. Pauly, Magn. Reson. Med., 1992, 28, 275. 37. G. H. Glover, A. T. Lee, and C. H. Meyers, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 197. 38. D. Le Bihan, P. Jezzard, R. Turner, C. A. Cuenod, L. Pannier, and A. Prinster, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 11. 39. P. A. Bandettini, A. Jesmanowicz, E. C. Wong, and J. S. Hyde, Magn. Reson. Med., 1993, 30, 161. 40. B. Biswal, P. A. Bandettini, A. Jesmanowicz, and J. S. Hyde, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 722. 41. R. M. Weisskoff, J. Baker, J. Belliveau, T. L. Davis, K. K. Kwong, M. S. Cohen, and B. R. Rosen, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 7. 42. P. Jezzard, D. Le Bihan, C. Cuenod, L. Pannier, A. Prinster, and R. Turner, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1392. 43. E. A. DeYoe, P. A. Bandettini, J. Neitz, D. L. Miller, and P. Winans, J. Neurosci. Meth., 1994, 54, 171. 44. S.-G. Kim, J. Ashe, A. P. Georgopoulos, H. Merkle, J. M. Ellermann, R. S. Menon, S. Ogawa, and K. Ugurbil, J. Neurophysiol., 1993, 69, 297.

5

45. S.-G. Kim, J. Ashe, K. Hendrich, J. M. Ellermann, H. Merkle, K. Ugurbil, and A. P. Georgopoulos, Science, 1993, 261, 615. 46. S. M. Rao, J. R. Binder, P. A. Bandettini, T. A. Hammeke, F. Z. Yetkin, A. Jesmanowicz, L. M. Lisk, G. L. Morris. W. M. Meuller, L. D. Estkowski, E. C. Wong, V. M. Haughton, and J. S. Hyde, Neurology, 1993, 43, 2311. 47. J. R. Binder, S. M. Rao, T. A. Hammeke, F. Z. Yetkin, A. Jesmanowicz, P. A. Bandettini, E. C. Wong, L. D. Estkowski, M. D. Goldstein, V. M. Haughton, and J. S. Hyde, Ann. Neurol., 1994, 35, 672. 48. G. McCarthy, A. M. Blamire, D. L. Rothman, R. Gruetter, and R. S. Shulman, Proc. Natl. Acad. Sci. USA, 1993, 90, 4952. 49. R. M. Hinke, X. Hu, A. E. Stillman, S.-G. Kim, H. Merkle, R. Salmi, and K. Ugurbil, Neuroreport, 1993, 4, 675. 50. D. Le Bihan, R. Turner, T. Zef®ro, C. A. Cuenod, P. Jezzard, and V. Bonnerot, Proc. Natl. Acad. Sci. USA, 1993, 90, 11802. 51. J. R. Binder, S. M. Rao, T. A. Hammeke, J. A. Frost, P. A. Bandettini, A. Jesmanowicz, and J. S. Hyde, Arch. Neurol., 1995, 52, 593. 52. E. C. Wong, P. A. Bandettini, and J. S. Hyde. Proc. Xth Annu. Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 105. 53. E. C. Wong, E. Boskamp, and J. S. Hyde, Proc. XIth Annu. Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 4015. 54. E. A. DeYoe, J. Neitz, D. Miller, and J. Wieser, Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p, 1394.

Acknowledgements This work was supported in part, by grants CA41464 and RR01008 from the National Institute of Health. P.A.B. thanks GE Medical Systems for ®nancial support. In adition, the support of R. Scott Hinks at GE Medical Systems and of Donald Dickerson, Lloyd D. Estkowski, Andrew S. Greene, Thomas A. Hammeke, Victor M. Haughton, Andre Jesmanowicz, George L. Morris, Wade M. Meuller, Joel B. Myklebust, David Miller, Jay Neitz, Steven M. Rao, Elliot Stein, Eric C. Wong, F. Zerrin Yetkin, and Jeffrey R. Zigun, at the Medical College of Wisconsin, are gratefully appreciated.

Biographical Sketches Peter A. Bandettini. b 1966. B.S., 1989, Ph.D., 1994, Biophysics, Medical College of Wisconsin, USA (supervisors James S. Hyde and R. Scott Hinks). Postdoctoral fellow, Massachusetts General Hospital NMR Center, 1994±95. Approx. 15 publications. Research specialties: functional MRI contrast mechanisms, postprocessing techniques, and applications. Jeffrey R. Binder. b 1958. B.M., 1980, M.D., 1986, University of Nebraska, USA. Neurology Residency, 1987±90. Neurological Institute of New York. Fellow in cerebrovascular disease, 1990±92. Neurological Institute (with J. P. Mohr). Currently Assistant Professor of Neurology, Medical College of Wisconsin, USA. Approx. 15 publications. Research specialties: behavioral neurology, neuroscience applications of functional MRI, cerebrovascular disease. Edgar A. DeYoe III. b 1950. B.S. Electrical Engineering, 1972; Ph.D. 1983, Experimental Psychology; Ph.D. 1983, Neuroscience, University of Rochester, USA (Advisor: Robert W. Doty). Postdoctoral fellowship at California Institute of Technology (Supervisor: David Van Essen). Currently assistant professor in Department of Cellular Biology and Anatomy with adjunct appointment to Department of Biophysics at the Medical College of Wisconsin, Milwaukee, WI. Approx. 21 publications. Research interests: Neural mechanisms of visual perceptionÐanatomy and physiology; Functional magnetic resonance imaging (fMRI) of the human visual system; development of systems for testing normal vision and visual system; development of systems for testing normal vision and visual dysfunction during fMRI.

6 HEMODYNAMIC CHANGES OWING TO SENSORY ACTIVATION OF THE BRAIN James S. Hyde. b 1932. B.S., 1954, Ph.D., 1959, Physics, M.I.T, Cambridge, MA, USA. Member of the scienti®c staff, Varian Associates, 1959±75, working under the direction of W. A. Anderson and M. E. Packard. Professor of Biophysics, Medical College of Wisconsin,

1975±present. Approx. 275 publications, 27 patents. Research specialties: ESR spectroscopy, ENDOR, musculoskeletal MRI, surface coils, functional MRI, electron and nuclear spin physics, and magnetic resonance instrumentation.

POSTOPERATIVE TRAUMA OBSERVED BY MRI

Postoperative Trauma Observed by MRI David N. F. Harris Royal Postgraduate Medical School, London, UK

1 INTRODUCTION This article reviews the use of magnetic resonance imaging (MRI) in patients undergoing major cardiac and vascular surgery, with particular emphasis on the problems presented when imaging patients immediately after major surgery. Up to 80% of patients develop neuropsychological defects following coronary artery bypass surgery (CABS), with 38% persisting at 1 year; this is thought to indicate permanent damage. Of CABS patients, 1±2% develop overt strokes with a higher incidence reported following valve surgery. The cause of the defects is not known, but cerebral emboli of both air and microaggregates, hypoperfusion, and release of cytokines from the trauma to blood cells by the cardiopulmonary bypass circuit have been implicated. It is not yet clear whether these defects are speci®cally related to cardiopulmonary bypass as the incidence of neuropsychological defects in noncardiac surgery varies widely in reported studies, and it is dif®cult to ®nd a satisfactory control group for CABS patients. Neuropsychological testing shows a high variability between patients, and performance on tests improves with practice (learning): different investigators also produce different results. Accuracy can be improved by using multiple tests and a battery of 10 is commonly used, but this reduces the sensitivity and makes assessment of subtle defects dif®cult. X-Ray computerized tomography (CAT) scanning is the most widely used test for detecting cerebral trauma; it shows abnormalities in patients who have suffered major neurological events, but often does not show such anomalies in patients with less severe de®cits. The development of MRI offers the attractive possibility for sensitive and objective assessment of cerebral structure and function noninvasively.

2 CARDIAC SURGERY 2.1 Focal Defects Schmidt et al.1 investigated patients with 1.5 T MRI before and 1±3 weeks after CABS and showed an increase in focal defectsÐinfarcts, lacunar lesions, and cortical atrophy. In 50 patients, Peden et al.2 showed a high incidence of defects in the preoperative scans of CABS patients, with a signi®cant increase in focal defects in 35% of patients 1 week after surgery; 20% of the defects persisted at 8 weeks. New defects were more common in those who had preoperative defects, and imaging defects were associated with neuropsychological de®cits.

2.2

1

Swelling

To assess the incidence in the early postoperative phase we looked at six patients before, immediately after the end of surgery before going to Intensive Care, and 1 week after surgery.3 All patients showed marked cerebral edema on the immediate postoperative images, which returned to normal when imaged 1±3 weeks later (Figure 1). The cause of this swelling was not known, but it was suggested that it might re¯ect cerebral ischemia at the end of bypass as shown by jugular bulb desaturation. 2.3

High-energy Phosphates

To see if there was any evidence of ischemia, we studied eight patients with MR spectroscopy immediately after cardiac surgery, four with 1H spectroscopy and four with 31P spectroscopy. There was no visible lactate either globally or regionally, the ratio of phosphocreatine to ATP was maintained, and there was no intracellular acidosis. However, phosphocreatine/inorganic phosphate was increased immediately after the operation, which suggests rebound replacement of energy stores following recovery from temporary cerebral ischemia during cardiopulmonary bypass: intraoperative studies would be needed to test this hypothesis further.4 3 3.1

MAJOR VASCULAR SURGERY Aortic Aneurysm

It has been noted that many patients become confused 2±3 d after successful aortic aneurysm surgery. This was thought to be due to hypoxia over the ®rst postoperative nights, but MR images in three patients who also exhibited confusion on neuropsychological testing showed cerebral swelling on the second and third postoperative day similar to that found after CABS, though to a smaller degree. One patient also showed cerebral swelling immediately after the operation which had returned to normal 14 h later. 3.2

Carotid Endarterectomy

We studied six patients immediately after carotid endarterectomy to see if cerebral edema was localized to the operative side. Four patients showed signi®cant swelling, but no lateralizing signs were seen on FLAIR imaging. With subtraction imaging, two patients with previous strokes showed deviation of the ventricles away from the infarct.

4

TECHNICAL REQUIREMENTS FOR EARLY POSTOPERATIVE SCANS

When imaging patients in the early postoperative period, great care must be taken to ensure the safety of the patients. As with any MR scan of the head, the patient must be still, with their head enclosed in a coil and inaccessible in the mag-

2 POSTOPERATIVE TRAUMA OBSERVED BY MRI

Figure 1 FLAIR (6000/160/2100) sequences before cardiac surgery (a) and 40 min after the end of cardiopulmonary bypass (b) in a 62-year-old patient. Sulci are clearly seen in (a) and are almost obliterated in (b)

net during the scan. Following any anesthetic there is a risk of nausea and vomiting, respiratory obstruction, or depression, as well as cardiovascular collapse. The incidence of complications depends on the length and severity of the operation, and patients are closely monitored in the recovery room by highly trained staff for between 30 min and 8 h following routine surgery so that such complications can be prevented or treated rapidly: this requires instant access to the patient. When discharged from the recovery area, the patient is considered to be safe under occasional and possibly unskilled observation. Any of the above complications could be fatal if they occurred unnoticed during a scan when access is by no means instant. There is therefore a `no-go' time after an operation, similar to the recovery time, during which it is dif®cult to examine patients safely with MR. After this time imaging can be performed, but it would be advisable to monitor the ECG and oxygen saturation (SaO2) throughout the scan within 24 h of any operation; this is clearly an arbitrary limit. If the operation was major or prolonged, then measurement of blood pressure should be considered, and it should be possible to transfuse intravenous ¯uids at any desired rate. Usually this would require a suitably trained member of staff (anesthetist, critical care nurse) present in the magnet room. If it is important to acquire MR scans within this time, it is safer to keep the patient under anesthesia throughout the scan and the transport to and from the MR unit; this requires the `metal check' to be performed before the operation. Criticallyill patients can be transferred safely, but it requires a protocol with close attention to detail and safety. For the images immediately following CABS and aortic aneurysm surgery,

patients were required to be hemodynamically stable, with no likelihood of requiring cardiac pacing or massive transfusion. The system described by Peden et al.2 was used with minor modi®cations. All cables were enclosed together in tubing so that `spaghetti knitting' and disconnections could be avoided. All intravenous extensions and infusions were attached and primed in the operating room to minimize hemodynamic changes, and anesthesia was maintained by infusion (fentanyl/ propofol). Suf®cient supplies of intravenous ¯uids (crystalloid, colloid, and blood) with a pressure infuser were available in the magnet room, and suitable vasoactive drugs were brought with the patient from the operating room (OR). The patient was transferred from the operating table to an MR compatible trolley and to the MR magnet couch using a ¯exible roller to ensure a controlled, smooth transfer; monitoring of ECG, invasive blood pressure, and oxygen saturation was interrupted for less than 10 s. Intermittent positive pressure ventilation was provided using a jet ventilator5 with pressure sensed at the endotracheal tube to avoid compression losses in tidal volume over the 5 m from head to ventilator. Capnography was used to con®rm the adequacy of ventilation and to warn of disconnections in the circuit as standard pressure sensing alarms do not work reliably over that distance. The patient's airway is not visible during the scan and disconnections are easily possible as the patient is moved into the magnet bore; if unnoticed this would be catastrophic. Two anesthesiologists remained in the magnet room throughout the procedure, continuing the same monitoring that had been provided in the operating room. With this protocol it has been possible to investigate patients immediately following major cardiac and vascular sur-

POSTOPERATIVE TRAUMA OBSERVED BY MRI

gery while providing the same excellence of care that they had received in the operating room (and a similar system should be considered for subsequent studies). This has produced new and interesting information on postoperative cerebral edema.

5 SUMMARY It is dif®cult to study cerebral function in the anesthetized, early postoperative, and critically ill patient; yet it is an area of great interest. MRI offers interesting information on cerebral function and structure, but these studies present dif®cult problems for patient safety. Monitoring should be available for ECG, invasive and noninvasive blood pressures, pulse oximetry, and capnography throughout the scans. Anesthesia/sedation must be maintained during transfer to and from the MR unit; this is usually best provided by intravenous infusion. Intermittent positive pressure ventilation must be available over a distance of 3±5 m, which requires care in the choice of ventilator. It is important to have extensions on all intravenous lines to enable ¯uids to be transfused and changed during a scan, and all lines and cables should be routed to avoid snagging during movement of the patient. As the patient's airway is hidden during scans, an anesthetically trained member of staff should be present throughout, and evacuation procedures established and practised. With these protocols, MRI scanning can be offered safely to critically-ill patients, and has provided interesting new information on cerebral edema immediately after cardiopulmonary bypass surgery.

6

3

RELATED ARTICLES

Cardiac Gating Practice; Patient Life Support and Monitoring Facilities for Whole Body MRI.

7

REFERENCES

1. R. Schmidt, F. Fazekas, H. Offenbacher, H. Machler, N. Freidi, F. Payer, B. Rigler, M. J. Harrison, and H. Leehner, Neurology, 1993, 43, 775. 2. C. J. Peden, D. K. Menon, A. S. Hall, J. Sargentoni, and J. G. Whitwam, Anaesthesia, 1992, 47, 508. 3. D. N. F. Harris, S. M. Bailey, P. L. C. Smith, K. M. Taylor, A. Oatridge, and G. M. Bydder, Lancet, 1993, 342, 586. 4. D. N. Harris, A. Oatridge, D. Dob, P. L. Smith, K. M. Taylor, and G. M. Bydder, Anesthesiology, 1998, 88, 340. 5. J. G. Whitwam, M. K. Chakrabarti, W. H. Konarzewski, and H. Askitopoulou, Br. J. Anaesth., 1983, 55, 1017.

Biographical Sketch D. N. F. Harris. b 1949. M.B., Ch.B., 1974, M.D., 1982, Bristol, FRCA, 1984. Research fellow, Nuclear Cardiology, Southampton, 1979±82. Senior lecturer, anesthesia, Royal Postgraduate Medical School, London, 1991±present. Approx. 50 publications. Research interests: cerebral damage following major surgery, using MRI to correlate with neuropsychological assessment of cerebral impairment.

ABDOMINAL MRA

Abdominal MRA Paolo Pavone, Andrea Laghi, Carlo Catalano, Valeria Panebianco, Francesco Fraioli, Isabella Baeli, and Roberto Passariello University of Rome, `La Sapienza', Italy

1 INTRODUCTION The study of abdominal vessels by MRA is a new and interesting perspective available in a routine clinical examination. The main problem which has limited such applications of MRA is the sensitivity of this method to motion artifacts due to breathing, cardiac motion, peristalsis, and pulsatile blood ¯ow. However, the improvement in technical equipment has made it possible to reduce these problems and to obtain an excellent evaluation of abdominal vessels.1,2 To date, both the time-of-¯ight (TOF) and phase contrast (PC) techniques are used and each has its own advantages and disadvantages.3±5 The TOF method has a faster acquisition time and a higher signal-to-noise ratio (SNR), but it is very sensitive to saturation effects; as a consequence it is only possible to work with small volumes, particularly when a 3D technique is employed.6 Recently, a new method of study using a dynamic intravenous injection of gadolinium-DTPA in conjunction with a 3D TOF acquisition technique has been developed.7 The use of gadolinium shortens the blood relaxation time and overcomes in-¯ow saturation; hence it is possible to image arteries with a low ¯ow velocity and large volumes. These vessels can therefore be imaged in any direction without saturation problem, i.e. using a coronal plane for the aorta and so reducing the acquisition time. However, the overlapping of venous structures occurs and presaturation pulses are not effective. The phase contrast technique can be used over large ®elds of view because of its lack of sensitivity to the spin saturation effect. Other advantages are the complete cancellation of stationary tissues and the possibility of quantifying ¯ow velocity. By comparison, PC techniques require longer acquisition times and are very sensitive to motion artifacts, especially if such motion is not constant during image acquisition. In the evaluation of abdominal vessels, contrast-enhanced MRA is currently considered the only technique able to provide anatomical images of all major vessels, allowing an optimal image quality, with improved diagnosis.6 Using this technique, all the datasets needed can be acquired during a breath-hold, provided that enhanced gradient systems are available to reduce repetition and echo time dramatically. This overcomes the old problem with the conventional TOF or PC methods in which long acquisition times and the complexity of ¯ow effects resulted in a myriad of artifacts, rendering a considerable fraction of examinations unreadable. Furthermore, arterial blood is clearly visualized because it contains the paramagnetic contrast agents. The presence of paramagnetic agents such as a gadolinium chelate rapidly reduces the T1 relaxation time of blood from 1400 ms to 200± 400 ms depending on the dose and the rate of injection.

1

Obviously the visualization of blood depends on the presence of contrast in the arterial system while the 3D imaging data are being collected; this makes the technique very sensitive to the timing of the intravenous contrast administration relative to data acquisition.7 Great efforts have been made in recent years to improve further the image quality of enhanced MRA. These have been aimed at the improvement of the pulse sequences on the one hand and the contrast agents used on the other. Improvement of the pulse sequence for enhanced MRA must involve one single and important issue: minimizing dephasing phenomena, thus reducing the signal loss related to phase dispersion of ¯owing spins, mostly evident in stenotic (jet ¯ow) and tortuous vessels (turbulent ¯ow). These effects are, in fact, still evident even when enhanced techniques are used, despite the fact that these newer sequences visualize directly the effect of the contrast agent (T1 shortening) rather than the ¯ow phenomena. Reduction of phase dispersion is achieved by using either optimized gradient rephasing pulses or, in a more effective way, by reducing the echo time of the pulse sequence. With current commercial equipment, a minimum echo time of 1.5 ms can be used, and the improvement achieved in moving to this from the 3 ms available on older equipment is really dramatic, with more consistent evidence about small vessels and better de®nition of the true diameter of every vessel imaged by MRA. Moving to shorter echo time has only one important requisite: that the gradient characteristics of the equipment are optimized. Current equipment must have a gradient strength of at least 20 mT mÿ1 and a rise time of 400±800 s.8 Specialized equipments for use in the cardiovascular system are being produced that have a gradient strength of 40 mT mÿ1, although they are currently being offered only for research purposes. Another way to improve the image quality in enhanced MRA is to use more effective contrast agents. Conventional gadolinium chelates have basic pharmacokinetics, with intravascular±extracellular diffusion. In practice, these agents pass rapidly through the capillary membrane and very rapidly (within 5 min) reach pseudoequilibium between the vascular and the interstitial concentration. There are two ways to improve the characteristics of the contrast agent for enhanced MRA: by using a contrast agent with the same pharmacokinetic characteristics but having an increased relaxivity (the capacity to in¯uence the local magnetism and, therefore, the T1 relaxation time of blood), or by using contrast agents with prolonged blood half-life, having a lower capacity to diffuse through the capillary membrane. The ®rst approach is used with a new gadolinium chelate Gd-Bopta (Multihance, Bracco). This contrast agent was initially proposed only for hepatobiliary use, because of its consistent biliary excretion, but has now also been applied for general purposes, because of its good characteristics. In particular, in the vascular system, this contrast agent is able to provide a very high enhancement of the vascular structure because its relaxivity is almost twice as great as that of other gadolinium chelates in human plasma (R1 9.7 versus 4.9 for Gd-DTPA). Early studies have demonstrated that evaluation of abdominal vessels by enhanced MRA is improved by this contrast agent. The second approach is to use contrast agents that are not able to cross the capillary membrane rapidly. The ®rst class of such contrast agents are the macromolecular com-

2 ABDOMINAL MRA pounds, where a long molecular chain with large molecular weight is bonded to gadolinium. Schering has proposed a gadolinium chelate covalently bound to a polymer composed of polylysin (Gadomer 17). This compound has a molecular weight of 30±35 kDa, with high relaxivity, good acute tolerance, low viscosity, and low osmolarity. In early studies, this contrast agent has been able to improve the enhancement of vessels and to provide prolonged evidence of the vascular system in enhanced MRA. The second class of contrast agent that is not able to cross the capillary membrane rapidly is represented by one single molecule: MS 325, ®rst synthesized by EPIX and currently produced by Mallinkrodt. This contrast agent is rapidly bound to human serum albumin after being injected into the vascular system, with a strong but short-lasting link. In fact, after a few hours all the contrast agent injected is free from this link and is excreted through the kidneys. The result is that this agent provides a very high enhancement of the vascular system that is prolonged in time. The advantage of using intravascular agents is that the whole vascular system can be imaged more safely, with images being acquired in different anatomic areas, without the need for giving repeated injections of contrast agent at each anatomic level. In conclusion, there is no doubt that contrast-enhanced MRA is the technique of choice for abdominal MRA. Hardware improvements are to be expected and the use of newer contrast agents will de®nitely contribute to increasing the image quality and clinical value of this technique.

2 CLINICAL APPLICATIONS 2.1 Abdominal Aorta The main clinical application of aortic MRA is the study of aortic aneurysms in terms of the extent and size of the pathology as well as the involvement of collateral vessels, identi®cation of either an aortic stenosis or a thrombus in an atherosclerotic vessel and evaluation of either the patency or the complications of a graft.9 In the study of aortic aneurysms MRA is complementary to conventional spin echo images. In fact, a complete evaluation of the diameter of the lesion is possible only after a study performed using spin echo sequences. However, MRA has a major role in de®ning the size of the true lumen, in depicting the neck of the aneurysm, and a potential role in the quanti®cation of the blood ¯ow. Figure 1 illustrates the utility of MRA in the evaluation of an aortic aneurysm. By reconstructing the images with maximum intensity projection (MIP), it is possible to obtain a clear de®nition of the relationship between the aneurysm and the other vessels arising from the aorta. One study10 showed a sensitivity of 67% and a speci®city of 96% for MRA in the identi®cation of proximal stenosis of the renal artery involved by the aneurysm (Figure 2), and a sensitivity of 67% and a speci®city of 100% for stenosis of the celiac trunk and of the superior mesenteric artery. Enhanced MRA is particularly useful in the evaluation of abdominal aortic aneurysms. In fact, the presence of blood turbulence has limited to a large extent the use of nonenhanced techniques, with areas of ¯ow void related to spin dephasing. With enhanced MRA all these problems are overcome, and the

use of a coronal acquisition makes it possible to image the whole aorta and iliaco-femular district in one single ®eld of view (FOV of 50 cm can easily be used). The absence of excretion toxicity from gadolinium compounds also permits a complete study in patients with suspected renal arterial stenosis or involvement. Currently, noninvasive 3D imaging studies (either CT or MRI) are recommended for the evaluation of abdominal aortic aneurysms, avoiding the use of invasive catheter angiography. In particular, the aim is to determine the treatment that is to be performed in each patient; endovascular placement of aortic stent grafts can be developed as the treatment of abdominal aortic aneurysms, and all the data needed to evaluate if the graft can be placed can be collected. A study has compared the results of MRA and CTA (CT X-ray angiography) for evaluating the information needed for graft placement.11 Both techniques agreed with catheter angiography (as gold standard) in the measurement of the abdominal aorta at the level of its largest diameter and at the level of the renal artery ostium. Another measurement that was obtained consistently with good agreement with both methods was the de®nition of the distance between the renal arterial ostium and the origin of the aneurysm. For tube graft placement, a distal nonaneurysmatic segment of at least 15 mm is mandatory. This distance was also measured equally well by CTA and MRA (Figures 3 and 4). Another important issue is the evaluation of accessory renal arteries. In initial studies, MRA had generally performed very poorly, with sensitivities lying between 28 and 92%. More recent studies have reported an accuracy of 100%. The importance of breath-hold MRA is shown in all these studies. In general, 12% of patients with abdominal aortic aneurysms present with accessory renal arteries, and information about them is crucial for proper treatment planning. CTA and MRA now also offer similar results in this ®eld. Other information that is provided by MRA concerns the patency of the mesenteric artery: in some cases the inferior mesenteric artery may be patent and this can cause revascularization of the aneurysmal lumen through reverse ¯ow. The status of the iliac vessels has also to be evaluated. The possibility of stenosis has to be considered, since the endovascular stent graft is of ®xed expanded diameter once introduced (18 F requires a minimum vessel diameter of 6 mm). Involvement of the iliac arteries through aneurysmal dilatation does not prevent the placement of a stent graft, but size and location should be considered carefully in deciding on the type of graft to be used and the best approach to its location. Material arising from thromboses adhering to the walls can result in underestimation of aortic diameter and misinterpretation of the proximal extent of an abdominal aortic aneurysm. Evaluation of images acquired in the axial and coronal planes with gradient echo sequences straight after the MRA sequences (when contrast agent is still enhancing the vascular lumen) permits a detailed evaluation of the thrombus. Some advantages for MRA do exist compared with CTA: the use of safer contrast agents, which can be used as well in patients with initial renal insuf®ciency; the use of nonionizing radiation; the possibility of acquiring images directly in the coronal plane, with a more consistent longitudinal (or z axis) resolution; and faster postprocessing analysis (as a 3D MRA dataset is composed of far fewer images than those provided by CTA). However, MRA also has some drawbacks, including

ABDOMINAL MRA

3

Figure 1 Abdominal aortic aneurysm; evaluation with the 3D TOF technique after gadolinium injection. (a) Single coronal slice depicting the true lumen of the aneurysm; the image also allows the thrombolytic part to be recognized. (b) MIP reconstruction of the same aneurysm; good visualization of the abdominal aorta from the renal arteries to the iliac vessels is obtained. (c) Magni®cation of the iliac bifurcation with evidence of involvement of the right common iliac artery and a tight stenosis at the origin of the left common iliac artery

the usual contraindications for MR in general (such as implanted pacemakers). Further, inaccuracy in the detection of small accessory renal arteries must be considered, although this problem may be overcome by continuing technical improvements that allow imaging with better spatial resolution and visualization of very small vessels. Another ®eld of application for MRA is in the evaluation of leaks after the placement of endovascular stent grafts.12 Follow-up of endovascular grafts is required in order to evalu-

ate the success of the treatment and the development of complications. In the early phase after placement of an endovascular graft, it is important to ensure that the procedure has led to technical success and that there is no leakage of blood between the graft and the native lumen of the aneurysm, which is better known as endoleak. The two major complications observed in the ®rst series reported by Moore13 were stent fractures (23%) and endoleaks (44%); 23% of the endoleaks had spontaneous closure. Stent fractures have been corrected fol-

4 ABDOMINAL MRA

Figure 2

Volume-rendering reconstruction clearly visualizes the celiac trunk, mesenteric artery and both renal arteries in a patient with aneurysm

lowing a modi®cation in the stent hook design by the manufacturer. However, the endoleaks remain a signi®cant problem and may be associated with an increase in aneurysmal diameter at 1 year.14 In the long-term follow-up of aneurysms treated with a stent graft, it is necessary to monitor the caliber of the lumen of the aneurysm at regular intervals. In fact there have been several case reports of ruptured aneurysms after stent-graft repair. A noninvasive technique is required to monitor all these potential complications, and catheter angiography should be limited to the role of guidance of inventional procedures needed to correct such complications. Ultrasound may be able to provide consistent information with high accuracy in de®ning the presence of leaks or in evaluating the caliber and patency of the aorta and of the iliac vessels.15 The value of CT has also been advocated;16 with current helical acquisition it is possible to reconstruct angiographic projection images. CT is

Figure 3 CT axial image with the aneurysm lumen surrounded by thrombolytic material

considered to be the required investigation for the EUROSTAR registry.17 With MRA, both types of endoleak can be easily detected by the evaluation of areas of increased signal intensity detected after contrast agent injection. Moreover, 3D images provided by MRA are able to de®ne both the patency and morphology of the stent graft and its relative position compared with the origin of the renal arteries.

2.2

Renal Arteries

Several types of examination have been used during the past few years for the evaluation of renovascular hypertension, but many have been abandoned because of their invasivity. Although digital subtraction angiography (DSA) and conventional angiography have become increasingly important because of the success of percutaneous procedures, they still remain too invasive as screening method procedures. It is now possible to propose MRA of the renal arteries as an accurate, noninvasive diagnostic procedure in the evaluation of renal artery stenosis, especially of the ostium of the renal vessel.18 An example of MRA of the renal arteries is shown in Figure 5. Recent data show the high accuracy of MRA, which is similar in this case to conventional angiography in detecting both the normal and accessory renal arteries as well as the anomalous vessels. According to the literature, MRA has a sensitivity varying between 89% and 94% and a speci®city of between 95% and 97%.19,20 The evaluation of renal arteries has been performed principally with DSA, ultrasound, and conventional CT during the past few years. Different etiologies are implicated in renovascular disease and, in most cases, the renal arteries are affected as part of the entire vascular system. Because of the continuous advances in percutaneous endoluminal therapies, the best possible imaging method is required in the detection of renovascular diseases. According to the literature, MRA allows the visualization of all the main arteries as well as accessory arteries. In the detection of stenosis, the MRA sensitivity and speci®city amounted

ABDOMINAL MRA

Figure 4

5

Maximum intensity projections (MIP) show the extension of the aneurysm and the renal involvement

to 93% and 90%, respectively, with a negative predictive value of 96%.21,22 In contrast to other noninvasive techniques such as scintigraphy or ultrasound, contrast-enhanced 3D MRA allows visualization of accessory renal arteries with a high degree of accuracy. The visualization of accessory arteries is especially important in the evaluation of renal transplant donors. Imaging must be completely accurate before surgical revascularization or nephrectomy in living-related renal donors can be contemplated. As far as atherosclerotic lesions are concerned, the diagnostic performance of 3D MRA is still limited, particularly when all forms of stenosis are considered; in fact, as seen in the literature, there have been a considerable number of falsepositive and false-negative ®ndings.23 Therefore, the gold standard diagnostic procedure for detecting arterial stenosis still remains angiography, though this is an invasive examination that exposes the patient to ionizing radiation as well as iodi-

nated contrast material, which may cause allergic reactions and/or nephrotoxicity. Therefore, according to the literature,24 contrast-enhanced 3D MRA is, as yet, only a promising approach to obtaining contrast arteriography but is comparable to arteriography in the assessment of renal artery morphology and pathologic states. It does not involve the risk of arterial catheterization, ionizing radiation, or nephrotoxic contrast material.

2.3

MR Venography (MRV)

Studies of abdominal veins are an interesting application of MRA. It is possible to evaluate the normal and pathologic anatomy of the portal system, including the splenic vein, the superior mesenteric vein (SMV), and the eventual collateral vessels.25 The relatively slow and homogeneous venous blood

6 ABDOMINAL MRA

Figure 5 Axial MIP reconstruction of renal arteries; good evaluation of the distal part of the vessels is also obtained

¯ow allows high-quality images to be obtained in almost all patients. The large FOV permits the imaging of the whole spleno-portal axis and the collateral vessels at the same time. This option is not available in any other imaging technique such as DSA, duplex scan, or color Doppler ultrasonography. The main technical dif®culties arising from respiratory and peristaltic motion, and to the different ¯ow direction and velocity in abdominal vessels, have now been overcome by new image acquisition procedures. The breath-hold coronal TOF technique is the preferred method with which to image the abdominal veins. An image can be acquired in about 8 s and in a few minutes all the abdominal vessels can be displayed. It is, therefore, possible to obtain a selected evaluation of the portal vein or any of the other vessels by acquiring the images in the plane corresponding to the vessel to be studied. Because of saturation effects, all the vessels not in that plane will not be depicted. To obtain a more selective evaluation of the venous system, it is possible to use more sophisticated techniques such as bolus tracking and selective presaturation. Those two options allow the clinician to de®ne accurately the ¯ow direction in a vessel and to distinguish a thrombus from ¯owing blood. Furthermore, they permit the selective suppression of the signal coming from speci®c arterial vessels or veins. After acquisition of the data, a postprocessing MIP technique can be used to reconstruct the images. Thus, it is possible to rotate the vessels and to show them on the monitor in the most useful projection and it is also possible to recognize strange anatomic variants in this way. Portal hypertension represents one of the major ®elds of application of MR venography (MRV) in the abdomen (Figure 6). Because of the large FOV, MRV allows the whole splenoportal axis and its collateral vessels to be shown at the same time in almost all patients. It is possible to evaluate the diameter of the portal vein and, with adequate techniques, to obtain quantitative information about the blood ¯ow. Compared with color Doppler ultrasonography, which can actually be considered to be the method of choice for the screening and follow-up of patients with portal hypertension, MRV is not operator-dependent; it is not in¯uenced by the presence of ascites, which is very common in these patients, or by colic

meteorism and patient habitus, all of which disturb a sonographic examination. Another recent clinical indication for MRV is the evaluation of complex venous disease in cases of liver transplantation. MRV is very accurate and has a good correlation with DSA and duplex ultrasound. Moreover, MRV displays images in different orientations: thus the reconstructed images can be easily interpreted by the surgeon and complex structures can be demonstrated better than by any alternative imaging method. Following liver transplantation, complications may arise as a result of anastomosis failure or stenosis. In this case the examination technique is ultrasonography, which has an important limitation in depicting the anastomosis. MRV can also show the condition through the paramagnetic effects of the material in the anastomosis, which creates a lack of signal. Additionally, anastomosis on the hepatic artery can be depicted by MRA in a high percentage of patients. Finally, another recent ®eld of application for MRV is the study of the patency of both surgical and percutaneous shunts of the ortho-systemic structure (TIPS). MRV makes it possible to obtain a noninvasive postsurgical follow-up. The anastomosis can be identi®ed, ¯ow direction established, and both size and patency of the shunts evaluated. After TIPS procedures the position of the stent can also be accurately determined. There are a few pitfalls and limitations in the use of MRV in the portal system. At a venous con¯uence, such as the junction of superior mesenteric and splenic veins, a signal void may occur simulating a thrombus. In order not to misunderstand this artifact, gradient echo images can be acquired in different orientations through the questionable area. Changing the slice orientation can also be useful sometimes so as to

Figure 6 2D TOF MR venography. (a) MIP reconstruction shows the inferior vena cava (IVC) with the hepatic vessels and the portal vein. (b) Good evaluation of the portal vein and its bifurcation. The IVC and one hepatic vessel are also depicted

ABDOMINAL MRA

avoid the in-plane saturation effect. Usually, this is not as common for the portal system as it is for the aorta or the inferior vena cava, but a verticalization of the portal axis together with slow ¯ow, as in a cirrhotic patient, may impair signal acquisition. In conclusion, MRV of the portal system is extremely accurate and complete for the evaluation of these vessels. The large ®eld of view, the insensitivity to patient habitus and the graphic presentation format make it very useful for preoperative assessment in patients with portal hypertension. 3 RELATED ARTICLES

7

19. T. M. Grist, T. W. Kennell, I. A. Sproat, E. M. Flath, J. C. McDermott, and M. M. Majkiwcyz, Radiology, 1993, 189P, 190. 20. T. F. Hany, D. A. Leung, T. Pfammatter, and J. F. Debatin, Invest. Radiol., 1998, 33, 653. 21. S. Miller, F. Schick, S. H. Duda, T. Nagele, U. Hahn, F. Teu¯, M. Muller-Schimp¯e, C. M. Erley, J. M. Albes, and C. D. Claussen, Magn. Reson. Imag., 1998, 16, 1005. 22. D. Kim, R. R. Edelman, K. C. Kent, D. J. Porter, and J. J. Skillman, Radiology, 1990, 174, 727. 23. M. R. Prince, JMRI, 1998, 8, 511. 24. D. B. Stafford-Johnson, C. A. Lerner, M. R. Prince, S. N. Kazanjian, D. L. Narasimham, A. B. Leichtman, and K. J. Cho. Magn. Reson. Imag., 1997, 15, 13. 25. H. B. Gehl, K. Bohndorf, K. C. Klose, and R. W. Gunther, J. Comput. Assist. Tomogr., 1990, 14, 619.

Liver, Pancreas, Spleen, and Kidney MRI; Peripheral Vasculature MRA; Phase Contrast MRA; Time-of-Flight Method of MRA. Biographical Sketches 4 REFERENCES 1. R. R. Edelman, H. P. Mattle, D. J. Atkinson, and H. M. Hoogewoud, Am. J. Roentgenol., 1990, 154, 937. 2. H. Bosmans, G. Marchal, P. van Hecke, and P. Vanhoenacker, Clin. Imag., 1992, 16, 152. 3. D. G. Nishimura, Magn. Reson. Med., 1990, 14, 194. 4. C. L. Dumoulin, S. P. Souza, M. F. Walker, and W. Wagle, Magn. Reson. Med., 1989, 9, 139. 5. C. L. Dumoulin, E. K. Yucel, P. Vock, S. P. Souza, F. Terrier, F. L. Steinberg, and H. Weymuller, J. Comput. Assist. Tomogr., 1990, 14, 779. 6. J. S. Lewin, G. Laub, and R. Hausmann, Radiology, 1991, 179, 261. 7. J. L. Creasy, R. R. Price, T. Presbrey, D. Goins, C. L. Partain, and R. M. Kessler, Radiology, 1990, 175, 280. 8. A. N. Shetty, K. G. Bis, T. G. Vrachliotis, M. Kirsch, A. Shirkhoda, and R. Ellwood, JMRI, 1998, 8, 603. 9. R. Passariello, P. Pavone, C. Catalano, A. Laghi, L. Marsili, and M. Di Girolamo, Ann. Chir. Gynaecol., 1993, 82, 87. 10. J. A. Kaufman, E. K. Yucel, A. C. Waltman, S. C. Geller, C. A. Athanosoulis, and M. R. Prince, Radiology, 1993, 189(P), 174. 11. A. Siegfried, M. D. Thurnher, R. Dorffner, P. Polterauer, and J. Lammer, Radiology, 1997, 205, 341. 12. L. Engellau, E. M. Larsson, U. Albrechtsson, T. Jonung, E. Ribbe, J. Thorne, Z. Zdanowski, and L. Norgren, Eur. J. Vasc. Endovasc. Surg., 1998, 15, 212. 13. W. S. Moore, J. Endovasc. Surg., 1997, 4, 182. 14. J. S. Matsumura, J. Vasc. Surg., 1998, 4, 606. 15. J. Golzarian, L. Dussaussois, H. T. Abada, P. A. Gevenois, D. van Gansbeke, J. Ferreira, and J. Struyven, Am. J. Roentgenol., 1998, 171, 329. 16. D. T. Sato, C. D. Goff, R. T. Gregory, K. D. Robinson, K. A. Carter, B. R. Herts, H. B. Vilsack, R. G. Gayle, F. N. Parent III, R. J. de Masi, and G. H. Maier, J. Vasc. Surg., 1998, 28, 657. 17. P. L. Harris, J. Endovasc. Surg., 1997, 4, 72. 18. M. Strotzer, C. M. Pruell, A. Geissler, S. M. Kohler, B. R. Kraemer, and J. Gmeinwieser, Radiology, 1993, 189P, 189.

P. Pavone, b 1957. M.D., 1981, Rome. Studies on CT and vascular and interventional radiology, 1981±87. Researcher on MRI at the University of L'Aquila, Italy, 1988±93; researcher at University of Rome `La Sapienza', 1993±present. Approx. 300 publications. Current research specialties; experimental and clinical studies of the abdominal organs and vascular system by MRI. A. Laghi. b 1967. M.D., 1992, Rome. Resident, University of Rome `La Sapienza', 1992±present. Approx. 60 publications. Current research specialties: experimental and clinical studies of the abdominal organs and vascular system by MRI. C. Catalano. b 1965. M.D., 1990, Rome. Residency, University of L'Aquila, 1990±1994. Fellow, University of Rome `La Sapienza', 1994±present. Approx. 120 publications. Current research specialties: experimental and clinical studies of the abdominal organs and vascular system by MRI. V. Panebianco. b 1967. M.D., 1994, Rome. Resident, University of Rome `La Sapienza', 1993±present. Approx. 30 publications. Current research specialties: clinical studies of the abdominal organs by MRI. F. Fraioli. b 1973. M.D., 1998, Rome. Residency University of Rome `La Sapienza', 1993±present. Approx. 10 publications. Current research specialties: experimental and clinical studies of the abdominal and vascular system by MRI. I. Baeli. b 1974. M.D. 1999, Rome. Residency University of Rome `La Sapienza', 1993±present. Approx. 10 publications. Current research specialties: experimental and clinical studies of the abdominal and vascular system by MRI. R. Passariello. b 1942. M.D. 1965, Rome. Residency University Padua, 1965±1967. Full Professor of Radiology, University of L'Aquila, 1986±1991. Full Professor of Radiology and Chairman of the 2nd Department of Radiology, University of Rome `La Sapienza', 1991±present. Approx. 500 publications. Organization of more than 65 Congresses. 1994±present Vice-President of the European Congress of Radiology '95 and '97. Chairman of the European Seminars on Diagnostic and Interventional Radiology, 1990±present. Member of the Executive Board of the NICER Programme. 1990±present.

ASSESSMENT OF REGIONAL BLOOD FLOW AND VOLUME BY KINETIC ANALYSIS OF CONTRAST-DILUTION CURVES

Assessment of Regional Blood Flow and Volume by Kinetic Analysis of Contrast-Dilution Curves Johannes C. BoÈck and Roland Felix Freie UniversitaÈt Berlin, Germany

1

By contrast, organ blood ¯ow can be quantitatively assessed using diffusible contrast media such as deuterium oxide5 or H217O.6 However, these techniques will not be addressed in depth in this article since clinical experience is still lacking, probably because the methodology requires an injection either directly into the tissue or into the feeding artery. A less invasive approach, intravenous injection, would require the measurement of the concentration±time curve in the feeding artery and deconvolution of the arterial and tissue curves.7 Such techniques offer the same advantages as positron emission tomography in that blood ¯ow can be measured in absolute terms (mL minÿ1 per 100 g of tissue).

1 INTRODUCTION

2

PRINCIPLES

The success of clinical magnetic resonance imaging (MRI) is based on its superb anatomic and contrast resolution, in particular in the central nervous system. However, MRI was initially unable to contribute to the ®eld of perfusion imaging because of two factors: the lack of MRI contrast media and insuf®cient temporal resolution of the pulse sequences. Contrast media are now available. The ®rst agent in clinical use was gadolinium±DTPA, a well-tolerated ionic paramagnetic contrast agent developed by Schering AG, Berlin and ®rst used in humans at our Institution in 1983.1 After intravenous injection, gadolinium±DTPA is distributed in the extracellular space with the notable exception of the central nervous system, where the extravasation of gadolinium±DTPA into the interstitial space is effectively prevented by the intact blood/brain barrier. No such barrier prevents leakage of gadolinium±DTPA in the normal myocardium. The limitation of temporal resolution has been overcome by the development of pulse sequences such as gradient echo and echo planar imaging. These pulse sequences allow for acquisition times ranging from 0.5±20 images per second, suf®cient to follow the ®rst passage of a contrast bolus through an organ. In 1986, Haase et al.2 published a report which pioneered contrast enhanced perfusion MRI. In an experimental model, they combined the intravenous bolus injection of the ®rst available paramagnetic contrast agent, gadolinium±DTPA, and a new rapid image acquisition strategy `fast low angle shot' (FLASH). When a series of images was acquired in the same slice position after contrast bolus injection (`FLASH MR movie') a reversible signal intensity decrease was observed in the brain which apparently coincided with the ®rst pass of the contrast medium through the brain. Unfortunately, the interpretation of the contrast kinetic is limited because of the physical properties of currently available contrast agents, which distribute either in the intravascular space (in the intact brain) or, in addition, in the interstitial space (in the myocardium but also in cerebral pathologies with disruption of the blood/brain barrier). From a rigorous mathematical standpoint only relative cerebral blood volume can be calculated on the basis of concentration±time curves measured in the brain3Ðcerebral blood ¯ow can only be estimated.3,4 Since gadolinium±DTPA is not a strictly intravascular contrast medium in other organs than the brain, both blood ¯ow and volume cannot be derived from gadolinium±DTPA kinetics measured in the myocardium and the kidneys.

Conventional contrast enhanced MRI using paramagnetic agents such as gadolinium±DTPA relies on dipole±dipole interactions between the nuclear spins of local protons and unpaired electrons of the contrast molecules. These interactions shorten the T1 relaxation times and thereby increase signal intensity in the interstitial space minutes (the time required for contrast extravasation) after injection. If, however, the contrast is rapidly injected as a bolus, and if an appropriate (susceptibility-sensitive) pulse sequence is used, then a signal intensity decrease is observed within seconds after injection. The underlying mechanism is the compartmentalization of the contrast material in the vasculature (at least in the brain) with consecutively high contrast concentration gradients between the intravascular and interstitial compartments. (Positive) susceptibility is the property of paramagnetic substances to produce an intrinsic magnetic ®eld when exposed to an extrinsic magnetic ®eld. Because the magnetic susceptibility of gadolinium±DTPA is different to that of biological tissues, the contrast concentration gradients translate into regional (intravoxel) magnetic ®eld gradients. Intravoxel magnetic ®eld inhomogeneity broadens the resonance frequency spectrum with subsequent dephasing and signal loss. Since gradient echo pulse sequences do not refocus static magnetic ®eld inhomogeneities, signal intensity is directly affected by this process, in particular when the echo time is long (on the order of 20 ms). Spin echo pulse sequences do refocus static inhomogeneities and are therefore only sensitive to dephasing of mobile spins (diffusion of water molecules); again, the sensitivity increases with the echo time. The signal intensity decrease is reversed as the contrast bolus passes the brain (®rst pass); however, a second (third, etc.) passage of the contrast (systemic contrast recirculation) is usually observed before the baseline signal intensity has been reached. For further computations (e.g. blood volume), indicator dilution theory requires a concentration±time curve which can be computed for every voxel on the basis of the measured signal intensities by logarithmic transformation:8 c…t† ˆ ÿln‰S…t†=S…t0 †Š=…k2 TE†

…1†

(c = concentration, S = signal intensity, t = time, t0 = time of injection, k2 = constant depending on system and tissue properties, TE = echo time).

2 ASSESSMENT OF REGIONAL BLOOD FLOW AND VOLUME BY KINETIC ANALYSIS OF CONTRAST-DILUTION CURVES

Figure 1 Sixty-one year old man with focal seizures of the right arm and speech arrest. Left parietal mass with homogeneous hyperperfusion pattern documented by MRI and con®rmed by the radionuclide reference technique. Histology: meningiomatous and angiomatous meningioma. (a) T2*-weighted ®rst-pass study (subtraction image); (b) radionuclide study (HMPAO±SPECT)

The postprocessing algorithm is based on the hypothesis that the area under the concentration±time curve recorded in every voxel re¯ects regional cerebral blood volume.9 This assumption is correct for the ®rst pass of the contrast bolus and requires numerical elimination of recirculating contrast. This is achieved with suf®cient accuracy by a nonlinear least-squares ®t procedure using a -variate or some other algebraic function as a model for the cerebral contrast transit in the absence of systemic contrast recirculation. Finally, a parameter image is generated in which the brightness of each voxel re¯ects the area of the concentration±time curve and thus regional cerebral blood volume. In the myocardium, the contrast extravasates into the interstitial space. This already occurs to a substantial degree during the ®rst pass;10 the area under the concentration±time curve therefore does not correspond to a well-de®ned biologic compartment, but re¯ects the intravascular and a variable portion of the interstitial compartment. Myocardial blood volume can therefore not be derived from the concentration±time curve of small molecules11 like gadolinium±DTPA. The use of a macromolecular agent with a molecular weight above 20 000 Da such as gadolinium±DTPA±polylysine12 would be required for this purpose, but macromolecular agents are not yet available for clinical use. As already mentioned, there is no rigorous theoretical concept for the construction of parameter images from concentration±time curves which re¯ect regional cerebral or myocardial blood ¯ow. Some approaches are based on mean transit time concepts although the correct regional mean transit times cannot be determined using intravascular contrast agents.3,4 In fact, the mean transit time calculated from the tis-

sue concentration±time curve alone (without consideration of the arterial input function) overestimates the true organ mean transit time (which is on the order of only a few seconds in the brain) by more than 100%. Correction for the dispersion of the contrast bolus outside of the brain (®nite injection duration, passage from the peripheral vein to the brain) would require numerical deconvolution of the tissue curves with the arterial input function. Although the arterial input function can be measured13 it is not the correct input function for every voxel. Finally, since the volume of distribution of gadolinium±DTPAtype contrast material is restricted, the inverse of the (correct) mean transit time would not represent blood ¯ow in terms of [mL minÿ1 per 100 g tissue] but blood ¯ow per unit blood volume [mL minÿ1 per ml blood]. Despite these limitations, some investigators have used transit time approaches with some success.14±17 Another approach is to calculate a subtraction image where the image acquired at the time of maximum cerebral contrast concentration is subtracted from an image acquired before any contrast has reached the brain on a voxel-by-voxel basis [Figure 1(a)].18 The underlying theoretical concept is valid under certain experimental conditions19 and requires the consideration of the arterial input function. Fortunately, the arterial input function can be neglected as long as the image contains its own reference (e.g. the normal contralateral hemisphere), since the arterial input function is identical for all voxels under considerations in this concept. Subtraction images appear to re¯ect regional cerebral blood ¯ow as assessed by an established radionuclide reference technique (HMPAO±SPECT) in the vast majority of patients with cerebral infarcts18 and in all patients with homogeneous

ASSESSMENT OF REGIONAL BLOOD FLOW AND VOLUME BY KINETIC ANALYSIS OF CONTRAST-DILUTION CURVES

intracranial tumors [Figures 1(a) and (b)].20 Discrepant results between the radionuclide method and ®rst-pass subtraction images were found in inhomogeneous tumors (vital tumor/ necrosis/edema). However, the discrepancy did not appear to result from the inability of MRI to depict adequately regional cerebral blood ¯ow in these tumors but from the lower spatial resolution of SPECT (voxel volume = 1500 mm3) versus MRI (voxel volume = 50 mm3).

3 CEREBRAL PERFUSION IMAGING Clinical experience with contrast enhanced MR perfusion imaging is steadily increasing. A number of recent reports focus on cerebral ischemia, intracranial tumors, and arteriovenous malformations. In cerebral infarcts, regional cerebral blood ¯ow is usually decreased but a hyperperfusion pattern is identi®ed in some cases.18,21 While the prognostic signi®cance of the hyperperfusion pattern in subacute infarcts has yet to be established, one might speculate that evidence of reperfusion indicates a better clinical outcome. The grade of malignancy of intraaxial tumors appears to be correlated with the perfusion pattern. Low-grade astrocytomas are characterized by decreased blood volume when compared with contralateral gray matter. High-grade astrocytomas show either a homogeneous increase of blood volume or signs of marked inhomogeneity. These ®ndings parallelÐto a certain degreeÐthe extent of the blood/brain barrier disruption as demonstrated by contrast enhancement on T1weighted (or cranial computerized tomographyÐCCT) images. However, it is important to remember that the blood/ brain barrier and regional cerebral perfusion are two distinct features which obey different underlying regulating and disturbing mechanisms.22 Multiple biopsies of intraaxial tumors have shown that highgrade tumors are composed of tissues with different histological grading. It is obvious that biopsies should be obtained from the most malignant part of the tumor. These areas are often characterized by increased metabolism, increased tumor blood ¯ow, and volume. Perfusion imaging should therefore be considered for optimal biopsy guiding.23 The differentiation of radiation necrosis after therapy of high-grade intraaxial tumors from recurrent tumor is cumbersome because both present with mass effect, necrosis and blood/brain barrier disruption. However, initial clinical results indicate that blood volume is homogeneously decreased in radiation necrosis, while an inhomogeneous pattern with areas of high regional blood volume can be demonstrated in recurrences.23 Arteriovenous malformations are quite easily demonstrated by MRI, including magnetic resonance angiography. However, due to inherent limitations of magnetic resonance angiography, the success of interventional therapy by embolization tends to be overestimated. In a recent study24 we showed that contrast enhanced perfusion imaging is by far the most sensitive MRI technique to demonstrate residual pathologic vasculature after embolization. Perfusion imaging is also highly useful to demonstrate reopening of previously occluded pathways and is therefore part of our standard protocol in the evaluation and follow-up of arteriovenous malformations.

4

535

MYOCARDIAL PERFUSION IMAGING

Initial experience with gadolinium±DTPA at our Institution demonstrated increased enhancement in acute infarcts but no signi®cant enhancement in subacute and chronic infarcts.25 Since spin echo pulse sequences with an acquisition time of more than 5 min were used in these studies, the enhancement did not re¯ect blood ¯ow but contrast extravasation into the interstitial space which is mainly a function of microvascular permeability. Other MR contrast media such as dysprosium, manganese, macromolecular gadolinium compounds, and superparamagnetic agents have been successfully studied.26,27 Atkinson and co-workers performed bolus injection experiments using a T1-weighted pulse sequence;28 they described a transient signal intensity increase during the ®rst pass of the contrast material in normal myocardium of experimental animals and volunteers, but no signal intensity change after experimental occlusion of the left coronary artery. In a subsequent study, similar effects were seen in patients with coronary artery disease.29 No attempt was made in these studies to quantify the effects in terms of myocardial blood ¯ow. Wilke et al. have used a kinetic model involving the mean transit time concept to obtain blood ¯ow estimates in an experimental model. Blood ¯ow was measured by a radiomicrosphere reference technique. A correlation coef®cient of r = 0.89 was found between the MRI estimate of ¯ow (the inverse of mean transit time) and the reference technique. It must be noted, however, that changes in myocardial blood volume are not accounted for in that model. Since myocardial blood volume is highly variable, myocardial blood ¯ow estimates based on mean transit time alone are likely to be much less reliable in a clinical setting.

5

PERFUSION IMAGING OF OTHER ORGANS

Next to the brain and heart, perfusion imaging of the kidneys might be of clinical interest. Imaging of renal blood ¯ow is quite complex because ¯ow and function (= excretion) are tightly related. Gadolinium±DTPA is freely ®ltrated in the glomerulus and neither secreted nor reabsorbed in the tubular system.30 Due to water reabsorption in the tubules, the concentration of gadolinium±DTPA increases by up to a factor of 100 from the initial plasma concentration. This wide range of concentrations is unique in the kidneys and has important effects on the observed signal intensity: at lower concentrations, signal intensity increases due to T1 shortening; at higher concentrations, signal intensity decreases due to T2 shortening. Obviously, this nonlinearity complicates the interpretation of signal intensity versus time curves measured in the kidneys. Nevertheless, three phases can be distinguished after contrast bolus injection: a vascular, a tubular, and a ductal phase.31 Due to the complexity of both renal physiology and the contrast mechanisms, assessment of renal blood ¯ow will remain one of the most challenging tasks in MR imaging of organ blood ¯ow. The methodology discussed in this article also applies to perfusion imaging of other tissues. It must be noted, however, that the quality of the primary data (the concentration versus time curve) depends on the contrast concentration obtained during the ®rst pass after bolus injection; tissues with low re-

4 ASSESSMENT OF REGIONAL BLOOD FLOW AND VOLUME BY KINETIC ANALYSIS OF CONTRAST-DILUTION CURVES gional blood ¯ow and volume are therefore unlikely to be candidates for the assessment of perfusion by contrast bolus experiments.

6 FUNCTIONAL IMAGING OF THE HUMAN CORTEX Functional imaging of the brain refers to the detection of stimulated cortical activity. The stimulus may be an optical, motor, or sensory paradigm; even ideation of motion or a visual pattern can act as a stimulus. Cortical stimulation entails alterations of regional cerebral blood ¯ow, blood volume, oxygenation, and metabolism which can be detected by MRI. Classically a domain of nuclear medicine, in particular of positron emission tomography, MRI has been very successful in detecting evidence of cortical activation.32 The methodology relies upon the above-described contrast bolus experiment which is performed under resting conditions and during stimulation. The computed blood volume images are subtracted from one another; the subtraction image represents areas of locally altered perfusion due to cortical stimulation.

7 RELATED ARTICLES Blood Flow: Quantitative Measurement by MRI; Hemodynamic Changes owing to Sensory Activation of the Brain Monitored by Echo-Planar Imaging; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Gadolinium Chelates: Chemistry, Safety, and Behavior; Susceptibility Effects in Whole Body Experiments.

8 REFERENCES 1. W. SchoÈrner, R. Felix, M. Laniado, L. Lange, H. J. Weinmann, C. Claussen, W. Fiegler, U. Speck, and E. Katzner, Fortschr. RoÈntgenstr., 1984, 140, 493. 2. A. Haase, D. Matthaei, W. HaÈnicke, and J. Frahm, Radiology, 1986, 160, 537. 3. N. A. Lassen, J. Cerebr. Blood Flow Metab., 1984, 4, 633. 4. R. M. Weisskoff, D. Chesler, J. L. Boxerman, and B. R. Rosen, Magn. Reson. Med., 1993, 29, 553. 5. J. J. H. Ackerman, C. S. Ewy, S. G. Kim, and R. A. Shalwitz, Ann. N. Y. Acad. Sci., 1987, 508, 89. 6. K. K. Kwong, A. L. Hopkins, J. W. Belliveau, D. A. Chesler, L. M. Porkka, R. C. McKinstry, D. A. Finelli, G. J. Hunter, J. B. Moore, B. R. Rosen, and R. G. Barr, Magn. Reson. Med., 1991, 22, 154. 7. N. A. Lassen and W. Perl, `Tracer Kinetic Methods in Medical Physiology', Raven Press, New York, 1979. 8. B. R. Rosen, J. W. Belliveau, J. R. Vevea, and T. J. Brady, Magn. Reson. Med., 1990, 14, 249. 9. L. Axel, Radiology, 1990, 177, 679. 10. J. A. Rumberger and M. R. Bell, Invest. Radiol., 1992, 27, S40. 11. J. M. Canty, R. M. Judd, A. S. Brody, and F. J. Klocke, Circulation, 1991, 84, 2071. 12. M. Saeed, M. F. Wendland, T. Masui, A. J. Connolly, N. Derugin, R. C. Brasch, and C. B. Higgins, Radiology, 1991, 180, 153. 13. W. H. Perman, M. H. Gado, K. B. Larson, and J. S. Perlmutter, Magn. Reson. Med., 1992, 28, 74.

14. J. W. Belliveau, B. R. Rosen, G. J. Hunter, E. Tasdemiroglu, R. MacFarlane, P. Boccalini, L. M. Hamberg, M. A. Moskowitz, and S. H. Simon, Stroke, 1993, 24, 444. 15. N. Wilke, C. Simm, J. Zhang, J. Ellermann, X. Ya, H. Merkle, G. Path, H. LuÈdemann, R. J. Bache, and K. Ugurbil, Magn. Reson. Med., 1993, 29, 485. 16. K. A. Rempp, G. Brix, F. Wenz, C. R. Becker, F. Guckel, and W. J. Lorenz, Radiology, 1994, 193, 637. 17. J. C. BoÈck, O. Henrikson, A. H. G. GoÈtze, W. Wlodarczyk, B. Sander, and R. Felix, Invest. Radiol., 1995, 30, 693. 18. J. C. BoÈck, B. Sander, J. Hierholzer, M. Cordes, J. Haustein, W. SchoÈrner, and R. Felix, Fortschr. RoÈntgenstr., 1992, 156, 382. 19. N. A. Mullani and K. L. Gould, J. Nucl. Med., 1983, 24, 577. 20. J. C. BoÈck, B. Sander, J. Hierholzer, J. Haustein, M. Scholz, K. H. Radke, W. SchoÈrner, W. Lanksch, and R. Felix, Fortschr. RoÈntgenstr., 1992, 157, 378. 21. R. R. Edelman, H. P. Mattle, D. J. Atkinson, T. Hill, J. P. Finn, C. Mayman, M. Ronthal, H. M. Hoogewoud, and J. Klee®eld, Radiology, 1990, 176, 211. 22. J. C. BoÈck, B. Sander, K. H. Radke, J. Haustein, C. Zwicker, and R. Felix, Adv. MRI Contrast, 1993, 1, 76. 23. B. R. Rosen, J. W. Belliveau, A. J. Aronen, D. Kennedy, B. R. Buchbinder, A. Fischman, M. Gruber, J. Glas, R. M. Weisskoff, M. S. Cohen, F. H. Hochberg, and T. J. Brady, Magn. Reson. Med., 1991, 22, 293. 24. J. C. BoÈck, H. P. Molsen, B. Sander, and R. Felix, Fortschr. RoÈntgenstr., 1992, 157, 471. 25. H. W. Eichstaedt, R. Felix, F. C. Dougherty, M. Langer, W. Rutsch, and H. Schmutzler, Clin. Cardiol., 1986, 9, 527. 26. A. D. Watson, S. M. Rocklage, and M. J. Carvlin, in `Magnetic Resonance Imaging', 2nd edn. ed. D. D. Stark and W. G. Bradley, Mosby Year Book, St. Louis, MO, 1992, Vol. 1, Chap. 14. 27. C. B. Higgins, G. Caputo, M. F. Wendland, and M. Saeed, Invest. Radiol., 1992, 27, S66. 28. D. J. Atkinson, D. Burstein, and R. Edelman, Radiology, 1990, 174, 757. 29. W. J. Manning, D. J. Atkinson, W. Grossman, S. Paulin, and R. R. Edelman, J. Am. Coll. Cardiol., 1991, 18, 959. 30. C. H. Lorenz, T. A. Powers, and C. L. Partain, Invest. Radiol., 1992, 27, S109. 31. P. L. Choyke, J. A. Frank, M. E. Girton, S. W. Inscoe, M. J. Carvlin, J. L. Black, H. A. Austin, and A. J. Dwyer, Radiology, 1989, 170, 713. 32. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen, Science, 1991, 254, 716.

Biographical Sketches Johannes C. BoÈck. b 1957. M.D., 1987, GoÈttingen. Introduced to MRI by R. Felix in 1989. Research fellow, Departments of Surgery, Radiology and Nuclear Medicine, University of California, San Francisco, 1987±89; Strahlenklinik und Poliklinik, Virchow-Klinikum, Medizinische FakultaÈt der Humboldt-UniversitaÈt zu Berlin, 1989±present. Approximately 60 publications. Research interests include cerebral perfusion imaging by NMR, MR contrast media, MR angiography, and pulmonary imaging by NMR. Roland Felix. b 1938. M.D., 1962, Munich. 1962±78; Departments of Surgery, Freiburg; Gynecology, Hamburg; Medicine, Mainz; Physiology, Bonn; Medicine, Kiel; Radiology, Bonn. Professor of Clinical Radiology and Chairman, Strahlenklinik und Poliklinik (Department of Diagnostic Radiology, Nuclear Medicine, and Radiooncology), Virchow-Klinikum, Medizinische FakultaÈt der Humboldt-UniversitaÈt zu Berlin, 1978±present. Approximately 900 publications. Research interests cover contrast enhanced CT and MRI, telecommunication, and radiooncology including therapeutic hyperthermia.

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

Cerebral Perfusion Imaging by Exogenous Contrast Agents Leif éstergaard

AÊrhus University Hospital, AÊrhus, Denmark

1 INTRODUCTION Cerebral blood ¯ow (CBF) and cerebral blood volume (CBV) are traditionally used as physiologic indices of cerebral function. Positron emission tomography (PET), single photon emission computed tomography (SPECT) and stable xenonenhanced computed tomography (Xe-CT) are frequently used for determining relative or absolute CBF, as well as ¯ow changes in response to pharmacologic stimuli or functional activation. With the increasing role of magnetic resonance imaging (MRI) in the diagnosis of cerebral pathologies, techniques to derive cerebral perfusion indices have also emerged. Rosen and co-workers1±4 derived maps of relative CBV by dynamic bolus tracking of susceptibility contrast agents, a technique known as perfusion imaging. Theoretical analysis of bolus-tracking experiments has subsequently been extended to allow measurements of CBF. In this article, the theoretical foundation of perfusion imaging by exogenous contrast agents is reviewed, together with areas of potential clinical use for the techniques.

2 THEORY 2.1 Contrast Mechanisms The derivation of physiologic parameters from dynamic MR images requires detailed tracer kinetic analysis of tissue concentration±time curves obtained during passage of a bolus of contrast agent. Establishing the relation between signal changes observed with a given pulse sequence and the corresponding tissue contrast agent concentration is, therefore, necessary to derive quantitative information. Consequently, the study of interactions between contrast agents and tissue represents the foundation of perfusion imaging and remains a ®eld of intense research. Although a full overview of this area is beyond the scope of this article, certain ®ndings are of central importance to the application and understanding of perfusion imaging. 2.1.1

Susceptibility Contrast

Most cerebral perfusion imaging is carried out using dynamic susceptibility contrast imaging: tracking the passage of a rapidly injected paramagnetic gadolinium-based chelate. In

1

the presence of an intact blood±brain barrier, the intravascular compartmentalization of a paramagnetic contrast agent with high magnetic moment creates large, microscopic susceptibility gradients, with resulting signal loss in T2 and T2 -weighted images. The dynamics in the presence of a leaky blood±brain barrier will be further explored in Section 2.7. The microscopic susceptibility gradients cause local changes in water resonance frequency and in water frequency line width, with resultant signal loss and phase distortions for pulse sequences without full refocusing of static ®eld inhomogeneities (gradient echoes, asymmetric spin echoes). For pulse sequences with full refocusing of static ®eld inhomogeneities (spin echoes), these phenomena do not cause appreciable signal loss. Instead, signal loss is observed at long echo times (TE) because there is suf®cient time for water to diffuse through areas of different magnetic ®elds (i.e. among contrast-®lled vessels) and thereby dephase. This diffusion-related signal loss is a complex function of TE, the distribution and density of vessel sizes, and the concentration and magnetic properties of the contrast agent. Weisskoff, Boxerman, Fisel and co-workers performed a detailed analysis of these effects in the context of cerebral physiology, using Monte Carlo modeling as well as empirical data.5±7 Two results are particularly important in terms of interpreting perfusion imaging data. First, owing to the properties of susceptibility contrast mechanisms, spin echo measurements are mainly sensitive to vessels of a similar size to the water diffusion length (~10 m), whereas gradient echo measurements are equally sensitive to all vessel sizes. Figure 1 illustrates the change in transverse relaxation rate for gradient echo (
R2 ) and spin echo (
R2) caused by different concentrations of gadolinium and deoxyhemoglobin. Second, an approximately linear relationship exists between tissue contrast agent concentration and change in the T2 relaxation rate
R2 …t† / Ct …t†

…1†

where for gradient and spin echo sequences, signal intensity S(t) depends in an exponential fashion upon
R2 and the longitudinal relaxation rate change
R1 S…t† ˆ S…t0 †…1 ÿ expTR

R1 …t† †expTE

R2 …t†

…2†

where S(t0) is the baseline signal intensity and TR is the repetition time. If R1 remains constant, this provides the desired relation between concentration and signal intensity   S…t† =TE …3† Ct …t† ˆ ÿk log S…t0 † The constant k depends on the magnetic properties of the contrast agent, as well as its distribution space (plasma fraction). The assumption of linearity in Equation 1 has since been further con®rmed by indirect measurements in vivo;8 preliminary studies suggest that, in the brain, the microvascular CBV `visible' by spin echo planar imaging is approximately 45% of the `total' CBV as observed by PET.9 It is particularly important to bear in mind that tissue and major vessel hematocrit differ in the brain because of rheologic effects.10,11 This must be taken into account when simultaneously applying noninvasively determined tissue and arterial

2 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

Figure 1 Changes in the transverse relaxation rates (as derived from Equations 1 and 3) for (a) gradient echo [
R2 ; echo time (TE) 60 ms] and (b) spin echo (
R2; TE 100 ms) sequences as a function of vessel size. Curves are shown for typical gadolinium (Gd) doses used in cerebral perfusion imaging (0.1±0.2 mmol kgÿ1) and typical cerebral vascular volume fraction. Also shown are corresponding curves for the weakly paramagnetic deoxyhemoglobin (Deoxy Hb) used in functional MRI. Whereas spin echo measurements peak for small vessel sizes, gradient echo values reach a plateau and remain equally sensitive to all vessel sizes. In addition,
R2 values exceed
R2 values for all vessel sizes. (With permission from Boxerman et al. 19955)

concentration levels (Sections 2.2. and 2.3). In general, microvascular hematocrit is a complicated function of vessel size, ¯ow, and pathophysiologic conditions and ranges from 40 to 100% of the systemic blood hematocrit.12

2.2

Cerebral Blood Volume Measurements

Figure 2 shows a typical dynamic susceptibility contrast imaging experiment. Upon injection of contrast agent into an

Figure 2 Typical bolus tracking experiment. (a) MRI images at different time points (indicated in seconds) during a Gd-chelate bolus passage. Arrows highlight symbols, which are de®ned in the graphs. (b) Following intravenous injection, contrast agent appears early in the arteries, causing appreciable signal loss. Tissue signal loss varies according to regional blood volume. This study was obtained from a patient with a brain tumor and shows high blood volume (CBV) in the temporal lobe (see Section 3.3). (c) The corresponding concentration±time curves are derived from Equations 1 and 3. (d) These tissue curves are integrated using Equation 4 to generate maps of CBV.

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

antecubital vein, the contrast agent reaches the brain, causing a substantial signal loss. The transit of the bolus is very rapid; consequently, rapid imaging (typically 1 image every 1.5 s) is required in order to capture the ®rst pass of the bolus using fast low angle shot (FLASH) or multislice echo planar image sequences. By detecting the arterial as well as the total tissue concentration as a function of time during a single transit, the CBV can be determined from the ratio of the areas under the tissue concentration±time and arterial concentration±time curves, respectively:13±16 R1 Ct …† d …4† CBV ˆ R ÿ1 1 ÿ1 Ca …† d where Ca() and Ct() are the arterial and tissue contrast agent concentrations respectively. It is inherently dif®cult to obtain arterial concentrations by dynamic susceptibility contrast MRI (see Section 2.5); consequently, uniform arterial concentration pro®les in all arterial blood reaching the tissue is often assumed. The denominator of Equation 4 is thereby constant across regions, and relative CBV can be determined by simply integrating the area under the concentration±time curve,2±4 often by the use of a gamma variate function to correct for tracer recirculation.17

3

function of transit times is denoted by h(t), the transport function. For an arterial input Ca(t) with a ®nite duration, the venous output tracer concentration, Cv(t), is described by the convolution of the local transport function by the arterial input function: Z Cv …t† ˆ Ca …t† h…t†  Ft

t 0

Ca …†h…t ÿ † d

…5†

where denotes convolution and Ft is tissue ¯ow. The mean transit time (MTT) for the tracer particles is de®ned as the ®rst moment of the transport function: R1 h…† d …6† MTT  Rÿ1 1 ÿ1 h…† d As pointed out by Weisskoff et al. the distinction between MTT and the ®rst moment of the tissue concentration±time curve is crucial in attempts to measure transit times using intravascular tracers.18 The calculation of MTT, therefore, requires knowledge of the transport function or CBF, as, by the central volume theorem.14 MTT ˆ

CBV Ft

…7†

2.3 The Transport Function and Mean Transit Time Figure 3 shows a schematic vascular residue. An in®nitely sharp input is injected into the bloodstream of a tissue element at time t=0. The tracer follows the bloodstream through arterioles, capillaries, and venules, ®nally leaving the tissue vasculature through the vein. The associated probability density

2.4

The Residue Function: Cerebral Blood Flow

For tomographic measurements of tissue tracer trations, the local retention is described by the residue R(t), the fraction of injected tracer still present in the ture at time t; this is de®ned in terms of the transport as   Z t h…† d R…t†  1 ÿ 0

concenfunction vasculafunction

…8†

Note that by the de®nition of h(t) as a probability density function, R(0)=1 and R(?)=0, just as R(t) is a positive, decreasing function of time. As shown in Figure 3, the concentration CVOI(t) of tracer within a given volume of interest can now be written Z CVOI …t† ˆ Ft

Figure 3 Idealized, in®nitely sharp bolus injection into the bloodstream of a schematic tissue volume. The tissue concentration as a function of time (Ct(t)) is in this case given by the ¯ow (Ft) multiplied by the residue function R(t). Notice R(0)=1 and R(?)=0. The tracer concentration in the vein is determined by the transport function, h(t), which is given by the time derivative of R(t): h(t)=ÿR'(t). In actual experiments, the bolus reaching the brain will have a ®nite duration. The tissue concentration±time curve then becomes a convolved signal of the impulse response above (i.e. FtR(t)) with the arterial input shape (Equation 9).

0

t

Ca …†R…t ÿ † d

…9†

The equation states that the tissue concentration±time curve is given by an impulse response function, FtR(t), convolved with the arterial input function, Ca(t). In other words (as R(0)=1): the initial height of the deconvolved concentration±time curve equals the ¯ow, Ft. There are two main approaches to solve Equation 9 to determine regional ¯ow. The model-dependent techniques use speci®c analytical expressions chosen to describe the shape of the residue function. In model-independent approaches, straight deconvolution is performed in every image pixel, solving Equation 9 for the impulse response function.

4 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS 2.4.1

Model-independent Approach

In the model-independent approach, Equation 9 is solved for the impulse response function by standard mathematical deconvolution techniques, typically a transform approach or a linear algebraic approach. In the Fourier transform approach,19,20 Equation 9 is rewritten   FfCt …t†g FfFt
R…t† Ca …t†g ˆ FfCt …t†g ) Ft
R…t† ˆ Fÿ1 FfCa …t†g …10† where F and Fÿ1 denote the discrete and inverse discrete Fourier transform, respectively. In the linear algebraic approach, Equation 9 is rewritten into a matrix equation.21 Assuming that tissue and arterial concentrations are measured at equidistant time points t1, t2=t1+
t, . . ., tN, the tissue concentration C(tj) at time tj in Equation 9 is reformulated as a matrix equation by noting Z Ct …tj † ˆ Ft

0

tj

Ca …†R…tj ÿ † d  Ft
t

j X

Ca …ti †R…tj ÿ ti †

iˆ0

This is equivalent to 0 1 0 Ca …t1 † 0 Ct …t1 † B Ct …t2 † C B Ca …t2 † C …t1 † a B C B @ . . . A ˆ Ft

t@ . . . ... Ct …tN † Ca …tN † Ca …tNÿ1 † 0 1 R…t1 † B R…t2 † C C B @ ... A R…tN †

... ... ... ...

1 0 0 C C ... A Ca …t1 †

…11†

which can then, theoretically, be inverted to yield Ft
R(t) Stable solutions to Equations 10 and 11 can only be obtained by applying techniques to suppress experimental noise. For the Fourier transform, this is achieved by applying a ®lter to the higher frequencies in the frequency (transformed) domain, assuming this can be done without losing physiologic information. For matrix equations such as Equation 11, noise is often suppressed by regularization (forcing the solution to satisfy a priori, user-de®ned conditions, or otherwise be well behaved)22 or by singular value decomposition.23,24 The optimal choice of some transform and linear algebraic approaches was studied by éstergaard et al. using Monte Carlo simulations.24 It was found that the Fourier transform has an inherent problem in reproducing the discontinuity of the residue function at t=0, bearing in mind that lim R…t† ˆ 1

t!0‡

whereas lim R…t† ˆ 0

t!0ÿ

This discontinuity is represented at high frequencies in the transformed data and is, therefore, easily lost in subsequent ®l-

tering. As reliable determination of the initial height of the response function is essential in determining ¯ow, this property, in theory, introduces a tendency for Ft to be overestimated, and biased by the underlying vascular structure. Despite this drawback, the Fourier transform approach has the attraction of, theoretically, being insensitive to any delays between the arterial input function and the tissue arrival, as may occur in cerebrovascular disease. Preliminary results indicate that, in normal volunteers, the Fourier transform dependence upon vascular structure does not lead to appreciable differences in rCBF estimates from those obtained by the singular value decomposition approach.25 In the linear algebraic approaches, regularization showed an inherent dependency on signal-to-noise ratio (and, therefore, on rCBV). Deconvolution by singular value decomposition showed, however, a remarkable independence of vascular structure and CBV, yielding reasonably accurate CBF estimates even at the signal-to-noise ratio of pixel-by-pixel clinical echo planar imaging measurements. The major disadvantage of the linear algebraic approaches is a tendency to underestimate ¯ow when tissue tracer arrival is delayed relative to the arterial input function, a problem not associated with the transform approaches. The deconvolution techniques make no assumptions regarding the vascular structure. Instead, regional vascular transittime characteristics can be determined along with tissue ¯ow by studying the residue function. Given the potential metabolic information of the residue function (see Section 2.6. on ¯ow heterogeneity), this approach may be desirable in some cases. Even though a straight delay of tracer arrival can, in theory, be accounted for, model-less approaches cannot distinguish tracer dispersion in feeding vessels from tracer retention in the capillary bed: large vessel dispersion will be interpreted as a low ¯ow, although actual tissue ¯ow is normal.26 This is a more fundamental limitation that cannot be circumvented unless a speci®c model of major vessel dispersion is assumed. This will be described further below. 2.4.2

Model-dependent Approach

Models of tracer transport and retention must be chosen very carefully in order not to lose generality and thereby bias the resulting ¯ow values.27 Larson et al. suggested an exponential residue model, assuming the microvasculature to behave like a single, well-mixed compartment.28 Although residue functions determined by model-less approaches (Section 2.4.1) often appear exponential, this model tends to bias resulting ¯ow values in cases where the underlying residue function is nonexponential.26 éstergaard et al. modi®ed and applied a model of macrovascular transport and microvascular retention in the brain.29 The model, originally introduced to describe tracer transport and retention in the heart,30±32 utilizes vascular transport operators, allowing detailed modeling of the delay and dispersion of the arterial input owing to the passage through the artery downstream of the measurement site. Vonken et al. recently suggested a promising, more general family of analytical functions to ®t residue functions, with the further advantage of allowing for delayed tracer arrival relative to the measured arterial input.33,34 Whereas the model-less approach offers simultaneous determination of ¯ow and vascular residue function, the vascular model approach requires a model of major vessel transport as

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

well as of microvascular retention. Major vessel dispersion and microvascular retention can then to some extent be distinguished, stabilizing CBF estimates. However, abnormal capillary perfusion patterns (and consequent deviation from the normal ¯ow heterogeneity) are likely to affect ¯ow estimates by this approach.18,27 2.5 Quanti®cation Issues

Assuming that the capillary bed can be subdivided into a number of parallel microvessels with identical PS, and with individual relative ¯ow rates f (¯ow divided by the mean ¯ow Ft), Equation 12 generalizes to Equation 13 where w( f ) is the ¯ow heterogeneity, the probability density function of relative ¯ows. In this situation, the extraction fraction becomes: Z 1 w…f †…1 ÿ eÿ…PS=fFt † † df …13† Eˆ 0

The formalism above produces absolute values for CBF and CBV provided arterial and tissue concentrations are experimentally determined in identical units. This, however, represents a number of practical problems in actual clinical applications. First, in order to obtain arterial and tissue concentrations in similar units, the hematocrit difference between the micro- and macrovasculature (and thereby the difference in plasma volume) must be taken into account (see Section 2.1.1).20 Second, absolute arterial tracer concentration measurements are dif®cult to obtain from image data because of the inherently limited resolution of MRI relative to vessel sizes. Several studies have applied FLASH-type imaging sequences, allowing measurements of arterial levels in a separate, low slice with a short TE; this avoids complete loss of vascular signal during the bolus passage.18,20,35 These studies yielded somewhat high values for absolute CBF, possibly as a result of the choice of the deconvolution approach, or because of partial volume and averaging effects. A recent study by Schreiber et al. obtained absolute values in good agreement with accepted ¯ow rates.36 In multi-slice echo planar imaging experiments, a single echo is generally used, optimizing tissue signal loss but often causing complete signal loss at major vessels. Therefore, smaller arterial branches with partial volume effects with surrounding tissue are used. The shape rather than the absolute amplitude of the arterial input function is obtained in these experiments. For comparison of results in different subjects, an internal reference must be chosen that is believed to have little intersubject variability, for example white matter or cerebellum. In an attempt to obtain absolute ¯ow values from echo planar imaging, éstergaard et al. assumed proportionality between the area of the arterial input function and the injected contrast dose, using water clearance PET as a calibration method. This approach provided reproducible absolute CBF measurements in animal hypercapnia studies9 and in humans.37 It may, however, be too crude a method for general use in patients with severe cardiac or cerebrovascular disease. 2.6 Flow Heterogeneity: Metabolic Signi®cance of the Residue Function As diffusible solutes such as oxygen pass through the capillary bed, they are extracted by the tissue. The extraction fraction E (fraction of substance extracted during a single pass) of a solute is given by:38,39 E ˆ …1 ÿ eÿ…PS=Ft † †

5

…12†

where P is the permeability and S the surface area of the capillary endothelium. When the mean ¯ow is very small compared with PS, extraction is near unity; when ¯ow approaches or exceeds PS, extraction becomes incomplete.

Assuming the PS product for a given molecule and tissue is essentially constant, the ¯ow and associated ¯ow heterogeneity become the prime determinants of solute extraction when PS is large (i.e. extraction is determined by ¯ow rather than permeability). In particular, if ¯ow is ®xed (e.g. in acute stroke), extraction can only be modi®ed by altering ¯ow heterogeneity. The residue function and the ¯ow heterogeneity are related through the distribution of transit times (see Section 2.3). The transport function is the slope of the residue function h…t† ˆ ÿ

dR dt

…14†

and the distribution of ¯ows is by the central volume theorem14 given by w…f † ˆ ÿ

T
h…T † f

…15†

where T is the associated transit time for a relative ¯ow of f. Combining Equations 14 and 15 will give ¯ow heterogeneity in terms of the residue function. Flow and ¯ow heterogeneity, in turn, can be used to determine the extraction fraction by Equation 13. 2.7

Bolus Tracking when the Blood±Brain Barrier is Leaky

For high-grade CNS tumors and in diseases such as multiple sclerosis, the blood±brain barrier breaks down, and vessels become permeable to standard gadolinium chelates. The compartmentalization necessary to perform dynamic susceptibility contrast studies partly breaks down and the T1-shortening properties of the contrast agents become prominent (Figure 4). This problem has been circumvented in three ways. First, dysprosium-based contrast agents, which have little T1 enhancement, can be used rather than gadolinium-based contrast agents. As this inevitably affects subsequent conventional diagnostic imaging based on traditional T1 enhancement, this approach is now uncommon. Second, a pre-dose of gadolinium-based contrast agent can be applied, saturating the interstitial space and thereby minimizing ®rst-pass clearance during the subsequent bolus passage.40 In the following, a third approach, namely modeling the simultaneous T1 and T2 enhancement, is described. 2.7.1

Tracer Extraction: Flow versus Permeability

The total tissue concentration as a function of time as contrast agent passes from the bloodstream into the surrounding extravascular space can be broken down into an intravascular, Civ(t), and an extravascular, Cev(t), component as follows:

6 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

Figure 4 Dynamic susceptibility contrast imaging in a patient with an astrocytoma grade III with leakage of gadolinium contrast across the blood± brain barrier. In normal tissue, signal loss is observed during the bolus passage (graph, lower left). Notice the T1 enhancement in the tumor owing to contrast leakage (upper right), dominating over the normal signal loss caused by the T2 shortening from compartmentalization of the gadolinium (lower right). If not corrected for, this would cause an underestimation of cerebral blood volume. (Courtesy Dr. J. D. Rabinov)

Z Ct …t† ˆ Cev …t† ‡ Civ …t† ˆ K1
Z ‡Ft

0

t

0

t

K1 ˆ Ft …1 ÿ eÿ…PS=Ft † †

Ca …†
eÿk2 …tÿ† d

Ca …†
R…t ÿ † d

…16†

where K1 is the unidirectional clearance rate of tracer from blood to tissue and k2 is the clearance rate from tissue to blood. The latter is given by K1/Vd, where Vd is the distribution space for the tracer; for most contrast agents this is extravascular, extracellular space. The last term in Equation 16 is that given in Equation 9 to describe the concentration of intravascular tracer within a volume of interest. The ®rst and second terms in Equation 16 are very similar; in transport into the extravascular space, the residue function is assumed to be that of a well-mixed compartment, namely an exponential. The terms K1 and Vd in Equation 16 (the latter as part of k2) are equivalent to Ft and CBV, respectively, in Equations 7 and 9. Neglecting for a while the intravascular term in Equation 16, the physiologic interpretation of the unidirectional clearance K1 depends somewhat on its magnitude. Bearing in mind that the extraction fraction is the unidirectional clearance divided by the tissue ¯ow, Equation 12 can be rewritten

…17†

When the term PS is much smaller than the mean ¯ow, transport into the tissue is limited by the vascular permeability and surface area, and, from Equation 17, K1 is approximately equal to PS, i.e. it is a measure of capillary surface area and capillary permeability. When PS is much greater than mean ¯ow, clearance is independent of permeability and only ¯ow through the tissues limits the appearance of tracer in the tissue. In this ¯ow-limited case, K1 approximates to mean ¯ow. Interpretation of measured clearance rates, therefore, requires careful inspection as to whether these are comparable to likely ¯ow rates. If, on the one hand, clearance values represent PS, they may serve to characterize vascular density and integrity. However, on the other hand, if they represent ¯ow, the correlation with vascular characteristics is less clear, and they may be an indirect measurement of the supply of nutrients to the tissue. 2.7.2

Correction for Permeability

As gadolinium-chelate leaks into the tissue (®rst term in Equation 16), the T1 shortening effects of the contrast agent

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

become dominant, ultimately overwhelming the signal loss resulting from the T2 decrease associated with compartmentalization of the agent. This causes a net increase in signal intensity (Figure 4). Unless the effects of contrast leakage on T1 are corrected for, the signal loss will result in underestimation of CBV. Weisskoff et al. introduced an approximation to Equation 16 allowing simultaneous mapping of CBV and the permeability of the blood±brain barrier to gadolinium-chelates.41 The intravascular concentration term in Equation 17 was assumed to be proportional to the vascular signal in a nonleaky reference volume, Cc(t). Assuming this vascular concentration constitutes the input function to the extravascular compartment in Equation 16, and further assuming that backdiffusion is negligible (k2=0) during the short time of a bolus transit, the combined change in transverse relaxation rate,
R2,total (as derived from Equations 1 and 3) can be shown to simplify to

3.1 3.1.1

7

Acute Stroke Background

where A is proportional to K1 and B to CBV, thus allowing simultaneous determination of permeability and CBV even in the presence of contrast leakage. éstergaard et al. presented an extension of this model, maintaining the second term in Equation 16 and allowing simultaneous measurement of permeability, CBF, and CBV.42

Acute stroke is the third leading cause of death in the Western World and the major cause of adult disability. As the incidence of the disease increases exponentially with age, the increase in mean age will cause the incidence to approximately double within the ®rst half of the 21st century. In light of these severe personal and socioeconomic consequences, therapeutic strategies are emerging, that seek to limit tissue damage in the acute phase of the disease. For all types of pharmacologic intervention, treatment must be given promptly, as neuronal death from ischemia progresses rapidly after symptom onset. Currently, recombinant tissue-type plasminogen activator is approved by the US Food and Drug Administration for administration within 3 h of symptom onset (later administration of the drug is associated with risk owing to intracranial hemorrhage). Studies by PET and diffusion-weighted imaging (DWI, see below), however, indicate that neuronal death continues beyond 3 h, probably as long as 24 h in some patients. There is, therefore, an urgent need for imaging techniques to guide treatment, especially in terms of pointing to the presence of tissue at risk of infarction and selecting patients who would bene®t from aggressive treatment. This poses a special challenge to diagnostic methods, as conventional CT and MRI are insensitive to the extent of neuronal death or extent of metabolic derangement within the ®rst critical hours after symptom onset.

2.8 Other Hemodynamic Indices

3.1.2

Z
R2;total …t† ˆ A

t 0

Cc …† d ‡ B
Cc …t†

…18†

The derivation of ¯ow and transit time from bolus tracking requires derivation of arterial input tracer levels. In some cases, this may not be practical, just as the inherent complexity of deconvolution approaches may preclude the use of these techniques in some situations. Parameters that can be derived directly from the tissue concentration±time curves include timeto-peak (time from injection to maximum concentration is reached), arrival time (arrival time of tracer in the pixel), full width at half maximum of the tissue bolus shape, and ®rst moment of the peak. Although the dependence of these parameters upon MTT and CBF is strongly affected by the vascular structure and the arterial input function,18 they often suf®ce to delineate pathologic changes and provide important qualitative information in many diseases. It appears, however, that the derivation of CBF, CBV, and MTT from kinetic principles somewhat improves speci®city and sensitivity of clinical studies, facilitating inter- and intrasubject comparison.43

3 APPLICATIONS Much of the success of cerebral perfusion imaging lies in the fact that it allows the association of high-quality structural conventional MRI with physiologic parameters of importance in diagnosing and assessing outcome in major diseases. In the following discussion, major areas of impact for perfusion imaging are outlined, along with the rationale of its use.

The Role of Diffusion Weighted Imaging

DWI (see Methods and Applications of Diffusion MRI) has become widely accepted as the method of choice for demonstrating severe ischemia in acute experimental situations as well as in human stroke. It is believed that as the sodium± potassium pump breaks down as a result of the severe energy depletion, there is a subsequent loss of water homeostasis and cells swell. As total water diffusivity is dominated by water in the extracellular space, this reduction in extracellular space results in a marked reduction in the apparent diffusion coef®cient of water, which corresponds to signal increase in DWI. Although there is evidence from animal studies that the apparent diffusion coef®cient decreases gradually with the severity and duration of ischemia, and may indeed be reversible if tissue is reperfused early on, there are only sporadic reports of spontaneous reversal of DWI abnormalities in humans. After onset of ischemia, irreversible tissue damage occurs rapidly, and abnormalities have been observed by DWI within minutes in animals44 as well as in humans.45 DWI abnormalities seen in patients imaged early after symptom onset often spread substantially within the following 24 h,46±49 demonstrating further neuronal loss. It is hoped that perfusion weighted imaging (PWI) will enable prediction of this subsequent spread of tissue damage and, thus, improve patient management. 3.1.3

Perfusion Weighted Imaging: Rationale and Findings

The role of PWI in cerebral ischemia lies mainly in identifying areas where blood ¯ow and oxygen supply are compromised to such an extent that subsequent tissue damage is imminent.

8 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

Figure 5 Changes in cerebral blood ¯ow (CBF), blood volume (CBV), metabolic rate for oxygen (CMRO2), and oxygen extraction fraction (OEF) as a function of perfusion pressure. A, perfusion pressure at which CBF can no longer be maintained. B, the perfusion pressure at which tissue oxygen demands can no longer be met by the blood supply. (With permission from Powers, 199151).

Figure 5 shows a schematic outline of the hemodynamic and metabolic events that can lead to irreversible tissue death owing to a sudden decrease of perfusion pressure in acute stroke.50 Through cerebral autoregulation, CBF remains constant at perfusion pressures above approximately 60 mmHg, through a compensatory increase in CBV. Below this perfusion pressure, CBF can no longer be maintained, and with a further decrease in CBF, CBV ultimately decreases. As cortical CBF falls below roughly half of normal values, oxygen supply is no longer suf®cient to drive metabolic needs, and the patient experiences neurological de®cits. Brain tissue can only withstand this energy depletion for a limited time. With prolonged ischemia or further decrease in CBF (down to roughly 25% of normal ¯ow rates), irreversible damage is imminent. MTT is the CBV:CBF ratio (Figure 5) and is roughly inversely proportional to the perfusion pressure.51±54 As MTT thereby increases steeply as perfusion pressure drops, it has been proposed as a sensitive marker (together with CBF and CBV) of the severity of the hemodynamic impairment in patients who have suffered an acute stroke or have carotid stenosis (Section 3.2). As discussed in Section 3.1.2, abnormalities by DWI increase within the ®rst 24 h after infarction. Studies have further established that abnormal perfusion patterns can be observed by PWI in patients with acute stroke; the volume of tissue showing PWI abnormalities often exceeds that of abnormal DWI, which is referred to as a perfusion±diffusion

mismatch. More importantly, infarct growth (i.e. growth of DWI lesion) almost exclusively occurs into areas of PWI±DWI mismatch. Sorensen et al. investigated a group of 23 patients using PWI and DWI within 12 h of symptom onset of stroke.55 In two thirds, the infarct subsequently grew signi®cantly and lesion volumes by DWI at follow up exceeded the initial values by more than 15%. One third of patients displayed a decrease CBV exceeding the DWI abnormality, and half the patients showed decreased CBF beyond the DWI-determined lesion. In both cases, more than 90% of patients displaying such PWI±DWI mismatches experienced lesion growth, demonstrated on subsequent follow-up scans (Figure 6). The study con®rmed the presence of increased CBV around DWIdetermined lesions in some patients, as implied by Figure 5. Furthermore, one patient was found to have a DWI abnormality exceeding in size the PWI abnormality, suggesting the combined modalities may be used in detecting spontaneous reperfusion prior to institution of treatment. Although DWI±PWI mismatch represents a predictor of infarct growth, the volume of MTT or CBF overestimated ®nal infarct size by almost a factor of two; by comparison CBV and DWI seemed to underestimate ®nal infarct size somewhat.55 In the effort to be able to predict ®nal infarct sizes, this has led to attempts not only to detect the presence of PWI abnormalities but also to build predictive models, thereby identifying levels of CBV, CBF, and MTT associated with extreme risk of subsequent infarction.56 It is likely that effective prediction will require a combination of these parameters, rather than a single parameter (c.f. Figure 5), together with data such as duration of symptoms, blood pressure, and temperature. The observation that a PWI±DWI volume mismatch suggests risk of subsequent infarction has been suggested as a tool in patient management. Sunshine et al. demonstrated that acute measurement by DWI and PWI is feasible in patients immediately after stroke.57 Based on DWI, time-to-peak PWI, and duration of symptoms, patients in this study were strati®ed into groups with no infarct, lacunar infarcts, and cortical infarcts; on the basis of this grouping, patients were selected for conservative treatment, intravenous thrombolysis, or angiography with intra-arterial thrombolysis, respectively. This triage procedure enabled the risk of thrombolysis or invasive diagnostic procedures to be avoided in some patients while maintaining a more aggressive approach in others. 3.1.4

Flow Heterogeneity in Acute Stroke

In Figure 5, the cerebral metabolic rate of oxygen and the oxygen extraction fraction are shown. With limited blood ¯ow, the relative lack of oxygen is compensated by extracting a higher fraction of the blood oxygen pool.58 Classically, oxygen extraction is measured by PET as the ratio between regional oxygen clearance and CBF. Using invasive techniques in animal models, it can be shown that a graded decrease in perfusion pressure causes progressive loss of rapidly ¯owing red blood cells in the brain, thus decreasing total ¯ow heterogeneity and causing increased oxygen extraction (see Section 2.6).59 The importance of heterogeneity in cerebral physiology is also reviewed by Kuschinsky et al.60 éstergaard et al. examined ¯ow heterogeneity in normal volunteers and in patients within 12 h of symptom onset in strokes.29,61 In normal brain tissue, the distribution of relative

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

9

Figure 6 MR images at two levels from a 78-year-old female, 3 h after acute onset of aphasia during a cardiac catheterization routine. A, G, Structural MR images. Diffusion weighted images (B, H) show irreversible cell death in a small anterior region (arrows). Perfusion weighted imaging showed low cerebral blood volume (CBV) (C, I) and blood ¯ow (CBF) in both anterior and posterior watershed areas. Note the mismatch between areas of low CBV and CBF, respectively, easily visible as a prolongation of plasma mean transit times (E, K). Note the correspondence with the ®nal infarct size on subacute diffusion weighted images (F, L) (With permission from Sorensen et al., 199955).

¯ows was found to be markedly skewed towards high ¯ow rates, in agreement with observations using invasive techniques. Within regions of increased MTT, subregions showed loss of the high-¯ow component of the ¯ow distribution, causing increased homogeneity of ¯ow velocities,61 in agreement with animal studies.59 In parametric maps quantifying statistically signi®cant acute deviation of ¯ow heterogeneity from that of normal tissue, areas of extreme homogenization of capillary ¯ows predicted ®nal infarct size on follow-up scans in 10 out of 11 patients (Figure 7). Although this ®nding is surprising and needs further validation, it suggests that heterogeneity measurements by high signal-to-noise ratio PWI may be a value in assessing acute stroke. 3.2 Carotid Stenosis High-grade stenosis of the internal carotid arteries is associated with high risk of ischemic events, accounting for a signi®cant fraction of acute strokes. Despite the high risk of stroke in patients with carotid stenosis, surgical treatment by carotid endarterectomy has often yielded only marginal reductions in the subsequent incidence of ischemic events, indicating a need for methods to identify subgroups of highrisk patients. The pathophysiologic events ultimately leading to neurologic symptoms and tissue damage are similar to those described for stroke (see Section 3.1.3). PET studies have demonstrated that severe stenosis causes a marked compensatory increase in CBV.62 The degree of decrease in the CBF:CBV ratio (1/MTT) has been used as an indicator of regional perfusion pressure: low values (down to 50% of normal ratios) corresponding to hazardous hemodynamic impairment.62,63 However, some results suggest that oxygen extraction fraction assessed by PET may correlate more closely

with the degree of hemodynamic impairment than does the CBF:CBV ratio.64 The hemodynamic severity of carotid stenosis may also be assessed by monitoring the brain's reserve capacity, i.e. its ability to respond to stimuli increasing CBF, such as breathing carbon dioxide or therapy with acetazolamide (a carbonic anhydrase inhibitor causing a similar increase in arterial carbon dioxide partial pressure). In a group of eight patients with known carotid occlusion, Schreiber et al. found decreased CBF in six and increased MTT in seven patients; ®ve patients showed a decreased cerebrovascular reserve capacity as determined by an acetazolamide challenge.36 The study suggests that MR, which is also capable of providing accurate delineation of vascular stenosis by MR angiography, may provide a powerful tool in assessing the hemodynamic severity of carotid stenosis, thus improving the identi®cation of patients who may bene®t from endarterectomy or stenting. 3.3

Cerebral Neoplasms

Tumor growth is limited by the ability of a neoplasm to stimulate surrounding tissue to form new blood vessels. Tumor angiogenesis by release of humoral factors (vascular endothelial growth factor) is one of the hallmarks of tumor growth and the target of novel approaches to treat human neoplasms. Noninvasive methods to assess the size, density, and integrity of tumor microvessels are, therefore, essential to detect malignancies, understand tumor growth, and to target and monitor new therapeutic approaches.65 3.3.1

Tumor Vessel Density

The speci®c sensitivity of spin echo, echo planar imaging to microvessels provides a unique potential to study the proliferation of capillary sized vessels in tumor angiogenesis. As the

10 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

Figure 7 Initial diffusion weighted images (DWI) from a patient 5.5 h after onset of symptoms compatible with acute stroke (a). The patient suffers from a partial anterior cerebral artery (ACA) occlusion: note the prolonged mean transit time (MTT) in ACA territory (b) and that no acute DWI abnormalities are seen, indicating potential reversibility of the symptoms. The graph shows plots of ¯ow heterogeneity probability density functions for normal MTT (red area indicated on MTT map) and two areas showing prolonged MTT (green, yellow areas). Note the narrow ¯ow distributions in areas displaying prolonged MTT relative to normal tissue. This presumably results in higher oxygen extraction fraction. Areas of abnormal ¯ow heterogeneity (1±5; de®ned as p<0.01 by a Kolmogorov±Smirnov test comparing the pixels ¯ow heterogeneity with that of normal tissue) are shown on a cerebral blood ¯ow map (c). Note the striking correspondence with the ®nal pattern of neuronal death (indicated by numbers to show areas in (b)) on the follow-up DWI (d).

ability to form new vessels is closely related to the growth potential and thereby aggressiveness of the tumor, the relation between histologic tumor grade and regional CBV has been studied extensively. Aronen et al. studied 19 patients with cerebral glioma and showed that CBV correlated with tumor grade as determined by biopsy or surgery.40 In addition, a positive correlation was found between CBV and microscopic vascularity as well as with mitotic activity (Figure 8). These studies also suggested a close correlation between microvascular CBV and the unidirectional clearance of ¯uorodeoxy glucose (FDG) as measured by PET, the gold standard for noninvasive staging and prediction of outcome in cerebral neoplasms (Figure 8). This correspondence may not be entirely coincidental: FDG transport across cerebral microvessels is limited by endothelial transporters and is, therefore, related to vascular surface area. Its uptake therefore to some extent represents a measurement of capillary density,66 as do spin echo CBV measurements.

Subsequent reports have con®rmed that CBV mapping may aid in noninvasive staging of tumors.67,68 and in the differential diagnosis of neoplasms and infectious changes.69 CBV mapping in conjunction with conventional imaging may aid diagnosis by allowing accurate biopsy planning31,67 and can be used for assessment of the effects of treatment such as radiotherapy.70 Consequently, MR perfusion imaging may provide a tool for the important, but often dif®cult, distinction between radiation necrosis and tumor regrowth.71,72 Dennie et al. recently proposed an experimental methodology whereby the combined use of gradient and spin echo techniques for PWI could be used for more accurate, quantitative characterization of vascular morphology in tumor angiogenesis.73 With further development of these techniques, there is reason to believe that CBV mapping may become an important supplement to conventional MRI, particularly in delineating and staging tumors, in de®ning appropriate biopsy sites, in the

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS

11

Figure 8 MR of the brain in a patient with a grade III astrocytoma. (a) The T2-weighted image showing a small, contrast-enhancing lesion within the tumor volume, indicating edema. (b) A postcontrast T1-weighted image. (c) The CBV map identi®es a region of signi®cantly increased microvascular blood volume within the region of contrast leakage. (d) A PET scan using ¯uorodeoxyglucose to assess capillary density. There is a striking correspondence between the CBV map (c) and the area of high ¯uorodeoxyglucose uptake. (Courtesy of Prof. Hannu Aronen)

assessment of angiogenesis, and, ®nally, in monitoring conventional as well as novel, anti-angiogenic therapies. 4 RELATED ARTICLES Anisotropically Restricted Diffusion in MRI; Diffusion: Clinical Utility of MRI Studies; Hemodynamic Changes owing to Sensory Activation of the Brain Monitored by Echo-Planar Imaging; Ischemic Stroke; Methods and Applications of Diffusion MRI. 5 REFERENCES 1. J.W. Belliveau, D.N.J. Kennedy, R.C. McKinstry, B.R. Buchbinder, R.M. Weisskoff, M.S. Cohen, J.M. Vevea, T.J. Brady, B.R. Rosen, Science, 1991, 254: 716.

2. B. R. Rosen, J. W. Belliveau, H. J. Aronen, D. Kennedy, B. R. Buchbinder, A. Fischman, M. Gruber, J. Glas, R. M. Weisskoff, M. S. Cohen, F. H. Hochberg, T. J. Brady, Magn. Reson. Med., 1991, 22: 293. 3. B. R. Rosen, J. W. Belliveau, B. R. Buchbinder, R. C. McKinstry, L. M. Porkka, D. N. Kennedy, M. S. Neuder, C. R. Fisel, H. J. Aronen, K. K. Kwong, R. M. Weiskoff, M. S. Cohen, T. J. Brady, Magn. Reson. Med., 1991, 19: 285. 4. B. R. Rosen, J. W. Belliveau, J. M. Vevea, T. J. Brady, Magn. Reson. Med., 1990, 14: 249. 5. J. L. Boxerman, L. M. Hamberg, B. R. Rosen, R. M. Weisskoff, 1995, Magn. Reson. Med., 34: 555. 6. C. R. Fisel, J. L. Ackerman, R. B. Buxton, L. Garrido, J. W. Belliveau, B. R. Rosen, T. J. Brady, Magn. Reson. Med., 1991, 17: 336. 7. R. M. Weisskoff, C. S. Zuo, J. L. Boxerman, B. R. Rosen, Magn. Reson. Med., 1994, 31: 601. 8. C. Z. Simonsen, L. éstergaard, P. Vestergaard-Poulsen, L. Rùhl, A. Bjùrnerud, C. Gyldensted, JMRI, 1999, 9: 342.

12 CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS 9. L. éstergaard, D. F. Smith, P. Vestergaard-Poulsen, S. B. Hansen, A. Gee, A. Gjedde, C. Gyldensted, J. Cereb. Blood Flow Metab., 1998, 18: 425. 10. A. A. Lammertsma, D. J. Brooks, R. P. Beaney, D. R. Turton, M. J. Kensett, J. D. Heather, J. Marshall, T. Jones, J. Cereb. Blood Flow Metab., 1984, 4: 317. 11. O. A. Larsen, N. A. Lassen, J. Apply. Phys., 1964, 19: 571. 12. P. Gaehtgens, Biorheology, 1980, 17: 183. 13. P. Meier, K. L. Zierler, J. Apply. Phys., 1954, 6: 731. 14. G. N. Stewart, J. Physiol. (Lond.), 1894, 15: 1. 15. K. L. Zierler, Circ. Res., 1962, 10: 393. 16. K. L. Zierler, Circ. Res., 1965, 16: 309 17. H. K. Thompson Jr, F. Starmer, R. E. Whalen, H. D. McIntosh, Circ. Res., 1964, 14: 502. 18. R. M. Weisskoff, D. Chesler, J. L. Boxerman, B. R. Rosen, Magn. Reson. Med., 1993, 29: 553. 19. G. T. Gobbel, J. R. Fike, Phys. Med. Biol., 1994, 39: 1833. 20. K. A. Rempp, G. Brix, F. Wenz, C. R. Becker, F. Guckel, W. J. Lorenz, Radiology, 1994, 193: 637. 21. M. E. Valentinuzzi, V. E. Montaldo, Med. Biol. Eng., 1975, 13: 123. 22. T. A. Bronikowski, C. A. Dawson, J. H. Linehan, Int. J. Comput., 1983, 14: 411. 23. H. S. Van Huffel, J. Vandewalle, R.M. De, J. L. Willems, Med. Biol. Eng. Comput., 1987, 25: 26. 24. H. L. Liu, Y. Pu, Y. Liu, L. Nickerson, T. Andrews, P. T. Fox, J. H. Gao, Magn. Reson. Med., 1999, 42: 167. 25. R. Wirestam, L. Andersson, L. éstergaard, F. Stahlberg, Proc. VIIth Annu. Mtg. (Int.) Soc. Magn. Reson. Med., Philadelphia, 1999, p. 605. 26. L. éstergaard, R. M. Weisskoff, D. A. Chesler, C. Gyldensted, B. R. Rosen, Magn. Reson. Med., 1996, 36: 715. 27. N. A. Lassen, J. Cereb. Blood Flow Metab., 1984, 4: 633. 28. K. B. Larson, W. H. Perman, J. S. Perlmutter, M. H. Gado, K. L. Zierler, J. Theor. Biol., 1994, 170: 1. 29. L. éstergaard, D. Chesler, R. M. Weisskoff, A. G. Sorensen, B. R. Rosen, J. Cereb. Blood Flow Metab., 1999, 19: 690. 30. R. B. King, A. Deussen, G. M. Raymond, J. B. Bassingthwaighte, Am. J. Physiol., 1993, 265: H2196. 31. R. B. King, G. M. Ramond, J. B. Bassingthwaighte, Ann. Biomed. Eng., 1996, 24: 352. 32. K. Kroll, N. Wilke, H. M. Jerosch, Y. Wang, Y. Zhang, R. J. Bache, J. B. Bassingthwaighte, Am. J. Physiol., 1996, 271: H1643. 33. E. P. Vonken, F. J. Beekman, C. J. Bakker, M. A. Viergever, Magn. Reson. Med., 1999, 41: 343. 34. E. P. Vonken, M. J. Osch, C. J. Bakker, M. A. Viergever, JMRI, 1999, 10: 109. 35. W. H. Perman, M. H. Gado, K. B. Larson, J. S. Perlmutter, Magn. Res. Med., 1992, 28: 74. 36. W. G. Schreiber, F. Guckel, P. Stritzke, P. Schmiedek, A. Schwartz, G. Brix, J. Cereb. Blood Flow Metab., 1998, 18: 1143. 37. L. éstergaard, P. Johannsen, P. H. Poulsen, P. VestergaardPoulsen, H. Asboe, A. Gee, S. B. Hansen, C. E. Cold, A. Gjedde, C. Gyldensted, J. Cereb. Blood Flow Metab., 1998, 18: 935. 38. C. Crone, Acta Physiol. Scand, 1963, 58: 292. 39. E. M. Renkin, Am. J. Physiol., 1959, 197: 1211. 40. H. J. Aronen, I. E. Gazit, D. N. Louis, B. R. Buchbinder, F. S. Pardo, R. M. Weisskoff, G. R. Harsh, G. R. Cosgrove, E. F. Halpern, F. H. Hochberg, Radiology, 1994, 191: 41. 41. R. M. Weisskoff, J. L. Boxerman, A. G. Sorensen, S. F. Kulke, T. A. Campbell, B. R. Rosen, Proc. Soc. Magn. Reson., 1994, 1: 279. 42. L. éstergaard, J. D. Rabinov, B. R. Rosen, C. Gyldensted, Proc. IVth Annu. Mtg. (Int.) Soc. Magn. Reson, Med., New York, 1996, p. 1307.

43. K. Yamada, A. G. Sorensen, G. Gonzalez, W. A. Copen, O. Wu, C. J. Bakker, L. éstergaard, B. R. Rosen, Proc. 36th Annu. Mtg. Soc. Neuroradiol., 1998, p. 213. 44. M. E. Moseley, Y. Cohen, J. Mintorovitch, L. Chileuitt, H. Shimizu, J. Kucharczyk, M. F. Wendland, P. R. Weinstein, Magn. Reson. Med., 1990, 14: 330 45. Y. Yoneda, K. Tokui, T. Hanihara, H. Kitagaki, M. Tabuchi, E. Mori, Ann. Neurol., 1999, 45: 794. 46. A. E. Baird, A. Ben®eld, G. Schlaug, B. Siewert, K. O. Lovblad, R. R. Edelman, S. Warach, Ann. Neurol,, 1997, 41: 581. 47. C. Beaulieu, A. de Crespigny, D. C. Tong, M. E. Moseley, G. W. Albers, M. P. Marks, Ann. Neurol., 1999, 46: 568. 48. J. O. Karonen, R. L. Vanninen, Y. Liu, L. éstergaard, J. T. Kuikka, J. Nuutinen, E. J. Vanninen, P. L. Partanen, P. A. Vainio, K. Korhonen, J. Perkio, R. Roivainen, J. Sivenius, H. J. Aronen, Stroke, 1999, 30: 1583. 49. L. H. Schwamm, W. J. Koroshetz, A. G. Sorensen, B. Wang, W. A. Copen, R. Budzik, G. Rordorf, F. S. Buonanno, P. W. Schaefer, R. G. Gonzalez, Stroke, 1998, 29: 2268. 50. W. J. Powers, Ann. Neurol., 1991, 29: 231. 51. W.-D. Heiss, I. Podreka, Cerebrovascular disease. In: H.N. Wagner, Z. Szabo, J.W. Buchanan (eds). Principles of Nuclear Medicine, 2nd edition. W.B. Saunders Co., Philadelphia, 1995, pp. 531. 52. J. C. Baron, M. G. Bousser, A. Rey, A. Guillard, D. Comar, P. Castigne, Stroke, 1981, 12: 454. 53. P. Schumann, O. Touzani, A. R. Young, J. C. Baron, R. Morello, E. T. MacKenzie, Brain, 1998, 121: 1369. 54. G. Zaharchuk, J. B. Mandeville, A. A. J. Bogdanov, R. Weissleder, B. R. Rosen, J. J. Marota, Stroke, 1999, 30: 2197. 55. A. G. Sorensen, W. A. Copen, L. éstergaard, F. S. Buonanno, R. G. Gonzalez, G. Rordorf, B. R. Rosen, L. H. Schwamm, R. M. Weisskoff, W. J. Koroshetz, Radiology, 1999, 210: 519. 56. O. Wu, A. G. Sorensen, D. Bakker, F. Buonanno, W. A. Copen, G. Gonzalez, C. Harmath, L. éstergaard, B. R. Rosen, L. H. Schwamm, G. Rordorf, R. M. Weisskoff, W. M. Wells, K. Yamada, G. Zaharchuk, W. J. Koroshetz, Proc. VIth Annu. Mtg. (Int.) Soc. Magn. Reson. Med., Sydney, 1998, p. 235. 57. J. L. Sunshine, R. W. Tarr, C. F. Lanzieri, D. M. Landis, W. R. Selman, J. S. Lewin, Radiology, 1999, 212: 325 58. R. J. Wise, S. Bernardi, R. S. Frackowiak, N. J. Legg, T. Jones, Brain, 1983, 106: 197. 59. A. G. Hudetz, G. Feher, J. P. Kampine, Microvasc. Res., 1996, 51: 131. 60. W. Kuschinsky, O. B. Paulson, Cerebrovasc. Brain Metab. Rev., 1992, 4: 261. 61. L. éstergaard, A. G. Sorensen, D. Chesler, R. M. Weisskoff, W. J. Koroshetz, C. Gyldensted, B. R. Rosen, J. Cereb. Blood Flow Metab., 1999, 19 (Suppl. 1), S621. 62. J. M. Gibbs, R. J. Wise, K. L. Leenders, T. Jones, Lancet, 1984, i: 310. 63. W. J. Powers, G. A. Press, R. L. J. Grubb, M. H. Gado, M. E. Raichle, Ann. Intern. Med., 1987, 106: 27. 64. M. Itoh, J. Hatazawa, C. Pozzelli, H. Fukuda, Y. Abe, T. Fujiwara, K. Kubota, K. Yamaguchi, T. Sato, T. Matsuzawa, Neuroradiology, 1987, 29: 416. 65. N. Ferrara, K. Alitalo, Nature Med., 1999, 5: 1359. 66. A. Gjedde, H. Kuwabara, A. M. Hakim, J. Cereb. Blood Flow Metab., 1990, 10: 317. 67. E. A. Knopp, S. Cha, G. Johnson, A. Mazumdar, J. G. Gol®nos, D. Zagzag, D. C. Miller, P. J. Kelly, I. I. Kricheff, Radiology, 1996, 211: 791. 68. T. Sugahara, Y. Korogi, M. Kochi, I. Ikushima, T. Hirai, T. Okuda, Y. Shigematsu, L. Liang, Y. Ge, Y. Ushio, M. Takahashi, Am. J. Roentgenol., 1998, 171: 1479. 69. T. M. Ernst, L. Chang, M. D. Witt, H. A. Aronow, M. E. Cornford, I. Walot, M. A. Goldberg, Radiology, 1998, 208: 663.

CEREBRAL PERFUSION IMAGING BY EXOGENOUS CONTRAST AGENTS 70. F. Wenz, K. Rempp, T. Hess, J. Debus, G. Brix, R. Engenhart, M. V. Knopp, K. G. Van Kaick, M. Wannenmacher, Am. J. Roentgenol., 1996, 166: 187. 71. H. J. Aronen, J. Glass, F. S. Pardo, J. W. Belliveau, H. L. Gruber, B. R. Buchbinder, I. E. Gazit, R. M. Linggood, A. J. Fischman, B. R. Rosen, Acta Radiol., 1995, 36: 520. 72. T. Siegal, R. Rubinstein, T. Tzuk-Shina, J. M. Gomori, J. Neurosurg., 1997, 86: 22. 73. J. Dennie, J. B. Mandeville, J. L. Boxerman, S. D. Packard, B. R. Rosen, R. M. Weisskoff, Magn. Reson. Med., 1998, 40: 793.

13

Biographical Sketch Leif éstergaard, b 1965. B.Sc. 1987, Physics and Mathematics, M.Sc., Ê rhus University, A Ê rhus, Denmark. 1992, Astrophysics, M.D., 1994, A Formerly Research Fellow, MGH-NMR Center, Department of Radiology, Massachusetts General Hospital, Charlestown, Massachusetts, USA. Presently Research Director, Magnetic Resonance Imaging, Department of Neuroradiology and Intern, Grenaa General Hospital, Grenaa, Denmark. Approx. 140 publications. Research interests: perfusion, diffusion, water exchange phenomena and functional MRI.

CSF VELOCITY IMAGING

CSF Velocity Imaging William G. Bradley, Jr Long Beach Memorial Medical Center, CA, USA

1 INTRODUCTION Cerebrospinal ¯uid (CSF) is produced by the choroid plexus within the ventricles of the brain at a rate of 500 ml per day. It ¯ows out of the foramina of Lushka and Magendie in the fourth ventricle and is absorbed by the arachnoid villi on either side of the superior sagittal sinus.1 Superimposed on this slow, steady egress of CSF from the ventricular system are much more prominent cardiac pulsations resulting from systolic expansion and diastolic contraction of the brain. When the local CSF velocity is suf®ciently high, e.g. through the narrow aqueduct of Sylvius, a `CSF ¯ow void'2,3 is normally produced, similar to that seen with rapidly ¯owing blood.4 When there is obstruction to CSF ¯ow, the ¯ow void is not observed.3 Obstruction proximal to the outlet foramina of the fourth ventricle is termed `obstructive hydrocephalus'; obstruction between the outlet foramina of the fourth ventricle and the arachoid villi is termed `communicating hydrocephalus'. Communicating hydrocephalus most often follows subarachoid hemorrhage or meningitis; however, a number of elderly patients may present with a radiographic pattern of communicating hydrocephalus without a history of either of these preceding events. Such hydrocephalus is termed `occult' or `idiopathic'.5,6 Since the mean intraventricular pressure is normal, it has also been called `normal pressure hydrocephalus' (NPH). Classically, NPH patients present with a clinical triad of gait apraxia, dementia, and incontinence.5±7 Although the mean intraventricular pressure is normal in these patients, the pulse pressure, i.e. the change in CSF pressure over the cardiac cycle, is much greater than normal. A `waterhammer' pulse has been described in such patients when their intraventricular pressure is monitored.8,9 Until now the diagnosis of NPH has been made on the basis of a radiographic picture of ventricles enlarged out of proportion to any cortical sulcal enlargement (to distinguish NPH from atrophy). Prior to the advent of MRI, this radiographic picture and the appropriate clinical triad alone were not suf®cient to predict which patients would respond to the primary treatment for NPH, i.e. ventriculoperitoneal (VP) shunting. Recently, the magnitude of the aqueductal CSF ¯ow void (Figure 1) has been used to predict which patients with clinical NPH will respond to shunting.10 The CSF ¯ow void is a normal phenomenon, resulting from the pulsatile motion of CSF back and forth through the ventricular system over the cardiac cycle.11 In the normal patient, there is space over the cortical convexities occupied by the cortical veins and the CSF in the subarachnoid space (Figure 2). When the brain expands during systole, it expands outward, compressing the cortical veins (leading to subsequent venous out¯ow), and it expands inward, compressing the lateral ventricles (leading

1

to out¯ow of CSF through the aqueduct which produces the normal ¯ow void).10 In communicating hydrocephalus, the brain has already expanded against the inner table of the calvarium, so further outward expansion is not possible. Thus, during systole all cerebral expansion is directed inward, compressing the lateral ventricles, leading to increased out¯ow of CSF through the aqueduct. This is the etiology for the increased CSF ¯ow void7 in patients with communicating hydrocephalus and normal cerebral blood ¯ow (Figure 2). When the blood supply to the brain decreases due to atrophy, the main force behind the CSF pump decreases and the CSF ¯ow void is consequently also reduced. In a recent study, there was a signi®cant correlation between an increased CSF ¯ow void, i.e. one extending from the third ventricle through the obex of the fourth ventricle, and a successful response to ventriculoperitoneal shunting.7 Patients with clinical NPH who did not have a hyperdynamic ¯ow void, on the other hand, failed to respond to shunting. This association was signi®cant at the p < 0.003 level.10 Thus, in the setting of an appropriate clinical triad for NPH and ventricular enlargement out of proportion to cortical sulcal dilatation, the presence of an increased aqueductal CSF ¯ow void on MRI should prompt the neurosurgeon to perform a VP shunt. It should be emphasized, however, that younger patients with communicating hydrocephalus and intact cerebral blood supply may also have hyperdynamic CSF ¯ow. Thus, the sign is not speci®c for NPH. Since NPH is a disease of elderly patients, it appears that symptoms follow a combined insult related to the ventricular enlargement and also to decreased perfusion of the periventricular region, i.e. the same process that leads to deep white matter ischemia and infarction.12

Figure 1 CSF ¯ow void in normal pressure hydrocephalus. Proton density-weighted axial image through the upper fourth ventricle demonstrates marked CSF ¯ow void (arrow)

2 CSF VELOCITY IMAGING Cortical venous space Brain Ventricles

Calvarium Jugular veins

Aqueduct

Venous outflow

Normal: Diastole

CSF outflow Normal: Systole

(a) Diastole

Systole

expected (e.g. the aqueduct), however, inferences can be made concerning obstruction.3 Axial, single slice, gradient echo techniques which maximize ¯ow-related enhancement4 (entry phenomenon) have also been used14 to evaluate aqueductal patency. However, the best observations of CSF ¯ow have been obtained using dedicated techniques, gated to the cardiac cycle. In 1985, Feinberg and Mark15,16 used MR `velocity density imaging' to measure the velocity of CSF through the aqueduct as a function of the cardiac cycle. Subsequently Edelman et al.17 used saturation pulses and bolus tracking techniques, while Axel and Dougherty.18 used spatial modulation of magnetization (SPAMM) to measure CSF velocity. Today, most investigators are using phase contrast techniques.19±21 The phase contrast CSF velocity imaging techniques19±21 can be divided on the basis of the form of cardiac gating used, i.e. prospective or retrospective,22 and on the basis of background phase determination.23 Most investigators19,20 determine the background phase from an additional acquisition. This approach necessitates obtaining two interleaved acquisitions and then subtracting them. We make the assumption of zero net ¯ow21 over the cardiac cycle and calculate the background phase (which entails a single acquisition in one-half the time). The relationship between the phase shift  and velocity v is given by: Z ˆ

(b)

Figure 2 Schematic representation of CSF ¯ow mechanism. (a) In normal patients, systolic expansion of cerebral hemispheres occurs outward, compressing cortical veins and, to a lesser extent, inward, compressing the lateral ventricles. (b) In communicating hydrocephalus, the brain has already expanded against the inner table of the calvarium. Since outward expansion of the brain is no longer possible, the entire increase in volume due to systolic expansion is directed inwards, towards the lateral ventricles. The combination of this greater inward displacement and the mild aqueductal enlargement that accompanies communicating hydrocephalus leads to hyperdynamic CSF ¯ow

2 MR TECHNIQUES USED TO MEASURE CSF FLOW A number of magnetic resonance (MR) observations and techniques have been employed to take advantage of signal changes associated with CSF ¯ow. The aqueductal CSF ¯ow void (Figure 1) noted on routine MR images was the ®rst indication of CSF motion.2,3 Subsequently, the CSF ¯ow void was shown to be related to the cardiac cycle.13 Although a CSF ¯ow void may be appreciated on routine spin echo and gradient echo MR images, the degree of signal loss is not consistent, nor does it change in a linear fashion with velocity.1 Thus by using the CSF ¯ow void sign alone, it may be dif®cult to appreciate when CSF ¯ow is mildly abnormal. When a CSF ¯ow void is not visualized at all in areas where it is normally

t 0

G…t†vt dt

…1†

where is the gyromagnetic ratio, G(t) is the gradient strength as a function of time t, and t is the elapsed time.24 Thus for an appropriate gradient pulse of strength G and duration t, a linear relationship should exist between the phase shift (from zero to ‹180 , depending on the direction of ¯ow) and the actual CSF velocity. A new parameter, the Venc, is the aliasing or `encoding' velocity, which leads to a 180 phase shift. With prospective cardiac gating, acquisition starts at a predetermined time interval after the R wave and continues at approximately 50±75 ms intervals to within about 200 ms of the next R wave (to allow for respiratory variation).19 Since there is no sampling of CSF motion during the ®nal 100 or 200 ms of the R±R interval (when ¯ow is in a rostral or retrograde direction), there appears to be a large net ¯ow of CSF in the systolic direction (i.e. craniocaudal) over this partially sampled cardiac cycle.21 With retrospective cardiac gating,22 the computer keeps track of the R wave and data are acquired throughout the cardiac cycle and then retrospectively `binned' into a predetermined number of frames.22 The entire cardiac cycle is sampled; therefore, there is no time for eddy currents to build up, as they do during the 200 ms dead time when prospective cardiac gating is used.21 We routinely use two retrospective cardiac gated techniques:21 a `routine resolution' sagittal acquisition in the midline (see Figure 3); and a `high-resolution' axial acquisition through the aqueduct (see Figure 4). In the sagittal technique, a 4 mm slice is acquired in the midsagittal plane (Figure 3) with velocity sensitization along the read-out axis (which is also the craniocaudal axis of the patient). Cardiac gating is performed with ECG leads rather than using ®nger plethysmography,

CSF VELOCITY IMAGING

3

Figure 4 Axial phase contrast technique. (a) High-resolution axial image obtained during diastole indicates upward ¯ow (black) in the aqueduct (arrow). (b) During systole, ¯ow is downward (white). The mean velocity (left upper), peak velocity (right upper), and volumetric ¯ow rates (left lower) through the aqueduct are shown as a function of the cardiac cycle

Figure 3 Sagittal phase contrast technique. (a) During systole, CSF ¯ow is downward (craniocaudal), as indicated by shades of white. (b) During diastole, CSF ¯ow is upward (caudocraniad), as indicated by shades of black. By placing a cursor over any point in the image, the velocity can be obtained as a function of the cardiac cycle

since the systolic motion of the brain may actually precede the systolic pulse in the ®nger.21 The basic technique is a ¯owcompensated, two-dimensional fast imaging with steady precession (FISP) with a repetition time (TR) of 70 ms, a echo delay time (TE) of 13 ms, and a 15 ¯ip angle (Figure 3). A 192  256 matrix is acquired over a 25-cm ®eld of view, providing

in-plane spatial resolution of approximately 1 mm. The aliasing velocity is 100 mm sÿ1. Such images are used routinely in the head to document midsagittal CSF ¯ow. Using a single excitation and 32 acquisitions before phase advance, this sagittal technique takes 7 min. When a cursor is placed over any point on the phase contrast velocity image, craniocaudal CSF velocities can be measured and plotted versus the phase of the cardiac cycle, allowing comparison of phase relationships of CSF motion at different points in the ventricular system and subarachnoid space21 (Figure 3). The measured velocity through the aqueduct on this sagittal acquisition also re¯ects the stationary tissue in the midbrain on either side of the aqueduct in a 4 mm thick slice. Therefore,

4 CSF VELOCITY IMAGING

Figure 5 In this 78-year-old man with clinical NPH, hyperdynamic ¯ow is demonstrated through the aqueduct with the qualitative sagittal technique

the velocity is artifactually reduced due to partial volume effects.21 To circumvent these effects, the high resolution axial technique was developed.21 In this technique, an angled axial slice is positioned perpendicular to the aqueduct so that the CSF is viewed en face, without inclusion of adjacent stationary tissue (Figure 4). Velocity sensitization is provided by the slice-select gradient, i.e. for through-plane ¯ow. We currently use a 512  512 (half Fourier) matrix over a 16 cm ®eld of view, providing pixels 0.3 mm on a side. Thirty such pixels can be placed in the average 2-mm diameter aqueduct. Like the sagittal technique, this is a modi®ed two-dimensional FISP with a 15 ¯ip angle, a TR of 100 ms, and a TE of 16 ms. With a single excitation and 32 acquisitions per phase step, it takes 14 min.21 An aliasing velocity of 200 mm sÿ1 is routinely used. Integrating the volumetric ¯ow rate over the cardiac cycle, an aqueductal CSF `stroke volume' can be calculated.21 Using a pulsatile ¯ow phantom velocity, volumetric ¯ow measurements have recently been validated.25 3 NORMAL CSF FLOW IN THE BRAIN In normals, the maximal downward wave of CSF movement through the aqueduct occurs 175±200 ms following the R wave; however, it is usually preceded by CSF ¯ow out of the fourth ventricle through the foramen of Magendie.20 Mixing occurs in the middle of the fourth ventricle, resulting in turbulence.21 Midway through the cardiac cycle, a diastolic cephalad (retrograde) wave of CSF can be seen pulsing through the aqueduct.20 Variable degrees of CSF ¯ow within the posterior third ventricle are observed and, occasionally, ¯ow at the level of the foramen of Monro is also seen in normals. Flow is less often observed in the lateral ventricles, however, due to the larger diameters involved. As a result of volume changes in the brain during the cardiac cycle, CSF also ¯ows to and fro in the basal cisterns.21 Flow out of the inferior portion of the fourth ventricle through the foramen of Magendie is not simply a continu-

Figure 6 Shunt malfunction in NPH. (a) A 73-year-old man shunted initially for NPH one year previously now presents with recurrent symptoms and `normal' ¯ow pattern through the aqueduct. (b) Two months later, when the shunt has been revised and the patient is again asymptomatic, the CSF ¯ow pattern through the aqueduct is reversed (as expected with normal shunt function)

ation of the ¯ow coming in from the aqueduct. Such asynchrony may result from pulsations of the choroid plexus of the fourth ventricle.20 In fact, the caudally directed CSF pulse wave is observed to start ®rst in the fourth ventricle, followed approximately 100 ms later by caudal ¯ow in the aqueduct.20 Cranial ¯ow of CSF back through the aqueduct may still be occurring when the ®rst downward pulsation through the foramen of Magendie is seen. Clearly, therefore, this cephalad aqueductal CSF ¯ow is not simply the result of retrograde ¯ow from the basal cisterns into the fourth ventricle and upwards.20 The probable reasons for these normal

CSF VELOCITY IMAGING

5

Figure 7 Normal shunt function. (a) A shunt tube (arrow) is noted in a 2-year-old boy with obstructive hydrocephalus. (b) This right parasagittal section demonstrates the position of the high-resolution acquisition plane perpendicular to the dark shunt tube (arrow). (c) A `magnitude' image from a high-resolution coronal acquisition, angled perpendicular to the shunt tube (small arrow); the image also demonstrates the quadrigeminal plate cistern (large arrow). (d) A phase contrast image again demonstrates the quadrigeminal plate cistern (large arrow) and high signal intensity (small arrow) in the shunt, corresponding to forward ¯ow

variations are multiple, but relate most likely to the size of the nearby vascular structures, the compliance of surrounding brain and spinal cord, the anatomy of the CSF spaces, the volume and vascularity of the choroid plexus, and the resultant systemic hydrodynamics.20

4

ABNORMAL CSF FLOW IN THE BRAIN

Using the qualitative sagittal technique, the hyperdynamic CSF ¯ow of NPH (Figure 5) can be distinguished from the hypodynamic ¯ow of atrophy. Even more accurate is the high

6 CSF VELOCITY IMAGING

CSF VELOCITY IMAGING

7

Figure 8 The `rebound sign' in an arachnoid cyst. The qualitative (sagittal) technique has been rotated into the coronal plane with velocity encoding along the craniocaudal read-out axis. White, downward ¯ow; black, upward ¯ow (no ¯ow is gray). (a) During systole, ¯ow in the fourth ventricle (large arrow) is down while ¯ow in the arachnoid cyst (small arrow) has stopped due to maximal deformation by the surrounding CSF. (b) 200 ms later, ¯ow in the fourth ventricle has stopped (turning gray), while ¯ow in the cyst has already started heading up (turning black). (c) 200 ms later, both the ¯ow in the cyst and that in the fourth ventricle are up. (d) 100 ms later, the cyst has become maximally deformed and has stopped ¯owing (turning gray), while ¯ow continues in the caudocraniad direction in the fourth ventricle. (e) By placing cursors on the fourth ventricle and the cyst, motion in the cyst is seen to be of lower amplitude and 90 phase-advanced relative to the ¯ow within the fourth ventricle

resolution axial technique which provides a measurement of the aqueductal CSF stroke volume. In a recent study of 20 patients with clinically suspected NPH, 14 were found to have hyperdynamic CSF ¯ow, i.e. stroke volumes greater than 40 L, while six were found to have hypodynamic ¯ow.26 Of the patients with hyperdynamic ¯ow, 13 of 14 responded favorably to the ventriculoperitoneal shunt. (The one patient who did not was known to have chronic multiple sclerosis.) Thus this technique has an excellent positive predictive value. Of the six patients who had hypodynamic ¯ow, three responded to shunting and three did not, probably re¯ecting the presence of concomitant atrophy in these patients. Interestingly, of the 14 patients with hyperdynamic ¯ow (as determined by the measured stroke volume), only seven (50%) had hyperdynamic CSF ¯ow voids on their routine MR images. This most likely re¯ects the current widespread use of ®rstorder ¯ow compensation on the proton density-weighted images which was not available on the 1984 images in the earlier study.10 Thus, quantitative CSF velocity imaging is an even better technique than routine MRI to determine the presence of shunt-responsive normal pressure hydrocephalus. CSF velocity imaging is also useful in the evaluation of potential shunt malfunction.27 In patients who have been shunted for communicating hydrocephalus, the normal pattern of CSF ¯ow through the aqueduct is exactly 180 out of phase with normal with respect to the cardiac cycle, i.e. CSF ¯ows caudo-

craniad during systole and craniocaudad during diastole (Figure 6). This probably re¯ects the fact that the shunt tube is the lowest resistance pathway to out¯ow of CSF following systolic expansion of both the cerebrum and the cerebellum. In patients with either obstructive or communicating hydrocephalus, the high-resolution axial technique can also be used to assess ¯ow through the shunt tube itself (Figure 7). In cases of normal shunt function, CSF motion tends to be intermittent and unidirectional, rather than sinusoidal (as is normally seen through the aqueduct), re¯ecting the presence of the one-way valve. Since the plane and position of both CSF velocity imaging techniques can be modi®ed, they can also be used to distinguish cysts from enlarged CSF spaces. Within a cyst, the motion of CSF appears to rebound against the ¯owing CSF around it. This `cyst rebound sign'28 is the combination of decreased CSF velocity (within the cyst) 90 phase-advanced relative to the motion of the surrounding CSF (Figure 8).

5

RELATED ARTICLES

Brain Parenchyma Motion Observed by MRI; Marker Grids for Observing Motion in MRI; Phase Contrast MRA; Whole Body Magnetic Resonance Angiography.

8 CSF VELOCITY IMAGING 6 REFERENCES 1. W. G. Bradley and R. M. Quencer, in `Magnetic Resonance Imaging', 2nd edn, eds D. D. Stark and W. G. Bradley. Mosby-Yearbook, St Louis, MO, 1992, Chap. 28. 2. W. G. Bradley, K. E. Kortman, and B. Burgoyne, Radiology, 1986, 159, 611. 3. J. L. Sherman and C. M. Citrin, Am. J. Neuroradiol., 1986, 7, 3. 4. W. G. Bradley and V. Waluch, Radiology, 1985, 154, 443. 5. S. Hakim, Thesis No. 957, Javerian University, School of Medicine, Bogota, Columbia, 1964. 6. R. D. Adams, C. M. Fisher, S. Hakim, R. G. Ojemann, and W. H. Sweet, N. Engl. J. Med., 1965, 273, 117. 7. S. Hakim, J. G. Venegas, and J. D. Burton, Surg. Neurol. 1976, 5, 187. 8. J. Ekstedt and H. Friden, in `Hydrocephalus' eds. K. Shapiro, A. Marmarou, and H. Portnoy, Raven, New York, 1984. 9. E. L. Foltz, in `Hydrocephalus', eds. K. Shapiro, A. Marmarou, and H. Portnoy, Raven, New York, 1984. 10. W. G. Bradley, A. R. Whittemore, K. E. Kortman, A. S. Watanabe, M. Homyak, L. M. Teresi, and S. J. Davis, Radiology, 1991, 178, 459. 11. G. H. DuBoulay, Br. J. Radiology., 1966, 39, 255. 12. W. G. Bradley, A. R. Whittemore, A. S. Watanabe, S. J. Davis, L. M. Teresi, and M. Homyak, Am. J. Neuroradiol., 1991, 12, 31. 13. C. M. Citrin, J. L. Sherman, R. E. Gangarosa, and D. Scanlon, Am. J. Roentgenol., 1987, 148, 205. 14. S. W. Atlas, A. S. Mark, and E. K. Fram, Radiology, 1988, 169, 449. 15. D. A. Feinberg, L. E. Crooks, P. Sheldon, J. Hoenninger, J. C. Watts, and M. Arakawa, Magn. Reson. Med., 1985, 2, 555. 16. D. A. Feinberg and A. S. Mark, Radiology, 1987, 163, 793. 17. R. R. Edelman, H. P. Mattle, J. Klee®eld, and M. S. Silver, Radiology 1989, 171, 551.

18. L. Axel and L. Dougherty, Radiology, 1989, 171, 841. 19. R. M. Quencer, M. J. Post, and R. S. Hinks, Radiology, 1990, 32, 371. 20. D. R. Enzmann and N. J. Pelc, Radiology, 1991, 178, 467. 21. W. R. Nitz, W. G. Bradley Jr., A. S. Watanabe, R. R. Lee, B. Burgoyne, R. M. O'Sullivan, and M. D. Herbst, Radiology, 1992, 183, 395. 22. T. A. Spraggins, Magn. Reson. Imaging, 1990, 8, 676. 23. D. N. Firmin, G. L. Nayler, P. J. Kilner, and D. B. Longmore, Magn. Reson. Med., 1990, 14, 230. 24. P. R. Moran, Magn. Reson. Imaging, 1982, 1, 197. 25. W. J. Mullin, D. Atkinson, R. Hashemi, J. Yu, and W. G. Bradley, J. Magn. Reson. Imaging, 1993, 3, 55. 26. W. G. Bradley, D. Scalzo, J. Queralt, W. N. Nitz, D. J. Atkinson, and P. Wong, Radiology, 1996, 198, 523. 27. W. G. Bradley, D. Atkinson, D.-Y. Chen, W. R. Nitz, and W. J. Mullin, Radiology, 1993, 189, 223. 28. M. D. Herbst, W. G. Bradley, W. R. Nitz, B. Burgoyne, R. R. Lee, and R. M. O'Sullivan, Radiology, 1991, 181, 287.

Biographical Sketch William G. Bradley, Jr. b 1948. B.S., Chemical Engineering Caltech, 1970; Ph.D., Chemical Engineering, Princeton, 1974; M.D. UCSF 1977; Internship UCSF, 1978; Radiology Residency UCSF, 1981. Introduced to NMR by J. D. Roberts in 1968 and to MRI by L. Crooks in 1979. Approx. 150 publications and 10 textbooks including `Magnetic Resonance Imaging' with David D. Stark. President, SMRI 1988±89. Currently Director of MRI and Radiology Research, Long Beach Memorial Medical Center, and Professor of Radiology, University of California at Irvine.

HEAD AND NECK STUDIES USING MRA

Head and Neck Studies Using MRA Paul M. Ruggieri, Jean A. Tkach, and Thomas J. Masaryk Cleveland Clinic Foundation, Cleveland, OH, USA

1 INTRODUCTION Magnetic resonance angiography (MRA) is most commonly applied to the evaluation of cerebrovascular disease as neurological examinations continue to be the mainstay of clinical MRI. In addition, the rapid blood ¯ow through the extra-/intracranial cerebral vasculature, the paucity of gross physiological motion, and the existing coil technology make these vessels ideally suited to vascular MRI. MRA is especially appealing since it provides a morphological representation of the vasculature as a relatively rapid, noninvasive alternative to the existing vascular imaging modalities. During the same examination, conventional spin echo images of the brain parenchyma are acquired to assess the consequences of the underlying vascular pathology. 1.1 Carotid MRA: Technical Considerations Two basic types of MRA sequences are commonly employed to study the cerebral vasculature: phase contrast (PC) and time-of-¯ight (TOF).1,2 The PC technique provides optimal suppression of the background stationary tissues via subtraction of alternating ¯ow encoded images, and can be designed to be highly sensitive to blood moving at speci®c velocities. Appropriately calibrated sequences can also be designed to measure ¯ow velocities. However, this technique demands longer acquisition and postprocessing times than the TOF studies of comparable spatial resolution. The phase sensitivity of this method also makes it more vulnerable to intravascular signal loss due to complex motion (e.g. carotid bifurcation stenosis) and/or local ®eld inhomogeneities (e.g. carotid siphon, densely calci®ed atherosclerotic plaque).3 Consequently, the PC study is not commonly used to evaluate the carotid bifurcations in routine clinical work. A 2D PC study is useful as a quick scout view for subsequent positioning of the 2D or 3D TOF volumes and to con®rm vessel patency. The 2D PC technique may also prove to be useful for making ¯ow velocity measurements in the region of a carotid stenosis (similar to duplex ultrasound studies). The TOF techniques take advantage of the rapid, constant in¯ow of fresh unsaturated spins to provide the necessary contrast relative to the adjacent stationary tissues. In lieu of subtraction mask acquisitions, angiographic images are rendered using computer postprocessing algorithms which may then create multiple MRA images from a single data set. The TOF methods are capable of minimizing intravoxel phase dispersion (and intravascular signal loss) by incorporating very

1

small voxels and very short echo times with motion refocusing gradients.4 Alternatively, the TOF sequences are not as sensitive to slow ¯ow as the PC techniques. The TOF images also have less effective background suppression. Because the TOF studies are inherently T1-weighted, and the commonly used postprocessing method (maximum intensity projection or MIP) cannot necessarily distinguish bright vessels from other tissues with short T1 relaxation times (e.g. fat, subacute hemorrhage), the vessels in the volume may be obscured by adjacent tissues. The MIP reconstruction technique may introduce additional artifacts in all MRA images. Small vessels may frequently be visible in the original slices but are not represented in the MRA image because of: (1) variation in background intensities; (2) marginal ¯ow-related enhancement; and (3) partial volume averaging artifacts.5 Neither the PC nor the TOF technique can provide the contrast and spatial resolution needed to identify very subtle contour abnormalities which would be important in con®rming mild vascular dysplasias (e.g. ®bromuscular disease) or vasculitis that would only be evident using conventional arteriography. These limitations are predominantly related to signal-to-noise ratio (S/N) issues and the hardware (e.g. gradients, computer memory, processing speed) constraints imposed on the MRA pulse sequences. The two techniques are also unable to provide the same dynamic information as with catheter angiography. 1.2

Carotid MRA: Clinical Signi®cance

MRA has the potential to make its greatest impact on routine clinical work in the evaluation of carotid bifurcation atherosclerotic disease. Justi®cation for imaging the carotid bifurcations is based on the conclusions of the North American Symptomatic Carotid Endarterectomy Trial (NASCET) and the European Carotid Surgery Trial (ECST). These controlled randomized clinical studies support the clinical utility of carotid endarterectomies in symptomatic patients with severe (>70%) carotid bifurcation stenosis, and set well de®ned standards for the preoperative evaluation of these patients (angiographic demonstration, speci®c measurements of the stenosis, signi®cant tandem lesions).6±10 Ideally, it would be preferable to replace the conventional catheter arteriogram with a noninvasive alternative, thereby avoiding the potential risks of the invasive study. As many of the symptomatic patients with cerebrovascular disease are evaluated by spin echo MRI to assess the severity of the parenchymal sequelae, MRA of the carotid and intracranial vasculature would be a logical (noninvasive) extension of such studies. However, this makes the assumption that MRA can provide the same information as the invasive study and has similar sensitivity and speci®city. The questions that remain to be proven in large scale, prospective clinical trials include the following: (1) can MRA characterize carotid stenoses of 70± 99% severity, (2) distinguish complete occlusion from severe stenosis, and (3) exclude tandem stenoses as accurately as catheter angiography? Among these questions, the issue with the least objective data is the ability of MRA to exclude the presence of tandem stenoses, most commonly in the carotid siphon. The presence of distal stenoses have been associated with an increased risk for cerebrovascular and cardiac complications, suggesting a subpopulation with more malignant

2 HEAD AND NECK STUDIES USING MRA atherosclerosis.11,12 Moreover, although 3D TOF and PC MRA continue to improve, there are frequently artifacts mimicking stenoses of the carotid siphons due to complex ¯ow and/or local ®eld inhomogeneities.13,14 In the evaluation of carotid bifurcation disease, MRA is currently most appropriately assigned the role of a screening study for signi®cant atherosclerotic disease which would be best followed up by conventional angiography, preoperatively. 1.3 Carotid MRA: Technical Considerations In 2D TOF imaging, the carotid arteries are visualized by acquiring a series of thin 2D axial gradient echo images sequentially (superior to inferior) against the direction of ¯ow. The high signal from the venous structures (particularly the jugular vein) can obscure the carotid bifurcation, but a traveling saturation pulse is placed superior to each excitation slice to eliminate the signal from the caudally directed venous ¯ow.15 The motion-induced dephasing (intravascular signal loss) is minimized through a combination of constant velocity ¯ow compensation gradients along the slice-select and frequency-encoding directions, the shortest possible echo times (8.0±9.0 ms with conventional gradients), and the smallest possible voxels (1.5±3.0 mm slice thickness and 1.0 mm inplane resolution). The major advantage of 2D TOF sequences relate to their sensitivity to slow ¯ow as the spins must only move a short distance (1.5±3.0 mm 2D slice) within each TR interval to ensure high intravascular signal in each slice. Consequently, these sequences are readily able to distinguish a severe stenosis from an occlusion. A disadvantage of the 2D technique is the stair-step misregistration artifact or vessel discontinuities in the ®nal MRA arteriogram images that arise due to gross patient motion during the acquisition of the individual slices. More importantly, the 2D sequences demand high gradient amplitudes to de®ne the thin 2D slices, which, in turn, place signi®cant restrictions on the minimum echo time and in-plane voxel dimensions. The high gradient amplitudes accentuate any spin dephasing due to motion which is not corrected by the ¯ow compensating gradients. As a result, severe stenoses are often seen as `¯ow voids' or vascular interruptions. In spite of this, 2D TOF MRA has been demonstrated to be a highly sensitive and very acceptable screening study.16 3D TOF sequences incorporate the same strategies for vessel visualization as their 2D counterparts, but the rf pulse excites a thick slab of tissue during each TR interval and the thin slices within the volume are de®ned by a second phaseencoding gradient along the slice-select direction. The examination time increases in proportion to the number of phaseencoding steps in the second direction (number of slices), but there is also a proportional increase in the S/N when contrasted to comparable size voxels in 2D sequences. As a result of the reduced gradient demands, the 3D sequences permit shorter echo times (4.5±7.0 ms) and higher spatial resolution (<1.0 mm3) with comparable refocusing pulses when contrasted to the 2D TOF sequences. These factors combine to minimize the degree of intravoxel phase dispersion and signal loss within the vessel.17 The phase errors increase quadratically with time in the setting of higher order motion terms so intravascular signal loss is especially likely within and immediately distal to a stenosis, within a focal tortuosity, and in some normal vessel

bifurcations. Since the greatest amount of dephasing occurs during the application of the gradients, the phase errors are most effectively suppressed by minimizing the duration, amplitude, and timing of the gradients.18 Additional ¯ow compensation gradients can be incorporated to correct for the higher order motion terms but only at the expense of signi®cantly stronger and longer gradients before the echo. The signal loss with the longer gradients more than offsets the minimal improvement derived from the supplementary motion compensation. Because these sequences are all gradient echo acquisitions, the intravascular signal is also reduced by T2* phenomena. This effect is similarly reduced with small voxels and short echo times.19 The rapid unidirectional ¯ow in the carotid arteries is particularly well suited to a 3D TOF acquisition. This technique also has its limitations which primarily relate to the saturation of ¯owing spins (e.g. slow antegrade ¯ow distal to a severe stenosis). Spatially variable rf pulses (lower ¯ip angle as the spins enter the volume and higher ¯ip angles distally in the volume) have proven particularly helpful in this regard. The vessel/soft tissue contrast is somewhat reduced at the caudal end of the volume, but the contrast increases superiorly and the spins can ¯ow further into the volume before becoming saturated without increasing the TR. Higher ¯ip angles at the cranial end also increase the saturation of the stationary tissues for a given TR. The advantages of this rf pulse are particularly noticeable with slow ¯ow such as in an elderly patient with poor cardiac output, a very small vessel lumen, or distal to a severe stenosis. Failure to adequately suppress the background stationary tissues is a limitation of any TOF technique, but particularly the 3D technique in view of the lower signal of the blood (relative to 2D TOF). Excitation with higher ¯ip angles improves the ¯ow-related enhancement but accentuates the problem of spin saturation in 3D acquisitions. Shorter repetition times enhance the background suppression while causing more saturation of the ¯owing spins. In the neck, fat is the most problematic, as its short T1 relaxation time and high spin density causes it to appear bright on the ®nal MRA images. Since there tends to be a large amount of perivascular and subcutaneous fat in older patients (the group most likely to be studied for carotid atherosclerotic disease), the fat can largely obscure the underlying vessels. Reconstructing a subvolume that is limited to the region immediately surrounding the vessels excludes much of the stationary tissue from the reconstruction but does not eliminate the problem. Incorporating additional fat suppression pulses into these sequences will reduce the signal of the fat without affecting the vessel signal and can signi®cantly improve the vessel/soft tissue contrast-to-noise level.20 In practice, these pulses can be unreliable in clinical studies since local ®eld inhomogeneities in the neck may cause inhomogeneous fat suppression or even water (blood) suppression. In light of the above arguments, a most promising clinical method is the stacked overlapping 3D volume technique, which combines many of the advantages of the 2D and 3D studies.21 Relatively small volumes minimize the problem of spin saturation while the lower gradient demands of the 3D sequences allow for very short echo times and small voxels. Moreover, the stacked volume approach can be implemented with higher ¯ip angles and/or shorter TR values than are possible with a large single 3D volume so the vessel/soft tissue contrast is

HEAD AND NECK STUDIES USING MRA

enhanced. If the stacked volumes incorporate the specialized rf pulses as discussed above, intermediate sized volumes can be used to improve the postprocessing, increase the anatomical coverage, and reduce the overall examination time. 1.4 Carotid MRA: Clinical ApplicationsÐCarotid Bifurcation A number of studies have evaluated the clinical ef®cacy of 2D TOF MRA for imaging the carotid bifurcations, the largest of which was a study by Laster et al. in which the images from 101 patients were compared with conventional catheter angiography and, in some cases, Doppler ultrasound.22,23 As in any MRA technique, the degree of lumenal narrowing tends to be accentuated compared with the conventional arteriogram images. Because of the limitations in echo time and voxel size with the 2D technique, the dephasing caused by higher order motion terms induces more prominent intravascular signal loss and therefore greater apparent narrowing of the lumen than with the 3D technique. In the case of a severe stenosis, there is frequently a ¯ow void or complete absence of signal in the involved segment so the morphology of the stenosis cannot be de®ned at all. The reader simply concludes that the vessel is severely stenotic because there is complete loss of signal in that segment with reappearance of the vessel distally. Whereas the MRA images agreed with the catheter studies in up to 75% of cases, overestimation of the stenosis was seen in up to 44%.24,25 As expected, the highest percentage of these errors was evident in those studies with relatively long echo times and large voxel dimensions. The most signi®cant error, however, was seen in those cases of moderate stenoses which were labeled as `severe' on MRA. This was a problem even with a slice thickness of 1.5 mm and echo times of approximately 9.0 ms.24 If MRA is used as the sole imaging study preoperatively, a carotid endarterectomy may be recommended inappropriately. It can be argued that Doppler ultrasound could be performed in conjunction with the MRA, but the ultrasound also has potential pitfalls such as operator dependence, machine variability, vessel tortuosity and plaque calci®cation leading to nondiagnostic studies, overestimation of stenoses, and (although less commonly with color ¯ow systems) misdiagnoses of complete occlusion at the bifurcation. Although it is dif®cult to depict the morphology of severe stenoses, the sensitivity of the 2D TOF technique to slow ¯ow generally makes it readily able to distinguish a severe stenosis from an occlusion if the slice thickness is suf®ciently small, the TR is suf®ciently long, and a long enough segment of the vessel is evaluated to include a segment in which the ¯ow has returned to a laminar pro®le.22 Nevertheless, in a few cases, MRA overestimates a critical stenosis as an occlusion which has important clinical implications as the patient is no longer considered a surgical candidate if the vessel is occluded. The 2D TOF studies have also been compared to Doppler ultrasound examinations of the carotid bifurcations. The only noticeable difference was that the MRA studies tended to be less likely to misdiagnose a severe stenosis as an occlusion, but there was ultimately no statistically signi®cant difference in accuracy between the two modalities.24,26 Both studies tended to overestimate the degree of carotid stenosis. Even when the

3

results of the two examinations are in agreement, they still do not necessarily correlate with the invasive arteriogram and may not have the accuracy to replace the conventional study.24±26 Other groups have primarily relied upon 3D TOF imaging in patient studies because of a lesser degree of ¯ow-related dephasing and, consequently, improved characterization of carotid stenoses. When single-volume 3D TOF MRA was compared to intra-arterial digital subtraction angiography (IADSA) using a caliper technique, ROC analysis showed that technically adequate 3D TOF MRA images can be interpreted consistently, show a very strong correlation with IADSA, and that 3D TOF MRA may be a sensitive screening examination for extracranial carotid stenoses.27 In spite of the advantages of the 3D sequence parameters, the more severe stenoses may still be accentuated due to higher order motion in the region of the stenosis. As alluded to above, another obvious limitation of the 3D technique is its potential to misrepresent a severe stenosis as an occlusion so 2D and 3D TOF carotid MRA investigations have been combined in patient examinations to avoid this pitfall.28 Even when the 2D and 3D TOF studies are combined, it seems unlikely at this time that MRA, duplex ultrasound, or the combination of the two can replace conventional angiography in patients who are being considered for surgery. It would be more appropriate for these modalities to serve as a screening examination, and, if a signi®cant stenosis is detected, an invasive diagnostic study can be performed for further evaluation prior to surgical intervention. Based on the guidelines set by the NASCET study, it is now not only necessary to measure accurately the lumenal narrowing but it is also necessary to exclude the presence of a tandem stenosis elsewhere in the carotid circulation.6 Moreover, it is important to measure the stenosis in a consistent fashion and to have the capability to distinguish different gradations of severe lumenal compromise since the NASCET study demonstrated a greater bene®t of the endarterectomies with increasing severity of the stenoses. While the 3D TOF technique characterizes the morphology of the stenotic segment more accurately than 2D studies, both tend to exaggerate the degree of narrowing and may show complete signal void within the stenosis (particularly in the 2D studies) (Figure 1). Further improvements should be realized with improved gradient capabilities which will permit shorter echo times, higher resolution, and reduced intravoxel dephasing at the site of severe stenoses.18,29 1.5

Carotid MRA: Clinical ApplicationsÐCarotid Dissection

Dissection of the carotid artery is not an infrequent cause of stroke but tends to occur in younger adults than strokes relating to atherosclerotic disease. If the clinical suspicion is low, a carotid dissection is only one possible explanation for the patient's clinical presentation, and/or there is a relative contraindication to conventional angiography. MRI in conjunction with MRA serves as a noninvasive alternative to evaluate for dissection and to identify simultaneously the extent of parenchymal sequelae. Catheter angiography, however, remains the gold standard, and is the most reliable means by which the underlying etiology can be identi®ed (e.g. ®bromuscular dysplasia). The conventional study may also be the only method

4 HEAD AND NECK STUDIES USING MRA

Figure 1 (a) Stacked overlapping volume acquisition of the carotid bifurcations shows no signi®cant saturation despite the tortuosity of the vessels and the proximal stenosis of the internal carotid artery. (b) The degree of stenosis (arrow) is less severe in the conventional arteriogram, likely due to the long gradient duration for the MRA acquisition

of fully appreciating the extent of collateral ®lling of the intracranial circulation if the alternative routes are circuitous with slow ¯ow in small vessels. Alternatively, it is also possible to have a false-negative catheter study if the vessel is completely occluded (non®lling) or the dissection is subadventitial (normal size and con®guration of the lumen) and the MRI/MRA study may be the best way to con®rm the diagnosis by demonstrating the perilumenal blood products.30 In routine T1- and T2weighted spin echo imaging the dissection is generally detected as a crescentic area of hyperintensity around a narrowed lumen of the cervical internal carotid artery.31 In particular, an axial T1-weighted spin echo study using fat saturation and an inferior spatial saturation pulse suppresses the signal of the perivascular fat and intraluminal ¯ow-related enhancement, thereby increasing the conspicuity of the hyperintense (short T1) periluminal hemorrhage.32 MRA is used to complement conventional spin echo imaging in the evaluation of patients with suspected carotid dissections.33 As in IADSA studies, the internal carotid artery may appear narrowed and there may be poor ¯ow-related enhancement proximally if the out¯ow through the internal carotid artery is severely impaired. The dissection is often imaged when the periluminal hemorrhage contains methemoglobin, so the vessel lumen can appear artifactually widened on the MRA since the hyperintense thrombus is similar in intensity to the

¯ow-related enhancement in the true lumen. Because of the variability in the state of evolution of the thrombus, the hemorrhagic by-products are either hyperintense or hypointense relative to the lumen. If the distinction between lumen and hemorrhage is not apparent on the projection arteriogram images, the two lumena and the intervening ¯ap are usually evident in the source images for the MRA. 1.6

Intracranial MRA: Technical Considerations

The intracranial circulation presents problems similar to those seen in the evaluation of the extracranial cerebral vasculature except the studies are further complicated by smaller vessels, ®eld inhomogeneities, and reduced through-plane ¯ow. The 3D sequences seem particularly well suited to this situation as they not only provide the high spatial resolution needed to visualize the smaller vessels but they also provide the reconstruction capabilities (maximum intensity projection) necessary to display the complex intracranial vascular geometry. Intracranially, local ®eld inhomogeneities exist at air±soft tissue and bone±soft tissue interfaces while higher order motion terms occur at focal vessel tortuosities, bifurcations, and lumenal stenoses. Both of those factors are limited by small voxels and short echo times. In comparison to the 2D sequences, the 3D acquisitions are better able to demonstrate the large component

HEAD AND NECK STUDIES USING MRA

of in-plane ¯ow intracranially (with appropriate adjustments in the TR and ¯ip angle to avoid saturation). The reduced ¯ow velocities and large component of inplane ¯ow are still somewhat limiting in the 3D acquisitions as these factors aggravate the problem of spin saturation, thereby reducing the vessel±soft tissue contrast particularly in the distal volume. Adjustments in the conventional imaging parameters provide only limited bene®t. The incorporation of magnetization transfer saturation in all intracranial TOF MRA studies has become routine.34 Since the off-resonance saturation pulse suppresses the signal of the bound water molecules, and the bound component is largely restricted to the brain parenchyma, the magnetization transfer saturation effectively enhances the vessel±soft tissue contrast. Additional vascular contrast may be obtained by incorporating the spatially varying rf pulses for excitation designed to limit the problem of spin saturation. Because of the slower ¯ow rates in the intracranial arteries, the most promising TOF technique seems to be the stacked volume approach, which combines the advantages of 2D and 3D TOF acquisitions.21 Phase contrast sequences are also commonly used to evaluate the intracranial circulation. The vessel geometry and small size of the arteries make the 3D acquisition necessary for most purposes but, for comparable spatial resolution and ¯ow sensitization along all three planes, the acquisition time is longer and the postprocessing more complex for the 3D PC technique. Alternatively, 2D acquisitions may provide similar ¯ow information relatively quickly, but at the expense of limited anatomical coverage. A problem common to both phase contrast techniques is the need for some a priori knowledge of the range of velocities present within a given patient or lesion. Inappropriate choice of the velocity-encoding gradient can cause poor visualization of the small and/or peripheral vessels with slow ¯ow if the chosen velocity-encoding gradient is too weak, or aliasing of ¯ow information and signal loss if the chosen velocity range is too low. On the other hand, the freedom to choose different velocity encodings provides options that are not possible or more cumbersome in a TOF acquisition such as the direction of blood ¯ow (e.g. collateral ¯ow in the circle of Willis).35,36 Alternatively, directional information is easily attainable in the PC acquisition without performing a separate study. The PC studies also readily provide the capability to measure ¯ow velocities and volume ¯ow rates of individual vessels. Lastly, if 2D PC studies are performed in a cardiac gated cine mode, the ¯ow dynamics in various vascular lesions can be studied at different, speci®ed velocity ranges which is only possible in a very limited fashion with TOF techniques. 1.6.1

Intracranial MRA: Aneurysms

The statistics relative to subarachnoid hemorrhage from a ruptured intracranial aneurysm remain devastatingÐgreater than 50% mortality and major morbidity.37 Given the low operative morbidity and mortality of surgery on unruptured aneurysms, it is appealing to consider the possibility of using MRA as a screening test. It would be especially applicable in patient groups where there is an increased incidence or a high anxiety level for an aneurysm, such as patients with a strong family history, polycystic kidney disease, aortic coarctation, ®bromuscular disease, and collagen vascular disease. Based on the decision analysis of Levy et al., the risk of an aneurysm

5

even in these groups does not warrant the screening of all these patients with conventional arteriography.38 It is possible, however, to perform an accurate noninvasive pretest that would increase the prevalence in subgroups which could then be studied by conventional arteriography for further evaluation. Based on the results of MRA in the evaluation of intracranial aneurysms, MRA exceeds the criteria of sensitivity and speci®city for the noninvasive pretest outlined by Levy et al.37,39 It is possible to use high resolution contrast enhanced computerized tomography (CT) to study these patients, but the sensitivity and speci®city have not been studied, and the requirement of intravenous iodinated contrast and the beam hardening artifact from the skull base are potentially limiting, which make MRA preferable.40 Currently, TOF MRA is frequently used as a screening test in these groups or in stable patients with a subacute onset of symptoms in which an intracranial aneurysm is only one of a number of potential explanations for the clinical presentation and the suspicion of an aneurysm is not high enough to warrant a conventional arteriogram (Figure 2). However, the patients who present acutely with a subarachnoid hemorrhage are best served by a nonenhanced head CT followed by conventional catheter angiography. Retrospective studies have been conducted to test the accuracy of the 3D TOF MRA method for the detection of intracranial aneurysms as compared to IADSA examinations.39,41,42 Aneurysms as small as 3±4 mm have been detected using 3D TOF and 3D PC techniques. In a series of 21 angiographically con®rmed aneurysms in 19 patients, the authors demonstrated an increase in sensitivity from 67% when evaluating the 3D TOF MRA projection images alone, to 86% when the MRA arteriogram images were evaluated along with the source images from the 3D data set and the spin echo images. Using the MIP images, the individual slices, and the spin echo study, the sensitivity was calculated to be 95% for the detection of at least one aneurysm in a patient such that the study would lead to the referral for a conventional arteriogram for further evaluation preoperatively. Recognizing the limitations of these MRA acquisitions becomes particularly important when choosing the method of postprocessing, interpreting the images, and recommending a conventional arteriogram as a follow-up study or in place of MRA. Larger aneurysms or aneurysms compressing the parent vessel may reduce the extent of in¯ow and washout of fresh unsaturated spins within the aneurysm lumen, which causes an apparent reduction of the size of the aneurysm on TOF studies.41 This is less of a problem with a PC acquisition if the preliminary spin echo study reveals a giant intracranial aneurysm, so that a low velocity encoding range can be selected for PC MRA.42 On the other hand, these lesions are also obvious on the spin echo study owing to the prominent ¯ow artifacts, intimate relationship with the adjacent basilar vessels, and (if present) the contained blood products. The stagnant ¯ow within a large aneurysm tends to promote thrombus formation which partially ®lls the aneurysm lumen. Since the TOF sequences are inherently T1 and spin density weighted, both the thrombus and the patent lumen may be visible on the MIP images, depending on the state of evolution of the thrombus just as in conventional IADSA. This will cause an underestimation of the overall aneurysm size in PC studies because of the strong background suppression. In the case of

6 HEAD AND NECK STUDIES USING MRA subarachnoid hemorrhage, the aneurysm and the parent vessel may not be observed altogether. Although the gross relationships with the parent vessel may be visible, identi®cation of the aneurysm neck may be dif®cult at times when evaluating the projection images and occasionally with conventional cerebral angiography. In these cases, evaluation of the 3D TOF source images and multiplanar reconstructions of the original slices create `angiotomograms' to delineate more clearly both the aneurysm and its point of origin. Alternatively, reconstructing vessels from small subvolumes that contain only the vessels of interest provides images with better de®nition (reduces the artifact inherent to the MIP algorithm) and eliminates the problems relating to vessel overlap.5 This is especially helpful with aneurysms arising from a tortuous carotid siphon or in determining which branch the aneurysm arises from if it occurs at a bifurcation (e.g. middle cerebral artery bifurcation/trifurcation). More importantly, routine evaluation of the source images and planar reconstructions of the axial slices reduces the likelihood of missing small aneurysms that may otherwise be missed by reviewing only the projection MRA images. Identifying the relationship of the aneurysm to the parent vessel may also be dif®cult if the aneurysm arises from an arterial segment that is incompletely visualized due to motion induced dephasing. This is most commonly seen in the carotid siphon, particularly if the supraclinoid segment is immediately apposed to the juxtasellar segment so the turn of the siphon is rather tight. Because of a greater tendency towards motion induced dephasing, this should be more problematic with PC than TOF acquisitions. Shorter gradient times and smaller voxels signi®cantly reduce the phase dispersion and signal loss in these vessels.19±21

MRA can also be applied to the long term postoperative follow-up of aneurysm patients treated via an endovascular route (e.g. balloon occlusion) or noninvasive follow-up of a known, untreated aneurysm (Figure 2). Susceptibility artifacts from the nonferromagnetic aneurysm clips preclude the evaluation of aneurysms treated by surgical clipping but would not prohibit the use of MRA to evaluate for intracranial aneurysms elsewhere. 1.6.2

Intracranial MRA: Arterial Occlusive Disease

Patients with symptoms of cerebral ischemia are frequently examined by MRI to assess the extent of parenchymal ischemic disease. The morphology and signal characteristics of the vessel lumena can only suggest stenosis and slow ¯ow or occlusion, whereas MRA can directly identify the site and severity of the underlying stenosis. Not infrequently, the patient has already had a Doppler ultrasound study of the carotid bifurcations, and the ischemic symptoms are not explained by the mild narrowing in the carotid bifurcations. MRA provides an additional noninvasive study which can be performed at the same time as the spin echo study to evaluate the morphology of the intracranial vessels. Alternatively, in a patient with severe stenosis, MRA provides the means by which the physician can evaluate the degree of collateral ¯ow to the circulation that is severely compromised in the extracranial circulation. As compared with transcranial Doppler ultrasound studies, MRA is better able to evaluate the posterior fossa circulation and provides an anatomical display of the intracranial vasculature that is not possible with the transcranial Doppler technique.43,44 The MRA ®ndings can alert the clinician to the nature and location of the occlusive disease and a conventional

Figure 2 (a) This 55-year-old woman is known to have a basilar tip aneurysm and is presenting for noninvasive follow-up. The aneurysm is seen as a rounded hypointense focus (arrow) in the interpeduncular cistern. (b) The 3D TOF MRA image con®rms the aneurysms (arrow), including its size and orientation with respect to the parent vessel

HEAD AND NECK STUDIES USING MRA

angiographic study could then be performed to con®rm the full extent of pathology prior to starting therapy if the MR/MRA ®ndings do not explain the clinical situation or if surgical or endovascular therapy is being considered. Alternatively, a negative study might avert the need for an invasive diagnostic examination altogether. The hope for the future is that MRA can effectively evaluate for tandem stenoses in the distal internal carotid arteries in patients being considered for carotid endarterectomies and, in combination with a similar noninvasive study of the carotid bifurcations, can obviate the need for an invasive study preoperatively (i.e. noninvasively meet the NASCET criteria). Initial studies have been restricted to the evaluation of the larger vessels such as the carotid siphon, the proximal anterior and middle cerebral arteries, the distal vertebral arteries, and the basilar trunk. This is, in part, related to the spatial resolution limitations of these sequences and the small vessel caliber of the intracranial vasculature. Narrowing of the distal peripheral vessels, typical of cerebral vasculitis, cannot be effectively evaluated with MRA, and requires a conventional arteriogram. Flow rates also impose restrictions in the TOF acquisitions as only the larger arteries of the circle of Willis will have ¯ow which is rapid enough that the ¯ow-related enhancement will likely `opacify' the vessel distal to the stenosis. This is not as problematic with the PC sequences if the physician has a priori knowledge about the expected range of ¯ow velocities in the vessels. 3D TOF MRA is readily able to display normal vessels and to distinguish normal from stenotic vessels. In one study, the sensitivity for identifying stenoses in the distal vertebral and basilar arteries was shown to be 100% whereas the speci®city of the study was not nearly as high.44 When 3D TOF MRA was compared with the IADSA studies, the degree of stenosis was exaggerated in as many as 63% of the MRA images. When correlated with the results of other clinical studies, this error was likely accentuated to some extent by the chosen voxel dimensions, the small but highly variable caliber of the distal vertebral arteries, and the separation of the stenoses into ®ve categories.44 Reviewing the original 3D source images was found to be helpful for distinguishing occlusion from slow ¯ow in a severely stenotic or hypoplastic vessel. This distinction was further aided by incorporating selected 2D TOF slices because of their improved sensitivity to slow ¯ow. Knowledge of the degree of collateral ¯ow between different vascular territories may have a signi®cant impact on the physician's therapeutic decisions. Often times, the method of cross ®lling is obvious on the standard TOF MRA acquisition. For example, if one internal carotid artery is occluded but there is still good ¯ow to the corresponding anterior circulation, the ¯ow is more likely through the patient's large anterior communicating artery than the diminutive posterior communicating arteries. It is not as easy to draw these conclusions when the potential collateral vessels are more uniform in size or they are all diminutive. The slow ¯ow and small size of leptomeningeal collateral vessels prevents their visualization. Evaluation of the source images from PC studies clearly demonstrates the direction of ¯ow in major vascular trunks, making the pattern of collateral ¯ow obvious.36 Moreover, PC studies also provide the potential to make ¯ow velocity and ¯ow volume measurements. Alternatively, examining the corresponding phase images from a TOF study can also con®rm the direction of

7

blood ¯ow in main vascular trunks. The relative contributions from the other vascular territories can also be identi®ed in a TOF study through the use of appropriately positioned saturation pulses.35 A narrow saturation slab, which is perpendicular to a 2D TOF slice or narrow 3D TOF volume, can be oriented to eliminate the ¯ow-related enhancement from any one or any combination of the main vascular trunks to identify the source of collateral ¯ow to a given vascular distribution. For example, selectively saturating the contralateral internal carotid artery will demonstrate the relative contribution from the posterior circulation to the remaining internal carotid artery distribution in question. 1.6.3

Intracranial MRA: Vascular Malformations

Intracranial vascular malformations are generally easy to identify on the routine spin echo study of the head. Capillary telangiectasias are usually discovered incidentally at autopsy while cavernous angiomas are most commonly detected as subcortical foci of mottled hyperintensity on the T1- and T2weighted sequences with a peripheral rim of ferritin and hemosiderin deposition on the T2 images from prior hemorrhage.45 In both cases, the ¯ow through these lesions is so slow that their abnormal vessels cannot be visualized by any MRA technique. (In fact, TOF MRA images can be misleading since the short T1 of hemorrhage in a cavernous angioma can mimic the high ¯ow vascular nidus of an arteriovenous malformation.) Venous angiomas are also readily identi®ed on the spin echo study as a caput of veins draining into a large medullary vein extending from a point near the ependymal surface to the cortex. While the sensitivity to slow ¯ow in the 2D TOF or PC studies would make these veins visible with MRA, their characteristic spin echo appearance makes this unnecessary. High ¯ow arteriovenous malformations are also easily diagnosed on the conventional spin echo parenchymal study. The vascular nidus appears as a cluster of serpiginous ¯ow voids because of the relatively fast ¯ow through these vessels. The factors that are considered important predictors of surgical resectability are also readily appreciated, including the size of the nidus, its relationship to eloquent areas of the brain, and the presence of deep venous drainage.46 If there is a direct arteriovenous ®stula, there will be no intervening nidus but the enlarged feeding arteries and draining veins will still be easily identi®ed. There may also be associated parenchymal ®ndings related to hemorrhage, ischemic steal, secondary venous thrombosis, or prior surgical, embolization and/or radiation therapy. The arteriovenous malformations can be identi®ed with TOF or PC MRA, but the images generally do little more than reinforce the 3D spatial relationships of the vascular nidus with the circle of Willis and identify the main feeding arteries and draining veins. Conversely, one situation in which MRA is essential in the work-up is in the evaluation of suspected dural vascular malformations which make up approximately 15% of the arteriovenous malformations. Frequently, the dural malformations are not visible at all on the spin echo study because of the small size and lack of contrast between the ¯ow voids and the adjacent hypointense skull base.47 The only clues to the presence of the malformation may be the clinical history, the presence of enlarged draining cortical veins, or secondary ®ndings such as dural venous thrombosis, subdural or parenchymal

8 HEAD AND NECK STUDIES USING MRA hematomas, and venous infarcts. On occasion, subtle peripheral ¯ow voids or focal prominence of a dural venous sinus may be present, but these are often identi®ed only in retrospect, once the MRA study makes the diagnosis obvious. MRA is often able to identify the vascular nidus along the dura, enlarged branches of the external carotid and/or vertebral arteries, and possibly enlarged draining veins or dural sinuses if there is a direct arteriovenous ®stula present. It is essential to evaluate the source images from the MRA acquisition, as the ®ndings may even be subtle on the projection MRA. Intravenous gadolinium can be used with 3D TOF MRA to evaluate better small feeding arteries and draining veins, but a study of dural ®stulas failed to identify venous constriction. A study of dural ®stulas with 3D TOF MRA was unable to identify venous constriction when gadolinium was used to reduce the problems of spin saturation and better identify the slow ¯ow in the small feeding arteries and draining veins.48 Ultimately, conventional catheter angiography will be performed on any patient with an arteriovenous malformation to characterize it prior to therapy. Only the invasive arteriogram provides the spatial resolution necessary to identify all of the afferent and efferent vessels. While the MRA acquisition can identify aneurysms along the afferent vessels, these too should be further evaluated by a conventional arteriogram preoperatively. Lastly, only the catheter study provides the temporal resolution to assess the arteriovenous circulation time, to identify individual vessels with the greatest degree of arteriovenous shunting and, in the case of a direct ®stula, to localize the point of communication between the arterial and venous systems. The MRA acquisitions can be designed to provide a limited degree of physiological information about the vascular malformations. Thin saturation slabs can be applied to a 2D or thin volume 3D TOF acquisition to eliminate the ¯ow-related enhancement from one vascular trunk and monitor the relative intensity and/or size of the malformation in the ®nal image. In this way, it is possible to determine if the malformation is fed by more than one vascular distribution (collateral ¯ow) and to assess the relative contributions from the different vascular territories.35 The PC acquisitions can also be designed to supply similar information by applying a series of different velocity-encoding gradients in a stepwise fashion to demonstrate different components of the malformation.35 A low velocity-encoding range (e.g. 10±20 cm sÿ1) would demonstrate draining veins with low to medium velocities whereas higher velocity-encoding ranges (e.g. 60±100 cm sÿ1) would preferentially demonstrate the nidus, larger veins, and feeding arteries with more rapid ¯ow.42 In this fashion, it is possible to provide a velocity ¯ow map of the vascular malformation. Such information may be clinically useful if serial studies indicate a change in ¯ow dynamics which may be predictive of hemorrhage, but this remains to be tested. In spite of the advantages, saturation of ¯owing spins is perhaps the most signi®cant limitation of the 3D TOF technique. This limits visualization of the small feeding arteries (slower ¯ow) distally in the volume, all but the largest early draining veins, and potentially part of the vascular nidus itself if there is relatively slow ¯ow through the nidus and it is located distally in the volume. The 2D TOF studies are less susceptible to spin saturation and provide better contrast-to-noise levels across the imaging volume. Alternatively, 2D TOF images are more lim-

ited by ¯ow-induced dephasing and may not visualize portions of the afferent and efferent vessels depending on the direction along which the slices are acquired relative to the course of these vessels. The problem of spin saturation is reduced with intravenous gadolinium, but this also increases the intensity of the background tissues, which may be quite limiting if the malformation is situated along the skull base (e.g. many dural vascular malformations). The best compromise between the TOF techniques is the stacked overlapping volume approach.21 If the malformation is located within the parenchyma, a paramagnetic contrast medium could be added to improve venous visualization, while the stacked volume approach improves arterial visualization.49 1.6.4

Intracranial MRA: Venous Sinus Thrombosis/Occlusion

Dural venous sinus thrombosis and secondary parenchymal ®ndings (e.g. edema, hemorrhagic venous infarct) can be detected on conventional spin echo images although the ®ndings are not always obvious and may be misleading. The most sensitive spin echo study to evaluate the venous sinuses is a T2 weighted sequence in which the slices are oriented perpendicular to the direction of ¯ow in the sinus in question (e.g. coronal for the superior sagittal and straight sinuses). With such a long echo time, blood ¯owing perpendicular to the slice normally causes a ¯ow void as it is least apt to experience the 90 and 180 pulses and remain in the imaging plane at readout unless the sinus is thrombosed or occluded. An occluded sinus will fail to demonstrate a ¯ow void or may be frankly hyperintense on the T1- and T2-weighted images. Slow ¯ow may also cause an intermediate to high signal within the sinus and simulate thrombus. In the case of acute/early subacute thrombus, the hypointense thrombus may simulate a ¯ow void on a T2weighted image.50 In the situations where the spin echo ®ndings of thrombosis are ambiguous or an adjacent mass appears to have invaded and occluded the sinus, MRA venograms can be most helpful in con®rming or excluding thrombosis/occlusion without resorting to an invasive study. A 2D technique would be the most sensitive TOF study to evaluate the slow ¯ow in the dural sinuses.51 Coronal slices acquired sequentially, from posterior to anterior, should ensure ¯ow-related enhancement in the major dural sinuses since the ¯ow is largely in the reverse direction except the posterior portion of the superior sagittal sinus. This same strategy can be followed with fewer slices and a shorter examination time if the slices were acquired in an oblique coronal plane (e.g. rotated 75 towards the sagittal plane). In view of the relatively slow normal velocities in these veins, the slice thickness for these 2D slices should be minimized, the TR maintained relatively long, and a low-to-intermediate ¯ip angle should be used (e.g. 30 ) to avoid spin saturation. Hyperintense thrombus could simulate ¯ow-related enhancement, but this is easily distinguished when correlated with the ®ndings on the spin echo study. The prior administration of intravenous gadolinium would make it dif®cult to distinguish a normal sinus from contrast enhancement along the periphery of a thrombosed sinus. A more important potential error relates to the limitations of the technique. There may be insuf®cient ¯ow-related enhancement in a small sinus with slow ¯ow, leading to a falsepositive study. Similarly, an oblique coronal acquisition will insure ¯ow-related enhancement in one transverse sinus, but

HEAD AND NECK STUDIES USING MRA

this is not necessarily the case in the opposite transverse sinus. In either case, evaluating the individual slices or reconstructing a small subvolume eliminates the reconstruction artifact and usually makes it possible to distinguish patency from occlusion. Phase contrast techniques are especially well suited to this application because of strong background suppression and the ability to adjust the sequence so it is sensitive to very slow ¯ow (10±20 cm sÿ1).52,53 While the sensitivity of the 2D TOF study is suf®cient for most routine clinical work, the sensitivity is somewhat greater with PC, and it may be possible to detect ¯ow in a recanalized sinus when the TOF images would suggest occlusion. The relatively large size of these sinuses, laminar ¯ow pro®les, and predominantly unidirectional ¯ow make it easy to study these veins with two quick 2D PC acquisitions. A thick 2D slab can be selected in the midline sagittal plane, with velocity sensitization along the anterior/posterior direction, to assess the patency of the deep and super®cial midline veins. A second thick axial slab in the posterior fossa with sensitization in a similar direction will evaluate the patency of the transverse and sigmoid sinuses.

2 RELATED ARTICLES Abdominal MRA; Peripheral Vasculature MRA; Phase Contrast MRA; Time-of-Flight Method of MRA; Whole Body Magnetic Resonance Angiography.

3 REFERENCES 1. C. L. Dumoulin, Neuroimaging Clin. N. Am., 1992, 2, 657. 2. G. A. Laub and W. A. Kaiser, J. Comput. Assist. Tomogr., 1988, 12, 377. 3. C. L. Dumoulin, S. P. Souza, M. F. Walker, and W. Wagle, Magn. Reson. Med., 1989, 9, 139. 4. G. W. Lenz, E. M. Haacke, T. J. Masaryk, and G. A. Laub, Radiology, 1988, 166, 875. 5. C. M. Anderson, D. Saloner, J. S. Tsuruda, L. G. Shapeero, and R. E. Lee, Am. J. Radiol., 1990, 154, 623. 6. The North American Symptomatic Carotid Endarterectomy Trial (NASCET) Steering Committee, Stroke, 1991, 22, 711. 7. V. J. Howard, J. Grizzle, H. C. Diener, R. W. Hobson, M. R. Mayberg, and J. F. Toole, Stroke, 1992, 23, 583. 8. European Carotid Surgery Trialists' Collaborative Group, Lancet, 1991, 337, 1235. 9. Editorial, Lancet, 1991, 337, 1255. 10. North American Symptomatic Carotid Endarterectomy Trial Investigators, Stroke, 1991, 22, 816. 11. J. J. Schuler, P. Flanigan, L. T. Lim, T. Keifer, L. R. Williams, and A. J. Behrend, Surgery, 1982, 92, 1058. 12. D. J. Marzewski, A. J. Furlan, P. St. Louis, J. R. Little, M. T. Modic, and G. Williams, Stroke, 1982, 13, 821. 13. T. J. Masaryk, M. T. Modic, J. S. Ross, P. M. Ruggieri, G. A. Laub, G. W. Lenz, E. M. Haacke, W. R. Selman, M. Wiznitzer, and S. I. Harik. Radiology, 1989, 171, 793. 14. J. Huston, D. A. Rufenacht, R. L. Ehman, and D. O. Wiebers, Radiology, 1991, 181, 721. 15. P. J. Keller, B. P. Drayer, E. K. Fram, K. D. Williams, C. L. Dumoulin, and S. P. Souza, Radiology, 1989, 173, 527. 16. J. E. Heiserman, B. P. Drayer, E. K. Fram, P. J. Keller, C. R. Bird, J. A. Hodak, and R. A. Flom, Radiology, 1992, 182, 761.

9

17. G. W. Lenz, E. M. Haacke, T. J. Masaryk, and G. A. Lamb, Radiology, 1988, 166, 875. 18. A. J. Evans, D. B. Richardson, R. Tien, J. R. MacFall, L. W. Hedlund, E. R. Heinz, O. Boyko, and H. D. Sostman, Am. J. Neuroradiol., 1993, 14, 721. 19. E. M. Haacke, J. A. Tkach, and T. B. Parrish, Radiology, 1989, 170, 457. 20. J. A. Tkach, P. M. Ruggieri, J. S. Ross, M. T. Modic, J. J. Dillinger, and T. J. Masaryk, J. Magn. Res. Imaging, 1993, 3, 811. 21. D. D. Blatter, D. L. Parker, S. S. Ahn, A. L. Bahr, R. O. Robison, R. B. Schwartz, F. A. Jolesz, and R. S. Boyer, Radiology, 1992, 183, 379. 22. R. E. Laster Jr., J. D. Acker, H. H. Halford III, and T. C. Nauert, Am. J. Neuroradiol., 1993, 14, 681. 23. B. C. Bowen, R. M. Quencer, P. Margosian, and P. M. Pattany, Am. J. Roentgenol., 1994, 162, 9. 24. R. L. Mittl, M. Broderick, J. P. Carpenter, H. I. Goldberg, J. Listerud, M. M. Mishkin, H. D. Berkowitz, and S. W. Atlas, Stroke, 1994, 25, 4. 25. T. S. Riles, E. M. Eidelman, A. W. Litt, R. S. Pinto, F. Oldford, and G. W. S. Schwartzenberg, Stroke, 1992, 23, 341. 26. J. Huston III, B. D. Lewis, D. O. Wiebers, F. B. Meyer, S. J. Riederer, and A. L. Weaver, Radiology, 1993, 186, 339. 27. T. J. Masaryk, J. S. Ross, M. T. Modic, G. W. Lenz, and E. M. Haacke, Radiology, 1988, 166, 461. 28. C. M. Anderson, D. Saloner, R. E. Lee, V. J. Griswold, L. G. Shapeero, J. H. Rapp, S. Nagarkar, X. Pan, and G. A. Gooding, Am. J. Neuroradiol., 1992, 13, 989. 29. J. A. Tkach, P. M. Ruggieri, J. J. Dillinger, J. S. Ross, M. T. Modic, and T. J. Masaryk, J. Magn. Reson. Imaging, 1993, 3, 365. 30. M. Assaf, P. J. Sweeney, G. Kosmorsky, and T. J. Masaryk, Can. J. Neurol. Sci., 1993, 20, 62. 31. H. I. Goldberg, R. I. Grossman, J. M. Gomori, A. K. Asbury, L. T. Bilaniuk, and R. A. Zimmerman, Radiology, 1986, 158, 157. 32. R. Pacini, J. Simon, L. Ketonen, D. Kido, and K. Kieburtz, Am. J. Neuroradiol., 1991, 12, 360. 33. C. Levy, J. P. Laissy, V. Raveau, P. Amarenco, V. Servois, M. G. Bousser, and J. M. Tubiana, Radiology, 1994, 190, 97. 34. R. R. Edelman S. S. Ahn, D. Chien, W. Li, A. Goldmann, M. Mantello, J. Kramer, and J. Klee®eld, Radiology, 1992, 184, 395. 35. R. R. Edelman, H. P. Mattle, G. V. O'Reilly, K. U. Wentz, C. Liu, and B. Zhao, Stroke, 1990, 21, 56. 36. P. Turski and F. Korosec. Neuroimaging Clin. N. Am., 1992, 2, 785. 37. T. J. Ingall, J. P. Whisnant, D. O. Wiebers, and W. M. O'Fallon, Stroke, 1989, 20, 718. 38. A. S. Levey, S. G. Pauker, and J. P. Kassirer, N. Engl. J. Med., 1983, 308, 986. 39. J. S. Ross, T. J. Masaryk, M. T. Modic, P. M. Ruggieri, E. M. Haacke, and W. R. Selman, Am. J. Neuroradiol., 1990, 11, 449. 40. A. B. Chapman, D. Rubinstein, R. Hughes, J. C. Stears, M. P. Earnest, A. M. Johnson, P. A. Gabow, and W. D. Kaehny. N. Engl. J. Med., 1992, 327, 916. 41. R. J. Sevick, J. S. Tsuruda, and P. Schmalbrock, J. Comput. Assist. Tomogr., 1990, 14, 874. 42. J. Huston, D. A. Rufenacht, R. L. Ehman, and D. O. Wiebers, Radiology, 1991, 181, 721. 43. J. E. Heiserman, B. P. Drayer, P. J. Keller, and E. K. Fram, Radiology, 1992, 185, 667. 44. K. U. Wentz, J. Rother, A. Schwartz, H. P. Mattle, R. Suchalla and R. R. Edelman, Radiology, 1994, 190, 105. 45. S. Atlas, Radiol. Clin. N. Am., 1988, 26, 821. 46. R. F. Spetzler, and N. A. Martin, J. Neurosurg., 1986, 65, 476. 47. J. K. DeMarco, W. P. Dillon, V. V. Halbach, and J. S. Tsuruda. Radiology, 1990, 175, 193.

10 HEAD AND NECK STUDIES USING MRA 48. J. Chen, J. S. Tsuruda, and V. V. Halbach, Radiology, 1992, 183, 265. 49. G. Marchal, H. Bosmans, L. Van Fraeyenhoven, G. Wilms, P. Van Hecke, C. Plets, and A. L. Baert, Radiology, 1990, 175, 443. 50. G. Sze, B. Simmons, G. Krol, R. Walker, R. D. Zimmerman, and M. D. Deck, Am. J. Neuroradiol., 1988, 9, 679. 51. H. P. Mattle, K. U. Wentz, R. R. Edelman, B. Wallner, J. P. Finn, P. Barnes, D. J. Atkinson, J. Klee®eld, and H. M. Hoogewoud, Radiology, 1991, 178, 453. 52. D. J. Rippe, O. B. Boyko, C. E. Spritzer, W. J. Meisler, C. L. Dumoulin, S. P. Souza, and E. R. Heinz, Am. J. Neuroradiol., 1990, 11, 199. 53. J. S. Tsuruda, A. Shimakawa, N. J. Pelc, and D. Saloner, Am. J. Neuroradiol., 1991, 12, 481.

Biographical Sketches Paul M. Ruggieri. b 1956. B.S., 1978, M.S., 1979, Bucknell University, Lewisburg, PA, USA. M.D., 1984, UMDNJ-Rutgers Medical School, Piscataway, NJ, USA. Radiology residency at University Hos-

pitals of Cleveland, Neuroradiology fellowship at Cleveland Clinic Foundation including pediatric neuroradiology at Cincinnati Children's Hospital, research fellowship with Siemens Medical Engineering Group in Erlangen, Germany (supervisor Gerhard Laub). 1990±present, Associate Staff, Sections of Neuroradiology and Pediatric Radiology, Cleveland Clinic Foundation, Cleveland, OH, USA. Approx. 25 publications. Current specialties: neuroradiology, pediatric neuroradiology, MRI ¯ow imaging. Jean A. Tkach. b 1961. B.S.E., 1982, M.S., 1985, Ph.D., 1988 (Biomedical Engineering), Case Western Reserve University, Cleveland, Ohio, USA. 1990±present, Head of MR Imaging Research at the Cleveland Clinic Foundation, Cleveland, Ohio, USA. Approx. 30 publications. Current research specialties: MRA, functional MRI, cardiac MRI, ¯ow quanti®cation, magnetization transfer saturation MRI and steady state MRI techniques. Thomas J. Masaryk. b 1955. B.S., 1978, M.D., 1981, Medical College of Ohio, Toledo, Ohio, USA. Radiology residency at the Cleveland Clinic Foundation, Neuroradiology fellowship at the Cleveland Clinic Foundation. 1990±present, Head of Neuroradiology, the Cleveland Clinic Foundation, Cleveland, Ohio, USA. Approx. 92 publications. Current specialties: neuroradiology, interventional neuroradiology, MRI ¯ow imaging.

PERIPHERAL VASCULATURE MRA

Peripheral Vasculature MRA Cathy Maldjian Mount Sinai Medical Center, New York, NY, USA

and Mitchell D. Schnall University of Pennsylvania Medical Center, Philadelphia, PA, USA

1 INTRODUCTION While conventional angiography has been the `gold standard' in the evaluation of peripheral vascular disease, recent technical advances have made magnetic resonance angiography (MRA) a favorable alternative in many instances. Multiplanar imaging capabilities and noninvasive acquisition of data provide distinct advantages over contrast angiography. Overnight hospital stays, and potential morbidity from contrast injection and from obtaining arterial access may be avoided with MRA. Furthermore, distal runoff may be obscured by inadequate contrast ®lling in up to 70% of cases.1±4 This has mainly been attributed to contrast dilution in intramuscular collaterals before reaching the reconstituted vessels of interest. MRA allows cross-sectional imaging, which facilitates the evaluation of the vessel lumen for stenotic lesions. Impaired detection of stenotic lesions due to interference from overlying bone is not a problem, although it may be in contrast angiography. Quantitation of ¯ow, in particular by cine-phase contrast, is also possible with MRA. Three-dimensional data sets may be obtained by MRA. Gravimetric layering of contrast, which may prevent ®lling in anterior branches of slow-¯owing vessels, is obviated with MRA. Pulsatility artifacts tend not be a problem in studying collateral reconstitution past a high level stenotic lesion, since these vessels often adopt a monophasic rather than a triphasic wave pattern. However, with the advent of digital subtraction angiography, a higher level of technical pro®ciency and accuracy can be achieved. Inplane resolution is superior with cut-®lm angiography compared with MRA. Skeletal bony anatomy, sequential demonstration of retrograde ¯ow, metal artifacts, and total length of vascular anatomy, may be better appreciated with contrast angiography. Also, contraindications to MR, such as pacemakers, de®brillators, and metallic prosthetic valves, prohibit its use in these instances. Various artifacts of MRA will be discussed later in further detail. Bearing all this in mind, MRA may play an integral part in the presurgical evaluation of lower-extremity ischemia by providing the surgeon with the arterial anatomy in a noninvasive manner. It may afford decreased intraoperative times. Stenosis, occlusions, and native target out¯ow vessels favorable for bypass procedures may be identi®ed noninvasively. In addition, reconstituted vessels may potentially be seen to better advantage.

2

1

TECHNIQUES

Currently, the predominant technique for imaging the peripheral vasculature consists of two-dimensional (2D) time-of¯ight (TOF). Single sections are imaged perpendicular to the plane of ¯ow. Short TR is utilized to saturate stationary spins, which imparts high intensity to ¯owing blood where spins are unsaturated. This ¯ow-related enhancement allows visualization of the peripheral vasculature [Figure 1(a)]. Further selection of the arterial or venous system can be accomplished by utilizing a saturation band [Figure 1(b)]. A saturation band positioned distally to the imaging section would provide an arterial study. This relationship would be reversed in the head and neck, where venous blood generally travels inferiorly rather than superiorly and arterial blood courses superiorly rather than inferiorly. During arterial diastole, there is a component of ¯ow reversal which allows arterial blood to enter the presaturation slab during systole and subsequently the imaging section during diastolic reversal of ¯ow if the slab is positioned close enough to the imaging section. This creates undesirable dark, horizontal, stripe artifacts. For this reason, a saturation gap is often utilized. A saturation gap is a distance placed between the saturation pulse and the

Figure 1 Single slice from a 2D TOF angiogram saturation bands demonstrating both arteries (larger (smaller arrows); (b) with inferior saturation. The have been saturated out and only arterial structures

taken: (a) without arrows) and veins venous structures remain

2 PERIPHERAL VASCULATURE MRA

Figure 2 MIP projection obtained from a 2D TOF arteriogram in the region of the calf, demonstrating high-signal arterial vessels (large arrow) and lower signal venous vessels (small arrow). The veins are visible due to the wide saturation gap (2 cm) used in this case

imaging plane. The larger the gap, the less likely that retrograde arterial ¯ow from the saturation slab will reach the imaging section. On the other hand, a larger gap would also enable venous blood within the gap to enter the imaging slice without ®rst being saturated. Therefore, some venous blood may also be inadvertently imaged and potentially confused for arterial ¯ow. This phenomenon is called `venous bleedthrough' (Figure 2). It is therefore evident that the size of the saturation gap is critical and that the optimal saturation gap size may vary depending upon the arterial and venous ¯ow rates at the particular site of interest. In addition to the above techniques, we also utilize ¯ow compensation at our institution. The 2D axial images obtained are stacked on top of each other to create a three-dimensional (3D) data set. Projection angiograms are created by the MIP algorithm. A ray is projected through this 3D data set, and each pixel in the projection is designated a value based on the maximum intensity encountered within the pixel. At our institution, we utilize a TR of 33±36 ms, a TE of 6.2±7.7 ms, and a 60 ¯ip angle. Our 2D TOF imaging protocol entails imaging at several stations. The pelvis is imaged with a 32 cm ®eld-of-view (FOV), 2.5 mm thickness, 128  256 matrix, and a body coil. No skip is utilized. Distal thigh, knee, leg, ankle, and foot are imaged with an extremity coil at a 16 cm ®eld of view. Ninety 2 mm images per station are obtained with the knee coil. The pelvis and one lower extremity are imaged in approximately 1.5 h. Station 1 is at the forefoot, where the dorsal and plantar arches form from the dorsalis pedis and posterior tibial arteries [Figure 3(a)]. Upper and lower margins of the coil are marked to ensure that all vessels are examined, which in the General Electric knee coil is up to 20 cm long. In patients with peripheral vascular disease, the ¯ow distally tends to be nonpulsatile, thus a

narrower saturation gap is utilized. This also limits venous bleed-through. The arches frequently cannot be seen, since they course horizontally and become saturated. Retrograde ¯ow, sometimes seen in the dorsalis pedis, susceptible to saturation by the inferior saturation slab, may create the impression of stenosis on the MIP. The second station examines the anterior tibial, posterior tibial, and peroneal arteries [Figure 3(b)]. Calcaneal and perforator branches arise from the peroneal artery, and these course posteriorly and anteriorly, respectively. The calcaneal artery (communicator branch) may function as a collateral to the plantar aspect of the foot, while the perforator branch coursing through the interosseous membrane may function as a dorsal collateral. The posterior tibial artery bifurcates into medial and lateral plantar arteries at the ankle. The third station ends at the mid-calf [Figure 3(c)]. The posterior tibial artery is most medial and usually anterior to the peroneal artery. The anterior tibial artery is located very super®cially at the talus, where the dorsalis pedis forms, and it travels in the tibialis anterior muscle within the anterior compartment. The peroneal artery parallels the ®bula. Intramuscular collaterals may accompany disease at the trifurcation. Venous bleed-through may mimic such arterial collaterals. The fourth station, at the knee, is where the popliteal artery and the trifurcation are evaluated [Figure 3(d)]. Basis images are critical, especially for assessing stenosis of the anterior tibial artery. This area represents the most variable for the number of vessels seen. Anatomic variation, occlusion, and collaterals all contribute to this phenomenon. The length of the region examined is also variable due to individual height discrepancies. The ®fth station examines the common femoral artery to the super®cial femoral artery (SFA)±popliteal artery junction [Figure 3(e)]. The SFA is medial and super®cial. The profunda femoral artery is posterior and lateral in location, and terminates at the mid-thigh with associated muscular branches. The SFA has no muscular branches. Hunter's canal is contained in this series. Stenoses at this level are common. Faster arterial in¯ow and venous out¯ow at this station are conducive to a shorter TR (33 ms) and a larger saturation gap (1 cm), respectively. The sixth station (pelvis) covers the distal abdominal aorta to the common femoral arteries [Figure 3(f)]. The aorta normally bifurcates into the common iliac arteries, which further bifurcate into internal and external iliac arteries. The common femoral arteries subsequently give rise to the deep and super®cial femoral arteries. The faster ¯owing veins at this level require an even larger saturation gap (2 cm) at this station.

3

CLINICAL DATA

Clinical data thus far has shown 2D TOF MRA to be a promising technique. Mulligan et al. studied 140 lesions and found a 71% correlation between 2D TOF MRA and conventional angiography of the abdominal aorta and distal-runoff vessels.5 MRA alone proved inaccurate in six out of 140 cases for planning a surgical approach. As this was one of the early studies in this area, these inaccuracies to some extent may re¯ect inadequacies of the early techniques. Later studies by other investigators produced more promising results, probably

PERIPHERAL VASCULATURE MRA

3

Figure 3 Two-dimensional TOF projection: (a) through the hind foot, demonstrating the dorsalis pedis artery dorsally; (b) anteriorly through the ankle. The anterior tibial (small arrow), peroneal (curved arrow), and posterior tibial arteries (large arrow) are demonstrated; (c) anteriorly in the region of the mid-calf, demonstrating the anterior tibial artery (small arrow) and tibial peroneal trunk bifurcating into posterior tibial (large arrow) and peroneal arteries (curved arrow); (d) through the region of the knee, demonstrating the popliteal artery and the take-off of the anterior tibial artery (arrow); (e) obtained with the body coil, demonstrating the super®cial femoral arteries bilaterally from upper thigh through Hunter's canal (large arrows). The distal profunda femoral arteries are also identi®ed (small arrows); (f) through the pelvis, demonstrating the aorta, common iliac arteries (largest arrows), internal (small curved arrows) and external iliac (larger curved arrows) arteries, and common deep (curved open arrow) and super®cial (open arrow) femoral arteries. Note the excellent visualization of the femoral artery bifurcation

4 PERIPHERAL VASCULATURE MRA

Figure 4 Representative images from the magnetic resonance arteriogram of a patient who presents with lymph-threatening ischemia. (a) Images through the pelvis demonstrating bilateral iliac artery occlusion and reconstruction of the SFA bilaterally (arrows). (b) Images obtained through the upper thighs. Bilateral super®cial femoral artery occlusions are demonstrated (arrows). (c) Images through the lower left calf. The presence of an anterior tibial artery crossing the ankle to form the dorsalis pedis (arrow) and a peroneal artery (curved arrow) with a calcaneal branch crossing the ankle to form the plantar arches is demonstrated. On the basis of these MR images, this patient went on to have revascularization procedure without contrast angiogram

due to technical advancements. Color-duplex ultrasound (US) demonstrated 93% concurrence for infrainguinal disease. While MRA was superior to US in evaluating the iliac regions due to limited visualization with US in that area, image quality with MRA was inconsistent. Workers concluded that both duplex US and MRA were suboptimal for surgical planning. Owen et al. found that most of the discordance between 2D TOF MRA and conventional angiography was due to enhanced detection of runoff vessels on MRA6 (Figure 4). Interventional planning was altered in 16% of cases (12 patients) due to improved visualization with MRA of stenosis and vessels not seen on conventional angiography. Carpenter et al. studied 51 patients with peripheral vascular disease.7 Two-dimensional TOF MRA evaluation demonstrated that in 48% of cases, MRA provided additional information to

contrast angiography without loss of other information. In 22% of cases, the MRA data were responsible for altering surgical treatment. Target vessels were observed in 18% of cases on MRA that were undetected on contrast angiography. Identi®cation of these target vessels enabled the surgeons to perform limb-salvaging procedures. Similarly, two cases (4%) demonstrated MRA ®ndings of target runoff vessels which were not seen on conventional angiography, where the information was used to formulate a better interventional plan for bypass grafting. These investigators demonstrated superior sensitivity of MRA compared with contrast angiography in evaluating distal runoff vessels.8 Other workers also found that 2D TOF suppressed veins better than the 3D techniques.9 In evaluating for stenosis, Chien et al.9 suggest black-blood techniques to prevent nonvisualization of turbulent spins from

PERIPHERAL VASCULATURE MRA

phase incoherence.5 Bright-blood methods, on the other hand, may overestimate a stenosis due to phase incoherence. Mulligan et al. evaluated stenosis from MIP, which simulates the unidimensionality of angiographic studies; however, axial-basis images are more accurate for such assessments.5 Also, since they utilize the large FOV, the in-plane resolution of the basis images is worsened. This also compromises the resolution of the MIP. Cambria et al. reported 98% concurrence between 2D TOF MRA and contrast angiography in 178 patients with a 100% concurrence below the inguinal ligament.10 Intraoperative con®rmation was obtained in the seven bypass grafts performed on the basis of the MRA data alone. Hertz et al. also reported a high correlation between 2D TOF MRA and contrast angiography in assessing arterial stenosis in 19 patients11 (Figure 5). For the popliteal to the dorsalis pedis area, Carpenter et al.7 showed MRA to be superior to conventional angiography, with a 48% discordance rate between the two techniques.6 This difference appeared to be more progressively pronounced the further distally one examined. Up to 50% of stenosis may be overlooked with conventional angiography if the plaque is en face with no available edge for imaging. We have found 2D TOF MRA to be accurate in assessing peripheral stenosis. Cross-sectional images are particularly useful, in that they may demonstrate more information than contrast angiography. We showed good agreement in blinded interobserver interpretation (91%).11 Also, lesions distal and proximal to stenosis were seen equally well with MRA, with no appreciable discrepancy between the two.11 We investigated 41 patients for aorto±iliac disease.12 Only six cases showed a discrepancy between MRA and contrast angiography. Two of these proved to be false-positive on MRA, and two proved to be false-negative. The other two underestimated a greater than 50% stenosis seen on angiography. In the pelvis and lower extremity, we obtained a 62% concordance in a 103-patient series, with no false-positives or false-negatives on MRA.13 Contrast angiography failed to show 162 open segments that were con®rmed surgically. We have found MRA to be reliable in assessing vessel patency. Baum et al. demonstrated a 0.7 correlation coef®cient between contrast angiography and 2D TOF MRA for determining the parameters of a stenosis.14 We compared contrast angiography with MRA results in 60 patients with peripheral vascular disease. ROC-curve analysis of one blinded observer demonstrated that location and patency yielded a high correlation between angiography and MRA at our institution.15

4 PHASE CONTRAST (PC) Phase contrast techniques utilize motion as an imaging tool, whereby ¯ow is detected in all three axes. A technique relying on phase discrepancy, which measures phase shift of moving spins to determine their velocities, was proposed by Bryant et al.16 Other investigators demonstrated that cardiac gating imparts accuracy to the calculated arterial ¯ow rates.17 Various phases of arterial ¯ow could also be better segregated and evaluated. Applications of PC in peripheral vascular disease soon followed. A 2D TOF with PC velocity-sensitive technique demonstrated signi®cantly increased velocities at systole in diseased peripheral vessels after angioplasty, whereas normal

5

Figure 5 Two-dimensional TOF arteriogram: (a) on a patient who had recently had an angioplasty, demonstrating recurrent stenosis with an intimal ¯ap (arrows); (b) obtained 1 week prior, demonstrating the stenosis (arrow)

vessels demonstrated no alteration in ¯ow.18 A one spatial- and one-dimensional-variant of PC may enhance tracing of the triphasic ¯ow in the popliteal artery.19 In our recent experience with two-axis cine-PC angiography for arterial wave patterns in diseased vessels, we have observed monophasicity beyond points of critical stenosis. Simultaneous measurements of velocities at several sites may be implemented with a combined excitation technique to obtain cine-PC data in one dimension, with cardiac gating to generate a second dimension (time).20 Measurements of arterial compliance are made

6 PERIPHERAL VASCULATURE MRA possible. Addressing arterial compliance is useful in atherosclerotic disease, where compliance is decreased and arterial waveforms are altered. Recently, investigators have demonstrated similar velocity measurements in common iliac arteries with a cine-PC technique with a surgically attached US ¯ow probe on the artery.21 Velocity-encoded cine MRA was compared with color-coded US in ten healthy subjects above and below the trifurcation and showed good correlation.22 The classic triphasic pattern was seen. These investigators suggest that velocity-encoded cine MRA may help to evaluate the hemodynamic signi®cance of a stenotic lesion.23 This information may complement anatomical information obtained from standard 2D TOF MRA. They suggest that a single MRA examination may derive all the information that previously might have been obtained by performing both contrast angiography and duplex US. ECG-triggered 2D PC techniques, where velocity encoding varies with the cardiac cycle, may improve the contrast-tonoise ratio, most notably in small vessels, where up to 260% improved signal has been noted.24 In summary, PC techniques appear to be very promising. 5 3D MRA While we do not advocate the use of 3D techniques, we will brie¯y describe their applications. While in 2D imaging, many slices are imaged in sequence, in 3D imaging a volume is imaged. Acquisition of data from a larger volume, as in 3D TOF, enhances spatial resolution and therefore increases S/N. Imaging times are also decreased. Flow compensation is easier with 3D rather than 2D TOF MRA due to the increased gradient amplitude required for slice selection in 2D TOF. In 3D techniques, resolution along the Y and Z axes is obtained by phase-encoding techniques. Therefore, the slice selection gradient is relatively unburdened. In contradistinction, an extra lobe on the gradient waveform would be necessary to implement ¯ow compensation with 2D techniques, and this would consequently increase the TE. Therefore, shorter TEs are possible with a 3D modality when utilizing ¯ow compensation in comparison with a 2D modality. So, less dephasing of moving spins occurs with 3D ¯ow compensation. Smaller voxel sizes may also be utilized with 3D imaging. The main disadvantages of 3D imaging techniques, which signi®cantly hinder their usefulness, include decreased sensitivity to slow ¯ows and prolonged reconstruction times of a 3D volume-based data set. In the 3D method, slow ¯ow may incur saturation as it traverses further into the saturation slab. By the same reasoning, venous blood would saturate more readily compared with arterial blood. Therefore 3D MRA may be of greater value in imaging fast ¯owing and small vessels. Its primary bene®ts have been seen in the area of neurovascular imaging. 6 ARTIFACTS Clip artifacts and orthopedic prostheses and artifacts related to susceptibility differences at various interfaces may simulate stenosis or occlusion of a vessel (Figure 6). Local magnetic ®eld variations may be large enough to cause phase dispersion, yielding a net signal intensity vector of zero. Frequently, there is an associated high-signal intensity rind near the metal or air

Figure 6 (a) Frontal projection from a 2D TOF arteriogram through the pelvis demonstrating complete absence of iliac arteries distal to the common iliac artery bilaterally. (b) Coronal scout image demonstrating large artifacts due to the presence of hip prostheses bilaterally

due to misregistration in frequency and slice selection directions as a result of magnetic susceptibility. Gradient echo sequences are particularly vulnerable to these effects. Due to the length of the examination movement may occur in some patients. Displacement may appear as one translocated, or multiple translocated, slices on the MIP (Figure 7). Motion may create the appearance of a vascular lesion, and individual sections must be reviewed. Peristalsis or respiratory motion may confer high signal intensity on bowel material by `refreshing stationary spins'. Muscle contraction may confer increased signal intensity to muscle in the same fashion as ¯owing motion in vessels produces ¯ow-related enhancement. As discussed previously, pulsatile ¯ow may generate dark, horizontal bands on the MIP, as each section is not obtained at identical times in the cardiac cycle. Thus, varying signal intensities are obtained in each section. This banding artifact, or `Venetian-blind effect,' is more pronounced in vessels with triphasic ¯ow (Figure 8). Cardiac gating and an increased saturation gap are two ways of counteracting these effects. Phase ghosting, also produced by vessel pulsatility, may be

PERIPHERAL VASCULATURE MRA

7

Figure 8 Oblique magnetic resonance arteriogram projection through the pelvis demonstrating high and low signal stripes (arrows) representing banding from pulsatile ¯ow

7 Figure 7 Lateral MIP projection demonstrating the effect of motion on the MR arteriogram. Displacement of segments of the vessel can simulate stenosis (arrows). Motion can be con®rmed by observing motion of the skin line (small arrows)

counteracted with EKG gating. Apparent absence of retrograde arterial ¯ow may occur with the use of a saturation slab which is primarily intended to saturate venous ¯ow. Arterial pulsatility may be con®rmed by removing the saturation slab or by acquiring velocity-encoded information. Turbulence artifacts near a site of stenosis, with its dephasing effects, causes high ¯ow with spin-phase incoherence. However, this is not much of a problem peripherally where ¯ow is less pulsatile. Inspection of axial images for high-grade stenosis would explain the ®ndings. Potential pitfalls of MIP images occur when a high signal intensity interferes with visualization of a blood vessel, such as from a blood clot. Conversely, a ®lling defect may become nonapparent if it is surrounded by high signal intensity ¯owing blood. Also, when section thickness exceeds that of in-plane resolution, normal horizontal ¯ow may appear obstructed on the MIP. In the same vein, pseudobeating may occur with tortuous oblique vessels. This may manifest as a `stair-step' artifact on the MIP. This may be counteracted by thinner slices, slice overlap, and reacquisition perpendicular to the vessel. Inplane or `horizontal' ¯ow refers to areas where, due to the orientation or tortuosity of a vessel, ¯ow occurs in a roughly horizontal rather than vertical plane. Since ¯ow in the horizontal plane would not be detected because of saturation, the vascular anatomy would be obscured or underrepresented as stenotic. Examples of this include the anterior tibial artery take-off, which is roughly horizontal or perpendicular to the popliteal artery, and tortuous iliac bifurcations.

RELATED ARTICLES

Abdominal MRA; Head and Neck Studies Using MRA; Phase Contrast MRA; Time-of-Flight Method of MRA.

8

REFERENCES

1. K. R. Patel, L. Semel, and R. H. Clauss, J. Vasc. Surg., 1988, 7, 531. 2. J. B. Ricco, W. H. Pearce, J. S. T. Yao, W. R. Flinn, and J. J. Bergan, Ann. Surg., 1983, 198, 646. 3. R. Scarpato, R. Gembarowicz, S. Farber, T. F. O'Donnell, J. J. Kelly, A. D. Callow, and R. A. Deterling, Arch. Surg., 1981, 116, 1053. 4. D. P. Flanigan, L. R. Williams, T. Keifer, J. J. Schuler, and A. J. Behrend, Surgery, 1982, 92, 627. 5. S. A. Mulligan, T. Matsuda, P. Lanzer, G. M. Gross, W. D. Routh, F. S. Keller, D. B. Koslin, L. L. Berland, M. D. Fields, and M. Doyle, Radiology, 1991, 178, 695. 6. R. S. Owen, R. A. Baum, J. P. Carpenter, G. A. Holland, and C. Cope, Radiology, 1993, 187, 627. 7. J. P. Carpenter, R. S. Owen, R. A. Baum, C. Cope, C. F. Barker, H. D. Berkowitz, M. A. Golden, and L. J. Perloff, J. Vasc. Surg., 1992, 16, 807. 8. R. S. Owen, M. Sheline, J. Listerud, and H. Y. Kressel, Proc. Xth Ann. Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 138. 9. D. Chien, A. Goldmann, and R. R. Edelman, Proc. XIth Ann. Mtg. Soc. Magn. Reson. Med., New York, 1992, p. 3111. 10. R. P. Cambria, E. K. Yucel, D. C. Brewster, G. L'Italien, J. P. Gertler, G. M. Lamuragha, J. A. Kaufman, A. C. Waltman, and W. M. Abbott, J. Vasc. Surg., 1993, 17, 1050. 11. S. M. Hertz, R. A. Baum, R. S. Owen, G. A. Holland, D. R. Logan, and J. P. Carpenter, Am. J. Surg., 1993, 166, 112. 12. R. A. Baum, G. A. Holland, D. R. Logan, J. P. Carpenter, and C. Cope, J. Vasc. Intervent. Radiol., 1993, 4, 59. 13. R. A. Baum, G. A. Holland, D. R. Logan, J. P. Carpenter, and C. Cope, J. Vasc. Intervent. Radiol., 1993, 4, 15.

8 PERIPHERAL VASCULATURE MRA 14. R. A. Baum, G. A. Holland, D. R. Logan, S. Hertz, J. P. Carpenter, R. S. Owen, and C. Cope, J. Vasc. Intervent. Radiol., 1993, 4, 60. 15. M. L. Schiebler, J. Listerud, G. A. Holland, R. Owen, R. Baum, and H. Kressel, Invest. Radiol., 1992, 27, S90. 16. D. J. Bryant, J. A. Payne, D. N. Firman, and D. B. Longmore, J. Comput. Assist. Tomogr., 1984, 8, 588. 17. G. L. Naylor, D. N. Firman, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 18. M. Koch, S. E. Maier, I. Baumgartner, K. D. Hagspiel, C. Von Weymarn, P. Boesinger, A. Bollinger, and G. K. Von Schultess, Proc. Xth Ann. Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 137. 19. V. Dousset, F. W. Wehrli, A. Louie, and J. Listerud, Radiology, 1991, 179, 437. 20. C. L. Dumoulin, D. J. Doorly, and C. G. Caro, Magn. Reson. Med., 1993, 29, 44. 21. L. R. Pelc, N. J. Pelc, S. C. Rayhill, L. J. Castro, G. H. Glover, R. J. Herfkens, D. C. Miller, and R. B. Jeffrey, Radiology, 1992, 185, 809. 22. G. R. Caputo, T. Masui, G. A. Gooding, J. M. Chang, and C. Higgins, Radiology, 1992, 182, 387.

23. G. R. Caputo, C. B. Higgins, Invest. Radiol., 1992, 27, S97± S102. 24. J. S. Swan, D. M. Weber, T. M. Grist, M. M. Wojtowycz, F. R. Kovosec, and C. A. Mistretta, Radiology, 1992, 184, 813.

Biographical Sketches Cathy Maldjian. b 1965. B.A. Columbia University, New York, 1986; M.D. University of Medicine and Dentistry of New Jersey, 1990. Diagnostic Radiology Residency, Mount Sinai Medical Center (New York), 1995. Magnetic Resonance Imaging Fellowship at the Hospital of the University of Pennsylvania following residency. Approx. 10 publications. Research interests include clinical applications of MRI. Mitchell D. Schnall. B.A., physics, University of Pennsylvania, 1982, M.D., Ph.D., University of Pennsylvania, 1986. Residency in Radiology at the Hospital of the University of Pennsylvania, 1991. Assistant Professor of Radiology at the University of Pennsylvania 1991±1994. Currently Associate Professor of Radiology and Chief of the MRI Section.

1

PHASE CONTRAST MRA

Phase Contrast MRA Charles L. Dumoulin General Electric Research and Development Center, Schenectady, NY, USA

and Patrick A. Turski University of Wisconsin, Madison, WI, USA

1 INTRODUCTION Recent advances in X-ray, ultrasonic, nuclear, and NMR technologies are making the role of diagnostic radiology increasingly important in the practice of modern medicine. The rapid development of magnetic resonance methods in particular has been most striking. MR has several advantages over X-ray, and MR methods have replaced X-ray methods for many applications. One example of this is the development of MR methods for the visualization of blood vessels within the body. Today this modality is known as magnetic resonance angiography (MRA). The signi®cance of MRA in medicine cannot be fully appreciated unless one is aware of the competing methods for the generation of blood vessel images. Before the advent of MRA, clinicians generated vascular images using ionizing radiation in the X-ray spectrum. While X-rays of an appropriate energy easily penetrate the body, there is essentially no difference in the X-ray attenuation of blood and surrounding tissue. To overcome this problem, diagnosticians replace blood in a selected vessel with an X-ray-opaque liquid. This contrast media is somewhat toxic and uncomfortable for the patient. In addition, it is typically delivered to the selected blood vessel by a catheter placed into the patient. In light of concerns regarding the ionizing radiation, the toxic effects of the contrast media and a low, but nonzero, mortality rate, the diagnostic bene®t of X-ray angiography must be weighed against the risks to the patient. An alternative to X-rays in vascular radiology is the use of ultrasound. Ultrasonic vascular images are made by observing the re¯ection of acoustic energy at high frequency. The acoustic attenuation and re¯ection properties of blood are different from surrounding tissue and frequently provide suf®cient contrast to permit the visualization of a vessel. Doppler frequency shifts in the ultrasonic signal can also be used to highlight moving blood. While ultrasonic measurements carry none of the risks of X-ray methods, they are limited in their applications to only a few clinical applications. For example, vessels that are obscured by bone cannot be observed. Vessels deep within the body, or hidden by a thick layer of adipose tissue, are also dif®cult or impossible to detect. Another aspect of ultrasonic detection is that the resulting images are heavily dependent on the skills of the operator, since the ultrasonic probe must be manipulated at the surface of the patient. Magnetic resonance angiography overcomes many of the problems of X-ray angiography and ultrasonic imaging. The

MRA techniques employed today fall into two classes of technique that are de®ned by the two different ways in which macroscopic motion of nuclear spins in¯uences the creation and detection of an NMR signal. The ®rst class of methods relies on the macroscopic displacement of longitudinal magnetization. The phenomenon was ®rst observed and used by Singer and others long before the advent of MR imaging.1±3 Today, MRA methods of this type are frequently referred to as `time-of-¯ight' methods (see Time-of-Flight Method of MRA). The second class of MRA methods relies on a very different phenomenon. These methods exploit the change in the phase of transverse spin magnetization which occurs when nuclear spins move along a magnetic ®eld gradient. The phenomenon was ®rst reported by Hahn4 in 1960, and was recognized as a major source of artifacts in the early development of MRI.5,6 The utility of motion-induced phase shifts for the detection of ¯ow was also recognized7,8 and many phase-sensitive MRA methods have since been proposed and demonstrated. This article will attempt to describe the more widely used methods of phase-sensitive MRA and provide an overview of the unique aspects of this class of imaging methods.

2

PRINCIPLES OF PHASE DETECTION OF MOTION

2.1

Bipolar Gradient Pulses

The fundamental phenomenon which is the source of many motion-induced artifacts in MRI and the source of motionbased contrast in MR angiograms is the effect that a magnetic ®eld gradient pulse has on transverse spin magnetization. This effect is illustrated in Figure 1. In Figure 1, longitudinal spin magnetization is ®rst generated by an excitation rf pulse. A ¯ow-encoding magnetic ®eld gradient pulse is then applied. This ¯ow-encoding pulse induces a phase shift which is proportional to velocity. Phase-sensitive MRA methods use this phase shift to discriminate between moving nuclear spins and those that are stationary.

z

z

y

y F

x (a) Create transverse magnetization

x (b) Apply flow-encoding magnetic field gradient pulses

(c) Phase shift now proportional to velocity

Figure 1 Velocity-induced phase shifts. All phase-sensitive velocity imaging and measurement with magnetic resonance follows these steps. First, transverse spin magnetization within the sample is generated by applying a radiofrequency pulse at the Larmor frequency. A ¯owencoding gradient pulse is then applied. This ¯ow-encoding gradient pulse induces a phase shift which is directly proportional to velocity

2 PHASE CONTRAST MRA T

Ag

–Ag

Figure 2 The bipolar ¯ow-encoding gradient pulse. The ®rst lobe dephases transverse magnetization by an amount proportional to position. The second lobe rephases the magnetization. If the magnetization moves in the interval between the pulses, however, the cancelation of the induced phase shift is incomplete. The residual phase shift is proportional to velocity

The simplest ¯ow-encoding magnetic ®eld gradient pulse is shown in Figure 2. This pulse consists of two lobes, separated by a time interval, T. These gradient lobes are applied in the same direction, but with opposite polarity. The unsigned area of each lobe, Ag, is identical. The phase shift induced by velocity, , can be expressed as:  ˆ VTAg

…1†

where is the gyromagnetic ratio of the nuclear spin, and V is the velocity of motion. The effect of a ¯ow-encoding gradient on transverse spin magnetization is illustrated in Figure 3. In Figure 3, two bodies

of nuclear spins are shown. The ®rst body is located some distance from the center of the magnetic ®eld gradient and remains stationary during the ¯ow-encoding gradient pulse. The second body of nuclear spins is also located along the magnetic ®eld gradient, but is moved during the time interval between the gradient lobes. Before application of the ®rst lobe of the ¯ow-encoding gradient, transverse spin magnetization is generated and exists with a phase shift, which for the sake of simplicity is assumed to be zero. The application of the ®rst gradient lobe causes the strength of the magnetic ®eld to be proportional to position for the duration of the lobe. Since the Larmor frequency of the transverse spin magnetization is directly proportional to the strength of the magnetic ®eld, during the application of the lobe, nuclear spins will have a resonant frequency that is proportional to location. After the lobe is ®nished, the magnetic ®eld is homogeneous once again (i.e. all transverse spin magnetization resonates at the same frequency), but the transverse spin magnetization will have acquired a phase shift. This phase shift is proportional to the strength and duration of the lobe, and is also proportional to the location of the nuclear spin along the applied magnetic ®eld gradient. The phase shift induced by the ®rst lobe in Figure 3 can be undone by the application of a second lobe having equal amplitude and duration, but having opposite polarity. If the nuclear spins do not move during the interval between the gradient lobes, the phase shifts induced by the ®rst lobe will be canceled by the second lobe. If, on the other hand, the nuclear spins move during the interval between gradient lobes, the phase shift induced by the ®rst lobe will not be exactly canceled by the phase shift induced by the second lobe. The residual phase shift will be proportional to the distance the nuclear spins have moved and hence will be proportional to the velocity of the nuclear spins. 2.2

First gradient pulse

Second gradient pulse

fS1

fM2

Phase fM1 fS2

Stationary spin

Moving spin Position

Figure 3 The effect of a bipolar gradient pulse on the phase of stationary and moving spin magnetization. The ®rst magnetic ®eld gradient lobe of the bipolar pulse induces a phase shift that is proportional to position. The second lobe also induces a phase shift, but this phase shift has the opposite sign. The phase shifts induced in stationary nuclei by the ®rst and second lobes, S1 and S2, cancel upon subtraction. The phase shifts for moving nuclei, M1 and M2, on the other hand, are not equal and thus do not entirely cancel. The residual phase shift is proportional to the nuclei's velocity

Higher Order Gradient Pulse Waveforms

A bipolar magnetic ®eld gradient pulse is only one member of a series of magnetic ®eld gradient pulse waveforms which induce phase shifts in transverse spin magnetization. Singlelobed gradient pulses induce phase shifts that are proportional to position and are used to encode position in conventional MRI. Bipolar gradient pulses are simply two spatial-encoding gradient pulses applied at slightly different times so that changes in position during the bipolar pulse give phase shifts that are proportional to velocity. Higher orders of motion such as acceleration and jerk can be detected with more complicated gradient waveforms, which can be constructed with linear combinations of simpler gradient waveforms. An illustration of this concept is shown in Figure 4. 2.3

Detecting Velocity with Phase

Once transverse spin magnetization has been given a phase shift that is proportional to velocity, several strategies can be employed to selectively detect signals arising from moving nuclear spins in the presence of signals from stationary spins. Since ¯ow-encoding gradient pulses are only one of many potential sources of phase shifts in transverse spin magnetization, the direct detection of phase itself is usually impractical. Fortunately, phase shifts induced by phenomena other than ¯ow can

3

PHASE CONTRAST MRA z

Ag + T

y x

Magnetic field gradient

z

y Time

FAcc = 4 gAT 2Ag

FAcc = 2 gAT 2Ag

Figure 4 Construction of an acceleration (Acc)-encoding magnetic ®eld gradient pulse waveform by combining two velocity-encoding waveforms. Note that the two different waveforms induce different phase shifts

easily be removed through a variety of modulation schemes. For example, an early method described by Wedeen et al.9 relied on the modulation of velocity that occurs in arteries. With this method, two complex images were obtained. Each image was acquired with a ¯ow-encoding gradient. The ®rst image was acquired during the period of peak velocity in the cardiac cycle (systole) and the second was acquired during a period of low ¯ow (diastole). During periods of high velocity, large phase shifts were induced in the MR signals. These phase shifts were frequently so large that distributions of velocity within an image voxel would cause attenuation of the MR signal by phase cancelation. Since the only aspect of the detected MR signal that was different in each acquisition was the phase cancelation arising from velocity, only signals from moving blood were retained upon subtraction of the two images. A major disadvantage of the cardiac gated scheme described above was that only pulsatile blood ¯ow could be visualized. One strategy that was designed to overcome this limitation employed an amplitude modulation of the ¯ow-encoding gradient waveform.10,11 As with Wedeen's technique, moving blood was distinguished from stationary tissue by phase cancelation arising from a distribution of velocities as shown in Figure 5. An alternative modulation scheme that has found widespread use12±16 is outlined in Figure 6. With this method, two sets of data are acquired under identical conditions, except for the polarity of the ¯ow-encoding gradient pulses. The amplitude of each ¯ow-encoding gradient is kept small to minimize phase cancelation caused by velocity distributions within a pixel, and yet is large enough to cause a detectable phase shift in the MR signal. Velocity information contained within MR signals detected with modulated ¯ow-encoding gradient pulses can be extracted in several ways.17 The simplest method is to subtract the magnitude of each image data set. This is only useful, however, if the data are amplitude-modulated (i.e. the modulation scheme causes phase cancelation of MR signals for one of the acquired data sets). More commonly, the extraction of velocity information is performed by complex arithmetic on the image data.

x

Figure 5 Use of phase cancelation to modulate the amplitude of the MR signal. With this strategy data are acquired under two conditions. The ®rst data set is acquired under conditions which cause loss of signal due to phase cancelation caused by motion. The second data set is acquired under identical conditions to the ®rst, except that the phase cancelation is not performed. The two acquired data sets are identical except for that part arising from motion

Two methods for the extraction of velocity information are currently in use. The ®rst method is the computation of the magnitude of the complex difference. A vector representation of the generation of complex difference data is shown in Figure 7. Images acquired in this way have an angiographic appearance in which stationary features have little or no signal intensity, and moving blood appears bright. Complex difference data are relatively insensitive to the presence of stationary spin magnetization within the voxel, but provide pixel intensities that are not linearly related to velocity as shown in Figure 8. The relationship between pixel intensity, I, and velocity is: I ˆ K abs‰sin… VTAg †Š

…2†

where K is a constant of proportionality. The problem of nonlinearity is frequently minimized by choosing ¯ow-encoding graz

y x

Magnetic field gradient

z

y x Time

Figure 6 Phase modulation of the MR signal. With this strategy data are acquired under two conditions. The ®rst data set is acquired with a ¯ow-encoding gradient pulse which induces a discrete phase shift due to motion. The second data set is acquired under identical conditions to the ®rst, except that the polarity of the ¯ow-encoding gradient pulse is reversed. The two acquired data sets are identical except for that part arising from motion

4 PHASE CONTRAST MRA Stationary tissue

First acquisition

Second acquisition Fflow 2

–

=

Fflow 1 Fobserved 2 No net magnitude

Moving blood

–

Fstationary Fobserved 1

Fstationary

= Fflow =

Figure 7 Vector representation of complex subtraction in phase contrast angiography. Two data sets are collected. The ®rst is collected with a bipolar gradient pulse (e.g. a positive lobe followed by a negative lobe) which induces a phase shift that is proportional to velocity. The second is collected in an identical fashion except that the polarity of the bipolar gradient is reversed (e.g. a negative lobe followed by a positive lobe). The complex difference of these two data sets is then calculated. Note that the complex difference of stationary tissue is zero, but that of moving blood is represented by a nonzero length vector

dient pulses that are only strong enough to cause a 1 rad phase shift for the highest expected velocity. Under this condition, the assumption: x  sin…x†

…3†

holds and the pixel intensity becomes approximately proportional to velocity. The second method for the extraction of velocity information from complex data is outlined in Figure 9. With this method, the phase of data in each data set is computed. The difference between the phase of the ®rst and second data set is then computed to give a phase difference data set. The intensity of each pixel in the ®nal image is set equal to the phase difference. The pixel intensity of phase difference data is linearly proportional to velocity, but its dependence on velocity is discontinuous and has a periodicity of  as shown in Figure

(Fobserved 2 – Fstationary) – (Fobserved 1 – Fstationary) 2

Figure 9 Vector representation of phase difference computations in phase contrast angiography. Two data sets are collected. The ®rst is collected with a bipolar gradient pulse (e.g. a positive lobe followed by a negative lobe) which induces a phase shift that is proportional to velocity. The second is collected in an identical fashion except that the polarity of the bipolar gradient is reversed (e.g. a negative lobe followed by a positive lobe). The difference of the phases of these two data sets is then calculated. Note that if the data contain contributions from both stationary tissue and ¯owing blood (as shown), then the phase shift of the stationary tissue must be removed before the phase difference due to ¯ow can be calculated

10. Note that when phase differences are computed it is important to remove static phase shifts arising from resonance offset conditions in order to maintain the expected relationship between velocity and phase. When pixel intensity is set equal to the phase difference of the acquired data, the images have a unique appearance. Stationary tissue and regions devoid of signal typically appear gray in color. Blood ¯owing in the direction of the ¯ow-encoding gradient pulse, however, has bright signal intensity which is proportional to velocity. Blood ¯owing in the opposite direction appears dark.

Pixel intensity

Pixel intensity

0 0.0

Velocity 0 0.0 Velocity

Figure 8 The relationship between velocity and image pixel intensity when complex differences are employed. The magnitude of the complex difference does not include directional information and is only approximately linear with respect to velocity as long as the velocityinduced phase shift is small. If the phase shift is too large, it becomes indistinguishable from a smaller phase shift (an effect called aliasing)

Figure 10 Relationship between velocity and image pixel intensity when phase differences are employed. Note that phase shifts are directly proportional to velocity. If the phase shift is too large, however, it becomes indistinguishable from a smaller phase shift. Note that phase difference displays produce a more linear response than magnitude displays, but have more severe discontinuities in the event of velocity aliasing

PHASE CONTRAST MRA

A phase difference display is frequently used when a quantitative velocity image is desired. To ensure the quantitative nature of the image, however, phase offsets arising from sources other than velocity must be removed. These sources include resonance offset conditions such as transmitter offsets, ®eld inhomogeneities, chemical shift phenomena, and eddy current effects.

5

Common carotid artery Velocity

0 Time

2.4 Strategies for Vessel Selectivity

Femoral artery Velocity

In clinical practice it is frequently useful to limit the number and nature of the blood vessels imaged in an angiogram. In Xray angiography this is accomplished by injecting the contrast media into the selected vessel using a catheter. Since MRA uses the motion of blood for contrast, a selective injection is not possible. Fortunately, several strategies exist to permit the selective detection of blood vessels. Perhaps the most frequently used strategy for selectively detecting vessels is the use of saturation zones around the region of interest. Saturation zones are regions that are subjected to strong rf pulses, typically having a ¯ip angle of between 90 and 135 . These pulses are applied repeatedly, and cause the longitudinal magnetization of tissue within the zone to become saturated. Blood traveling through the zone also becomes saturated. Blood leaving the saturation zone and entering the region of interest is made invisible since transverse spin magnetization cannot be created when longitudinal magnetization has been destroyed. Once the blood has been in the region of interest for a while, however, its longitudinal magnetization reaches a new steady state, and the blood may become visible. Saturation zones are useful in many clinical applications, particularly those in which the targeted vessel crosses from the saturation zone to the region of interest only once. A second mechanism that can be employed to selectively detect vessels with phase-sensitive MRA relies on differences in blood velocity. This is possible because the size of the ¯owencoding gradient pulses can be tailored to cause maximum phase shifts for a selected range of velocities. Smaller ¯owencoding gradient pulses will induce phase shifts in only the highest velocities, and have little effect on slowly moving blood. Since arteries typically carry relatively high velocity blood and veins have lower velocity blood, phase-sensitive angiograms using small ¯ow-encoding gradient pulses usually highlight arterial structures. Venous structures can be highlighted by increasing the strength of the ¯ow-encoding gradient pulses. Increasing the strength of the ¯ow-encoding pulses will cause the phase shifts arising from higher velocities to exceed  and the pixel intensity of arterial structures will no longer be proportional to velocity. In the body, arterial blood tends to be pulsatile while venous blood ¯ows more smoothly. The nature of arterial pulsatility depends on the location of the artery within the body and is heavily in¯uenced by the presence of stenosis. This is illustrated in Figure 11. Data acquisition strategies that highlight pulsatile ¯ow and suppress steady ¯ow can be devised if data are acquired synchronously with the cardiac cycle. For example, pulsatile ¯ow can be highlighted in an angiogram by acquiring images at multiple points in the cardiac cycle and computing the standard deviation of each pixel.

0 Time

Downstream from a stenosis

Velocity

0 Time

Figure 11 Blood velocity during the cardiac cycle in several locations within the body. Note that the velocity pro®le is dramatically different in different vessels and is in¯uenced by vessel disease

2.5

Multiplexed Detection of Multiple Flow Components

The ¯ow-encoding gradient pulses used in phase-sensitive MRA induce phase shifts that are proportional only to the component of velocity that is parallel to the direction of the applied magnetic ®eld gradient. Consequently, to construct an angiogram that contains information for all components of velocity, data from three orthogonal ¯ow-sensitive directions must be acquired. Since each ¯ow-sensitive direction requires that data be collected twice, a total of six acquisitions are required. The information present in a six-excitation phase contrast angiogram has only four components (i.e. stationary tissue and velocity in three orthogonal directions). It is possible to acquire this information more ef®ciently using a four-excitation multiplexing scheme. One such scheme18±20 is based on the Hadamard transform. With this method four excitations are performed, each having ¯ow-encoding gradient pulses applied in all three orthogonal directions simultaneously. The combination of ¯ow-encoding pulse polarities is unique in each excitation. One possible combination of ¯ow-encoding pulse polarities is: Gradient axis Excitation

X

Y

Z

1

‡

‡

‡

2

ÿ

ÿ

‡

3

ÿ

‡

ÿ

4

‡

ÿ

ÿ

6 PHASE CONTRAST MRA where + and ÿ denote the polarity of the applied ¯ow-encoding gradient pulse. The four components of information can be extracted by computing appropriate linear combinations of the acquired data. For the modulation scheme shown above, the linear combinations take the form: Stationary tissue ˆ Ex 1 ‡ Ex 2 ‡ Ex 3 ‡ Ex 4

…4†

X flow component ˆ ÿEx 1 ‡ Ex 2 ‡ Ex 3 ÿ Ex 4

…5†

Y flow component ˆ ÿEx 1 ‡ Ex 2 ÿ Ex 3 ‡ Ex 4

…6†

Z flow component ˆ ÿEx 1 ÿ Ex 2 ‡ Ex 3 ‡ Ex 4

…7†

where Exi represents data from the ith excitation.

3 IMPORTANT ATTRIBUTES OF PHASE CONTRAST MRA There are several important attributes that distinguish phasesensitive MRA from time-of-¯ight MRA and from X-ray angiography. For example, each datum collected in a phasesensitive procedure is sensitive only to a single component of ¯ow. Thus, if a complete angiographic image is required, data must be collected at least four times to obtain all three velocity components and the stationary tissue component. This is in contrast to the single detection required for most time-of-¯ight procedures. A phase-sensitive angiogram does not necessarily take four times longer to collect than a time-of-¯ight angiogram, however, since the optimum repetition time, TR, for a phase-sensitive angiogram is the minimum available on the system (e.g. 16 ms) whereas the optimum repetition time for a time-of-¯ight angiogram is longer (e.g. 40±50 ms) to permit the in¯ow of nonsaturated spin magnetization. A second attribute of phase-sensitive angiography is that image acquisition is relatively independent of the ®eld-of-view. This is because the source of image contrast in a phase-sensitive angiogram is motion. With phase-sensitive MRA, images of slowly moving blood in small vessels can be acquired within small ®elds-of-view as easily as higher velocity blood travelling in larger vessels within a large ®eld-of-view. Timeof-¯ight angiography, however, is constrained by the requirements of vessel geometry and blood velocity to ensure that the in¯ow of relaxed blood into the region of interest is optimized. Because the only source of image contrast in a phase-sensitive MR angiogram is motion, excellent suppression of background signals from stationary tissue is possible. Full suppression of background tissue with time-of-¯ight methods, however, is dif®cult since the source of image contrast in timeof-¯ight methods is differential levels of spin saturation. Consequently, the saturation of spin magnetization to suppress stationary tissue signal in time-of-¯ight angiography also suppresses signals from moving blood. Another consequence of the time-of-¯ight contrast mechanism is that tissue having short T1 [e.g. mucus and neoplasms after an injection of Gd-diethylenetriaminepentaacetic acid (Gd-DTPA)] can appear with the same pixel intensity as moving blood. Different levels of spin

saturation among tissues can also occur in phase-sensitive angiography, but this does not affect the suppression of stationary tissue, and partially saturated blood appears with only a reduced signal intensity. One attribute of phase-sensitive methods that has found extensive use is their quantitative nature. Several nonangiographic methods are currently in use. These include thin-slice methods in which vessels are imaged in cross section and the intensity of each pixel of the image is made proportional to the ¯ow-induced phase shift.21±23 An alternative approach is the conversion of one image dimension into a velocity dimension.24±27 Images acquired in this fashion are not affected by artifacts arising from background phase shifts and are similar in appearance to Doppler ultrasonic images. Another characteristic of phase-sensitive MR angiographic methods is that they are heavily dependent on the quality of the instrument. This is particularly true with respect to phase stability since phase shifts are used to detect and measure motion. Perhaps the most common source of phase instability in an MRI system is eddy currents. These eddy currents are induced in metallic structures within the magnet by magnetic ®eld gradient pulses. Phase instability can also occur in the transmitter and receiver subsystems of the instrument. Fortunately, most modern MRI systems are stable enough to perform phase-sensitive angiography.

4

4.1

PHASE-SENSITIVE FLOW IMAGING PULSE SEQUENCES

Two-Dimensional Phase Contrast MRA

The earliest implementations of phase-sensitive MRA9,12 were similar to X-ray methods in that they provided twodimensional (2D) projections of the region of interest. In these methods the thickness of the imaged volume was often made to be the full thickness of the patient. If slice selection was employed at all, it was often used to excite a thick slab of magnetization orthogonal to the image plane. Today 2D phase contrast MRA is rarely performed as a full projection. Instead, one or more thick slices are typically obtained from the region of interest. The speed of 2D phase contrast procedures can be advantageously used in a number of ways. For example, speed can be used to shorten the overall patient examination time. Quite often, however, it is useful to acquire additional data to provide more diagnostic information. With 2D phase contrast MRA, this might include additional view angles or slice locations. Alternatively, certain types of artifacts can be minimized by acquiring additional data. For example, phase-encoding artifacts arising from changes in blood velocity during the cardiac cycle can be minimized by detecting the average velocity-induced phase shift rather than one detected at an arbitrary instant in time. This can be done by acquiring data with a relatively large NEX. For example, with a repetition time, TR, of 50 ms and a heart rate of 60 beats per minute, choosing NEX to be about 20 will cause velocity variations during the cardiac cycle to be averaged.

PHASE CONTRAST MRA

7

4.2 Cine Phase Contrast MRA An early variant of 2D phase contrast MRA was the use of cardiac gating and the detection of images at multiple points in the cardiac cycle.14,15 Cine MRA is particularly useful in imaging peripheral arteries since blood ¯ow in these vessels is typically triphasic and moves forward, backward, and remains still at different points of each cardiac cycle. Arterial ¯ow dynamics can be exploited to differentiate arterial ¯ow from venous ¯ow. Thin-slice cine phase contrast imaging in which the phase of each pixel is displayed21±23 has become widely used in the past few years. These images provide a quantitative measure of blood velocities during the cardiac cycle and have been applied in the head,28 aorta,29 renal,30 mesenteric,31 and peripheral vessels.32 4.3 Three-Dimensional (3D) Phase Contrast MRA Two-dimensional phase contrast angiograms have several limitations that can be overcome by the collection of the data in three dimensions.16 Unlike a projection angiogram, a typical 3D angiogram has approximately isotropic voxels. Phase variations across a voxel are minimized because the voxel dimensions are small. Perhaps the most signi®cant advantage, however, is that the data can be analyzed retrospectively in a number of ways. For example, 2D projections can be generated for any view angle. Selected subsets of the acquired data can be extracted to remove features that interfere with the presentation of features of interest. With 3D MRA it is also easy to generate cross-sections of vessels. 4.4 Fourier Velocity Encoding Fourier velocity-encoding pulse sequences are a form of phase-sensitive velocity imaging in which the velocity of spin magnetization is represented as a positional displacement rather than as a phase shift in the image. These pulse sequences are equivalent to conventional spin warp imaging pulse sequences, except that the spatially localizing phase-encoding gradient pulses are replaced with bipolar ¯ow-encoding gradient waveforms. In other words, a Fourier velocity-encoded image has a phase-encoded dimension in which displacement of signal depends on spin velocity rather than on spin position. Pixel intensity in these images is determined by the sensitivity of the imaging system and the strength of the detected MR signal. The phase of each pixel, however, is determined by resonance offset conditions such as transmitter offsets, ®eld inhomogeneities, and susceptibility effects. As with conventional MR imaging, this phase information is typically ignored. Figure 12 shows an example of Fourier velocity-encoding applied in the human abdomen. In this image the horizontal dimension is a spatial dimension corresponding to the subject's left/right axis. The vertical axis, however, is a velocity dimension corresponding to the component of velocity parallel with the subject's inferior/superior axis. Excitation of spin magnetization was limited in this image to a 5 mm axial slice. Note that the distribution of velocities within the aorta is narrow, but the distribution of velocities within the vena cava is relatively broad.

Figure 12 Fourier velocity-encoded image of the abdomen. In this image one spatial dimension and one velocity dimension have been collected. Images of this form can be used to measure the distributions of velocity within a vessel. Since they use a phase-encoding mechanism to detect velocity, these images have a velocity precision and accuracy equivalent to that of spatial measurements in conventional imaging. (Reproduced by permission from C. L. Dumoulin et al.27)

5

POSTPROCESSING

Many of the MRA methods in current use acquire data in three dimensions. For example, a 3D phase contrast angiogram has three spatial dimensions. Cine 2D phase contrast angiograms are also 3D, but consist of two spatial dimensions and a temporal dimension. Unfortunately, 3D data cannot be displayed directly with the hardware available in most institutions. Rather, a 2D subset of the 3D data set must be extracted and put into a suitable form such as ®lm for diagnosis. Several methods currently in use are outlined below. 5.1

Maximum Intensity Projection

Filming thin-slice extractions of a 3D data set is straightforward and often useful. Projection images in which one of the three dimensions is collapsed, however, are usually more helpful. The simplest and most commonly used projection algorithm is the Maximum Intensity Projection (MIP) method. With MIP a `ray' is sent through the 3D data set to create each pixel in the projection. The maximum pixel intensity value found along the ray determines the pixel value in the projection image as shown in Figure 13. An important aspect of MIP is that vessels close to the viewer are indistinguishable from vessels further away. Depth information can be recovered, however, by sequentially displaying several MIP projections at a variety of view angles to create the illusion of a 3D object. 5.2

Subvolume Reconstruction

In an ideal MR angiogram the pixel intensities of blood vessels and stationary tissue are different enough to guarantee

8 PHASE CONTRAST MRA an occlusion or high-grade stenosis, however, may no longer be pulsatile and a standard deviation projection may suppress rather than enhance the appearance of the vessel.

6

Phase contrast MRA is frequently used in clinical evaluations to display pathologic ¯ow conditions. The principle reasons for using phase contrast methods are: (1) to improve vessel detection by subtracting the background signal and (2) to utilize the velocity/phase relationship to provide physiological information regarding blood ¯ow. The ¯exibility of phase contrast methods makes them somewhat more complex to use than time-of-¯ight methods in the clinical environment. Nevertheless, phase contrast methods are becoming increasingly valuable for the evaluation of patients with suspected vascular disease.

a

b

(a)

APPLICATIONS

(b)

Figure 13 Projection of the maximum intensity pixel to extract a 2D image from a 3D data set. In this procedure `rays' are sent through the 3D object. A 2D projection is created by reducing the data encountered by each `ray' to a single value. With the maximum itensity projection (MIP) algorithm, the value of the most intense pixel is chosen for the projection image. Here two `rays' are shown. The ®rst `ray' (a) encounters the bright signal in a vessel and selects the most intense pixel. The second `ray' (b) also selects the brightest pixel, but this pixel contains noise

that only vessels are detected with an MIP algorithm. Complete suppression of stationary tissue, particularly for time-of-¯ight procedures, is often impossible, however, and the MIP algorithm projects the most intense pixels of the stationary tissue background. This makes small vascular features having relatively low pixel intensity more dif®cult to detect. Fortunately, the appearance of nonsuppressed tissue can be minimized by applying the MIP algorithm only over a selected subvolume of data. In addition to improving vessel contrast in the projection image, subvolume reconstructions can be used to remove overlapping blood vessels to enhance the appearance of the vessels of interest. 5.3 Temporal Filtering Data that are acquired with a temporal dimension can also be subjected to projection procedures. The appearance of all vessels within a dynamic image can be enhanced by applying the MIP algorithm in the temporal dimension. Alternatively, the standard deviation of pixel intensity, rather than the maximum pixel intensity, along the ray can be determined to highlight dynamic features, such as arteries. Features that are more constant during the cardiac cycle, such as veins and background, are suppressed. Arterial blood ¯ow downstream from

6.1

The Role of Flow-Encoding Strength in Phase Contrast MRA

The amplitude of the bipolar ¯ow-encoding gradient pulse determines the amount of velocity encoding. A larger amplitude will sensitize the MR angiogram to low velocities. Conversely, a small amplitude will produce an angiogram that emphasizes higher velocities.33 One can therefore create phase contrast angiograms that display pathologic conditions by selecting a velocity encoding that closely approximates the pathologic ¯ow state. Encoding for slower ¯ow velocities can improve detection of slow ¯ow within venous structures such as the dural sinuses or deep cerebral veins. Encoding for slow ¯ow can also improve visualization of small arterial structures as illustrated in Figure 14. In general, when one wants to optimize visualization of intracranial arterial structures a velocity encoding of 30 cm sÿ1 is preferable to improve detection of the slow blood ¯ow along the vessel wall and to help de®ne small vessels and pathologic vessels with reduced ¯ow velocities. Similarly when phase contrast techniques are used to identify aneurysms, it is important to use a low-velocity encoding in order to optimize visualization of slow vortex ¯ow in the central portions of the aneurysm.35 Phase wrap (or ¯ow aliasing) is not a serious problem in phase contrast `speed' imaging (i.e. when the magnitude of the data is displayed), because the higher velocity spins typically occur in the center of the vessel and the vessel morphology will still be well delineated by visualization of the slower ¯ow along the edges of the vessel. By selecting a higher velocity encoding (100±150 cm sÿ1) and using phase-difference reconstruction, images can be obtained that preserve the quantitative phase/velocity relationships and emphasize higher velocity ¯ow structures. This can be advantageous for identifying high-¯ow arteries that supply arteriovenous malformations or higher velocity in¯ow jets associated with intracranial aneurysms. Phase-difference images also permit the sign of the ¯ow-induced phase shift to be encoded as pixel intensity. This is particularly useful for determining the direction of ¯ow within vessels.

PHASE CONTRAST MRA

6.2.1

9

Small Arteries

Frequently small arteries with minimal ¯ow are dif®cult to detect. This is due to saturation of slowly moving blood as it passes through the imaging volume. An incorrect diagnosis of vascular occlusion might be made when in fact the vessel is patent. By administering intravenous contrast material the signal intensity associated with the slowly moving blood increases by a factor of two, making the vessel easier to detect. This increase in signal results in an increase of the dynamic range of detectable velocities. An example of increased sensitivity to slow ¯ow after injection of a T1 relaxation agent is shown in Figure 15. 6.2.2

Figure 14 Coronal 2D phase contrast angiogram of the vertebral arteries at two velocity encodings. (a) This image was obtained with a velocity encoding of 40 cm sÿ1. The right vertebral artery is not visualized and was initially thought to be occluded. (b) The angiogram was repeated with a velocity encoding of 10 cm sÿ1. The slow ¯ow in the small, right vertebral artery is now visualized (large arrows). The slow ¯ow along the wall of the left vertebral artery is also demonstrated (small arrows). Note that the central higher velocity ¯ow has aliased and appears as reduced signal intensity (open arrow). (Reproduced by permission of Raven Press from C. A. Anderson et al.34)

Imaging venous structures is improved by the use of intravenous contrast material by decreasing the saturation of slow venous ¯ow especially along the wall of the vein. This is particularly important in cases of dural sinus thrombosis where collateral ¯ow may be dif®cult to detect due to its very slow ¯ow rates. 6.2.3

Blood ¯owing into the region of interest reaches a steady state having reduced signal intensity after experiencing multiple rf pulses. This saturation of slowly moving blood is particularly problematic for time-of-¯ight MRA since the intensity of blood is ultimately reduced to that of the surrounding background tissue. Saturation of slowly moving blood can also be a problem with phase contrast MRA, but since the background signal intensity in a phase contrast angiogram is zero, saturation merely reduces the signal-to-noise ratio of the blood signal and the vessel is usually still visible. Nevertheless, slowly moving blood would be easier to detect if the equilibrium magnetization could be increased. This can be accomplished by shortening the T1 of blood by administering a relaxation agent such as Gd-DTPA. With a T1 relaxation agent in the bloodstream, ¯owing blood reaches a steady state, which gives a stronger MR signal. Higher concentrations of the relaxation agent result in a shorter T1 and a corresponding increase in the steady state magnetization. For Gd-DTPA this relationship holds for concentrations up to 0.5 mM kgÿ1. At higher doses the T2 shortening effects become dominant and the MR signal intensity begins to decrease.35,36 The effects of intravenous contrast enhancement for phase contrast methods is related to velocity encoding. At higher velocity encodings slow ¯ow has relatively low signal intensity and is poorly visualized. In this instance, intravenous contrast material will signi®cantly improve the detection of slow ¯ow. This is particularly advantageous in four clinical situations.

Complicated Flow

Vortex ¯ow within aneurysms may also result in signal loss due to saturation effects. By shortening the T1 of blood, the lumen of the aneurysm may be better de®ned. 6.2.4

6.2 MRA and T1 Relaxation Agents

Venous Structures

Increased Spatial Resolution

Increasing spatial resolution results in reduced signal-tonoise for most MRA acquisitions. Intravenous contrast enhancement enables more highly resolved matrices and smaller ®elds-of-view to be used by increasing the signal associated with small vessels.

Figure 15 Gd-DTPA-enhanced phase contrast MRA. (a) The nonenhanced axial phase contrast angiogram was obtained using a 256  256 matrix and a ®eld-of-view of 20 cm. (b) After the administration of Gd-DTPA at a dose of 0.1 mM kgÿ1, the scan was repeated using identical imaging parameters. Smaller arterial branches (arrows) as well as venous structures (open arrow) can now be identi®ed

10 PHASE CONTRAST MRA 6.3 Clinical Applications in the Head and Neck Phase contrast angiography has proven useful in the head,37,38 neck,39 abdomen,40,41 and peripheral vessels.42,43 Rather than provide a complete atlas of examples, only applications in the head will be discussed below. 6.3.1

Cerebral Ischemia

The ability to quantify ¯ow velocity and ¯ow volumes with phase contrast MRA has added a new dimension to the MRA evaluation of neurological disease. In addition to demonstrating vascular morphology, phase contrast MRA can now determine the adequacy and character of blood ¯ow for speci®c vessels. The goals of the MRA examination are to identify compromised blood ¯ow and to determine the severity and etiology of the ¯ow reduction. Usually, reduced or absent ¯ow within the internal carotid, vertebral, basilar, anterior, middle, and posterior cerebral arteries can be identi®ed with a high degree of accuracy. There are four pathologic ¯ow conditions that are generally recognized as leading to cerebral ischemia or infarction. The ®rst is global hypoperfusion due to sustained hypotension. The second is thrombosis of the small perforating arteries leading to infarction of deep structures such as the basal ganglia or white matter. The third category is embolic disease in which atherosclerotic material or thrombus dislodges from a preexisting atheromatous lesion, producing ¯ow compromise distally in smaller branch vessels. The last condition is the complete thrombosis of a major vessel resulting in decreased or absent distal perfusion.44,45 MRA can frequently determine which of these mechanisms is responsible for the patient's ischemic event in a fashion analogous to X-ray angiography.46±48 Patients may present with anterior circulation, large artery ischemic events (i.e. motor or sensory de®cits) requiring attention to the carotid bifurcations (Figure 16), the proximal middle cerebral arteries, and possibly the aortic arch. Other patients may present with ischemic events related to the larger posterior circulation vessels (i.e. ataxia, dizziness, nausea, cranial nerve dysfunction). In this instance, the proximal and distal vertebral arteries and basilar artery must be evaluated (Figure 17). For patients presenting with embolic events, the embolic material may arise from the heart or the proximal arterial system. If the transient ischemic attacks occur in multiple vascular territories, a cardiac source for emboli should be sought. If previous transient ischemic attacks have occurred in the same vascular territory, a localized proximal vascular lesion is the most likely etiology.49 Intracranial MRA examination can be approached in several ways. The most commonly used method is a 3D time-of-¯ight acquisition which includes the circle of Willis, the proximal middle cerebral arteries, and the distal basilar artery. This approach is effective for evaluating the proximal vessels. Unfortunately, time-of-¯ight angiography is not particularly sensitive to slow ¯ow, and there may be instances in which vessel patency is overlooked due to saturation of the slow ¯ow distal to an area of severe stenosis. An alternative approach is to obtain a 2D or a 3D phase contrast angiogram with low velocity encoding (e.g. 20 cm sÿ1).50 These images can de®ne the vascular morphology in the region of reduced ¯ow. In addition, cine phase contrast studies can be performed to display ¯ow direction50 and to permit quantitative measurements of ¯ow For

Figure 16 Right internal carotid artery stenosis demonstrated by 2D phase contrast MRA. (a) The coronal 2D phase contrast image demonstrates a stenosis of the proximal right internal carotid artery. Flow is symmetric in both internal carotid arteries. Sagittal 2D phase contrast MRA images clearly de®ne (b) the normal left internal carotid artery, and (c) the stenotic proximal right carotid artery (arrow)

and velocity. In selective instances, this can be helpful for de®ning the hemodynamic effects of vascular lesions. MRA evaluation of the extracranial carotid arteries is currently a source of controversy.51,52 Most investigators rely on single slab or multiple slab 3D time-of-¯ight MRA to de®ne

PHASE CONTRAST MRA

11

Figure 17 Basilar artery occlusion demonstrated by phase contrast MRA. (a) The axial and (b) the coronal phase contrast angiograms indicate absence of ¯ow within the basilar artery and only minimal ¯ow in the right and left vertebral arteries (arrows)

the degree of vessel compromise.53 Some authors have recommended performing a 2D time-of-¯ight study whenever a carotid occlusion is encountered to avoid missing a nearly occluded internal carotid artery.54,55 Unfortunately, high velocities caused by a severe stenosis may cause a signal loss in 2D time-of-¯ight angiography. Carefully controlled trials will ultimately be needed to de®ne precisely the accuracy of MRA for de®ning carotid stenosis. There is no question, however, that MRA is capable of providing high-resolution images that are adequate for many clinical management decisions. 6.3.2

Vascular Malformations

Arteriovenous malformations (AVMs) are most commonly supratentorial where a large component of the AVM is located super®cially in the pia. The malformation occasionally has a wedge shape with the apex of the wedge directed toward the ventricular system. The amount of parenchymal component of the AVM is variable. Occasionally vascular malformations may have both dural and pial components and are thus termed mixed AVMs. Malformations completely restricted to the dura are termed dural AVMs. Spinal vascular malformations are most often dural in location and frequently contain direct arteriovenous ®stulas.56 The nidus of the AVM may be variable and dif®cult to characterize accurately. The nidus has been de®ned as `that part of the malformation interposed between the recognizable feeding arteries and the larger terminal draining veins'. The nidus represents the area of arteriovenous shunting within the malformation. It is therefore the low-resistance portion of the malformation and is thus responsible for the hemodynamic changes related to the malformation. The venous drainage begins as the multiple vascular channels coalesce into one large

Figure 18 MRA of an AVM prior to stereotactic radiosurgery. (a) The phase contrast angiogram (velocity encoding = 80 cm sÿ1) demonstrates enlargement of the right pericallosal artery (arrow). (b) A second phase contrast angiogram (velocity encoding = 20 cm sÿ1) reveals deep venous drainage into the vein of Galen. Cortical venous drainage along the medial aspect of the right parietal lobe is also noted (arrows). (c) Following the phase contrast angiograms a 3D time-of¯ight acquisition was obtained. The axial projection demonstrates the pericallosal feeder to the AVM, as well as the AVM nidus (arrows)

venous structure. The ability to de®ne the size and location of the AVM nidus is essential for the proper selection of patients for surgery, endovascular therapy, or focused radiation therapy. For patients with AVMs, the MR angiographic evaluation attempts to (1) identify the arterial supply to the AVM (Figure 18), (2) determine the size and location of the AVM nidus, (3) de®ne the location and morphology of the venous drainage (Figure 19), (4) identify high ¯ow aneurysms, and (5) attempt to detect the presence of ®stulas within the AVM. The patient is initially evaluated by obtaining spin echo images. Typically sagittal T1-weighted and axial fast spin echo T2-weighted images are obtained. Sagittal 2D phase contrast angiograms are acquired with multiple velocity encodings (e.g. 80 and 20 cm

12 PHASE CONTRAST MRA and the venous drainage is delineated at the lower velocity encodings. High ¯ow rates may also be quantitated by cine phase contrast MRA techniques. 6.3.3

Intracranial Aneurysms

MRA is also becoming an important modality for identifying and characterizing intracranial aneurysms. Although MRA has outstanding potential as a noninvasive method for identifying intracranial aneurysms, the present limitations of MRA require that it be used cautiously in this patient population. Failure to detect an intracranial aneurysm may result in inappropriate therapy and subsequent devastating subarachnoid hemorrhage. The identi®cation of intracranial aneurysms prior to hemorrhage is of critical importance because nearly half of all patients presenting with acute subarachnoid hemorrhage will die within the ®rst 30 days of rupture. Although the morphology of intracranial aneurysms is best delineated by 3D time-of-¯ight MRA, additional information regarding the ¯ow dynamics of the aneurysm can be obtained by using phase contrast techniques57 as shown in Figure 20. A useful approach is to obtain a 2D phase contrast angiogram in a plane that pro®les the aneurysm and parent vessel. This is conveniently done by reviewing a 3D time-of-¯ight angiogram. The velocity encoding of the phase contrast angiogram can be selected to emphasize the faster in¯ow jet or the slower circulating central vortex ¯ow. Another advantage of the phase contrast angiogram is that high signal intensity thrombus does not appear in the angiographic image, thus better delineating the residual lumen. In addition, time-resolved cine phase contrast angiograms can be acquired to de®ne variations in the size and con®guration of the aneurysm during the cardiac cycle (Figure 21). Occasionally, a portion of the aneurysm wall will bulge during systole, suggesting that this portion of the aneurysm wall may be more likely to rupture. Thus phase contrast

Figure 19 Occipital lobe AVM. (a) The phase contrast MRA study demonstrated an enlarged angular branch of the middle cerebral artery and enlargement of the right posterior cerebral artery. There are dilated cortical veins overlying the vascular malformation, with super®cial venous drainage extending into the transverse sinus (arrow). (b) The 3D time-of-¯ight study was obtained without intravenous contrast material. The nidus of the vascular malformation is identi®ed (arrow) and the posterior cerebral and middle cerebral artery feeders are clearly delineated

sÿ1). By generating images that emphasize fast and slow velocities it is possible qualitatively to assess blood ¯ow within the AVM. Additional advantages of this approach are that the arterial supply is well de®ned at the higher velocity encodings

Figure 20 Giant middle cerebral artery aneurysm. (a) The axial 2D phase contrast MR angiogram demonstrates slow vortical ¯ow within the aneurysm (velocity encoding = 20 cm sÿ1). (b) The axial 3D timeof-¯ight angiogram displays only the high velocity in¯ow jet (arrow); the slow ¯ow within the aneurysm is not visible due to saturation

PHASE CONTRAST MRA

13

Phase contrast angiograms can also be obtained in the axial plane to detect ¯ow in the lateral sinuses and jugular veins. In this projection, the con¯uence of the dural sinuses is well visualized as well as the transverse and sigmoid segments of the lateral sinus. Phase contrast angiography has several desirable features that favor successful imaging of the dural sinuses. Phase contrast angiograms are sensitive to ¯ow in all directions facilitating imaging of complex geometries such as the sigmoid sinus. There is also better visualization of inplane ¯ow due to the subtraction of background tissue. Most importantly, high signal intensity thrombus is subtracted from the ¯ow image along with the other stationary soft tissues. The result is an MR venogram that is sensitive to slow ¯ow and only displays moving blood.

7

CONCLUSIONS

Cardiovascular disease is the leading cause of death in the industrialized world. Consequently, considerable medical resources are employed each year in its diagnosis and treatment. Before the advent of magnetic resonance imaging, diagnosticians relied primarily on invasive X-ray methods to visualize vascular anatomy. Unfortunately, X-ray methods are uncomfortable and present a risk to patient health. A frequently used alternative to X-ray angiography is ultrasonic imaging. While ultrasonic methods are risk-free, they are limited to a few clinical applications, and are heavily dependent on the skill of the operator. Magnetic resonance angiography has become an important imaging modality for the diagnosis of vascular disease because it overcomes many of the limitations of X-ray and ultrasonic imaging methods. For example, MRA presents the patient with the same degree of risk as conventional MRI. Unlike

Figure 21 Change in aneurysm size during the cardiac cycle. (a) The cine 2D phase contrast image was obtained during systole and represents one of 16 images obtained during the cardiac cycle. Note the high signal intensity in¯ow jet (arrow). (b) During diastole the slow ¯ow along the wall of the aneurysm can be appreciated (arrow)

MRA plays an important adjunctive role in 3D time-of-¯ight MRA for the presurgical workup of patients with intracranial aneurysms. 6.3.4

Cerebral Venous Thrombosis

Complex difference 2D phase contrast angiography is a fast and effective method for imaging blood ¯ow within the dural sinuses. The phase contrast angiogram can be made sensitive to venous ¯ow rates by selecting a relatively low velocity encoding (e.g. 20 cm sÿ1) as illustrated in Figure 22. Clinically useful images that clearly demonstrate the dural sinuses can be obtained in approximately 4 min. A midline sagittal phase contrast angiogram displays ¯ow in the sagittal sinus as well as the internal cerebral veins, vein of Galen and straight sinus.

Figure 22 Sagittal sinus thrombosis and recanalization. (a) The sagittal phase contrast angiogram obtained at a velocity encoding of 30 cm sÿ1 reveals absence of flow in the sagittal sinus and straight sinus (arrowheads). (b) The follow-up phase contrast angiogram obtained 1 month following anticoagulant therapy demonstrates recanalization of the sagittal sinus (arrowheads). There is, however, persistent thrombus within the straight sinus (arrow). Note that the inferior sagittal sinus is visible (open arrowhead)

14 PHASE CONTRAST MRA ultrasonic methods, MRA is applicable to a wide variety of blood vessels within the body and does not require a highly skilled operator. Magnetic resonance angiography has several features that make it more suitable than X-ray angiography for many clinical situations. For example, patients with compromised renal function may be better served with MRA since the injection of Xray contrast media employed in X-ray angiography can cause renal failure. The absence of ionization radiation is an attractive aspect of MRA and is particularly important for pediatric applications. Magnetic resonance angiography also provides information that is dif®cult or impossible to obtain with X-rays. In addition to providing cross-sectional images of blood vessels, MRA can provide details of blood ¯ow dynamics and quantitative measurements of blood velocity. Perhaps the most signi®cant aspect of MRA, however, is that the technology is still young and prospects for future development in applications and techniques are excellent.

8 RELATED ARTICLES Abdominal MRA; Time-of-Flight Method of MRA; Whole Body Magnetic Resonance Angiography.

9 REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

J. R. Singer, Science, 1959, 130, 1652. O. Morse and J. R. Singer, Science, 1970, 170, 440. T. Grover and J. R. Singer, J. Appl. Phys., 1971, 42, 938. E. L. Hahn, J. Geophys. Res., 1960, 65, 776. V. Waluch and W. G. Bradley, J. Comput. Assist. Tomogr., 1984, 8, 594. P. M. Pattany, J. J. Phillips, L. C. Chiu, J. D. Lipcomon, J. L. Duerk, J. M. McNally, and S. N. Mohapaka, J. Comput. Assist. Tomogr., 1987, 11, 369. P. R. Moran, Magn. Reson. Imag., 1982, 1, 197. M. O'Donnell, Med. Phys., 1985, 12, 59. V. J. Wedeen, R. A. Meuli, R. R. Edelman, S. C. Geller, L. R. Frank, T. J. Breely, and B. R. Rosen, Science, 1985, 230, 946. L. Axel and D. Morton, J. Comput. Assist. Tomogr., 1987, 11, 31. G. A. Laub and W. A. Kaiser, J. Comput. Assist. Tomogr., 1988, 12, 377. C. L. Dumoulin and H. R. Hart, Radiology, 1986, 161, 717. C. L. Dumoulin, S. P. Souza, and H. R. Hart, Magn. Reson. Med., 1987, 5, 238. C. L. Dumoulin, S. P. Souza, M. F. Walker, and E. Yoshitome, Magn. Reson. Med., 1988, 6, 275. S. P. Souza and C. L. Dumoulin, Dyn. Cardiovasc. Imaging, 1987, 1, 126. C. L. Dumoulin, S. P. Souza, M. F. Walker, and W. Wagle, Magn. Reson. Med., 1989, 9, 139. M. A. Bernstein and Y. Ikezaki, JMRI, 1991, 1, 725. C. L. Dumoulin, S. P. Souza, R. D. Darrow, N. J. Pelc, W. J. Adams, and S. A. Ash, JMRI, 1991, 1, 399. N. J. Pelc, M. A. Bernstein, A. Shimakawa, and G. H. Glover, JMRI, 1991, 1, 405. R. Hausmann, J. S. Lewin, and G. Laub, JMRI, 1991, 1, 415. P. van Dijk, J. Comput. Assist. Tomogr., 1984, 8, 429. D. J. Bryant, J. A. Payne, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1984, 8, 588.

23. I. R. Young, G. M. Bydder, and J. A. Payne, Magn. Reson. Med., 1986, 3, 175. 24. D. A. Feinberg, L. E. Crooks, P. Sheldon, J. Hoenninger 3rd, J. Watts, and M. Arakawa, Magn. Reson. Med., 1985, 2, 555. 25. J. Hennig, M. Muri, P. Brunner, and H. Friedburg, Radiology, 1988, 166, 237. 26. S. P. Souza, F. L. Steinberg, C. Caro, C. Dumoulin, and E. K. Yucel, Proc. VIIIth. Ann Mtg. Soc. Magn. Reson. Med., Amsterdam, 1989, p. 102. 27. C. L. Dumoulin, S. P. Souza, C. J. Hardy, and S. A. Ash, Magn. Reson. Med., 1991, 21, 242. 28. M. C. Henry-Feugeas, I. Idy-Peretti, B. Blanchet, D. Hassme, G. Zannoli, and E. Schouman-Clarys, Magn. Reson. Imag., 1993, 11, 1107. 29. G. L. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 30. J. F. Debatin, R. H. Ting, H. Wegmuller, F. G. Sommer, J. O. Fredrickson, T. J. Brasnen, B. S. Bowmen, B. D. Myers, R. J. Herfkens, and N. J. Pelc, Radiology, 1994, 190, 371. 31. K. C. P. Li, W. S. Whitney, C. H. McDonnell, J. O. Fredrikson, N. J. Pelc, R. L. Dalman, and R. B. Jeffrey, Jr., Radiology, 1994, 190, 175. 32. M. B. M. Hofman, M. Kouwenhoven, M. Sprenger, A. C. van Roseum, J. Volk, and N. Westerhof, Magn. Reson. Med., 1993, 29, 648. 33. W. A. Wagle, C. L. Dumoulin, S. P. Souza, and H. Cline, Am. J. Neuroradiol., 1989, 10, 911. 34. C. A. Anderson, R. R. Edelman, and P. A. Turski, `Clinical Magnetic Resonance Angiography', Raven Press, New York, 1993. 35. W. Lin, E. M. Haacke, A. S. Smith, and M. E. Clampitt, JMRI, 1992, 2, 277. 36. G. Marchal, H. Bosmans, and S. McLachlan, in `Magnetic Resonance Angiography', eds. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottschalk, Saunders, Philadelphia, 1992, p. 305. 37. J. R. Pernicone, J. E. Siebert, E. J. Potchen, A. Pera, C. L. Dumoulin, and S. P. Souza, Am. J. Roentgenol., 1990, 155, 167. 38. F. Nussel, H. Wegmuller, and P. Huber, Neuroradiology, 1991, 33, 56. 39. D. K. Kido, J. B. Barsotti, L. Z. Rice, B. M. Lottenberg, R. J. Panzer, S. P. Souza, and C. L. Dumoulin, Neuroradiology, 1991, 33, 48. 40. P. Vock, F. Terrier, H. WegmuÈller, F. Nahler, P. Gertsch, S. P. Souza, and C. L. Dumoulin, Br. J. Radiol., 1991, 64, 10. 41. C. L. Dumoulin, F. L. Steinberg, E. K. Yucel, and R. J. Darrow, J. Comput. Assist. Tomogr., 1993, 17, 328. 42. F. L. Steinberg, E. K. Yucel, C. L. Dumoulin, and S. P. Souza, Magn. Reson. Med., 1990, 14, 315. 43. P. Lanzer, D. Bohning, J. Groen, G. Gross, N. Nanola, and G. Pohost, Magn. Reson. Med., 1990, 15, 372. 44. L. R. Caplan and S. M. Wolpert, Am. J. Neuroradiol., 1991, 12, 593. 45. H. J. M. Barnett, N. Engl. J. Med., 1991, 325, 445. 46. S. Warach, W. Li, M. Ronthal, and R. R. Edelman, Radiology, 1992, 182, 41. 47. D. D. Blatter, D. L. Parker, S. S. Ahn, A. L. Bahr, R. O. Robinson, R. B. Schwartz, F. A. Jolesz, and R. S. Bayer, Radiology, 1992, 183, 379. 48. J. E. Heiserman, B. P. Drayer, P. J. Keller, and E. K. Fram, Radiology, 1992, 185, 667. 49. L. R. Caplan, J. Am. Med. Assoc., 1991, 266, 2413. 50. D. Crosby, P. Turski, and W. Davis, Neuroimaging Clin. North Am., 1992, 2(3), 509. 51. T. J. Masaryk and N. A. Obuchowski, Radiology, 1993, 186, 325. 52. J. F. Polak, R. L. Bajakian, D. H. O'Leary, M. R. Anderson, M. C. Donaldson, and F. A. Jolesz, Radiology, 1992, 182, 35. 53. A. M. Masaryk, J. S. Ross, M. C. DiCello, M. T. Modic, L. Paranandi, and T. J. Masaryk, Radiology, 1991, 179, 797.

PHASE CONTRAST MRA 54. C. Anderson, D. Saloner, R. E. Lee, V. S. Growidel, L. G. Shapeero, J. M. Rapp, S. Nogarkar, X. Pan, and G. A. Gooding, Am. J. Neuroradiol., 1992, 13, 989. 55. J. E. Heiserman, B. P. Drayer, E. K. Fram, P. S. Keller, C. R. Birol, J. A. Hodak, and R. A. Flom, Radiology, 1992, 182, 761. 56. W. F. McCormick, in `Intracranial Arteriovenous Malformations', eds. C. B. Wilson and B. M. Stein, Williams, and Wilkins, Baltimore, 1984. 57. J. Huston, D. A. Rufenacht, R. L. Ehman, and D. O. Wiebers, Radiology, 1991, 181, 721.

15

Biographical Sketches Charles L. Dumoulin. b 1956. B.S., 1977; Ph.D., 1981, Florida State University. Adjunct Faculty in Chemistry at Syracuse University, 1981±1984. Research scientist at General Electric Research and Development Center 1984±present. Associate Professor of Radiology, Albany Medical College, 1987±present. Approx. 60 publications. Research interests include analytical chemistry, NMR spectroscopy, MRI, and magnetic resonance ¯ow measurement and imaging. Patrick A. Turski. b 1950. B.S., 1971, chemistry, University of Illinois; M.D., 1975, Rush Medical College, Chicago, IL. Internship in internal medicine and radiology residency at the University of Wisconsin. Training in Neuroradiology at the University of California, and the Foundation Rothschild Cancer Center, Paris. Currently Professor of Radiology and Neurosciences, University of Wisconsin Medical School and Chief of the Neuroradiology Section at the University of Wisconsin Hospital and Clinic.

1

TIME-OF-FLIGHT METHOD OF MRA

Time-of-Flight Method of MRA

s

Gerhard Laub Siemens AG., Erlangen, Germany

vth

v

Figure 2 Relative signal enhancement of in¯ow enhancement as a function of blood ¯ow velocity. As velocity increases, the signal gradually increases to reach a plateau beyond which no further signal enhancement occurs

1 BASIC PHYSICS Most of the techniques currently used for magnetic resonance angiography (MRA) are based on one of the following basic physical principles: time-of-¯ight and phase shift effects. Time-of-¯ight (TOF) is related to the macroscopic motion of spins with a different history into the imaging volume.1,2 Phase shift effects are based on the motion of spins along the directions of the magnetic ®eld gradients which are used for the spatial encoding of the spins.3,4 Both effects are incorporated into imaging sequences for the purpose of ¯ow visualization. Figure 1 demonstrates the idea of in¯ow enhancement which is the basic effect for TOF MRA. The blood ¯ow is assumed to be perpendicular to the imaging plane [or volume in the case of a three-dimensional (3D) study]. For repetition times shorter than the longitudinal relaxation time T1, the signal of tissue within the slice will be reduced due to partial saturation effects. Blood ¯ow will move spins from outside the slice which have not been subjected to the spatially selective rf pulses into the imaging volume. These unsaturated or fully relaxed spins have full equilibrium magnetization and, therefore, upon entering the slice, will produce a much stronger signal than stationary spins assuming that a gradient echo sequence is applied to avoid out¯ow effects otherwise seen with spin echo sequences.5,6 This effect has also been referred to as the `entry slice phenomenon'.

A simple model helps to get a more quantitative aspect of in¯ow enhancement. According to Figure 1 a critical velocity vth can be de®ned as vth ˆ s=TR

Spins ¯owing at velocities smaller than vth will see several rf pulses depending on their actual speed, and consequently, the longitudinal magnetization will decay. Accordingly, less signal will be produced. With velocities larger than vth the spins will see only one rf pulse, resulting in maximal signal enhancement. The overall signal enhancement as a function of the spin's velocity is plotted in Figure 2. The second ¯ow effect, ¯ow-induced phase shifts, is a consequence of the phase memory of the spin system. When traveling along the direction of magnetic ®eld gradients, the excited spins will experience additional phase shifts which depend on the motion of spins. Techniques which make use of the ¯ow-induced phase shifts are usually referred to as phase contrast MRA.7,8 In the case of TOF MRA phase effects are eliminated as much as possible by using a ¯ow-compensated gradient waveform in combination with short echo times as shown in Figure 3. Both the slice select and the readout gradients are ¯ow-compensated to minimize ¯ow-induced phase shifts. The signal from blood ¯owing into the slice is maxia

TR << T1

a

…1†

TE

Saturated spins Gslice

Ds = v • TR Flow

1

v Gphase Unsaturated spins

Vessel Np Slice Gread

Figure 1 Principle of in¯ow enhancement. The longitudinal magnetization Mz of spins in the imaging slice is reduced due to partial saturation for TR << T1. In¯owing spins have equilibrium magnetization Mo and, therefore, will produce more signal intensity. The displacement of the spins for each TR interval is
s

Figure 3 Typical timing diagram of a ¯ow-compensated gradient echo sequence. Asymmetric echo sampling is used for shorter echo times

2 TIME-OF-FLIGHT METHOD OF MRA

I Imax

Figure 5 Schematic representation of the maximum intensity projection (MIP) method

Figure 4 Four images acquired at different positions in the neck. A FLASH sequence with TR/TE/a = 30 ms/9 ms/40 is used to saturate stationary spins in the slice and produce high signal from in¯owing blood. Slice thickness is 2.5 mm with 0.5 mm overlap, and a total of 60 slices is acquired to cover a section of 120 mm of the neck vessels

mized by a proper selection of sequence parameters. In most cases a pulse repetition time TR of 30±40 ms along with tip angles of 40 results in suf®cient saturation of biological tissues, as demonstrated in Figure 4. According to equation (1) thin slices help to avoid spin saturation within the slice in the case of slow blood ¯ow. For a TR of 30 ms, and a slice thickness of 3 mm, a blood ¯ow velocity of 10 cm sÿ1 is suf®cient to replace all of the spins in the slice within one TR interval, resulting in good vessel/background contrast. Additional presaturation pulses are applied on either side of the slice to suppress either arterial or venous in¯ow enhancement.

must be a 3D data set in which the structures to be extracted are associated with a characteristic range of signal intensity levels. In this case a projective image can be calculated by penetrating the data volume with a set of parallel projection rays and selecting along each of these rays only the data point that represents the intensity maximum as demonstrated in Figure 5. The in¯ow enhancement and proper pulse sequence parameters (¯ip angle, pulse repetition time, and ¯ow compensation on) ensure that the maximum intensity is always associated with a blood vessel, as long as the projection ray intersects at least one. All of the other projection rays will just pick up a background pixel out of the 3D data set. As a result, Figure 6 demonstrates a complete projection image calculated from one 3D data set.

2 POSTPROCESSING OF 3D DATA SETS In principle, with this technique any vessel segment can be imaged by cutting through the vessel perpendicular to the ¯ow direction. With repetitive increments of the slice position a 3D data set of the complete vessel tree can be measured. (see Figure 4 for example). For the observer, this form of representation requires experience in order to obtain the correct 3D spatial impression. Obviously, postprocessing methods should be used to extract two-dimensional (2D) projections of vessel structures from the 3D volume information. With these methods spatial impressions can be obtained in two ways: by showing a sequence of projective images with different projection angles, or by coding of the depth information onto the surface of the displayed objects.9,10 Since the surfaces of most vessels are relatively small, the ®rst method, multiple projections with different angles, has proven more useful in practice. The starting point for this method

Figure 6 Calculated MIP from the 3D data set of the carotid artery. Sixty slices were acquired in sequential order and transverse orientation with 160  256 matrix and 3/4 rectangular ®eld-of-view. Total acquisition time was 4 min 48 s for all 60 slices

TIME-OF-FLIGHT METHOD OF MRA

3

Mx v

a = 20° (const.) 0.3 a = 10° … 30° Vessel

1

0.2

N

Figure 7 Three-dimensional fast FFT imaging techniques in TOF MRA. The whole volume of slab thickness D is excited by an rf pulse, and is then subdivided into N partitions by the use of an additional phase-encoding gradient. The contrast of blood vessels is maximal at the entrance plane of the imaging volume, and decreases constantly toward the exit plane due to spin saturation

Stat. signal 0.1 TR = 40 ms T1 = 1000 ms 0.0

By varying the projection angle multiple projective images can be obtained retrospectively which allow the observer to obtain the correct spatial impression of the 3D information. By displaying a number of projections with projection increments of only a few degrees in a rapid fashion, the impression of a continuously rotated object will be generated which allows a correct 3D visualization of such complex structures as a vessel tree. Improvement of the spatial resolution along with an improvement of the signal-to-noise ratio can be achieved with 3D imaging techniques. As shown in Figure 7, the whole volume is excited simultaneously, and will then be subdivided into thin partitions or slices by using an additional phaseencoding scheme in the slice select direction. Unlike 2D imaging, where the slice resolution is de®ned by the excitation pro®le of the radiofrequency pulse, the slice resolution is de®ned by spatially-encoding magnetic ®eld gradients and can be less than 1 mm. In 3D, or volume imaging, complete compensation of the ¯ow-induced phase shifts is also necessary to avoid ¯ow artifacts in the form of signal losses and ghosting.11,12 For this purpose ¯ow compensation is applied in conjunction with short echo times similar to 2D techniques. The acquisition time TA for a 3D data set is de®ned as TA ˆ TR
Np
Ns

…2†

where TR denotes the pulse repetition time of the sequence, and Np, Ns are the number of phase-encoding steps in the phase-encoding and slab select directions, respectively. In most of the clinical protocols the repetition times vary between 20 and 40 ms and the typical matrix size is 192  256 with a 34 rectangular ®eld of view and 64 partitions along the slab select direction, resulting in acquisition times of between 4 and 8 min. One of the major limitations of 3D TOF MRA is the loss of vessel contrast as spins penetrate into the imaging volume. This effect is due to progressive saturation when spins experience the rf excitation pulses in the imaging volume. As can be seen from Figure 8 the transverse magnetization decreases with the number of rf pulses, and the difference between blood and stationary tissue is reduced rapidly. A more effective use of the magnetization is possible when using variable ¯ip angles across the slab.13 At the entrance plane when spins enter with equili-

0

10

20

30

No. of rf pulses

Figure 8 Transverse magnetization of spins as a function of rf pulses. For constant ¯ip angles throughout the imaging volume there is a constant decrease until the magnetization has approached its steady state value. In the case of a variable ¯ip angle the signal reaches its maximum after about 12 excitations. As a result the contrast of blood is maintained over a longer distance in the imaging volume. Stat = stationary

brium magnetization a relatively small ¯ip angle of, for example, 10 will still provide suf®cient signal, without reducing the longitudinal magnetization too much. Next, the ¯ip angle is increased a little to maintain, or even increase, the signal from blood spins on their way through the volume, and so forth. In principle, it is possible to shape the ¯ip angle distribution over the entire imaging volume according to the speci®c ¯ow velocity and vessel coverage. A further improvement is possible by the application of magnetization transfer pulses (magnetization transfer contrast, MTC).14 The idea is to use off-resonance rf pulses which do not directly affect the mobile protons which usually create the signal in MRI. Protons with restricted mobility, however, do get saturated, and because of magnetization or chemical exchange processes some of the magnetization will be transferred in some biological tissues, such as gray or white matter, resulting in a partial saturation. Blood will not be affected by the MTC pulses, and as a result, there will be more contrast between blood and background. This technique is particularly useful in the brain as demonstrated in Figure 9, as the background signal from gray and white matter tissue is signi®cantly reduced. 3

DISCUSSION

The results of preliminary clinical studies indicate that this MRA technique can provide accurate, reproducible ¯ow images in different anatomical regions of the body.15 Applications of TOF MRA for the assessment of intracranial vascular disease in the brain are shown in Figure 10. This technique is noninvasive and can be performed in conjunction with 2D spin echo or

4 TIME-OF-FLIGHT METHOD OF MRA that this capability may increase both the sensitivity and the speci®city of MRA in some patients with cerebrovascular disease when MRA is performed in conjunction with routine spin echo imaging of the brain. While the acceptance of the 3D in¯ow MRA technique will depend on its sensitivity and speci®city in disease diagnosis, as determined by means of prospective clinical trials relative to other screening modalities, the preliminary clinical results suggest that this MRA technique can serve as a screening tool for the identi®cation of normal vasculature as well as stenosis and/or occlusions produced by atherosclerotic disease. It can be performed in conjunction with MRI of the brain, with only a small increase in examination time, and provides both vascular and parenchymal evaluation of cerebrovascular disease in a single setting.

4

RELATED ARTICLES

Head and Neck Studies Using MRA; Phase Contrast MRA; Whole Body Magnetic Resonance Angiography. Figure 9 Maximum intensity projection image of the intracranial vasculature. The images were acquired with a 3D FISP sequence including MTC partial saturation and TONE (TR/TE/a = 40 ms/6.5 ms/ 10±30 ). Total scan time was 8±20 min

gradient echo imaging, with only a 5±10 min extension of the examination time. The unique advantages of this technique include the capacity to provide multiple projections of anatomically complex vascular abnormalities with a single data acquisition. It appears

5

REFERENCES

1. W. S. Hinshaw, P. A. Bottomley, and G. N. Holland, Nature (London), 1977, 270, 722. 2. L. Axel, Am. J. R., 1984, 143, 1157. 3. E. L. Hahn, Phys. Rev., 1950, 80, 580. 4. P. R. Moran, Magn. Reson. Imag., 1982, 1, 197. 5. F. W. Wehrli, A. Shimakawa, G. T. Gullberg, and J. R. MacFall, Radiology, 1986, 160, 781. 6. L. E. Crooks, C. M. Mills, P. L. Davis, M. Brant-Zawadzki, J. Hoenninger, M. Arakawa, J. Watts, and L. Kaufman, Radiology, 1982, 144, 843.

Figure 10 Clinical applications of TOF MRA. Left: arteriovenous malformation. Right: angioma. Both images were acquired with a 3D FISP sequence in 9 min. Axial MIPs are shown to demonstrate the vascular lesions

TIME-OF-FLIGHT METHOD OF MRA 7. L. Axel and D. Morton, J Comput. Assist. Tomogr., 1987, 11, 31. 8. C. L. Dumoulin, S. P. Souza, and M. F. Walker, Magn. Reson. Med., 1989, 9, 139. 9. G. A. Laub, Magn. Reson. Med., 1990, 14, 222. 10. H. E. Cline, C. L. Dumoulin, W. E. Lorensen, H. E. Clini, C. L. Dumoulin, W. E. Lorensen, S. P. Sonzo, and A. J. Adams, Magn. Reson. Med., 1991, 18, 384. 11. T. J. Masaryk, M. T. Modic, J. S. Ross, P. M. Ruggier, G. A. Laub, G. W. Lenz, E. M. Haacke, W. R. Selmon, M. Wiznitzer, and S. I. Sarik, Radiology, 1989, 171, 793. 12. T. J. Masaryk, M. T. Modic, P. M. Ruggieri, J. S. Ross, G. Loub, G. L. S. Kenz, J. A. Tkach, E. M. Haacke, W. R. Selmon, and S. I. Hank, Radiology, 1989, 171, 801. 13. D. Atkinson, M. Brant-Zawadzki, and G. Laub, Radiology, 1994, 190, 890.

5

14. S. D. Wolff and R. S. Balaban, Magn. Reson. Med., 1989, 10, 135. 15. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottschalk, `Magnetic Resonance Angiography: Concepts and Applications', Mosby, St. Louis, MO, 1993.

Biographical Sketch Gerhard A. Laub. b 1951. Ph.D., 1982, University of Stuttgart, Germany. Faculty of Physical Electronics, University of Stuttgart, 1982± 1985. MR applications development, Siemens Medical Systems, Erlangen, 1986±present. Research interests include the theory and application of ¯ow phenomena in MRI, postprocessing of MR data, and fast imaging techniques.

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

Whole Body Magnetic Resonance Angiography E. Mark Haacke, Weili Lin, and Debiao Li Washington University, St. Louis, MO, USA

1 INTRODUCTION The ®rst seeds of whole body magnetic resonance began with the simple concept of spatial encoding using gradients in Paul Lauterbur's seminal 1973 paper.1 It only took several years for it to be implemented on systems large enough for the human body, and developed rapidly owing to the several decades of previous study in chemistry and physics. Many of the concepts from steady-state free precession2 to adiabatic fast passage3 would play a role in the understanding of how to deal with motion [usually of blood, but equally simply cerebrospinal ¯uid (CSF) ¯ow] during a magnetic resonance experiment. The key to the present day success of MR angiography (MRA) lay in the discovery of how moving spins could be motion compensated.4±8 The advent of faster and faster computers and the role of image processing were key to the advances in MRA.9 Although the basic concepts of magnetic resonance whole body ¯ow imaging are well understood today, there is still room for basic evaluation and applications from the biological to clinical sciences. In this short article on whole body ¯ow imaging, we will review the history behind bulk ¯ow studies in the eras of preimaging and during imaging to show how technological and scienti®c advances led to the formation of dealing with motion as we understand it today. We begin by studying the effects of motion on the signal in terms of amplitude and phase, how it could be dealt with or corrected, the ongoing debate of whether phase contrast or time-of-¯ight (TOF) imaging is to be preferred, and what future directions remain to be investigated. This overview should serve as a good introduction to the more technical aspects of each of these areas. 2 THE EFFECTS OF MOTION The ®rst measurements and theory for magnetic resonance were based on the assumption that the spins were stationary. It was several years before the issues of motion were ®rst addressed by Suryan10 in 1951, and several more years before the ¯urry of papers appeared in the mid-1950s on diffusion.11 At this time a real effort was made to incorporate the effects of diffusion on the signal measured, and it was not long before the theory for the equations of motion were modi®ed to deal with these phenomena.12 Bulk effects in blood were considered toward the end of the decade by Singer13 and Bowman,14 who could be credited with the ®rst TOF magnetic resonance experiments on blood. This TOF effect relates to the fact that the in¯owing blood appeared brighter than it would have had it remained in the excitation region, and, hence, it was viewed as

1

having an effective T1 smaller than that of stationary blood. It must be remembered that all the experiments evaluating the signals were from projections over the entire excited region (whether it was a small vial of material or a mouse tail) and that no imaging per se was done at that time. This fact had severe consequences on the subsequent development of ¯ow methods and the ability to assess accurately what was happening. Nevertheless, brave attempts were made to detect and quantify ¯ow. The effects were well enough understood from the amplitude changes that it was possible to design a ¯owmeter using magnetic resonance equipment.15 Two papers have signi®cant historical relevance. The ®rst is that by Hahn in 1960, where he recognized the fact that motion through a gradient would lead to phase dispersion at the echo.16 For a simple bipolar pulse (peak gradient amplitude, G, negative lobe of duration , positive lobe following of duration , constant velocity ¯ow v) the phase at the echo is Gv 2 where is the gyromagnetic ratio. The key points to extract from this simple formula are the linear behavior with v and the quadratic behavior with time. The second key paper in retrospect is that of Packer17 in 1969 (later rediscovered by Waluch and Bradley18). He investigated the signal behavior when multiple pulses were used in the presence of gradients. He discovered the most important fact that the ¯owing spins rephased at the even echoes. This was the precursor to future imaging papers which rediscovered this phenomenon4±8 and others which extended the concept to correct for the fact that motion during the gradients led to a phase change. The ®rst was subsequently revisited in the imaging world by Moran19 in 1982 and the second by Constantinesco4 in 1984. These two papers set the foundation for future advances in the ®eld of MRA. 2.1

Time-of-Flight Effects

Early experiments with spin echo imaging noticed the `¯ow void' effect or lack of signal from blood. This was at the time considered anomalous (or paradoxical) enhancement since blood has a spin density of 0.8 relative to water. It was quickly realized that those spins that did not see both the 90 and 180 pulses were not refocused and did not contribute to the signal. This is known today as the wash-out effect, and it depends very much on the direction of the ¯ow relative to the slice acquisitions and the ¯ow rate itself. The same phenomenon using gradient echoes leads to a wash-in effect where stationary tissue is saturated and in¯owing blood is brighter than naively anticipated. This concept dominates today for much of MRA methodology to enhance the signal from moving blood. For a fast (short TR) gradient echo sequence, when a 2D slice or plane is excited, blood which was excited ¯ows out of the slice and sees no more rf pulses during the experiment (or at least we will assume so, an assumption which is usually valid). For 2D imaging and constant blood ¯ow, the blood will be imaged at the same location in each case. The signal seen will then be a sum over all sections of the blood each of which has seen a different number of rf pulses. For a 3D scan, the signal will be that of the blood in a given section which has seen that number of rf pulses associated with the time from its entry into the 3D volume and the repeat time.20 Speci®cally, the number of rf pulses is n = [t/TR] where t = d/v with d the distance traveled, v the speed and [ ] the integer part of the argument.

2 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

2.2 2D versus 3D TOF From the above arguments, it is clear that thin slices are best to obtain maximum signal for slow ¯ow21 (for example, with TR = 40 ms and a 2 mm thick slice, ¯ow greater than 5 cm sÿ1 presents with a maximum signal). This is still considered the best way to image veins in the leg.22 In these images, both arteries and veins will be bright. To avoid confusion between the two, a saturation pulse is usually applied upstream to saturate arteries or downstream to saturate veins (the latter is used when imaging the carotid arteries in the neck). However, 2D TOF suffers from longer echo times (although some cures for this may exist through higher gradient strengths23 or rf design24), and a longer acquisition time for thin slices due to the need to overlap slices for full spatial coverage. Saturation pulses can saturate retrograde ¯ow, and through-plane resolution is not as good as a 3D acquisition. On the other hand, slow ¯ow is easily seen and a small region of interest can be rapidly covered. The use of 3D imaging overcomes many of the above problems as it allows for short echo times and thin contiguous slices. It has the problem that blood appears nonuniformly throughout the region of interest because of the spin saturation. This is partly overcome using a variable ¯ip angle spatially25 (obtained by applying an rf pulse designed to increase along one direction, usually the direction of ¯ow, so that saturation effects are reduced; this method is referred to as TONE or tilted optimized nonsaturating excitation), multiple thin slabs,26 magnetization transfer saturation or MTS,27,28 and contrast agents.29,30 The use of MTS is good because it does not require injection of a contrast agent and it can suppress the background brain parenchyma27,28 by about 25% or myocardium31 by up to 50% while leaving the signal from blood virtually intact. MTS pulses have two problems: they add more power deposition to the sequences so care must be taken in their design and implementation, and they do not suppress lipid signal. The advantages of using a contrast agent are that the scan can be run faster, or better resolution can be obtained, or a larger ®eld-of-view (FOV) can be obtained without saturation effects. The problems are that both arteries and veins become bright, and the signal from regions of the blood±brain barrier breakdown can obscure vessel information; for both these reasons, vessel tracking is needed32±34 to separate the vessels of interest from other vessels or tissues.

1.0 TR/T1 = 50/1200

n=1

Signal amplitude

0.8

0.6 n=2 0.4 n=5 n = 10 0.2 n = inf 0.0 0

10

20

30

40 50 Flip angle

60

70

80

90

Figure 1 Blood signal as a function of ¯ip angle for different numbers of rf pulses (n) for T1/TR = 24

2.3

Phase Effects

The presence of motion during the application of a gradient pulse causes a phase shift of the signal from the moving tissue at the echo. For a simple bipolar pulse of duration and amplitude G, the phase is simply Gv 2 where v is the speed along the direction of the gradient G. In fact, for any symmetrically placed gradient pulses about a 180 pulse at time t' [i.e. a time reversed pulse where G(t') = G(2t' ÿ t) for t > t'], the phase for constant velocity spins will be zero at t = 2t'. This shift in phase is not necessarily a problem. For constant velocity plug ¯ow for every phase encoding step (where the ¯ow is only along the read direction), the ®nal image in the region of the ¯ow will have a nonzero phase as described above and the correct amplitude on a magnitude image. In practice, for nonpulsatile laminar ¯ow, spins develop a different phase as a 1.0 TR/T1 = 50/300

n=1

0.8 Signal amplitude

For blood which ¯ows through the entire slice, each new rf pulse excites unsaturated or fresh blood, and hence this blood has the maximum possible signal. For a ¯ip angle of 90 , the signal is a maximum at 1.0 units relative to the spin density of the blood. Figures 1 and 2 show the expected signal response for two different T1 values for blood, the ®rst of 1200 ms for pristine blood and the second for 300 ms for blood doped with a suf®cient amount of T1 reducing contrast agent. From these curves we see two most important features for imaging blood. The ®rst is that at low ¯ip angles of about 15±20 the best contrast between blood and stationary tissue is achieved at 1.5 T even for fairly slow ¯ow (hence most TOF imaging in the head is done with these low values). The second is that the use of contrast agents can dramatically enhance the contrast between even stationary blood and background tissue (assuming that gray matter is the background tissue with a T1 of about 950 ms at 1.5 T, quite close to that of blood).

0.6

n=2

0.4

n=5 n = 10 n = inf

0.2

0.0 0

10

20

30

40 50 Flip angle

60

70

80

90

Figure 2 As in Figure 1 but for T1/TR = 6 for blood with a triple dose of gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA)

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

function of position, and the signal in a voxel is reduced accordingly. For example, in a 1D analogy, if the ¯ow varies linearly across a voxel, the signal will be reduced to (sin )/ of its original value for a phase dispersion of 2. For a phase dispersion of 180 , the signal will fall to 2/ of its initial value (only a 30% signal loss). Both velocity and acceleration can create large velocity gradients in blood ¯ow (in areas near stenoses or curved ¯ow). Interestingly, the loss of signal has been recognized quite early, and in spin echoes was partly due to the out¯ow of the blood between the 90 and 180 pulses, and created a ¯ow void. Today the dephasing and in-¯ow/out-¯ow phenomena are also used to create what are referred to as `black blood' images. These are often very useful in delineating vessel wall. The disadvantage of this method is the poorer contrast-to-noise ratio of blood versus background in rapid imaging. The in¯ow enhancement (wash-in effect) of the bright blood methods offers a potentially much higher contrast-tonoise ratio between vessel and surrounding tissue. Even for plug ¯ow, phase is a problem if the ¯ow is pulsatile. Now the phase changes from line to line and creates what is referred to as ghosting. This can be understood as a periodic change in signal which creates a periodic effect on the sampling and hence creates aliasing of the image. The graphs Figures 1 and 2 indicate how severe a problem this is when the ¯ip angle is large, as now the signal jumps rapidly depending on how many rf pulses the blood sees (and therefore on how fast the blood is ¯owing20). Clearly, this problem is avoided at low ¯ip angles where the signal becomes more spin density weighted and the differences in ¯ow are less observable. Pulsatility can be handled quite well by triggering off the cardiac cycle. Now if a bipolar pulse is added to a conventional sequence and G is chosen so that the phase changes by less than  for all velocities, it is easy to use just the phase image to determine the blood velocity in the direction of the gradient. In principle, three such experiments must be run to determine the full velocity vector. It then becomes possible to predict from the sequence structure and the geometry how many rf pulses would be seen during the experiment and therefore what the magnitude of the signal should be. This would allow for a self-consistency check between phase and magnitude of the images. Little work has yet been done in this direction, although when it is accomplished it will have brought together the initial historical focus on amplitude ¯ow imaging discussed in the introduction with the ¯ow quanti®cation efforts begun after the advent of MRI.

3 MOTION COMPENSATION The real beginnings of MRA came with the recognition that there were effects on the signal from ¯ow. These were only brought under control by the redesign of the gradient structure to ®x the phase at the echo to be zero even in the presence of motion. First-order motion compensation refers to compensating for constant velocity ¯ow4±8 so that the phase at the echo is zero. This is effected by using at least three gradient lobes in both the read and slice select direction so that the zeroth and ®rst moment of the gradients is zero at the echo. Higher order terms such as acceleration and jerk are then usually not zero (although even order terms can be in some cases35,36). The choice of timings and

3

amplitudes makes this a complicated optimization problem.37 It has been suggested20 and shown38,39 that the best choice to make in the design of these sequences when acceleration effects are present is to keep the duration of the gradients as short as possible. In an uncompensated sequence, acceleration effects grow as time cubed, but in a velocity compensated sequence they grow as only time squared. This is still fast enough to warrant collecting the data with an asymmetric echo40 rather than adding a fourth pulse and compensating for acceleration as well. In summary, the sequence is designed so that Z

TE

0

Gr …t† dt ˆ 0

…1†

tGr …t† dt ˆ 0

…2†

and Z

TE 0

for the readout gradient, Gr and Z T

Gpe …t† dt ˆ T
Gpe

…3†

and Z T

tGpe …t† dt ˆ 0

…4†

for the phase-encoding gradient.41 Here, T is the time from the beginning of the ®rst phase-encoding table to the end of the second phase-encoding table, and Gpe is the usual phase-encoding increment. An example 3D sequence which has partial motion compensation in all axes is given in Figure 3. This sequence is ideal for phase-encoding in that it also compensates for position shift or spin misregistration. For example, since the phase encoding gradient is shifted temporally from the echo, the spins are encoded at a different location than they are read out at and, therefore, they appear shifted in the image from their actual location. (This gives the appearance of a black hole where the vessels should have been and the vessels appear extra bright because they now lie on top of some other tissueÐa fact which has ironically proved useful for visualizing vessels.) By velocity compensating in the phase-encoding direction as discussed above, there will be no misregistration artifacts; an important point if the magnetic resonance images are to be used as a reference for neurosurgical applications.

4

FLOW MEASUREMENT

The observation of ¯ow is quite a different challenge to its measurement. Early attempts to measure ¯ow rates used the effective change in T1, as the faster the ¯ow, the less saturated it was and hence the shorter the effective T1. These amplitude approaches rarely proved useful. The concept of ¯ow to visualize and quantify blood ¯ow also began rather early. In 1959, Singer described a method of quantifying blood ¯ow in a mouse's tail by using a receiver downstream of the transmit-

4 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY rf pulse

Slice select

throughout the cardiac cycle or just continuously to watch the in¯ow of blood. This is best accomplished using a rapid 2D acquisition approach such as a short TR method or echo planar imaging with blood inverted outside the slice. In these cases, the blood signal changes from negative to positive as time progresses and the longitudinal magnetization recovers (this approach is limited by the T1 of blood to a few seconds). 4.1

Phase-encode

Readout

FE ADC

TE

Figure 3 A 3D sequence diagram showing velocity compensation at the echo in both read and phase-encoding directions. The slice select tab ®le could have added to it a second tab ®le and the partition encoding could then be velocity compensated. FE refers to the time from the onset of the read gradient to the echo. ADC refers to the analogue to digital conversion of the signal to ®nite sampled data during the acquisition period

ter.13 The time delay between the spin excitation and signal reception was used to measure the ¯ow velocity. The method was then further modi®ed and applied to human arms and ®ngers.42,43 It was used again in imaging for adiabatic excitation of in¯owing blood44 and for exciting and refocusing just blood using a 180 pulse (orthogonal to and in the direction of the blood ¯ow) in a spin echo experiment.45,46 This gave an image showing how far the blood had moved between its encoding after the 90 pulse and where it was ®nally read out. For laminar ¯ow, the fastest ¯ow moved furthest into the image, and the standard crescent shape was seen in this type of image. The same type of approach was later implemented using fast imaging sequences by employing a saturation pulse instead of a 180 pulse.47 Although this is well described as part of the same concept, in retrospect one did not follow on the heels of the other but was rediscovered from a different perspective when saturation pulses were used to eliminate signal from unwanted tissue. By nulling the blood before the conventional excitation pulse and acquiring the data in a cardiac triggered mode, the saturated blood could be followed throughout the cardiac cycle. The complement of the bright blood image using the spin echo approach was seen with the crescent shape now being dark. Another variant of this method can be used to create a bright bolus of blood by using a magnetization preparation type of sequence to null or reduce the signal from the background tissue and then dynamically to acquire the data

1D Velocity Information

One of the concerns about MRA is that it does not give functional information. In this section, different methods of giving velocity information are discussed. One method projected to give velocity information quickly is the Fourier ¯ow quanti®cation technique ®rst introduced by Moran.19 He recognized that because of the linear dependence of the phase on the velocity and the gradient that a set of experiments changing the strength of an applied bipolar gradient, for example, would phase encode the velocity in that direction. Collecting in a 1D velocity and 1D transverse spatial direction through the neck gives a nice display as a function of the cardiac cycle of the different positive and negative velocity components in all the vessels cutting across the excited region.48 The resulting image shows velocity versus position with the amplitude of the signal representing the number of spins at that location with that velocity. This can also be accomplished in a much simpler way by exciting just a plane49 or cylinder50 and projecting all information along the read axis. The ¯ow is then sensitized to one direction (usually along the slice select direction using the bipolar gradient scheme discussed earlier), and the phase is used to map directly the velocity of all spins summed over the projection. This is often used in the aorta or to see heart valves in motion. It can have a time resolution of 20 ms and does not require cardiac triggering. It has dif®culties if arteries and veins overlap as vessel phase data become confounded. We will return to this method in the future directions section. 4.2

2D or 3D Phase Contrast Imaging

The most common and best investigated approach to ¯ow quanti®cation today is phase contrast imaging pioneered by Dumoulin.51,52 By not fully compensating the blood motion, a remnant phase remains which is proportional to the velocity. Using a sequence with one bipolar pulse giving a positive phase and a second sequence with a polarity inverted bipolar pulse giving a negative phase, the subtraction eliminates stationary tissue and leaves tissue with a nonzero velocity. This is the conventional approach used in 2D imaging and goes back to the early work of Moran19 and, somewhat later, Bryant.53 However, the bipolar pulse can be applied in all three directions and not at all so that the experiment is repeated four times and full 3D velocity information can be extracted. If the gradient amplitude chosen is too small, aliasing of the phase will occur. If the resolution is high enough, the aliasing can be unwrapped, but this is rarely done today (because of the processing time and dif®culties in passing through or around regions in the image with no signal and because the phase is complicated by ®eld inhomogeneity phase effects in gradient echo imaging). This is actually the best way to collect the data in terms of the signal-to-noise ratio (S/N) as a higher phase has a higher S/N. Phase contrast images can be acquired in two or

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

three dimensions. Their main disadvantage is the length of time it takes to acquire information in all three directions and the fact that this is a necessity. 3D phase contrast imaging is less sensitive to in¯ow but still loses signal from the saturation phenomena. Nevertheless, small vessels with very slow ¯ow (much less than 5 cm sÿ1) can be seen quite well with this approach, as can CSF.54 Today the use of 2D phase contrast imaging is the most common to obtain ¯ow rates by using the velocity information orthogonal to the excited plane and a high-resolution cross-section for the vessels. Since many pixels should be available for a given area, this limits its application to vessels with diameters several times the available pixel size. A further limitation comes from the fact that a ®xed maximum phase is usually encoded for, so that any values of velocity above this maximum lead to aliasing in the image. This implies that, if there is a broad range of velocity information present, several scans would need to be run to obtain the optimal S/N.55 5 MRA OF THE ENTIRE BODY 5.1 Contrast Optimization Both TOF and phase contrast methods need to be optimized depending on the regions being imaged. The optimization is not just one of getting the best S/N, but rather the best contrast-to-noise ratio or, one step further, object visibility. Phase contrast offers the best suppression of background tissue and the best S/N when the subtraction of two phase images is performed. The lower velocities are the hardest to see as the image intensity is proportional to the velocity. Nevertheless, phase contrast can do a good job with slow ¯ow if the gradient causing dephasing is made very large. It does suffer from velocity changes in time (pulsatility) which causes ghosting (except in a cine mode where the images are gated to the cardiac cycle). TOF can give good contrast in a variety of ways. These will be discussed individually in the different body part discussions to appear below. Suf®ce it to say at this time that the major improvements in this approach come from magnetization transfer contrast and the use of contrast agents. Lastly, there are other contrast mechanisms available such as black blood56,57 and steady-state free precession imaging (to map the tissue properties of blood rather than its ¯ow58). In the latter case the challenge is to make the signal insensitive to the ¯ow and then using spin density or T1/T2 to recognize blood from background tissue. In this section we will address the different areas of the body presently covered by MRA applications. 5.2 Head and Neck MRA Currently MRA is most widely used for imaging the head and neck. Nevertheless, as mentioned above, there still remain problems. The four most dominant ones include: loss of signal from disturbed or very fast ¯ow (such as in regions of curved ¯ow or tight stenoses); spin saturation caused by stationary blood or very slow ¯ow (such as for obstructed ¯ow or ¯ow in the carotid bulb, for example); insuf®cient resolution (such as in the intracranial or carotid bifurcation studies to verify accurately if a stenosis is 75%

5

or higher; and an insuf®cient S/N (for when high resolution is used). Solutions to these problems are still under development, and for the previous four points include: the use of shorter gradient timings to generate the echo; the use of spatially varying (or TONE) pulses, MTS, fat saturation and contrast agents to improve the contrast-to-noise ratio; the use of a smaller FOV or partial Fourier reconstruction methods to improve resolution; and the designing of new coils and, again, the use of contrast agents for a better S/N. Fat saturation itself has been a major problem since it requires a very uniform ®eld to avoid saturating water instead. Without the use of segmentation methods fat otherwise blocks information in a maximum intensity projection (MIP) because of its remnant high signal (especially after the use of an MTS pulse, which suppresses gray matter/white matter in the head or muscle in the neck). Some other methods are also under development which will use many of the above improvements but whose purpose is to get `black blood' images for highlighting vessel wall to examine plaque with high resolution and a spectroscopic analysis, as well as methods for functional imaging to see dynamic local blood ¯ow with and without contrast agents. Ideally, one would like to cover a large region of interest very quickly, but this usually leads to blood saturation. The combination of the above methods would make it feasible to cover the entire head or neck in 5±10 minutes. At the moment the high contrast and high resolution aspects are accomplished using the multiple thin slab method in a transverse orientation. This latter method, the use of the MIP algorithm, and the use of contrast agents all also require the use of vessel tracking or blood segmentation postprocessing to extract the best representation of the blood vessels from the images. The multislab technique requires signi®cant overlapping of adjacent 3D slabs because of the tapering off of the rf pro®le at the slab edges. Using segmentation it becomes possible to extract the vessels from the background in most slices even given the spatial variation of signal intensity, and to digitize the images. Merging the newly processed data now eliminates the border artifacts otherwise evident in the ®nal MIP images (Figure 4). The MIP itself causes a loss in the contrast-to-noise ratio, but if applied to the segmented data over a dilated region about the vessel tracked information this becomes less of a problem and, in fact, opens the door to other types of projections such as a straight sum, for example. These methods are becoming automated, so that the user only needs to input a single seed point in most cases [this is true for the major vessels in the head (Figure 5) and neck (Figure 6), and even for the cardiac system]. Other types of displays such as surface or volume rendering can also be used to advantage, and these will be dealt with in other sections in this treatise. 5.3

Cardiovascular Imaging

The application of MRA techniques to the heart is relatively new compared to that in the head and neck, owing to the special challenges and technical dif®culties in magnetic resonance ¯ow imaging of the heart. The motion of the heart during cardiac and respiratory cycles is the major obstacle for consistent magnetic resonance images. Some kind of motion compensation is usually needed to overcome this problem. Prospective or retrospective electrocardiogram (ECG) triggering is

6 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

Figure 4 Intracranial study with a high-resolution two slab acquisition (a) without and (b) with vessel tracking. The MIP has obscure information in (a) which is brought out after vessel tracking in (b). The image parameters were: TR = 35 ms, MAT = 160  256 with a 5/8 rectangular FOV, FA = 15 and slice thickness = 1 mm with 64 partitions

used in almost all imaging cases to obtain cine images (images obtained throughout the cardiac cycle one after the other) or to minimize the cardiac motion effects. The respiratory motion effects are usually reduced or eliminated by prospective or retrospective respiratory gating, multiple signal averaging, or breath-holding. The major areas of magnetic resonance ¯ow imaging application in the heart include the coronary arteries, great vessels, cardiac chambers and valves. 5.3.1

Coronary Arteries

In addition to their movement, coronary arteries have a small size. Thus, a high S/N and spatial resolution are required to visualize the coronary arteries reliably. Further, the coronary arteries are surrounded by epicardial fat, which tends to have a strong signal. Some kind of fat saturation is needed to increase the conspicuity of the coronary arteries. Early attempts at coronary MRA used multislice, multiphase, spin echo,59 2D gradient echo cine,60,61 and inversion±recovery techniques.62 Recent efforts have been on segmented 2D gradient echo acquisition,63,64 fast spiral echo planar acquisition,65 and 3D acquisition.66±69 2D scans are usually acquired within a single breath-hold, thus eliminating the respiratory motion effects. However, 2D images have relatively thick slices and evaluation

Figure 5 Comparison study of the intracranial vascular system (a) without and (b) with the contrast agent Gd-DTPA. A single dose of 0.1 mmol kgÿ1 was used. Saturated slow ¯ow blood is now enhanced, especially for smaller vessels

of images may be complicated by the discontinuity of the slices acquired from different breath-holds since coronary arteries are highly tortuous. The acquisition of the 2D images also requires more patient cooperation and experience from the operator. Fast 3D scans with fat saturation create thin and contiguous slices, and can be easily postprocessed to visualize the coronary arteries (Figure 7). 3D scans have a better S/N and higher resolution than 2D images, and do not require breath-holding. However, image blurring will occur, reducing the effective resolution. Both hardware (echo planar system,

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

7

¯ow patterns and to evaluate the functional signi®cance of great vessel diseases. The same methods were also used to assess the vena caval ¯ow.87 5.3.3

Cardiac Chambers and Valves

Cine gradient echo sequences have been used to visualize the cardiac pools to evaluate cardiac function.88 Magnetization transfer saturation is applied to enhance the blood/ myocardium contrast.89 Cardiac parameters such as ejection fraction, stroke volume, and cardiac output can then be calculated from the images. A segmented gradient echo sequence was also used to acquire cine images within a breath-hold to eliminate the motion of the heart induced by respiration. Signal void jets in the cardiac pools may indicate abnormalities of valves such as regurgitation and stenosis. Phase images of the cine scan can provide additional information about the motion of blood and myocardium during the cardiac cycle. Speci®cally, jet velocity mapping was used to assess valve stenosis.90 5.3.4

Figure 6 High resolution carotid artery study (a) without and (b) with vessel tracking. The imaging parameters were: TR = 31 ms, TE = 7 ms, FA = 20 , MAT = 256  512 and TH = 1 mm

high/faster gradient capability, better coils, respiratory gating) and software advancement (sequence development and postprocessing such as retrospective respiratory gating70,71) will further improve the performance of coronary MRA. These improvements will lead to faster coronary artery imaging using 3D retrospective72±75 or 3D breath-hold76±79 techniques and greater coverage of the heart. Using a segmented gradient echo sequence, Burstein ®rst reported on the feasibility of measuring coronary artery ¯ow in isolated and in vivo hearts.80 A phase contrast modi®cation of this method allowed quanti®cation of the coronary artery ¯ow velocity within a breath-hold.81 TOF quanti®cation of the coronary ¯ow with echo planar imaging was demonstrated to provide real time cine images.82 Currently the spatial resolution of these techniques is still not adequate for accurate blood volume and, therefore, blood ¯ow calculation. 5.3.2

Great Vessels

The anatomy of great vessels can be demonstrated by spin echo images.83,84 The blood is dark due to wash-out effects, and the vessel wall is highlighted. 2D and 3D TOF MRA sequences have also been used to evaluate the great vessels. Common great vessel diseases include dissection and congenital diseases. 3D sequences may have the advantage when complicated vascular abnormality is involved because of the postprocessing capability of 3D images. Cine phase contrast methods are commonly used to evaluate the blood ¯ow in the aorta.85,86 Blood ¯ow information is important to study the

Pulmonary Arteries

Pulmonary artery ¯ow imaging is a dif®cult task because of motion, magnetic susceptibility variations due to air±tissue interfaces in the lung, and the overlapping of arteries and veins. TOF gradient echo techniques are used in imaging the pulmonary vasculature. 2D scans are run within a single breathhold.91±93 These images have an enhanced blood signal due to adequate in¯ow and minimum motion artifacts. A 3D syncopated sequence was proposed91±94 to visualize the pulmonary vasculature. Multiple signal averaging is used to eliminate the ghosting artifacts due to motion and demonstrate the location where vessels stay most of the time. By adding an inversion pulse before data acquisition, a `black blood' scan is obtained. A combination of `bright blood' and `black blood' scans will be able to differentiate normal blood and emboli, which should have high intensity in both scans. The advantage of 3D imaging is that the images can be postprocessed by MIP to evaluate the whole vasculature. An example MIP image generated from an ECG-triggered 3D `bright blood' scan is shown in Figure 8. Patients with a single lung transplantation are studied using phase contrast velocity mapping.95 The blood volume and ¯ow pattern of the native and transplanted lungs are compared and could be used to monitor these patients. 5.4

Renal Arteries

Image artifacts associated with cardiac, respiratory, and bowel motion, and the inherent complex ¯ow patterns and directions of the renal arteries are the major challenges for renal artery ¯ow imaging. Reasonable success has been achieved in imaging the renal arteries using both 2D96±99 and 3D phase contrast methods.100 These methods are sensitive to slow ¯ow and can achieve good background suppression. However, they are sensitive to pulsatile and nonuniform ¯ow and tissue motion, and are time consuming. 2D TOF methods96,101 have been applied to acquire MRA images of the renal arteries with breath-holding. The major problems with these techniques are the thick slices and the possible slice misregistration between separate breath-holds.

8 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

Figure 7 Three slabs (each 64 mm thick with 32 partitions) were acquired using a 3D technique. First a multiplanar reconstruction (MPR) (b) was obtained along the line illustrated in (a). The origin of the right coronary artery at the aortic root was clearly identi®ed in (b). A second MPR (d) was obtained along the line illustrated in (c). A segment of the right coronary artery is visualized in (d)

3D TOF imaging102±106 has also been used, but the motion induced blurring and ghosting severely degrade the image quality. Background suppression increases the blood±tissue contrast and helps to delineate vascular details better. It also reduces vessel boundary blurring and ghosting artifacts from the physiological motion of the background tissue. Magnetization

transfer saturation can be used to suppress background with little effect on blood signal. While MTS is quite effective in enhancing vascular contrast in evaluating intracranial arteries, it suffers from high power deposition when used in body imaging. A selective inversion recovery, rapid gradient echo sequence107 was proposed to suppress the background and background induced motion artifacts. Data acquisition takes

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY

(a)

Figure 8 An MIP image of the pulmonary vasculature generated from an ECG-triggered 3D scan with 32 partitions, each 2 mm thick

place during diastole so that blood intravoxel dephasing due to pulsatile ¯ow is minimized. An MIP image generated from this sequence is shown in Figure 9. The contrast agent gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA) has been successfully used in enhancing the vascular contrast in imaging the abdominal arteries. A continuous injection instead of a bolus was utilized to enhance arteries but not veins.108 Two studies measured the renal blood ¯ow using cine phase contrast methods.109,110 The ¯ow measurements correlate closely with renal clearance of p-aminohippurate. The same sequence was also used to measure the volumetric ¯ow rates in the portal venous system.111 Technical improvement and better hardware are needed to improve the spatial resolution further. 5.5 Peripheral Imaging Imaging of slow ¯ow is dif®cult because of the saturation of blood. The use of an intravascular contrast agent makes it

(c)

Figure 10 An example set of 3D limited volume-of-interest maximum intensity projection images in the leg post contrast agent injection. (a) Image collected during the arterial phase over 30 s with 50 slices and a resolution of 1 mm  1 mm  2 mm. (b) Image collected afterward for a period of roughly 10 min with 100 slices and a resolution of 1 mm  1 mm  1 mm. (c) Image is a processed version of the central image using the phase difference between veins and arteries to suppress the venous signal

possible to collect rapid, 3D, projective, coronal images of both legs in roughly 30 s. With 25 mT mÿ1 gradients and a rise time of 300 s, an in-plane resolution of 1 mm by 1 mm and a through plane resolution of 2 mm with 32 slices is possible in roughly 30 s with a TR of 5 ms.112,113 Higher resolution covering the entire leg is possible for scans of several minutes but then arteries and veins become equally bright. Figure 10a shows a rapid early ®rst-pass scan of the vessels in which only arteries are enhanced, followed by a steady-state scan (Figure 10b) showing the enhanced veins as well (because the contrast agent has reached the same equilibrium value in both arteries and veins). Three methods have been proposed to deal with this problem. The ®rst is vessel tracking (mentioned above) and the second and third are developing methods using the T2 and phase differences between arteries and veins. The shorter T2 of venous blood can be used to suppress signals from veins while keeping those from arteries bright. This was accomplished in the 1980s using long echoes.114 In the late 1990s, using a T2 preparation pulse and a spiral acquisition, reasonably fast highresolution data can be collected115 where arteries are signi®cantly brighter than veins and surrounding muscle. Alternatively, the phase can be used to suppress venous signal.116 This is possible because the venous blood has a different susceptibility from the surrounding tissue for vessels parallel to the static ®eld (this difference is caused by the deoxyhemoglobin in the veins). Figure 10c shows the suppression of venous blood using this method (see below). 5.6

Figure 9 An MIP image of renal arteries from a 3D diastolic acquisition and selective inversion recovery background suppression. The original data set has 32 partitions, each 1.25 mm thick

(b)

9

Venographic Imaging in the Brain

Angiography is a general term for imaging vessels, including both arteries and veins. Arteries are usually the focus of MRA studies, since they carry the crucial morphologic information as to the supply of blood to the tissue. However, if one

10 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY for TOF imaging, faster imaging, and phase unwrapped images for an improved S/N for phase contrast quantitative ¯ow imaging are major areas for continued research. In the latter category, major new directions include the use of intravascular or higher relaxivity contrast agents, echo planar ¯ow imaging, ¯uid dynamic applications, the use of phase coil arrays, the acquisition of larger matrix sizes (1024 or even 2048) and better postprocessing and display techniques. As far as whether TOF or phase contrast is better, this may become a moot point as both methods become less competitive and more complementary due to the increased speed of acquisition possible with new systems today. With both high resolution images and functional information, MRA promises to be the ideal way to obtain most vascular information noninvasively.

7

RELATED ARTICLES

Abdominal MRA; Head and Neck Studies Using MRA; Phase Contrast MRA; Time-of-Flight Method of MRA. Figure 11 A minimum intensity projection image from a TE = 40 ms, TR = 57 ms, 3D, gradient echo image projected over seven images. The resolution is 0.5 mm  0.5 mm  2.0 mm. The acquisition time was 10 min and 32 partitions (slices) were collected in total. The peripheral

is interested in whether or not the tissue actually performed its function and used the oxygen delivered to it, then the venous image is important. The use of intravascular contrast agents makes it possible to visualize both arteries and veins. How to separate them remains the problem. However, one method exists that can highlight small venous structures even without a contrast agent. This method uses a high resolution, long TE, 3D gradient echo approach to reduce susceptibility effects caused by static ®eld inhomogeneities, and to highlight the ®eld differences between arteries and veins117,118 caused by the differences in deoxyhemoglobin content. The ®eld difference between veins and arteries is about 0.3 p.p.m. for veins parallel to the static ®eld and leads to an out-of-phase relationship between them that occurs for a TE of roughly 50 ms. This  phase shift leads to a cancellation effect between the background tissue and the veins. Further, the phase image itself reveals the same  phase shift and can be used to suppress the venous signal simply by scaling the phase image to unity where the phase is zero and to zero when the phase is ; the resulting phase mask is then multiplied by the magnitude image. Figure 11 demonstrates the level of cancellation and the visualization of small venous structures in the brain. This approach has proven valuable for the study of venous lesions and other vascular abnormalities.

6 FUTURE DIRECTIONS Future developments depend on what remains unanswered using the present techniques and what new directions remain to be investigated. In the former category, the use of short (possibly velocity compensated) gradient echoes, enhanced contrast

8

REFERENCES

1. P. C. Lauterbur, Nature, 1973, 242, 190. 2. S. Patz and R. C. Hawkes, Magn. Reson. Med., 1986, 3, 140. 3. W. T. Dixon, L. N. Du, D. D. Faul, M. Gado, and S. Rossnick, Magn. Reson. Med., 1986, 3, 454. 4. A. Constantinesco, J. J. Mallet, A. Bonmartin, C. Lallot, and A. Briguet, Magn. Reson. Imaging, 1984, 2, 335. 5. D. G. Nishimura, A. Macovski, and J. M. Pauly, IEEE Trans. Med. Imaging, 1986, 5, 140. 6. E. M. Haacke and G. W. Lenz, Am. J. Radiol., 1987, 148, 1251. 7. L. Axel and D. Morton, J. Comput. Assist. Tomogr., 1987, 11, 31. 8. G. L. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 9. G. A. Laub and W. A. Kaiser, J. Comput. Assist. Tomogr., 1988, 12(3), 377. 10. G. Suryan, Proc. Indian Acad. Sci., 1951, 33, 107. 11. H. Y. Carr and E. M. Purcell, Phys. Rev., 1954, 94, 630. 12. H. C. Torrey, Phys. Rev., 1956, 104, 563. 13. J. R. Singer, Science, 1959, 130, 1652. 14. R. L. Bowman, and V. Kudravcev, IRE Trans. Med. Elec., 1959, 6, 267. 15. R. E. Halbach, J. H. Battocletti, S. X. Salles-Cunha, and A. Sances, Med. Phys., 1981, 8, 444. 16. E. L. Hahn, J. Geophys. Res. 1960, 65, 776. 17. K. J. Packer, Mol. Phys., 1969, 17, 355. 18. V. Waluch and W. G. Bradley, J. Comput. Assist. Tomogr., 1984, 8, 594. 19. P. R. Moran, Magn. Reson. Imaging, 1982, 1, 197. 20. E. M. Haacke, T. J. Masaryk, P. A. Wielopolski, F. R. Zypman, J. A. Thach, S. Amartus, J. Mitchell, M. Clampitt, and C. Paschal, Magn. Reson. Med., 1990, 14, 202. 21. P. J. Keller, B. P. Drayer, E. K. Fram, K. D. Williams, C. L. Dumoulin, and S. P. Souza, Radiology, 1989, 173, 527. 22. R. S. Owen, J. P. Carpenter, R. A. Baum, L. J. Perloff, and C. Cope, N. Engl. J. Med., 1992, 326, 1577. 23. J. A. Tkach, P. M. Ruggieri, J. J. Dillinger, J. S. Rose, M. T. Modic, and T. J. Masaryk, J. Magn. Reson. Imaging, 1993, 3, 365.

WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY 24. J. I. Jackson, D. G. Nishimura, and A. Macovski, Magn. Reson. Med., 1992, 25, 128. 25. D. Purdy, G. Cadena, and G. Laub, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 882. 26. D. L. Parker, C. Yuan, and D. D. Blatter, Magn. Reson. Med., 1991, 17, 434. 27. G. B. Pike, B. S. Hu, G. H. Glover, and D. R. Enzmann, Magn. Reson. Med., 1992, 25, 372. 28. R. R. Edelman, S. S. Ahn, D. Chien, W. Li, A. Goldmann, M. Mantello, J. Krames, and J. Klee®eld, Radiology, 1992, 184, 395. 29. G. Marchal, H. Bosmans, P. van Hecke, U. Speck, P. Aerts, P. Vanhoenacker, and A. L. Baert, Am. J. Roentgenol., 1990, 155, 407. 30. W. Lin, E. M. Haacke, A. S. Smith, and M. E. Clampitt, J. Magn. Reson. Imaging, 1992, 2, 277. 31. T. D. Scholz, T. L. Ceckler, and R. S. Balaban, Magn. Reson. Med. 1993, 29, 352. 32. H. E. Cline, C. L. Dumoulin, W. E. Lorensen, S. P. Souza, and W. J. Adams, Magn. Reson. Med. 1991, 18, 384. 33. W. Lin, E. M. Haacke, T. J. Masaryk, and A. S. Smith, JMRI, 1992, 2, 519. 34. D. Saloner, W. A. Hanson, J. S. Tsuruda, R. van Tyen, C. M. Anderson, and R. E. Lee, JMRI, 1991, 1, 423. 35. P. M. Pattany, R. Marino, and J. M. McNally, Magn. Reson. Imaging, 1986, 4, 154. 36. G. L. Naylor, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 37. L. R. Frank, A. P. Crawley, and R. B. Buxton, Magn. Reson. Med., 1992, 25, 299. 38. O. P. Simonetti, J. L. Duerk, and V. Chankong, Magn. Reson. Med., 1993, 23, 498. 39. S. N. Urchuk and D. B. Plewes, JMRI, 1992, 2, 453. 40. P. J. Kilner, D. N. Firmin, R. S. Rees, J. Martinez, D. J. Pennell, R. H. Mohiaddin, S. R. Underwood, and D. B. Longmar, Radiology, 1991, 178(1), 229. 41. E. M. Haacke, E. D. Lindskog, and W. Lin, J. Magn. Reson. 1991, 92, 126. 42. J. R. Singer, Electronics, 1960, 33, 77. 43. O. C. Morse, and J. R. Singer, Science, 1970, 170, 440. 44. W. T. Dixon, L. N. Du, and M. Gado, Magn. Reson. Med., 1986, 3, 454. 45. L. Axel, A. Shimakawa, and J. MacFall, Magn. Reson. Imaging, 1986, 4, 199. 46. K. Shimizu, T. Matsuda, T. Sakurai, A. Fujita, H. Ohara, S. Okamura, S. Hashimoto, H. Mano, C. Kawai, and M. Kiri, Radiology, 1986, 159, 195. 47. R. R. Edelman, H. P. Mattle, J. Klee®eld, and M. S. Silver, Radiology, 1989, 171, 551. 48. J. Hennig, M. Mueri, P. Brunner, and H. Friedburg, Radiology, 1988, 166, 237. 49. E. M. Haacke, A. S. Smith, W. Lin, J. S. Lewin, D. A. Finelli, and J. L. Duerk, Top. Magn. Reson. Imaging, 1991, 3, 34. 50. C. L. Dumoulin, S. P. Souza, C. J. Hardy, and S. A. Ash, Magn. Reson. Med. 1991, 21, 242. 51. C. L. Dumoulin, S. P. Souza, M. F. Walker, and E. Yoshitome, Magn. Reson. Med., 1988, 6, 275. 52. C. L. Dumoulin, S. P. Souza, M. F. Walker, and W. Wagle, Magn. Reson. Med., 1989, 9, 139. 53. D. J. Bryant, J. A. Payne, D. N. Firmin, and D. B. Longmore, J. Comp. Assist. Tomogr., 1984, 8, 588. 54. C. Thomsen, F. Stahlberg, M. Stubgaard, and B. Nordell, Radiology, 1990, 177, 659.

11

55. N. J. Pelc, M. A. Bernstein, A. Shimakawa, and G. H. Glover, J. Magn. Reson. Imaging, 1991, 1, 405. 56. R. R. Edelman, D. Chien, and D. Kim, Radiology, 1991, 181, 655. 57. W. Lin, R. R. Edelman, E. M. Haacke, in `Magnetic Resonance Angiography: Concepts and Applications', ed. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottshalk, Mosby-Yearbook, St. Louis, Ch. 8, 1993. 58. D. LeBihan, R. Turner, and J. R. MacFall, Magn. Reson. Med., 1989, 10, 324. 59. S. Paulin, G. K. von Schulthess, E. Fossel, and H. P. Krayenbuehl, Am. J. Radiol., 1987, 148, 665. 60. Z. H. Cho, C. W. Mun, and R. M. Friedenberg, Magn. Reson. Med., 1991, 20, 134. 61. C. L. Dumoulin, S. P. Souza, R. D. Darrow, and W. J. Adams, J. Comput. Assist. Tomogr. 1991, 15, 705. 62. S. J. Wang, B. S. Hu, A. Macovski, and D. G. Nishimura, Magn. Reson. Med., 1991, 18, 417. 63. R. R. Edelman, W. J. Manning, D. Burstein, and S. Paulin, Radiology, 1991, 181, 641. 64. W. J. Manning, W. Li, and R. R. Edelman, N. Engl. J. Med., 1993, 328, 828. 65. C. H. Meyer, B. S. Hu, D. G. Nishimura, and A. Macovski, Magn. Reson. Med., 1992, 28, 202. 66. R. J. Al®di, T. J. Masaryk, E. M. Haacke, G. W. Lenz, J. S. Ross, M. T. Modic, A. D. Nelson, J. P. Li Puma, and A. M. Cohen, Am. J. Roentgenol., 1987, 149, 1097. 67. G. W. Lenz, E. M. Haacke, and R. D. White, Magn. Reson. Imaging, 1989, 7, 445. 68. D. Li, C. B. Paschal, E. M. Haacke, and L. P. Alder, Radiology, 1993, 181, 401. 69. C. B. Paschal, E. M. Haacke, and L. P. Adler, JMRI, 1993, 3, 491. 70. M. Hofman, C. B. Paschal, D. Li, and E. M. Haacke, J. Comput. Assist. Tomogr., 1995, 19, 56. 71. Y. Wang, R. C. Grimm, and P. J. Rossman, Magn. Reson. Med., 1995, 35, 11. 72. M. B. M. Hofman, C. B. Paschal, D. Li, E. M. Haacke, A. C. Van Rossum, and M. Sprenger, J. Comput. Assist. Tomogr., 1995, 19, 56. 73. Y. Wang, P. J. Rossman, R. C. Grimm, S. J. Riederer, and R. L. Ehman, Radiology, 1996, 198, 55. 74. D. Li, S. Kaushikkar, E. M. Haacke, et al., Radiology, 1996, 201, 857. 75. P. K. Woodard, D. Li, E. M. Haacke, et al., Am. J. Roentgenol., 1998, 170, 883. 76. P. A. Wielopolski, W. J. Manning, and R. R. Edelman, JMRI, 1995, 5, 403. 77. J. W. Goldfarb and R. R. Edelman, Radiology, 1998, 206, 830. 78. J. Zheng, D. Li, K. T. Bae, P. K. Woodard, and E. M. Haacke, J. Cardiovasc. Magn. Reson., 1998, 1, 33. 79. D. Li, B. Dolan, R. C. Walovitch, and R. B. Lauffer, Magn. Reson, Med., 1998, 39, 1014. 80. D. Burstein, J. Magn. Reson. Imaging, 1991, 1, 337. 81. R. R. Edelman, W. J. Manning, E. Gervino, and W. Li, J. Magn. Reson. Imaging, 1993, 3, 699. 82. B. P. Poncelet, R. M. Weisskoff, V. J. Wedeen, T. J. Brady, and H. Kantor, Magn. Reson. Med., 1993, 30, 447. 83. C. B. Higgins, B. F. Byrd, M. T. McNamara, P. Lanzer, M. J. Lipton, E. Botvinick, N. B. Schiller, L. E. Crooks, and L. Kaufman, Radiology, 1985, 155, 671. 84. R. D. White, C. B. Paschal, J. A. Tkach, and M. J. Carvlin, Top. Magn. Reson. Imaging, 1990, 2(2), 31.

12 WHOLE BODY MAGNETIC RESONANCE ANGIOGRAPHY 85. H. G. Bogren, R. H. Mohiaddin, R. K. Klipstein, D. N. Firmin, R. S. Underwood, R. S. Rees, and D. B. Longmore, Am. Heart J., 1989, 118, 234. 86. H. G. Bogren, R. K. Klipstein, D. N. Firmin, R. H. Mohiaddin, S. R. Underwood, R. S. Rees, and D. B. Longmore, Am. Heart J., 1989, 117, 1214. 87. R. H. Mohiaddin, S. L. Wann, R. Underwood, D. N. Firmin, R. S. Rees, and D. B. Longmore, Radiology, 1990, 177, 537. 88. U. Sechtem, P. W. P¯ugfelder, R. D. White, R. G. Gould, W. Holt, M. J. Lipton, and C. B. Higgins, Am. J. Roentgenol., 1987, 148, 239. 89. S. Kaushikkar, D. Li, W. Lin, E. M. Haacke, D. Wilson, L. Adler, R. Stengall, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 502. 90. P. J. Kilner, D. N. Firmin, R. S. O. Rees, J. Martinez, D. J. Pennell, R. H. Mohiaddin, S. R. Underwood, and D. B. Longman, Radiology, 1991, 178, 229. 91. H. Hatabu, W. B. Gefter, H. Y. Kressel, L. Axel, and R. E. Lenkinski, Radiology, 1989, 171, 391. 92. T. K. F. Foo, J. R. MacFall, C. E. Hayes, H. D. Sostman, and B. E. Slayman, Radiology, 1992, 183, 473. 93. T. M. Grist, H. D. Sostman, J. R. MacFall, T. K. Foo, C. E. Spritzer, and L. W. Wittig, Radiology, 1993, 189, 523. 94. P. A. Wielopolski, E. M. Haacke, and L. P. Adler, Radiology, 1992, 183, 465. 95. R. H. Mohiaddin, R. Paz, S. Theodoropoulos, D. L. Firmin, D. B. Longmore, and M. H. Yacoub,, J. Thorac. Cardiovasc. Surg., 1991, 101, 1016. 96. J. F. Debatin, C. E. Spritzer, T. M. Grist, C. Beam, L. P. Svetkey, G. E. Newman, and H. D. Sostman, Am. J. Roentgenol., 1991, 157, 981. 97. C. L. Dumoulin, E. K. Yucel, P. Vock, S. P. Souza, F. Terrier, F. L. Steinberg, and H. Weymueller, J. Comput. Assist. Tomogr. 1990, 14, 779. 98. F. L. Steinberg, E. K. Yucel, C. L. Dumoulin, and S. P. Souza, Magn. Reson. Med., 1990, 14, 315. 99. E. Farrugia, B. F. King, and T. S. Larson, Mayo Clin. Proc., 1993, 68, 157. 100. P. Vock, F. Terrier, H. Wegmueller, F. Mahler, P. Gertsch, S. P. Souza, and C. L. Dumoulin, Br. J. Radiol., 1991, 64, 10. 101. D. Kim, R. R. Edelman, K. C. Kent, D. H. Porter, and J. J. Skillman, Radiology, 1989, 174, 727. 102. J. Borrello, D. Li, T. Vesely, E. P. Vinning, J. J. Brown, and E. M. Haacke, Radiology, 1995, 197, 793. 103. J. S. Lewin, G. Laub, and R. Hausmann, Radiology, 1991, 179, 261. 104. P. A. Hardy, J. A. Tkach, G. K. Lammert, and M. G. Zelch, J. Magn. Reson. Imaging, 1993, 3, 58. 105. H. J. Smith and S. J. Bakke, Acta Radiol., 1993, 34, 150.

106. E. K. Yucel, J. A. Kaufman, M. Prince, H. Bazari, L. S. T. Fang and A. C. Waltman, Magn. Reson. Imaging, 1993, 11, 925. 107. D. Li, E. M. Haacke, J. P. Mugler, S. Berr, and J. R. Brookeman, Magn. Reson. Med., 1994, 31, 414. 108. M. R. Prince, E. K. Yucel, J. A. Kaufman, D. C. Harrison, and S. C. Geller, JMRI, 1993, 3, 877. 109. B. Lundin, T. G. Cooper, R. A. Meyer, and E. J. Potchen, Magn. Reson. Imaging, 1993, 11, 51. 110. R. L. Wolf, B. F. King, V. E. Torres, D. M. Wilson, and E. L. Ehman, Am. J. Roentgenol., 1993, 161, 995. 111. D. J. Burkart, C. D. Johnson, M. J. Morton, R. L. Wolf, and R. L. Ehman, Am. J. Roentgenol., 1993, 160, 1113. 112. M. R. Prince, D. Li, Narasimham, W. T. Jacoby, et al., Am. J. Roentgenol., 1996, 166, 1387. 113. T. M. Grist, F. R. Korosec, D. C. Peters, et al., Radiology, 1998, 207, 539. 114. P. J. Keller and D. Saloner, in `Magnetic Resonance Angiography: Concepts and Applications', eds. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottshalk, Mosby-Yearbook, St Louis, PA, 1993, Ch. 7. 115. J. H. Brittain, E. W. Olcott, A. Szuba, G. E. Gold, G. A. Wright, P. Irarrazaval, and D. G. Nishimura, Magn. Reson, Med., 1997, 38, 343. 116. Y. Wang, Y. Yu, T. Bae, D. Li, W. Lin, and E. M. Haacke, 1998 Xth Annual Workshop on MRA, Park City, Utah, p. 24. 117. J. R. Reichenbach, R. Feiwell, K. Kuppusamy, M. Bahn, and E. M. Haacke, Magn. Reson, Imag., 1997, 16, 281. 118. J. R. Reichenbach, R. Venkatesan, D. J. Schillinger, D. K. Kido, and E. M. Haacke, Radiology, 1997, 204, 272.

Biographical Sketches E. Mark Haacke. b 1951. B.S., 1973, M.Sc., 1975, Ph.D., 1978, University of Toronto. Introduced to NMR by Neil Holland with Picker Int. in 1983. Faculty at Case Western Reserve University, Department of Radiology, 1985±93. Faculty at Washington University, Mallinckrodt Institute of Radiology, 1993±present. Approx. 100 publications. Research interests include MR in functional imaging, angiography, cardiovascular imaging, fast imaging methods, and image reconstruction methods. Weili Lin. b 1963. M.S., 1990, Ph.D., 1993 Case Western Reserve University. Faculty at Washington University, 1993±present. Approx. 50 publications. Research interests include MR in functional imaging, angiography, signal processing, and stroke. Debiao Li. b 1962. Ph.D., 1992, University of Virginia. Research associate at Case Western Reserve University, 1992±93. Faculty at Washington University, 1993±present. Approx. 50 publications. Research interests include MR in cardiac imaging, angiography, and fast imaging techniques.

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Blood Flow: Quantitative Measurement by MRI David N. Firmin

3

1

EARLY QUANTITATIVE IN VIVO STUDIES

The quest for a method quantitatively to image blood ¯ow in vivo was ®rst realized in the early 1980s, initially using time-of-¯ight methods and soon afterwards phase methods.

National Heart and Lung Institute, University of London, UK

and

3.1

Raad H. Mohiaddin

Singer and Crooks6 initially used a saturation wash-in technique in an attempt to measure ¯ow in the internal jugular veins. They employed the same principles as ®rst Garroway10 and then Thulborn et al.11 had previously used in vitro. The method involved ®rst saturating the slice and then leaving an interval before acquiring image data from the slice; the longer the time interval  the greater the in¯ow of fully magnetized spins and the higher the signal on the image. The two main problems with these methods were that only low ¯ow velocities could realistically be measured with practical slice thicknesses, and also many other factors could affect the amplitude of the blood signal. Later work on this type of method tended to be more related to angiography, although some quantitative ¯ow measurements were demonstrated.12,13 However, the advent of 3D imaging techniques has enabled ungated measurements of constant ¯ows or averaged pulsatile ¯ows to be carried out by using the time-of-¯ight principles as described below. The ®rst true time-of-¯ight (TOF) method of blood ¯ow imaging was described by Feinberg et al.14 Their approach was to use a variation on a double echo spin echo sequence; the ®rst 180 selected slice was displaced by 3 mm from the 90 selected slice, and the second was displaced by 9 mm. The ®rst 180 selection overlapped suf®ciently with the 90 selection to produce a good anatomical image. The second 180 pulse selection did not overlap with that of the 90 or the ®rst 180 pulse selection, and hence produced no anatomical image but gave a high signal from the blood that had experienced all the preceding rf pulses (i.e. that had passed between the different selected planes). Flow in the carotid and vertebral arteries of a volunteer's neck was identi®ed with the technique. One-dimensional methods that directly imaged blood movement over a known time were described by both Shimizu et al.15 and Axel et al.16 The methods consisted of slice selection and frequency encoding being applied in the same axis. In this way material that had moved in this axis between selection and reading would be displaced relative to the stationary material. This technique is therefore making use of an effect that is often seen as a problem in other methods of ¯ow imaging. With the ¯ow information encoded in one axis as described, the other one or two axes may be spatially encoded by use of phase-encoding gradients. Shimizu et al.15 tested the technique on a U-tube phantom where, by varying the velocity and the echo time (TE) to produce different ¯ow displacements, they found very good correlation between these displacements and true ¯ow. The technique was demonstrated in vivo17 where it was repeated rapidly throughout the heart cycle so that pulsatile ¯ow information could be acquired in a reasonable time. One of the main advantages of the technique was its ability to image very high velocities.

Royal Brompton Hospital, London, UK

1 INTRODUCTION The study of ¯ow with nuclear magnetic resonance (NMR) has a long history, although the early research work merely set out to analyze the ¯ow effects on the signal without detailing a particular application. Hahn,1 for example, was the ®rst to note the effect of diffusion while the ®rst to study the effect of coherent ¯ow was Suryan.2 Several years later methods of ¯ow measurement were presented, a number of which were designed for the measurement of blood ¯ow. The methods all used either ¯ow-related phase shifts or time-of-¯ight principles, where the latter could involve tagging or saturation and washin effects. Following the introduction of NMR imaging techniques, a number of methods were described, again applying the same principles, but for obtaining ¯ow information in the form of an image.

2 EARLY QUALITATIVE IN VIVO STUDIES Much of the initial work concentrated on understanding the appearance of a ¯ow on an image and the artifacts it can cause; these can often be indicative of the type of ¯ow present and can therefore give important information in the diagnosis of a particular disorder. Examples of ¯ow phenomena studied include: signal dephasing,3 wash-in/wash-out with different sequences,4±6 even echo rephasing,7 and signal misregistration effects.8 Pulsatile ¯ow can also alter the appearance of the image by the formation of nonstructured ¯ow artifacts. These result because the motion of blood during the application of imaging gradients results in a phase shift of the blood signal which is added to the spatial phase encoding. If the motion of blood changes from one acquisition to another, the resultant phase shift will introduce a varying error to the spatial phase encoding. The Fourier transform will then get `confused' and the blood signal will end up being spread out along the phase encoding axis. Perman et al.9 investigated this type of ¯ow artifact and demonstrated that even echo rephasing corrected the error and removed the artifact as long as the motion was reasonably simple (i.e. it did not contain signi®cant acceleration or other high-order derivatives of position).

Time-of-Flight Methods

2 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 1 In¯ow of arterial blood (up) and venous blood (down) into a presaturation band in a 2D cardiac gated image. The saturation has taken place just above the carotid bifurcation during systole. The time between the saturation pulse and the echo time is 18 ms. The distance the blood has moved is slightly more than 1.2 cm, resulting in a velocity of about 60 cm s 1. (Reproduced with permission of Aspen Publishers from E. M. Haacke, A. S. Smith, W. Lin, J. S. Lewin, D. A. Finelli, and J. L. Duerk, Top. Magn. Reson. Imaging, 1991, 3, 34)

Methods have also been developed combining the time-of¯ight and wash-in/wash-out principles. Two general approaches using slice saturation in an orthogonal plane to that of the image have been used. The ®rst utilizes the pulse to saturate tissues within a band across the image so that magnetized blood ¯owing into the region can be imaged and ¯ow quanti®ed by measuring the distance of in¯ow over the time between saturation and image acquisition. Results have been demonstrated using two-dimensional cardiac gated (Figure 1) and three-dimensional ungated methods (Figure 2). The second approach is to saturate a band of tissue, for example in a transverse plane, and then follow the progress of this dark band in the coronal or sagittal planes.18 The main limitation of these saturation techniques is that they are limited by the T1Ã value of the tissue. The signal from the saturated blood or tissue will increase towards equilibrium with time, eventually making it dif®cult to measure accurately the distances traveled. Also, motion during the sampling gradients can result in signal distortion and thus affect the image.19 For arterial ¯ow measurements where cardiac gating is required, only two-dimensional images can realistically be acquired. This limits our ability to study ¯ow details as the ¯ow pro®le can only be acquired in one dimension at the most. 3.2 Phase Methods In 1982, Moran20 suggested the introduction of bipolar velocity phase-encoding pulses, such as had previously been used

Figure 2 In¯ow during a 3D TOF experiment carried out in the upper neck/head region reveals many arterial vessels. The peak ¯ow represented in the internal carotid is more than 1 m s 1. (Reproduced with permission of Mosby-Year Book, Inc. from D. N. Firmin, C. L. Dumoulin, and R. H. Mohiaddin, in `Magnetic Resonance Angiography, Concepts and Applications', ed. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottschalk, 1993)

in nonimaging experiments, in order to introduce velocity phase-encoding to the data. Phase methods of ¯ow imaging that were later developed using this suggested theory fell broadly into two categories: 1. Fourier ¯ow imaging methods that phase-encoded ¯ow velocity so that a number of ¯ow phase-encoding steps would separate the different ¯ow velocities present in a particular region. 2. Phase-mapping methods that mapped the phase of the signal directly in order to measure the ¯ow. Both categories of methods are based on the same theory that the NMR signal from a sample of spins moving in the direction of a bipolar magnetic gradient pulse pair (Figure 3) will accumulate a phase shift given by:  ˆ v
Ag

…1†

where is the gyromagnetic ratio, v is the velocity,
is the time between the centers of the lobes of the gradient pulse, and

Ag

D

Figure 3 A bipolar magnetic gradient pro®le as used to produce velocity-dependent phase shifts

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Ag is the area of one gradient lobe. The phase shift is therefore directly related to the velocity for a particular gradient wave form. 3.2.1

Fourier Flow Methods

The ®rst to demonstrate experimentally the method of Fourier ¯ow imaging were Redpath et al.21 when they used eight ¯ow phase-encoding steps to image a circle of ¯uid-®lled tubing that was rotated in the image plane. Different segments of the circle were seen on the eight resultant images, each corresponding to a different velocity range in the direction of the ¯ow phase-encoding gradient. Feinberg et al.22 applied the method both in vitro and in vivo, but increased the velocity resolution and simpli®ed the reconstruction by increasing the number of ¯ow phase-encoding steps and removing the spatial phase encoding. The phantom results proved the accuracy of the method and the in vivo study, while showing the ¯ow in the descending aorta, also highlighted the problem of a very high signal from stationary tissue imaged in the spatial dimension. In 1988, Hennig et al.23 described a development of this method where the signal from stationary tissue was saturated and the sequence repeated much more rapidly. The main problem with the technique, particularly when pulsatile ¯ow is being studied, is that the time required to obtain a reasonably high resolution in the velocity axis precludes the use of more than one spatial dimension. 3.2.2

Phase Mapping Methods

In 1984, both van Dijk24 and Bryant et al.25 demonstrated the use of velocity phase mapping to measure ¯ow velocity directly from the phase of the signal originating from each imaging voxel. These phase±mapping methods were made clinically more useful, partly by the use of a ®eld echo sequence26,27 and, most importantly, by the introduction of velocity compensated gradient waveforms.28,29 The improvements enabled the sequence to be repeated rapidly and prevented the characteristic signal loss due to shear ¯ow when uncompensated gradient waveforms were employed. The technique has been validated in vivo30 and has already provided useful information in clinical and physiological ¯ow studies.31 The accuracy of the techniques depends on numerous factors,32 although validation of the phase velocity mapping method has been demonstrated both in vitro and in vivo. In vitro measurements have been made by imaging a disk of doped water (1 mM CuSO4) rotating at known rate and by comparing measurements of the ¯ow of doped water through tubes with true ¯ows. The in vivo validation30 showed a very good correlation between the measurement of the stroke volume of the heart carried out by the phase mapping technique and a previously validated multislice volume method. 3.2.3

3

combining a phase mapping type approach with a very fast imaging method such as echo planar.33 The more obvious approach in many respects is to remove the slow spatial phase-encoding process and to image in only one spatial dimension. One such method, RACE (Real time ACquisition and velocity Evaluation),33 was developed to measure ¯ow perpendicular to the slice. The technique can be repeated rapidly throughout the cardiac cycle in order to give near real time ¯ow information. Figure 4 shows examples of ¯ow images acquired using the RACE technique. One problem with this type of approach is that data are acquired from a projection through the patient; this means that any signal overlapping with the ¯ow signal will combine and introduce errors to the ¯ow measurement. Several strategies have been suggested for localizing the signal in order to avoid this: they include spatial presaturation, projection dephasing (applying a gradient to suppress stationary tissue), collecting a cylinder of data, and multiple oblique measurements. For 2D spatial resolution with velocity information, a very rapid technique such as echo planar is required. The echo planar method is technically dif®cult and not possible on many current standard machines. However, methods have been developed to minimize the technical dif®culties and Figure 5 shows echo planar velocity images of a normal neck acquired in 40 ms of two consecutive cardiac cycles using such a method. Figure 6 shows a comparison of echo planar velocity measurements with those obtained using ®eld even echo rephasing (FEER)28 velocity mapping. The two techniques agree well although it should be remembered that one is an average over time and the other is not. An additional problem is that the conventional echo planar sequence has a relatively high phase sensitivity to ¯ow in the phase-encode (blip) direction, even if additional ¯ow compensation is applied. This has been used to advantage for more qualitative ¯ow imaging, showing ¯ow disturbances for example. Recent studies have concentrated on reducing the phase sensitivity and increasing the image matrices; one such approach is to use spiral echo planar where the phase sensitivity is reduced because the sampling starts at the center of k-space.34±36 An alternative is to combine phase mapping with subsecond FLASH techniques; here the limitation is the duration of the acquisition (>100 ms) and the relatively low spatial resolution. Less rapid and higher resolution acquisitions can be made on standard MR imaging equipment by acquiring k-space in an interleaved fashion. FLASH, echo planar, or spiral imaging may be acquired in such a so-called segmented k-space approach. Cardiac gated images can be acquired over a period of approximately 16 cardiac cycles so that, although dynamic changes in ¯ow cannot be monitored with this approach, images can be acquired in a breath-hold. The problems of respiration that have precluded imaging details such as the coronary arteries can thus be reduced or removed, and more reliable coronary ¯ow measurements made feasible.37

Rapid Flow Imaging Methods

The previously described methods acquire data over a long period of time in comparison to the cardiac cycle, and rapid variations in ¯ow cannot be followed. In order to study ¯ow dynamics, very rapid ¯ow imaging techniques have been developed, either by imaging only one spatial dimension or by

4

CLINICAL OBSERVATIONS AND APPLICATIONS

As described in the previous sections, magnetic resonance imaging (MRI) offers several methods for measuring blood

4 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 4 RACE pro®le through (a) ®ve cardiac cycles in the neck and (b) four cardiac cycles in the abdomen. The carotid velocity is similar to that seen with Doppler ultrasound, never quite dropping to zero. The aortic ¯ow is rapid and drops to zero and then returns to a fairly constant low ¯ow rate. (Reproduced with permission of Aspen Publishers from E. M. Haacke, A. S. Smith, W. Lin, J. S. Lewin, D. A. Finelli, and J. L. Duerk, Top. Magn. Reson. Imaging, 1991, 3, 34)

¯ow which have greatly enhanced the potential capability of MRI as a physiological tool in cardiology. As yet the majority of clinical use has been with the method of phase shift velocity mapping, and for this reason alone many of the following observations and applications will concentrate on this approach.

4.1

Thoracic Aorta

Quantitative analysis of aortic ¯ow using MRI has been a subject of considerable interest in health and disease. Normal systolic ¯ow in the ascending aorta is plug ¯ow with a skewed velocity pro®le which has higher velocities around

Figure 5 (a) A magnitude reconstruction of echo planar data acquired with one excitation. The selected region includes the carotid arteries (short arrows) and jugular veins (long arrows) which can be seen exhibiting high blood signal. (b) An example of a velocity phase map reconstructed from data acquired with one excitation of each of a modi®ed and unmodi®ed EPI ¯ow sequence. Flow in the carotid arteries can be seen tending toward white whilst that in the jugular veins tends toward black with stationary material mid-gray. (Reproduced with permission of Academic Press from D. N. Firmin, G. L. Nayler, P. J. Kilner, and D. B. Longmore, Magn. Reson. Med., 1990, 14, 230)

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

5

Mean velocity (cm s–1)

70 60

Echo Planar

50

FEER

40 30 20 10 0

0

100

200

300 400 Time (ms)

500

600

700

Figure 6 A comparison of the ¯ow versus time plots for the right carotid artery of a normal volunteer obtained from echo planar velocity images and FEER velocity images. The close comparison between the two plots suggests that echo planar velocity measurements are as accurate as the FEER measurements. It should be noted, however, that the two types of ¯ow information are somewhat different. (Reproduced with permission of Academic Press from D. N. Firmin, G. L. Nayler, P. J. Kilner, and D. B. Longmore, Magn. Reson. Med., 1990, 14, 230)

4.1.1

Aortic Flow Wave Velocity

Aortic ¯ow wave velocity can be calculated by MR velocity mapping from the delay between the leading edge or onset of the ¯ow wave in the ascending and descending limbs of the thoracic aorta.41,42 This parameter is closely related to aortic compliance41 which may prove to be useful for the detection and monitoring of arterial disease. 4.1.2

Aortic Dissection

Aortic dissection is readily detected by spin echo imaging and its extent can be displayed, including the involvement of other vessels.43 However, a thin intimal ¯ap may not be shown in these images unless static blood in the false lumen provides natural contrast with the true lumen. If there is any doubt, then the ¯ap will be more easily seen using a gradient echo sequence, and velocity mapping will con®rm the diagnosis by demonstrating the differential ¯ow velocities in each lumen (Figure 10).44

(c) 45 40 35 30 Flow (L min–1)

the inside of the arch. Throughout diastole the blood continues to move with simultaneous forward and reverse channels as is demonstrated in Figure 7.38 In patients with coronary artery disease, the reverse ¯ow channel is smaller and may enter any of the coronary sinuses.39 In aortic valve regurgitation, the magnitude of the reverse ¯ow is understandably increased (Figure 8 and Figure 9) and aortic or pulmonary regurgitation may be quanti®ed from the back¯ow of blood in the proximal great vessels assessed by velocity mapping.40

AA MPA DA SVC

25 20 15 10 5 0 –5

0

200

400 Time (ms)

600

800

Figure 7 (a) A mid-systolic gradient echo image in a transverse plane through the great vessels of a healthy volunteer with (b) the corresponding velocity map. (c) The instantaneous ¯ow volume curves of the ascending aorta (AA), main pulmonary artery (MPA), descending aorta (DA), and superior vena cava (SVC) calculated from the complete cine velocity map acquired in the same plane: 1, ascending aorta; 2, main pulmonary artery; 3, left atrium; 4, descending aorta; 5, superior vena cava. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

6 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 9 Fourier velocity measurements using a thin-slice excitation in the abdomen of (a) a healthy volunteer and (b) a volunteer with a bicuspid aortic valve regurgitation. (Reprinted with permission of Mosby Publishers from D. N. Firmin, C. L. Dumoulin, and R. H. Mohiaddin, in `Magnetic Resonance Angiography, Concepts and Applications', ed. E. J. Potchen, E. M. Haacke, J. E. Siebert, and A. Gottschalk, 1993) (b)

This abnormal pattern (Figure 11) may be caused by re¯ected waves from the distal vasculature which has a high impedance. Patients with a single lung transplant are unique in that the cardiac output is ejected into two pulmonary vascular beds with different characteristics, and in these patients the differential blood ¯ow depends on the relative resistance in each lung. Velocity mapping can assess the total and differential pulmonary blood ¯ow which may be useful for monitoring these patients (Figure 12).47

60 50

AA DA

Flow (L min–1)

40 30 20 10 0 –10

4.3

–20 –30

0

200

400

600

Time (ms)

Figure 8 (a) Gradient echo images in a coronal plane acquired during ventricular diastole in a patient with Marfan's syndrome and aortic valve regurgitation. (b) Flow volume curves in the ascending aorta (AA) and descending thoracic aorta (DA) measured from the complete cine acquisition acquired in a transverse plane perpendicular to the AA and DA. Note the large retrograde net ¯ow during diastole. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

4.2 Central Pulmonary Arteries The retrosternal position of the central pulmonary arteries makes it dif®cult to assess pulmonary blood ¯ow by Doppler echocardiography, especially in the presence of skeletal or lung abnormalities. Pulmonary ¯ow pro®les have been studied less well in patients but MR velocity mapping has con®rmed an abnormally early forward systolic peak and increased reverse diastolic ¯ow in patients with pulmonary hypertension.45,46

Caval Veins

The caval veins are relatively large, and reliable velocity maps and ¯ow measurements can readily be obtained.48 The normal pattern of caval ¯ow has two forward peaks in ventricular systole and diastole (Figure 7), but this pattern is disturbed by disease. Any condition that causes impaired ®lling of the right ventricle reduces the diastolic peak, a pattern seen in constrictive and restrictive cardiac disease (Figure 13).48 Tricuspid regurgitation attenuates the systolic peak of caval ¯ow, sometimes to the extent that reverse ¯ow occurs (Figure 14).48 MR is commonly requested for the assessment of pericardial disease and therefore the ability to measure caval ¯ow is an important adjunct, providing an estimate of the functional signi®cance of the disease. In patients with obstruction of the superior vena cava, absence of ¯ow can be con®rmed and reverse ¯ow in the azygous vein can be measured (Figure 15).48

4.4

Pulmonary Veins

Normal pulmonary venous ¯ow measured by MR velocity mapping shows two peaks of forward ¯ow, one during ventricular systole and the other in diastole.49,50 A small back ¯ow also occurs during atrial systole. A non-

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

7

of using a 2D selective rf pulse to excite a cylinder of tissue. Spatial resolution can then be obtained along the axis of the cylinder by use of the frequency-encoding gradient. The image, acquired in 9 s, displays ¯ow in the portal vein and inferior vena cava.

4.6

Valvular Stenosis

A stenotic valve may be assessed by measuring the ¯ow velocity in the jet of blood passing through the stenosis. In the case of a given ¯ow, an increasingly narrow stenosis leads to an increase in the velocity of ¯ow through the ori®ce. The relationship between the velocity of the jet and the difference between the pressures on either side of the stenosis can be approximated by the modi®ed Bernoulli equation which in its simplest form is:
P ˆ 4V 2

…2†

where
P is the pressure drop across the stenosis in mmHg and V is the velocity in m s 1. Accurate velocity mapping of stenotic jets by MRI requires the use of very short echo times.49,52,53 The velocity map may be through-plane with the jet passing perpendicularly through the chosen imaging plane, or in-plane when the imaging plane is chosen to encompass the length of the jet (Figure 17). Inplane imaging yields a greater number of pixels for the analysis of velocity, but if the jet is small in-plane imaging is less reliable because of partial volume effects and movement of the jet out of the imaging plane. It is preferable to acquire data in both planes.

Figure 10 (a) A spin echo image in an oblique plane through the ascending aorta, aortic arch, and descending thoracic aorta showing a dilated atherosclerotic thoracic aorta with an intimal ¯ap (arrows) separating the true lumen [t] from the thrombosed false lumen [f]. (b) The systolic velocity image shows high velocity in the true lumen and zero velocity in the false lumen: 1, Left ventricle; 2, ascending aorta; 3, right pulmonary artery; 4, left atrium. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

compliant left ventricle produces high left atrial pressure during atrial systole, causing the retrograde ¯ow in the pulmonary veins to become larger than the ¯ow through the mitral valve. An attenuated systolic forward ¯ow peak has been demonstrated in patients with mitral valve regurgitation and the degree of this attenuation correlates well with the severity of regurgitation.51

4.5 Portal Vein The nature of the Fourier ¯ow imaging technique, in that it only resolves one spatial dimension, requires that some form of signal localization is applied. Figure 16 demonstrates a method

4.7

Flow in the Coronary Arteries

The development of fast imaging techniques has improved the ability of MRI to image directly the proximal portions of both the left and right coronary arteries, but these are not comparable in quality with X-ray angiograms.54,55 It is doubtful however that the place for MR is simply as another method of demonstrating coronary anatomy, subject to interpretation but revealing no information about ¯ow in the vessels. The feasibility of MRI for the measurement of ¯ow in the epicardial coronary arteries has been demonstrated (Figure 18).37,56 Coronary artery bypass grafts can be assessed using MR velocity mapping.57 However, ¯ow measurement within the grafts is not always possible because of signal loss due to sternal suture and clips left after surgery.

4.8

Peripheral Arteries

Peripheral atherosclerosis produces clinical problems either by reducing blood ¯ow or by the release of emboli from ulcerated plaques. Arterial stenosis can be detected and its severity can be assessed from measurement of changes in the velocity pro®le across the stenosis using MR velocity mapping (Figure 19). The ¯ow ratio in paired vessels, like the iliac arteries, is greater than 0.85 in healthy volunteers (Figure 20) and less

8 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI Normal subject 53 yrs

400

Pulmonary hypertension

400

+

+ +

350

+

300 +

+

250 +

Flow (mL s–1)

Flow (mL s–1)

300

+

200 +

150

+

100

+

200

+

+ +

100 +

+

+

+

+

+ + +

0

+

50 +

–50

+

+

0 0

200

+

+

400

+

+

–100 –200

600

0

200

Time (ms) (a)

400

600

Time (ms) (b)

Figure 11 Main pulmonary artery ¯ow volume curves measured by cine resonance velocity mapping (a) in a normal subject and (b) in a patient with pulmonary arterial hypertension: & = net ¯ow; + = forward ¯ow, ^ = reverse ¯ow. In the patient studied, the net ¯ow, forward, and reverse ¯ow were irregular and the reverse ¯ow was relatively large. (Reproduced with permission of Mosby-Year Book, Inc. from H. G. Bogren, R. H. Klipstein, R. H. Mohiaddin, D. N. Firmin, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, Am. Heart J., 1989, 118, 990)

(b) 35 MPA RPA LPA

25

Flow (L min–1)

20 15 10 5 0 –5 –10

0

200

400

600

Time (ms) (c) 16 14

MPA RPA LPA

Flow (L min–1)

12 10 8 6 4 2 0

0

200

400 Time (ms)

600

800

Figure 12 (a) A spin echo image of the pulmonary artery bifurcation of a patient with left lung transplantation. (b) Flow curves of the main (MPA), right (RPA), and left (LPA) pulmonary arteries of the same patient calculated from the complete cine velocity mapping. Blood ¯ow in the transplanted left pulmonary artery is qualitatively and quantitatively different from that in the native right pulmonary artery. (c) Flow-volume curves of the MPA, RPA, and LPA pulmonary arteries calculated from the complete cine velocity mapping acquired in a healthy volunteer. Flow in the RPA and LPA is qualitatively and quantitatively similar: 1, main pulmonary artery; 2, right pulmonary artery; 3, left pulmonary artery; 4, ascending aorta; 5, descending aorta. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

9

(b) 6 5 SVC Flow (L min–1)

4 3 2 1 0 –1

0

200

400

600

Time (ms)

Figure 13 (a) A spin echo image in a transverse plane at mid-ventricular level in a patient with constrictive pericarditis showing pericardial thickening (arrows). (b) Superior vena caval ¯ow curve of the previous patient measured from the complete cine velocity map acquisition throughout the cardiac cycle. The diastolic peak is attenuated which implies impaired right ventricular ®lling: 1, left ventricle; 2, right ventricle; 3, right atrium. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

12

+

15

Desc Ao + IVC

+

10

Flow (L min–1)

Flow (L min–1)

9

+

6 + 3 + + + 0

+ + 200

+

+

+

+

0

+ +

+ +

4

+ +

+ +

+ +

+ +

+ +

SVC 6

2

+

0

Azygos

8

+

–3

+ IVC

SVC

12

+

+

–2 + 0 400 Time (ms)

600

800

Figure 14 Superior and inferior vena caval ¯ow curves in a patient with tricuspid valve regurgitation. The systolic peak is attenuated and there is retrograde ¯ow in the inferior vena cava in systole. (Reproduced with permission of RSNA Publications from R. H. Mohiaddin, S. L. Wann, S. R. Underwood, D. N. Firmin, R. S. O. Rees, and D. B. Longmore, Radiology, 1990, 177, 537)

100

200 Time (ms)

300

400

Figure 15 Flow curves in the superior and inferior venae cavae in a patient with a rhabdomyosarcoma and obstruction of the superior vena cava showing high ¯ow with a normal pattern in the inferior vena cava, absent ¯ow in the superior vena cava, and retrograde ¯ow in the azygos vein. (Reproduced with permission of RSNA Publications from R. H. Mohiaddin, S. L. Wann, S. R. Underwood, D. N. Firmin, R. S. O. Rees, and D. B. Longmore, Radiology, 1990, 177, 537)

10 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI (a)

Excitation cylinder Vessel

Readout Flow sensitivity

Scan geometry

Figure 16 Images from a velocity quanti®cation protocol. (a) Geometry of the procedure. The patient is placed in a supine position. The axis of the cylindric excitation is in the patient's anterioposterior (AP) direction. The direction of the ¯ow-encoding gradient is oblique to the major axes and is chosen to be aligned with the portal vein. The frequency-encoding axis is chosen to be coincident with the axis of the cylindric excitation pulse. (b) Coronal pilot image (acquisition time 20s for ®ve images). (c) Quantitative velocity image in which the horizontal axis is in velocity units and the vertical axis represents spatial displacement of the measured ¯ow along the AP dimension. (Reprinted with permission of Mosby Publishers from D. N. Firmin, C. L. Dumoulin, and R. H. Mohiaddin, in `Magnetic Resonance Angiography, Concepts and Applications', ed. E. J. Potchen, E. M. Haacke, J. E. Siebert and A. Gottschalk, 1993)

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 17 (a) A gradient echo image and (b) a corresponding velocity map in a patient with aortic stenosis. The unusual oblique plane was necessary to orient the abnormal jet direction vertically for velocity encoding. Velocity pro®le displayed in the center of the jet recorded a peak velocity of 3.3 m s 1 (44 mmHg): 1, the left ventricle; 2, ascending aorta; 3, pulmonary artery; 4, right atrium. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin and D. B. Longmore, Circulation, 1993, 88, 264)

than 0.85 in patients with tight stenosis of one iliac artery [Figure 21(a)].58 In patients studied pre- and postangioplasty, the preangioplasty ratio was 0.31 and postangioplasty improved to 0.79 [Figure 21(b)]. This ®nding was supported by improvement in the walking distance and in the peripheral Doppler pressure measurement (the posterior tibial/brachial artery index was right = 1, left = 0.88 on the preangioplasty recordings and right = 1, left = 0.96 on the postangioplasty recordings).58

11

Figure 18 (a) In-plane early diastolic magnitude images and (b) velocity map of a normal left anterior descending coronary artery (LAD). The peak ¯ow was measured at 12 cm s 1. (LV, left ventricle; RV, right ventricle). (Reproduced with permission of Williams & Wilkins from J. Keegan, D. N. Firmin, P. D. Gatehouse, and D. B. Longmore, Magn. Reson. Med., 1994, 31, 526)

4.9 Congenital Heart Disease MR velocity mapping has been very successful in grownup patients with congenital heart disease.52,59,60,61 Intra and extra cardiac shunting can be measured in a number of ways by MR velocity mapping. Flow directly through atrial and ventricular defects can be visualized [Figures 22(a) and (b)], but the best method has been to measure the pulmonary (Qp) to systemic (Qs) ratio directly from aortic and pulmonary ¯ow [Figure 22(c)].60 Other structures of interest for the measurement of ¯ow are surgically created shunts and conduits, and baf¯e obstruction following the Mustard procedure (see section 5, *). The technique is also useful for following up patients after Fontan's operation and its modi®cations (see section 5, {). The hemodynamic signi®cance of aortic coarctation or recoarctation can be assessed noninvasively by MR velocity mapping.59 The modi®ed Bernoulli equation can be used to calculate the pressure difference across the diseased

12 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 19 Multiple atheromatous plaques causing stenoses of both common iliac arteries and the origin of the left internal and external iliac arteries (arrows). (a) Spin echo image. (b) Magnetic resonance velocity map in the same plane as (a) showing velocity pro®les across [1] the abdominal aorta, [2] the right, and [3] the left iliac arteries. There is increased peak velocity in both iliac arteries compared with the aorta. The peak velocity is greater in the left iliac artery which demonstrates that the stenosis on the left is greater than that on the right. (Reproduced with permission from R. H. Mohiaddin, C. Sampson, D. N. Firmin, and D. B. Longmore, Eur. J. Vasc. Surg., 1991, 5, 383)

segment from peak jet velocity in the coarctation. Abnormalities in aortic volume ¯ow and aortic ¯ow waveforms distal to the coarctation can also be measured and could represent an additional index for monitoring the hemodynamic signi®cance of coarctation or recoarctation in patients (Figure 23).59

1.4 1.2

Abd A RIA LIA

Flow (L s–1)

1.0 0.8

5

0.6 0.4 0.2 0.0 –0.2

200

400 Time (ms)

600

Figure 20 Flow curve in the abdominal aorta, and right and left iliac arteries of a volunteer. Flow is qualitatively similar in the three arteries and quantitatively similar in the R and L iliac arteries. (Reproduced with permission from R. H. Mohiaddin, C. Sampson, D. N. Firmin, and D. B. Longmore, Eur. J. Vasc. Surg., 1991, 5, 383)

NOTES

*Mustard's operation was designed to redirect the in¯ow streams of blood in patients with congenital disease of transposition of the great arteries. The atrial septum is removed and a saddle-shaped baf¯e (Dacron or pericardial) is placed within the atria to direct the systemic venous blood through the mitral valve to the left ventricle, and pulmonary artery and pulmonary venous blood to traverse the tricuspid valve to the right ventricle and aorta. {Fontan's operation was designed to redirect systemic venous blood in patients with tricuspid atresia to the pulmonary arterial circulation. The right atrial appendage is usually connected to the pulmonary trunk. Modi®cations of Fontan's operation have been used for other complex cardiac malformations.

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

13

Figure 21 Flow in the abdominal aorta and R and L iliac arteries of a patient (a) before angioplasty and (b) after angioplasty. The ¯ow in the left artery has improved and it is interesting to note that the ¯ow in the right artery has decreased. (Reproduced with permission from R. H. Mohiaddin, C. Sampson, D. N. Firmin, and D. B. Longmore, Eur. J. Vasc. Surg., 1991, 5, 383)

(c) 50 MPA (9.0 L/min) Aorta (4.7 L/min)

Flow (L/min)

40 30 20 10 0 –10

0

200

400 Time (ms)

600

Figure 22 (a) Gradient echo image showing a large atrial septal defect (arrow). (b) Velocity map of the same slice encoded from bottom to top of the image (posterior to anterior). The left-to-right shunt through the defect is seen black. (c) Flow volume curve of the main pulmonary artery (Qp) and the ascending aorta (Qs) in the same patient (Qp/Qs ratio = 1.9) : 1, left atrium; 2, right atrium; 3, aorta; 4, main pulmonary artery. (Reproduced with permission of the American Heart Association from R. H. Mohiaddin, and D. B. Longmore, Circulation, 1993, 88, 264)

14 BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI

Figure 23 (a) A spin echo image (TE 40 ms) acquired in a sagittal plane in a patient with aortic coarctation (large arrow) and (b) a corresponding velocity map acquired during systole with velocity encoding displayed vertically on the image. The sagittal plane was rotated during velocity map acquisition to align the velocity-encoding direction with the jet. The velocity maps indicate zero velocity as midgray, caudal velocities in lighter shades of gray, and cranial velocities in darker shades of gray. Coarctation jet velocity is seen in white with a velocity pro®le (m s 1) displayed in the center of the jet. A reverse ¯ow is seen in black anterior to the jet: 1, descending aorta, left subclavian artery (small arrow); 2, right ventricle; 3, left ventricle; 4, pulmonary trunk; 5, left atrium. (Reproduced with permission of Elsevier Science from R. H. Mohiaddin, P. J. Kilner, R. S. O. Rees, and D. B. Longmore, J. Am. Coll. Cardiol., 1993, 22, 1515)

6 RELATED ARTICLES Time-of-Flight Method of MRA; Whole Body Magnetic Resonance Angiography.

7 REFERENCES 1. E. L. Hahn, Phys. Rev., 1950, 80, 580. 2. G. Suryan, Proc. Indian Acad. Sci., 1951, 33A, 107. 3. J. P. Grant and C. Back, Med. Phys., 1982, 9, 188.

4. C. R. George, G. Jacobs, W. J. MacIntyre, R. J. Lorig, R. T. Go, Y. Nose, and T. F. Meaney, Radiology, 1984, 151, 421. 5. W. G. Bradley, V. Waluch, K. S. Lai, E. J. Fernandez, and C. Spahter, AJR, 1984, 143, 1167. 6. J. R. Singer and L. E. Crooks, Science, 1983, 221, 654. 7. V. Waluch and W. G. Bradley, J. Comput. Assist. Tomogr., 1984, 8, 594. 8. G. K. Von Schulthess and C. B. Higgins, Radiology, 1985, 157, 687. 9. W. H. Perman, P. R. Moran, R. A. Moran, and M. A. Bernstein, J. Comput. Assist. Tomogr., 1986, 10, 473. 10. A. N. Garroway, J. Phys. D., 1974, 7, L159. 11. K. R. Thulborn, J. C. Waterton, and G. K. Radda, J. Magn. Reson., 1981, 45, 188. 12. F. W. Wehrli, A. Shimakawa, G. T. Gullberg and J. R. MacFall, Radiology, 1986, 160, 781. 13. J. Henning, M. Mueri, H. Friedburg, and P. Brunner, J. Comput. Assist. Tomogr., 1987, 11, 872. 14. D. A. Feinberg, L. E. Crooks, J. Hoenninger, M. Arakawa, and J. Watts, Radiology, 1984, 153, 177. 15. K. Shimizu, T. Matsuda, T. Sakurai, A. Fujita, H. Ohara, S. Okamura, S. Hashimoto, H. Mano, C. Kawai, and M. Kiri, Radiology, 1986, 159, 195. 16. L. Axel, A. Shimakawa, and J. MacFall, Magn. Reson. Imaging, 1986, 4, 199. 17. T. Matsuda, K. Shimizu, T. Sakurai, A. Fujita, H. Ohara, S. Okamura, S. Hashimoto, S. Tamaki, and C. Kawai, Radiology, 1987, 162, 857. 18. R. R. Edelman, H. P. Mattle, J. Klee®eld, and M. S. Silver, Radiology, 1989, 171, 551. 19. S. H. Izen and E. M. Haacke, IEEE Trans. Med. Imaging, 1990, MI-9, 450. 20. P. R. Moran, Magn. Rescon. Imaging, 1982, 1, 197. 21. T. W. Redpath, D. G. Norris, R. A. Jones, and M. S. Hutchinson, Phys. Med. Biol., 1984, 29, 891. 22. D. A. Feinberg, L. E. Crooks, P. Sheldon, J. Hoenninger, III, J. Watts, and M. Arakawa, Magn. Reson. Med., 1985, 2, 555. 23. J. Hennig, M. Mueri, P. Brunner, and H. Friedburg, Radiology, 1988, 166, 237. 24. P. van Dijk, J. Comput. Assist. Tomogr., 1984, 8, 429. 25. D. J. Bryant, J. A. Payne, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1984, 8, 588. 26. I. R. Young, G. M. Bydder, and J. A. Payne, Magn. Reson. Med., 1986, 3, 175. 27. J. P. Ridgway, and M. A. Smith, Br. J. Radiol., 1986, 59, 603. 28. G. L. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 29. E. M. Haacke, and G. W. Lenz, AJR, 1987, 148, 1251. 30. D. N. Firmin, G. L. Nayler, R. H. Klipstein, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, J. Comput. Assist. Tomogr., 1987, 11, 751. 31. S. R. Underwood, D. N. Firmin, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. Heart J., 1987, 57, 404. 32. D. N. Firmin, G. L. Nayler, P. J. Kilner, and D. B. Longmore, Magn. Reson. Med., 1990, 14, 230. 33. D. N. Firmin, R. H. Klipstein, G. L. Houns®eld, M. P. Paley, and D. B. Longmore, Magn. Reson. Med., 1989, 12, 316. 34. C. B. Ahn, J. H. Kim, and Z. H. Cho, Trans. Med. Imaging, 1986, MI-5, 2. 35. C. H. Meyer, B. S. Hu, D. G. Nishimura, and A. Macovski, Magn. Reson. Imaging, 1992, 28, 202. 36. P. D. Gatehouse, D. N. Firmin, S. Collins, and D. B. Longmore, Magn. Reson. Med., 1994, 31, 1. 37. J. Keegan, D. N. Firmin, P. D. Gatehouse, and D. B. Longmore, Magn. Reson. Med., 1994, 31, 526.

BLOOD FLOW: QUANTITATIVE MEASUREMENT BY MRI 38. R. H. Klipstein, D. N. Firmin, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, Br. Heart J., 1987, 58, 316. 39. H. G. Bogren, R. H. Mohiaddin, R. H. Klipstein, D. N. Firmin, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, Am. Heart J., 1989, 118, 234. 40. M. C. Dulce, G. H. Mostbeck, M. O'Sullivan, M. Cheitlin, G. R. Caputo, and C. B. Higgins, Radiology, 1992, 185, 235. 41. R. H. Mohiaddin, D. N. Firmin, and D. B. Longmore, J. Appl. Physiol., 1993, 74, 492. 42. C. L. Dumoulin, D. J. Doorly, and C. G. Caro, Magn. Reson. Med., 1993, 29, 44. 43. C. A. Nienaber, R. P. Spielmann, Y. von Kodolitsch, V. Siglow, A. Piepho, T. Jaup, V. Nicolas, P. Weber, H. Triebel, and W. Bleifeld, Circulation, 1992, 85, 434. 44. H. G. Bogren, S. R. Underwood, D. N. Firmin, R. H. Mohiaddin, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. J. Radiol., 1988, 61, 456. 45. H. G. Bogren, R. H. Klipstein, R. H. Mohiaddin, D. N. Firmin, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, Am. Heart J., 1989, 118, 990. 46. C. Kondo, G. R. Caputo, T. Masui, E. Foster, M. O'Sullivan, M. S. Stulbag, J. Golden, K. Catterjee, and C. B. Higgins, Radiology, 1992, 183, 751. 47. R. H. Mohiaddin, R. Paz, S. Theodoropulus, D. N. Firmin, D. B. Longmore, and M. H. Yacoub, J. Thorac. Cardiovasc. Surg., 1991, 101, 1016. 48. R. H. Mohiaddin, S. L. Wann, S. R. Underwood, D. N. Firmin, R. S. O. Rees, and D. B. Longmore, Radiology, 1990, 177, 537. 49. R. H. Mohiaddin, M. Amanuma, P. J. Kilner, D. J. Pennell, C. C. Manzara, and D. B. Longmore, J. Comput. Assist. Tomogr., 1991, 15, 237. 50. R. H. Mohiaddin, M. Amanuma, and D. B. Longmore, Am. J. Noninvasive Cardiol., 1992, 6, 13. 51. A. van Rossum, K. H. Sprenger, and F. C. Peels, Eur. Heart J., 1991, 12, 117. 52. P. J. Kilner, D. N. Firmin, R. S. O. Rees, J. E. Martinez, D. J. Pennell, R. H. Mohiaddin, S. R. Underwood, and D. B. Longmore, Radiology, 1991, 178, 229.

15

53. P. J. Kilner, C. C. Manzara, R. H. Mohiaddin, P. J. Pennell, M. G. St John Sutton, D. N. Firmin, and D. B. Longmore, Circulation, 1993, 87, 1239. 54. R. R. Edelman, W. J. Manning, D. Burstein, and S. Paulin, Radiology, 1991, 181, 641. 55. S. J. Wang, D. G. Nishimura, and A. Macovski, Magn. Reson. Med., 1992, 23, 109. 56. R. R. Edelman, W. J. Manning, E. Gervino, and W. Li, J. Magn. Reson. Imaging, 1993, 3, 699. 57. J. F. Debatin, A. Strong, H. D. Sostman, R. Negro-Vilar, S. S. Paine, J. M. Douglas, and N. J. Pelc, J. Magn. Reson. Imag., 1993, 3, 443. 58. R. H. Mohiaddin, C. Sampson, D. N. Firmin, and D. B. Longmore, Eur. J. Vasc. Surg., 1991, 5, 383. 59. R. H. Mohiaddin, P. J. Kilner, R. S. O. Rees, and D. B. Longmore, J. Am. Coll. Cardiol., 1993, 22, 1515. 60. R. S. O. Rees, D. N. Firmin, R. H. Mohiaddin, S. R. Underwood, and D. B. Longmore, Am. J. Cardiol., 1989, 64, 953. 61. J. E. Martinez, R. H. Mohiaddin, P. J. Kilner, K. Khaw, R. S. O. Rees, J. Somerville, and D. B. Longmore, J. Am. Coll. Cardiol., 1992, 20, 338.

Biographical Sketches David N. Firmin. b 1955. B.Sc., 1978, M.Phil., 1982, Ph.D., 1989, Magnetic Resonance Imaging of Blood Flow, University of London. Introduced to NMR By D. B. Longmore, National Heart and Chest Hospital, London. Senior Lecturer, National Heart and Lung Institute, University of London, 1982±present. Approx. 210 publications. Research specialties: rapid, ¯ow, and cardiovascular imaging by use of magnetic resonance. Raad H. Mohiaddin. b 1957. After graduation with M.B., Ch.B., 1981, Mosul-Iraq, received training in cardiology, M.Sc., 1985, Ph.D., 1994, Structural and Functional Evaluation of Atherosclerotic Vascular Disease by Magnetic Resonance Imaging, University of London, London. Introduced to NMR by D. B. Longmore, National Heart and Chest Hospital, London. Approx. 200 publications. Research specialties: magnetic resonance imaging as applied to congenital and acquired cardiovascular diseases.

BREAST MRI

Breast MRI Sylvia H. Heywang-KoÈbrunner University of Halle, Germany

Hans Oellinger UniversitaÈts-Klinikum Rudolf-Virchow, Freie UniversitaÈt Berlin, Germany

1 INTRODUCTION

At present, various combinations of pulse sequences are still being tested. These use differences of T1, of T2, and of the reasonance frequency of silicon compared with the surrounding fat, ®brosis, or glandular tissue.29±33 In general, a combination of at least three pulse sequences is recommended. These should, for example, include a T1-weighted pulse sequence with and without fat suppression, and one with silicone or water suppression as well as a fast high-resolution T2-weighted pulse sequence. The suppression of fat, water, or silicone can be obtained by spectrally selective presaturation pulses or by signal nulling with IR sequences. In order to cut potential ruptures orthogonally, these pulse sequences are usually applied in different planes (coronal, transverse, and/or sagittal). For the detection of discrete ruptures, at least one of the pulse sequences should allow slice thicknesses below 2±3 mm. For the other pulse sequences, a slice thickness below 5 mm appears optimal. The implant itself is generally covered by a ®ne shell, which in intact implants can usually not be recognized since it is directly attached to the capsule of ®brous tissue which surrounds the implant. When the shell of the implant is ruptured, silicone can leak. Depending on whether the leakage of silicone is con®ned to the area surrounded by the ®brous capsule, or whether the leak is beyond the ®brous capsule (about 5% of the ruptures), the rupture is called intra± or extracapsular. The following signs indicating a rupture have so far been described (Figure 1): 1. Linguine sign: The shell is ruptured, silicone remains predominantly within the ®brous capsule, and the collapsed shell ¯oats within the silicone. The tomographic slices show wavy dark lines within the silicone, which represent the collapsed implant shell ¯oating within the silicone. The wavy lines are called `linguine'.

Breast Fibrous capsule "Linguine" = shell of ruptured implant (a)


,




, 

On the basis of the high soft tissue contrast of MRI and based on suspected differences of T1 and T2 values between benign and malignant breast tissues, MRI was initially considered to be a particularly promising tool for breast imaging. Unfortunately, the expected tissue characterization by MRI could not be demonstrated despite intense further research; signal intensities on various pulse sequences, and in vivo calculated T1 and T2 values, varied signi®cantly in both benign and malignant tissues. Both MR parameters and signal intensities seemed to correlate well with the water/content, cells or ®brosis within the corresponding tissues, but did not allow a reliable distinction between benign and malignant tissues. This has proven true until today, even after the application of more sophisticated evaluation such as computer-aided evaluation methods.1±5 For this reason, plain MRI (= MRI without contrast agent) does not play a role in the detection and diagnosis of breast malignancy nowadays. The disappointing results mentioned above, however, initiated work concerning the use of contrast agents for MR of the breast and the very encouraging results caused its further exploration.6±8 Today contrast-enhanced MRI of the breast is developing as an important additional tool for breast diagnostics, providing new information which is different from that of conventional imaging and is thus most valuable in certain problem areas of conventional breast diagnostics.9±28 In one special area, however, both tomographic imaging and the high soft tissue contrast of plain MRI has proved helpful; this special area concerns the evaluation of the integrity of silicone implants.29±33 This chapter will provide an overview of these two major applications of breast MRI; plain MRI for the evaluation of implant integrity and contrast-enhanced (CE) MRI of the breast.

1

Silicone outside and inside the implant shell

Fibrous capsule

(b)

2 PLAIN MRI Ð DETECTION AND DIAGNOSIS OF IMPLANT FAILURE

Based on the latest knowledge concerning potential hazards caused by leaking or ruptured implants and free silicone, the detection of possible implant failure has become an important new demand. The capabilities of conventional methods such as mammography and ultrasound for detecting extra- and particularly intracapsular leakage are very limited. Hence the use of MRI in this ®eld has been evaluated.

Inner silicone lumen with droplets of saline Outside saline lumen

(c)

Figure 1 Morphological MR signs of implant rupture: (a) `linguine sign'; (b) `keyhole sign' or `reverse c-sign' (right), and (c) `droplet sign' in double lumen implant

2 BREAST MRI 2. When silicone enters between the shell and the ®brous capsule, silicone becomes visible outside the shell. Depending on the shape of the shell in this area, this sign is called `csign' or `keyhole sign'. 3. In double lumen implants, saline may enter the silicone lumen of the implant. Since saline and silicone gel do not mix, small droplets become visible; this sign is called the `droplet sign' or the `salad oil' sign. 4. Finally small droplets of silicone can be visualized outside the capsule. This sign indicates extracapsular rupture. To date, with careful examination techniques using several pulse sequences and thin slices, very good sensitivities (above 80%) and speci®cities (above 90%) have been reported.29±33 3 CONTRAST-ENHANCED MRI OF THE BREAST Based on the unsatisfying results of plain MRI mentioned above, CE MRI of the female breast has been performed and investigated by us since 1985.6±8,16,17,21,22,28 Other workers have followed and contributed signi®cantly.9±15,18±20,23±27 Today, contrast-enhanced MRI is gaining increasing importance as an additional tool for breast diagnostics. 3.1 Principles of Contrast-Enhanced MRI of the Breast The complete breast is examined with thin and contiguous slices before and after the i.v. application of Gd±DTPA. By comparison of the corresponding pre- and postcontrast slices, tissue enhancement can be detected or excluded. The fact that the large majority, possibly all, invasive breast malignancies do enhance is supported by the following fundamental data:34±40 (i) an increased number and size of vessels has also been encountered histologically in and around breast malignancies, as studies with special vascular staining show; (ii) according to in vivo animal experiments, vessel sprouting seems to be a prerequisite for tumor growth beyond a size of about 2±3 mm; (iii) tumor growth seems to depend on both increased vascularity and increased vascular permeability; (iv) both increased vascularity and vascular permeability are induced by socalled angiogenesis and permeability factors, numbers of which have already been identi®ed; (v) cells from both invasive and numerous in situ carcinomas were able to induce vascular sprouting when transplanted on an animal cornea, while cells from normal breast tissue could not induce such changes; (vi) it has also been reported that cells from several benign entities were equally well able to produce identical or similar angiogenesis and permeability factors leading to similar vascular changes as in malignancy. Benign changes with similar capabilities included some proliferative dysplasias, healing wounds, and benign tumors. The above-mentioned observations support may of the MR ®ndings. They may also serve to explain both the reported excellent sensitivity and the limited speci®city of CE MRI. 3.2 Techniques For prognostic reasons, it is generally agreed that breast imaging techniques should be able to image carcinomas of 5

mm and above reliably. Furthermore, it is also desirable to detect in situ carcinomas which usually grow within the ducts where they may extend within areas of very variable size. The ducts themselves, however, rarely exceed 1±3 mm in diameter and may be surrounded by uninvolved nonenhancing tissue. Hence with a tomographic imaging modality such as MRI it is important to minimize slice thickness so as to improve resolution and contrast. In order to achieve suf®cient signal-to-noise ratio, a dedicated single or double breast coil has to be used. Cardiac motion artifacts also need to be eliminated. At present, the best way to achieve the latter is by switching phase and frequency encoding gradients, so that the artifacts cross the axillas instead of the breast. An alternative is the use of coronal slices, where most of the artifacts neither cross the breast nor the axillas. (The use of local presaturation is not useful, since generally larger areas are extinguished than would be disturbed by artifacts. In our experience ¯ow compensation increases echo time without suf®cient effect on artifact elimination.) In order to obtain suf®ciently small slices without gaps, 3D imaging is strongly recommended. In this way it is possible, for example, to image the breast with 32 transverse slices of 4 mm slice thickness within slightly more than 1 min. By using a rectangular ®eld of view and coronal slices, either the imaging time can be further reduced by a factor of two, or 64 slices of 2-mm slice thickness are possible within the same imaging time. Thus with present 3D techniques both high temporal resolution or high spatial resolution are possible. However, when choosing between high spatial and high temporal resolution it should be remembered that with a 5-mm slice thickness in the worst case a 5-mm lesion may only be half contained within two neighboring slices. This means that only half of the signal increase can be measured in each neighboring slice. Hence 2-mm or thinner slices will probably allow better results, particularly with regard to the detection of intraductal enhancement. The optimum temporal resolution is still controversial.9±17 As far as the choice of pulse sequences is concerned, the highest sensitivity to Gd±DTPA and hence the best contrast of enhancing lesions can be achieved with 3D gradient echo sequences such as FLASH or spoiled GRASS or with some fat saturation techniques like RODEO or spoiled GRASS with fat saturation.23±25 Due to their low sensitivity to Gd±DTPA, SE sequences can no longer be recommended. Since enhancement might be confused with a fat lobule on the postcontrast image alone, elimination of the fat signal has been suggested as a means of improving the detectability of enhancing lesions. Elimination of the fat signal can be achieved in two different ways. 1. On T1-weighted fat suppression sequences, ideally only enhancing tissues display high signal intensity while dysplasia and suppressed fat exhibit low or no signal intensity. The disadvantage of fat suppression techniques is that they are more sensitive to magnetic ®eld inhomogeneities. Within such inhomogeneities, which may be caused by the patient or the breast coil, enhancement might be underestimated or missed. Thus, greater technical prerequisites need to be ful®lled, i.e. more experience and very careful quality control are necessary. 2. Since fat does not enhance, like all other nonenhancing tissues it is ideally eliminated on subtraction images. Software has now been developed which allows routine subtraction of

BREAST MRI

all corresponding pre- and postcontrast slices (2  64 slices) within about 1 min. The major disadvantage which has been ascribed to this technique relates to problems which occur when the patient moves. In our experience, this problem occurs with less than 5±10% of all patients with currently available very short imaging times. Motion and its inherent problems are at least readily recognized with this technique Three-dimensional MIP reconstruction is possible for both fat suppressed and subtracted images. It has been suggested as a means of improving the distinction of enhancing vessels from enhancing ducts.27 As far as optimum dosage is concerned, to date only one dose comparison study has been published.28 In this study, we demonstrated that signi®cantly better results were obtained with FLASH-3D using the higher dosage of 0.16 mmol of Gd± DTPA per kg as compared with the widely used lower dosage of 0.1 mmol of Gd±DTPA per kg. In fact, three out of 54 carcinomatous foci would have been overlooked with the lower dosage. Histologically, these three foci were all smaller than the applied slice thickness of 4 mm. 3.3 Results The results of CE MRI of the breast are based worldwide on the examination of over 5000 patients.6±28 Despite still remaining differences in technique and image interpretation, the following results have been con®rmed: (a) the great majority of invasive carcinomas are enhanced (see Figure 2); (b) no signi®cant enhancement is usually seen in normal breast tissue, in nonproliferative dysplasia, sclerosed benign tumors, old scarring, and cysts; and (c) variable enhancement, which might cause false positive calls, has been encountered in some proliferative dysplasias, in nonsclerosed benign tumors, and in in¯ammatory changes. In over 700 biopsy-proven tissues which we have examined with FLASH-3D, about 75% of the dysplasias enhanced only a little and it was possible to de®ne a threshold so that all carcinomas enhanced more than by this threshold.16,17 We also evaluated the shape of enhancement and roughly estimated the speed of enhancement in these 700 tissues. In a similar manner to mammography, we found that an irregular Table 1

outline of an enhancing lesion was indeed highly suggestive of malignancy, but neither diffuse nor well-circumscribed enhancement could reliably exclude malignancy (for example, 16% of the diffusely enhancing tissues and 21% of our postoperatively examined well circumscribed lesions, in fact, turned out to be malignant). When the speed of enhancement was estimated from two postcontrast studies (®rst measurement: 1±5 min post injection; second measurement 6±11 min post injection), fast enhancement was highly suggestive of malignancy but unfortunately delayed enhancement could not rule out malignancy reliably. This result has been con®rmed by another separate dynamic study in 79 patients, where selected slices were imaged every 30 s after i.v. injection of Gd± DTPA.7,16,17 As to the potential of dynamic contrast studies, controversial results are still being discussed. While Kaiser9,10 reports a sensitivity and speci®city of 97% with dynamic contrast-enhanced MRI, other authors have not been able to reproduce these results, reporting both fast enhancing benign tissues and (what is much more important) slowly enhancing malignancies.12±17 In our experience the use of a slow enhancement speed as a criterion of benignity is comparable to increasing the threshold above which enhancement is considered to be possibly malignant, i.e., speci®city can be improved at the cost of sensitivity. Since the development of very fast pulse sequences such as Turbo-FLASH or echo planar imaging, it has become possible to image the ®rst pass of a contrast agent within a tissue. Initial results with Turbo-FLASH have been promising.26 The optimum use of this additional information needs to be carefully determined, in view of its effect on both sensitivity and speci®city. When the results of various authors are compared6±28 (Table 1), excellent sensitivity values overall have been reported for CE MRI. Indeed, the sensitivity exceeds that of conventional imaging. Differences can be explained by different techniques (sometimes thick slices with gaps, the choice of pulse sequence, the dosage of Gd±DTPA), by different guidelines for interpretation, and by different patient selection. On comparison with conventional imaging alone, the sensitivity was however, improved in general. This diagnostic gain was, however, most prominent within mammographically dense and/or distorted tissue, while in fatty breasts and in cases with microcalci®cations little or no diagnostic gain was achieved.

Sensitivities and Speci®cities of Enhanced MRI, as Reported in the Literature

Author

Year

Sensitivity of MRI (%)

Speci®city of MRI (%)

Sensitivity of Conv.a (%)

Speci®city of Conv.a (%)

Allgayer (1)c Fischer (2) Gilles (3) Harms (3) Heywang-KoÈbrunner (1) Heywang-KoÈbrunner (3) Kaiser (1) Lewis-Jones (1) Oellinger (4)

1991 1992 1993 1993 1993 1993 1992 1991 1993

90 95 93.3 94 98.6b 99.5b 97b 100 88

40 89 63.1 37 65.0 28.0 97 91 80

ni ni ni ni 60 82 ni 100 83

ni ni ni ni 45 18 ni 60 75

a

Conv. = Sensitivity/Speci®city of conventional imaging. Sensitivity/Speci®city based on MR and conventional imaging. c (1), problem cases; (2) retrospective study; (3) preoperative patients; (4) evaluation of multicentricity. ni = not indicated. b

3

4 BREAST MRI

Figure 2 Mammography versus MRI in cases of very dense breasts. (a) Mammography of a 43 year old patient, cranio-caudal view. Very dense right breast, no radiological evidence of malignancy. (b) MR image before injection of the contrast agent; anatomical region of the upper quadrants. Basically fatty tissue and small areas of parenchyma in the upper inner quadrants of both sides. (c) MR image after application of the contrast agent gadolinium±DTPA with moderately strong signal intensity rise in the parenchyma of the inner quadrant of the patient's right breast (see arrow). Same position as in the precontrast MR image shown in (b). (d) The use of the subtraction method (postcontrast minus precontrast) allows the postcontrast enhancement of carcinomas to be seen much better. This method is especially useful in the case of small carcinomas. The histology revealed an invasive ductal carcinoma

BREAST MRI

With respect to microcalci®cations and in situ carcinomas, the following may be stated. Mostly due to patient preselection, experience concerning the capabilities of MRI in this ®eld is still limited to date. Some authors reported misses. Although our in situ carcinomas did enhance, some did so diffusely and many slowly. Thus, detection of in situ carcinomas and their differentiation from proliferative dysplasias is certainly also a question of threshold and interpretation. Those 50±60% of in situ carcinomas which contain microcalci®cations are already excellently visualized by mammography. Even though improvement of the mammographic speci®city in this ®eld is still desirable, we do not recommend MRI in this case for the following reasons: 1. It is not known as yet whether all in situ carcinomas are suf®ciently vascularized to be detectable by contrastenhanced MRI. 2. Further research must show whether in situ carcinomas can be included in the enhancement thresholds of MRI without risking too many false positive calls. In our own experience, numerous false positive calls occurred particularly in cases with mammographic microcalci®cations. 3. Due to the additional possibility of partial volume effects, excellent thin slice techniques will be necessary together with extremely careful application of any threshold. Thus, we do not recommend reliance on MRI in this ®eld at present and we do not recommend application of MRI as a means of answering these questions. Instead diagnostic decisions should be based on careful analysis of the mammogram and possibly other diagnostic tests such as transcutaneous biopsy. As far as noncalci®ed in situ carcinomas are concerned, it is important to mention that seven out of our 21 in situ carcinomas were detected by MRI alone. The in situ carcinomas detected by MRI frequently became apparent as a focal area of enhancement and sometimes enhancement with a ductlike shape. While the sensitivity results of CE MRI have been very satisfying, improvement of speci®city has been a major concern (Table 1). Overall the following possibilities are being discussed: (i) All available information including that of shape and speed of enhancement should be used for the ®nal diagnosis. Because of the existence of both diffusely and slowly enhancing carcinomas, it is necessary to warn against overinterpretation of shape or speed of enhancement. Optimization of interpretation guidelines in this ®eld (using optimum techniques) will therefore be an important task for the near future. (ii) In addition, all available information including that of mammography and clinical examination should be used for the ®nal diagnosis. MRI is not a stand-alone method! (iii) In order to avoid unnecessary biopsy of MR-detected small benign enhancing lesions, MR follow-up of benignappearing lesions (predominantly well-circumscribed round or oval lesions, which are likely to be small ®broadenomas) is recommended. MR-guided stereotaxic transcutaneaous biopsy should be considered for indeterminate benign-appearing lesions. First experiences with such devices are presently being acquired. (iv) Finally unnecessary costs should be avoided and optimum diagnostic gain attempted by using MRI exclusively for those indications where de®nite advantages compared with conventional imaging can be expected.

5

On the basis of our own experiences over 2000 contrastenhanced MR studies of the breast, we do not recommend contrast-enhanced MRI for the following indications: (a) evaluation of microcalci®cations (see above), and (b) evaluation of those cases where diffuse enhancement is to be expected in any event, since in these cases malignancy can never be excluded in principle. Such cases include clinically obvious in¯ammatory changes, patients with secretory disease, and patients with known enhancing dysplasia from previous studies. As an additional method however, CE MRI has been most helpful in diagnostic problems within mammographically dense or distorted tissue. Most interesting indications in these breasts include the search for primary tumors (not detected by conventional imaging) and the exclusion of multicentricity or contralateral involvement in patients with small tumors and very dense tissue (see Figure 2).15±17,24 Contrast-enhanced MRI is particularly useful for patients with severe scarring after surgery, after radiation therapy, and after silicone implant.16,20±22 It must, however, be added that MRI should not be used within the ®rst 3±6 months after surgery or within the ®rst 12 months after radiotherapy (RT), since posttherapeutic diffuse enhancement usually impairs evaluation during that period. 3.4

Conclusions and Outlook

The use of a contrast agent is obligatory for MR detection and diagnosis of malignancy. Contrast-enhanced MRI is gaining increasing importance as an additional tool for breast diagnostics. However, because of the increasing interest in the method, optimization and standardization of interpretation is an urgent task for the near future. In order to develop optimized criteria for interpretation, an international multicenter study is presently being started. The study is supported by Siemens Corporation, Schering, and Berlex Corporations, and includes only centers with a high reputation in both breast imaging and MRI based on histologically proven data only. Both the optimization of interpretation guidelines, and in a second prospective trial the determination of wellfounded and generally recognized accuracy data, are the major goals of this study. These data, as well as further research by other groups, will help to de®ne the optimum role which MRI should play in integrated diagnostic work-up. MRI is thus gaining its place in breast diagnostics. It must be our concern to use MRI optimally only for those indications where de®nite diagnostic gain can be achieved for the patient without risking false negative calls and without causing an unacceptable number of false positive calls, which might cause unnecessary concern, costs for additional clari®cation, or even surgery. Further development and improvements of the technique are still proceeding.

4

RELATED ARTICLES

Cardiac Gating Practice; Echo-Planar Imaging; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Gadolinium Chelates: Chemistry, Safety, and Behavior; Whole Body Magnetic Resonance: Fast Low-Angle Acquisition Methods.

6 BREAST MRI 5 REFERENCES 1. W. A. Murphy and J. K. Gohagan, `Magnetic Resonance Imaging', ed. D. D. Stark and W. G. Bradley, Jr., C. V. Mosby, St. Louis, MO, 1987, pp. 861±886. 2. J. I. Wiener, A. C. Chako, C. W. Merten, S. Gross, E. L. Coffey, and H. L. Stein, Radiology, 1986, 160, 299. 3. F. S. Alcorn, D. A. Turner, J. W. Clark, et al., Radiographics, 1985, 5, 631. 4. S. H. Heywang, R. Bassermann, G. Fenzl, W. Nathrath, D. Hahn, R. Beck, I. Krischke, and W. Eiermann, Eur. J. Radiol., 1987, 3, 175. 5. J. K. Gohagan, A. E. Tome, E. Spitznagel, et al., Radiology, 1994. 6. S. H. Heywang, D. Hahn, H. Schmid, I. Krischke, W. Eiermann, R. Bassermann, and J. Lissner, J. Comput. Assist. Tomogr., 1986, 10, 199. 7. S. H. Heywang, T. Hilbertz, E. Pruss, A. Wolf, W. Permanetter, W. Eiermann, and J. Lissner, Digitale Bilddiagn., 1988, 7. 8. S. H. Heywang, A. Wolf, E. Pruss, T. Hilbertz, W. Eiermann, and W. Permanetter, Radiology, 1989, 171, 95. 9. W. A. Kaiser and E. Zeitler, Radiology, 1989, 170, 681. 10. W. A. Kaiser, Diagn. Imaging Int., 1992, 44. 11. J. P. Stack, A. M. Redmond, M. B. Codd, P. A. Derran, and J. T. Ennis, Radiology, 1990, 174, 491. 12. F. W. Flickinger, J. D. Allison, R. Sherry, and J. C. Wright, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 955. 13. M. D. Schnall, S. Orel, and L. Muenz, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 120. 14. U. Fischer, D. von Heyden, R. Vosshenrich, I. Vieweg, and E. Grabbe, RoeFo, 1993, 158, 287. 15. H. Oellinger, S. Heins, B. Sander, et al., Eur. Radiol., 1993, 3, 223. 16. S. H. Heywang-KoÈbrunner, `Contrast-enhanced MRI of the Breast', Karger, Basel/Munchen, 1990. 17. S. H. Heywang-KoÈbrunner, Electromedica, 1991, 61, 43. 18. B. Allgayer, P. Lukas, W. Loos, and K. MuÈhlbauer, Roentgenpraxis, 1991, 44, 368. 19. D. Rubens, S. Totterman, A. K. Chacko, K. Kothari, W. LoganYoung, J. Szumowski, J. H. Simon, and E. Zacharich, AJR, 1991, 157, 267. 20. H. G. Lewis-Jones, G. H. Whitehouse, and S. J. Leinster, Clin. Radiol., 1991, 43, 197. 21. S. H. Heywang-KoÈbrunner, A. Schlegel, R. Beck, T. Wendt, W. Kellner, B. Lammetzsch, M. Untch, and W. B. Nathrath, J. Comput. Assist. Tomogr., 1993, 17, 891. 22. S. H. Heywang-KoÈbrunner, T. Hilbertz, R. Beck et al., J. Comput. Assist. Tomogr., 1990, 14, 348. 23. W. B. Pierce, S. E. Harms, D. P. Flamig, R. H. Giffey, W. P. Evans, and J. G. Hugins, Radiology, 1991, 181, 757. 24. S. E. Harms, D. P. Flamig, K. L. Hesley, M. D. Meiches, R. A. Jensen, W. P. Evans, D. A. Savina, and R. V. Wells, Radiology, 1993, 187, 493. 25. W. Whitney, R. J. Herfkens, J. Silverman, D. Ikeda, J. Brumbaugh, S. Jeffrey, L. Esserman, J. Frederickson, C. Meyer, and G.

26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40.

Glover, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 856. C. Boetes, R. D. Mus, J. O. Barentsz, J. H. Hendriks, R. Holland, and J. H. Ruijs, Radiology, 1993, 189, 301. H. Oellinger, B. Sander, U. Quednau, J. Hadijuana, H. Schoenegg, and F. Felix, JMRI, 1993, 3, 109. S. H. Heywang-KoÈbrunner, J. Haustein, C. Pohl, W. M. Bauer, W. Eirmann, W. Permanetter, Radiology, 1994, 191, 639. R. F. Brem, C. M. C. Tempany, and E. A. Zerhouni, J. Comput. Assist. Tomogr., 1992 16, 157. D. P. Gorczyca, S. Sinha, and C. Y. Ahn, Radiology, 1992, 185, 407. L. I. Everson, H. Parantainen, A. E. Stillman, T. Dethi, M. C. Toshager, and B. Cunningham, Radiology, 1993, 189, 301. W. A. Berg, N. D. Anderson, B. W. Chang, E. A. Zerhouni, and T. E. Kuhlman, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 123. S. Mukundan, W. T. Dixon, R. C. Nelson, D. C. Monticciolo, and J. Bostwick, Radiology, 1993, 189, 301. A. Ottinetti and A. Sapino, Breast Cancer Res. Treat., 1988, 11, 241. C. H. Blood and B. R. Zetter, Biochim. Biophys. Acta, 1990, 1032, 89, H. M. Jensen, J. Chen, M. R. De Vault, and A. E. Lewis, Science, 1982, 218, 293. N. Weidner, J. P. Semple, W. R. Welch, and J. Folkman, N. Engl. J. Med., 1991, 324, 1. M. A. Gimbrone, S. B. Leapman, R. S. Cotran, and J. Folkman, J. Exp. Med., 1972, 136, 261. I. F. Tannock, Br. Cancer Res., 1968, 22, 258. D. T. Connolly, D. M. Heuvelman, R. Nelson, J. V. Olander, B. L. Eppley, J. J. Del®no, N. R. Siegel, R. M. Leimgruber, and J. Feder, J. Clin. Invest., 1989, 84, 1470.

Biographical Sketches S. H. Heywang-KoÈbrunner. b 1956. Dr.med., 1982 University of Munich. Fellowship in breast imaging in the US (Georgetown University, Emory University, Betty Ford Breast Diagnosis Center, Washington DC) 1982±83. Radiology residency with special training in CT and MRI, University of Munich (Prof. Dr. J. Lissner); staff member and assistant professor, University of Munich, 1990±92, assistant professor, University of Leipzig, 1993, associate professor, University of Halle, 1994±present. Approx. 150 publications. Research specialties: since 1985, investigation of CE MRI of the breast clinical MRI, breast imaging. H. Oellinger. b 1948. Dr.med., 1989, Free University of Berlin. M.S.M.E. New Jersey Institute of Technology, Fulbright Fellow, 1974±76; site engineer for conventional and nuclear power plants (Siemens, Erlangen, Germany) 1976±82; Radiology Department (Strahlenklinik und Poliklinik, FUB, Prof. Dr. Dr h.c.. R. Felix) with special training in CT and MRI, 1989±present. Research interests: clinical MRI, breast and pelvic imaging.

CARDIAC GATING PRACTICE

Cardiac Gating Practice David N. Firmin National Heart and Lung Institute, University of London, UK

1 INTRODUCTION Conventional magnetic resonance images or spectra are acquired over a period that is much longer than the cardiac cycle and for this reason some method of referencing to the cardiac cycle is required. If cardiac images or spectra were to be acquired without any reference to the cardiac motion, then the position of the heart would vary from one view or signal acquisition to another and artifacts and blurring of the image or signal contamination of the spectrum would result. Although the referencing techniques are often referred to as cardiac gating, strictly speaking they would be better referred to as cardiac synchronization. Gating is a technique used in radionuclide ventriculography, for example, when the acquisition occurs continuously but the cardiac events are used to time the opening and closing of an acquisition `gate' or `gates', so that the image or images that are built up relate to speci®c parts of the cardiac cycle. There have been two approaches to cardiac synchronization of the MR acquisition: the ®rst, known as prospective cardiac gating or cardiac triggering, was initially developed and used for early cardiac MR scanning and can be used for any type of MR imaging sequence. The second, known as retrospective gating, has been developed for synchronizing fast repetition cine acquisitions to the cardiac cycle. Both techniques have their own advantages and disadvantages as will be discussed below.

2 REQUIREMENT OF CARDIAC GATING The importance of synchronizing the MR acquisition to the cardiac cycle is illustrated in Figure 1 which shows an example of spin echo images of the heart acquired with (a) and without (b) cardiac triggering. In this case, lack of gating generates artifacts resulting from view-to-view variations in the position of the heart and blood ¯owing through it. Cardiac synchronization is not only important for cardiac imaging; other organs in the body pulsate due to pressure changes in the arterial vascular systems and image quality has been shown to improve with cardiac gating. Single shot techniques such as echo planar imaging, where the acquisition time is relatively short in comparison to the cardiac cycle, require cardiac triggering when the relationship between the image and the time in the cardiac cycle is of interest, as would be the case for echo planar imaging of the heart.1 Also for imaging of the spinal cord where CSF pulsates with the cardiac cycle2 and techniques such as diffusion imaging3 are particularly sensitive to motion, some form of cardiac triggering or gating is required.

3

1

METHODS OF CARDIAC MONITORING

There have been a number of approaches to monitoring the cardiac cycle including various forms of pulse sensor, the electrocardiogram (ECG), and magnetic resonance methods of rapidly imaging the position of the heart. The pulse sensors include those that monitor the changes in infrared re¯ection or absorption in the vascular bed and are normally placed on the ®nger or the ear lobe and continuous wave Doppler probes directed at any `near surface' artery. Both of these methods monitor the pulsatility of blood ¯ow and can be set up to detect the slope or the amplitude of the waveform. In general, pulse sensors are not used unless the electrocardiographic waveform is unusable. The main problem with the methods are that timing errors can occur, both because the waveforms are less well de®ned than the ECG, in that there are no high-frequency portions to compare with the QRS complex, and because physiological factors can affect the shape of the waveform. An additional problem is that there can be a signi®cant delay between the beginning of cardiac contraction and the associated ¯ow pulse that is being detected. One advantage, however, is that the signal is affected less by the static magnetic ®eld, and the rf and gradient pulses that can interfere with the ECG waveform; the latter two are problematic, especially when sequences are repeated very rapidly. The safety implications are similar for ECG and pulse sensor monitoring although the hazards are somewhat greater for the ECG system because of the requirement of electrical contacts made directly to the skin. It is possible for the transmitted rf signal to be coupled into the leads of the system to such an extent that their temperature is raised enough to cause local burning. To avoid this risk, large loops are avoided and the leads are kept away from the patient's skin as much as possible. Transmission of the electrocardiogram out of the magnet also requires care as cables running out of the magnet can act as an aerial and transmit radiofrequency noise into the scanning region. To overcome this, the electrocardiogram can either be converted to an optical signal and carried out along a ®beroptic cable4 or transmitted by radiotelemetry at a frequency that does not interfere with the MR signal.5 In 1990, Spraggins described a method of cardiac gating that required no external monitoring device. The method involved alternating between the acquisition of image and timing data, the latter being nonphase-encoded and designed to give a 1D image to monitor the changing size of the heart, that could later be used to reorder the image data retrospectively.6

4

ARTIFACTS ON THE ELECTROCARDIOGRAM

Recording an electrocardiogram in a magnetic resonance scanner during a scan is not a simple matter and the waveform bears little relationship to the one that would be obtained away from the magnetic ®eld. The static ®eld has a number of effects, and the changing gradient and rf ®elds applied during the imaging introduce additional artifacts. The ¯ow of blood (a conductor) through the heart and great vessels in the presence of the static magnetic ®eld results in a varying induced potential that is superimposed on the electrocardiogram. The main effect is during the systolic part of the heart cycle when blood

2 CARDIAC GATING PRACTICE artifact caused by the presence of the static magnetic ®eld is that of induced potentials that result if the ECG leads are moved during the scan. Some care is required to choose electrode sites that minimize the potential of such a problem. The switched magnetic ®eld gradients that are used during all imaging sequences also introduce rapidly changing potentials that can severely interfere with the electrocardiogram. The size of these depends on the rate of change of magnetic ¯ux through the conductor loop presented to the ®eld and upon the area of the loop. The size of the artifacts can be reduced, therefore, by twisting the cables up to the patient and placing the electrodes close together. The latter of these is somewhat self-defeating in that placing the electrodes closer together also reduces the amplitude of the detected QRS complex. Most ECG ampli®ers that are used in MR systems are modi®ed so that they detect the rapidly changing potentials and clamp or hold the waveform for their duration. For the majority of magnetic resonance sequences, this allows the ECG waveform to be monitored as the gradient switching occupies a relatively small proportion of time. However, for very fast sequences such as echo planar imaging, the gradients are switched so rapidly that the waveform is obscured throughout the acquisition. The radiofrequency pulse may also cause problems if it couples into the ECG leads or ampli®er circuitry. Some ampli®er systems are controlled by a microprocessor and such systems have to be very well shielded from the rf to avoid signal disruption.

5

Figure 1 Illustration of the importance of cardiac gating when imaging the heart. The ®gure compares transverse spin echo images of the heart (a) ungated and (b) gated to the end-systolic part of the cardiac cycle

is ¯owing most rapidly out of the heart and in the aorta, and the size of the T wave is most noticeably affected. The effect is governed by the Faraday±Neumann law which states that if a conductor cuts a magnetic ¯ux, , the induced potential difference, V, is proportional to the rate of change of ¯ux, d/dt, so that the effect is greatest when there is a large component of ¯ow perpendicular to the ®eld (crossing lines of ¯ux); the direction of the induced potential is perpendicular to both of these. The size of the potential is also directly related to the strength of the magnetic ®eld so that the artifacts are much more pronounced at high ®eld strengths. As far as the detection of the QRS complex of the electrocardiogram is concerned, generally the detection systems monitor the rate of change of the signal and not the overall amplitude. Even though the increased T wave is accompanied by an increased rate of change of the signal, the rate of change found during the QRS complex is normally signi®cantly greater. Another

ELECTROCARDIOGRAM ELECTRODE AND LEAD POSITIONING

There a number of factors that are relevant when choosing the positions to place the ECG electrodes. These factors include the requirement to get a good reliably detectable QRS signal, to minimize artifacts due to breathing, to minimize the potential for artifacts on the image due to the metallic content of the ECG electrodes, and to minimize any potential coupling between the rf signals and the cables of the ECG system. Dimick and colleagues have shown that the vector of the blood ¯ow induced potential normally points anteriorly and slightly down the body.7 The QRS complex vector, on the other hand, usually points to the left, posteriorly and down so that a left-sided or posterior lead is recommended to maximize the ratio between this signal and the artifactual one. One approach that is used to achieve this is to place the two active electrodes one on the right shoulder and the other on the lower left side of the chest, with the common electrode often positioned on the left shoulder (Figure 2). An alternative is to place the electrodes in similar positions on the back; although the detected signal is not quite as strong, the potential for motion artifacts due to respiration are reduced. The material of which the electrodes and cables are made is also relevant. These are not normally ferromagnetic, but their metallic content can induce artifacts by altering the magnetic ®eld locally. In spin echo imaging the artifacts are very super®cial and can normally be ignored, but ®eld echo images are much more affected (Figure 3). Positioning of an electrode or lead over the region of interest should therefore be avoided unless specialized electrodes and leads containing minimal or no metal (i.e. carbon compounds) are used.

CARDIAC GATING PRACTICE

RA

3

LA

LL

Figure 2 Typical positions of the electrodes for recording the electrocardiogram during magnetic resonance imaging. Similar positions on the back can be used, at the expense of a weaker signal, to reduce motion artifacts due to respiration. (Reproduced by permission of Blackwell Scienti®c Publications, from `Magnetic Resonance of the Cardiovascular System', ed. S. R. Underwood and D. N. Firmin)

6 PROSPECTIVE AND RETROSPECTIVE GATING Both prospective and retrospective forms of gating involve monitoring the ECG or other cardiac-related signal with the regular detection of a prominent part, such as the QRS complex of the ECG or the arterial ¯ow pulse of a pulsimeter waveform. With the prospective system, the detection is used to initiate one or a number of sequence repetitions following a prede®ned delay and separated by a repetition time TR [Figure 4(a)].4,8 The sequence repetitions can be of the same phase encoding acquisition for a number of different slices, the same phase encoding acquisition for a number of frames from a cine scan, or a number of different phase encoding acquisitions contributing towards one image. If the sequence can be repeated very quickly, then a complete image can be acquired within a fraction of a cardiac cycle; in this case, a delay is used following the cardiac detection to the start of the sequence repetitions to de®ne the period of the cardiac cycle over which the image is acquired. For extremely fast techniques such as echo planar, several images can be acquired within a cardiac cycle and, rather as for the phase encoding steps of the slower methods, either multiple slices or multiple cine frames can be acquired at de®ned times following the cardiac detection. One problem with this prospective method of triggering is that of cardiac arrythmia, which invalidates the assumption that the cardiac cycles used to construct the image are identical. Artifacts can result for two different reasons: ®rstly the position and shape of the heart may vary from beat to beat, and secondly the time between excitation can be different from one beat to the next, a fact that can cause an amplitude modulation of the acquired signal. In practice, however, normal sinus arrhythmia and infrequent ectopic beats of the heart do not seriously inter-

Figure 3 Artifacts produced by metallic electrocardiogram electrode positions on the chest near the apex of the heart (large arrow). On the spin echo image (a) the artifacts are very much smaller than on the gradient echo image (b). (Reproduced by permission of Blackwell Scienti®c Publications, from `Magnetic Resonance of the Cardiovascular System', ed. S. R. Underwood and D. N. Firmin)

fere with image quality. For cine ®eld gradient echo imaging, however, the enforced requirement to leave a period of delay between the last frame and the next trigger results in a relatively high signal (sometimes known as the lightning artifact) on the ®rst frame of the cine because of the increased longitudinal relaxation that occurs during the delay. This also means that part of the heart cycle cannot be imaged, this being the end-diastolic part when using the R-wave of the ECG as a trigger. The retrospective method of cardiac gating was developed to enable cine ®eld echo images to be acquired throughout the entire cardiac cycle, to remove the lightning artifact and to reduce the other artifacts that result because of cardiac arrhythmia.9±11 The method involves the continuous repetition of the sequence at a constant TR whilst also monitoring and recording a cardiac waveform. There are two ways of implementing the technique: ®rstly by advancing the phase encoding at a regular

4 CARDIAC GATING PRACTICE

ECG

ECG

Acquisition timing

Acquisition timing

Image data timing

Image data timing

Phase encoding

1

2

3

4

Syst. Diast. Syst. Diast. Syst.

1 3 5 7 9 1 3 5 7 91 3 5

1

Acquisition timing

1 3 5 7 91 3 5 7 91 3 5 7 9 1 3 5 7 91 3 5 7 9

Phase encoding

1

2

3

4

(b)

ECG

Acquisition timing

Phase encoding

9 1 3 5 7 91 3 5 7 9

2

3

4

5

Figure 5 Diagram to illustrate how retrospective gating with data interpolation can be used to compensate for the nonproportionate variability in the cardiac cycle. The reconstructed systolic frames have a constant TR whereas for the diastolic frames the TR is varied

ECG

Image data timing

7

Syst. Diast. Syst. Diast.

Phase encoding

5

(a)

Image data timing

Diast.

1 3 5 7 91 3 5 7 91 3 5 7 9 1 3 5 7 91 3 5 7 9

1

2

3

4

5

(c)

Figure 4 Diagram comparing the different methods of cardiac gating for a cine acquisition. Cardiac triggering (a) cannot adequately cover the diastolic part of the heart cycle and has a variable delay between the last frame on one heart cycle and the ®rst frame of the next. For the ®rst retrospective gating method (b) the phase encoding is incremented at regular intevals that are ca. 20% longer than the average cardiac cycle. The majority of heart cycles are therefore oversampled and data for one phase encoding may well be made up from two different heart cycles. For the second retrospective gating method (c) the phase encoding is incremented on the ¯y immediately after the R wave detection

interval that is longer than the longest cardiac cycle [Figure 4(b)] or secondly by changing the phase encoding step immediately after each cardiac detection [Figure 4(c)]. The ®rst method is the easier to implement; the cardiac cycle is monitored for a period of approximately 30 s and the mean cardiac cycle duration calculated. Then, from the TR of the prede®ned cine sequence, the number of time frames that will ®t into the mean cardiac cycle plus 20% is calculated; the phase encoding is then stepped regularly after this number of frames. The phase encoding steps now occur at arbitrary times in the cardiac cycle and data reordering is necessary before image reconstruction. The disadvantage of this approach is that the scan time is

extended by approximately 20% and the data for particular phase encoding steps are likely to be acquired over two different cardiac cycles. The second method of retrospective gating overcomes these problems: the phase encoding gradient is incremented immediately after each cardiac detection and, depending on the cardiac duration, a variable number of sequence repetitions and data acquisitions may be employed for each phase encoding gradient. The technique requires very ¯exible hardware enabling the phase encoding gradient to be incremented on the ¯y, an option not available on the majority of systems. Both retrospective methods require data interpolation to reconstruct a de®ned number of images within the cardiac cycle at time points that may lie between those that were actually acquired. Normally, simple linear data interpolation is used although least square interpolation schemes using variable time windows have been investigated; the wider the time window the higher the S/N, but the greater the temporal averaging and consequential image blurring. The simple interpolation method is generally acceptable if image frames are acquired with a reasonably high temporal resolution. The fact that the cardiac cycle varies in a nonproportionate manner, such that the systolic part varies in length less than the diastolic part, can be taken into account with this method of interpolation before reconstruction (Figure 5). A heart in atrial ®brillation or exhibiting frequent ectopic beats can present a problem for either prospective or retrospective gating, although for cardiac triggering and the second

1000 ms accept

700 ms accept

1300 ms reject

1000 ms reject

1000 ms accept

Figure 6 Arrhythmia rejection. The user de®nes a window of beat lengths that will be accepted. The data acquisition on the beat following one that falls outside this window is rejected and the effect is to reject the acquisition of the ectopic beat and the following sinus beat. (Reproduced by permission of Blackwell Scienti®c Publications, from `Magnetic Resonance of the Cardiovascular System', ed. S. R. Underwood and D. N. Firmin)

CARDIAC GATING PRACTICE

ECG

Acquisition timing Velocity encoding

1212121212121212121212121212121212121212121212

Phase encoding

1

2

3

(a)

ECG

Acquisition timing Velocity encoding

1111111111111111222222222222222211111111111111

Phase encoding

1

1

2

(b)

5

method of phase velocity mapping12±14 has been shown to be particularly useful clinically.15 The method of gating has been shown to have implications on the accuracy of pulsatile blood ¯ow measurement using phase velocity mapping.16,17 The phase velocity mapping technique, when used to measure ¯ow in a particular direction, involves the simultaneous acquisition of two scans with a different phase sensitivity to velocity in that direction. Phase images are reconstructed and these are subtracted to form the phase velocity map. When prospective gating is used, the data are acquired for the different velocity sensitivities on successive cardiac cycles. This enables high temporal resolution ¯ow data to be acquired throughout the majority of the cardiac cycle. With retrospective gating, in order to minimize the scan time, a method of alternately acquiring the different velocity sensitivity data on successive sequence repetitions has been used [Figure 7(a)]. This has the effect of doubling the TR of the sequences and has been shown to result in a high-frequency ®ltering of the pulsatile blood ¯ow waveform. Methods of reducing or removing this error include using a shorter TR, minimizing the interpolation window, and running two separate scans with the order of velocity encoding reversed before resorting the data appropriately.16 Alternatively, the method of acquisition could be changed so that the velocity sensitivity of the sequence is changed less frequently, on successive cardiac cycles or on successive periods 20% longer than the mean cardiac cycle, depending on the method of retrospective gating that is being used [Figure 7(b) and (c)].

ECG

8 Acquisition timing Velocity encoding Phase encoding

1111111111111222222222211111111111111111222222

1

1

2

2

RELATED ARTICLES

Cardiovascular NMR to Study Function; Echo-Planar Imaging; Health and Safety Aspects of Human MR Studies; Heart: Clinical Applications of MRI; Motion Artifacts: Mechanism and Control; NMR Spectroscopy of the Human Heart; Pulsatility Artifacts Due to Blood Flow and Tissue Motion and Their Control; Whole Body Magnetic Resonance Artifacts.

(c)

Figure 7 Illustration of the methods of acquiring velocity encoded data with retrospective gating. The velocity encoding is normally alternated on successive acquisitions (a); however, this can result in some high-frequency ®ltering of the pulsatile ¯ow waveform. Alternating the velocity encoding after a ®xed number of phase encoding steps removes the problem but is inef®cient (b), and ef®ciency can be improved by alternating velocity encoding on the ¯y on successive cardiac cycles (c)

method of retrospective gating it is possible to suspend data acquisition and incrementing the phase encoding for a number of beats after a cycle of abnormal length (Figure 6).

7 IMPLICATIONS OF THE GATING TYPE ON THE ACCURACY OF BLOOD FLOW MAPPING The potential of magnetic resonance to measure blood ¯ow is of particular importance to studies of the cardiovascular system. Although many techniques have been described, the

9

REFERENCES

1. R. J. Ordidge, A. Howseman, R. Coxon, R. Turner, B. Chapman, P. Glover, M. Stehling, and P. Mans®eld, Magn. Reson. Med., 1989, 10, 227. 2. G. Bergstrand, M. Bergstrom, B. Nordell, F. Stahlberg, A. A. Ericsson, A. Hemmingsson, G. Sperber, K. A. Thuomas, and B. Jung, J. Comput. Assist. Tomogr., 1985, 9, 1003. 3. J. V. Hajnal, M. Doran, A. S. Hall, A. G. Collins, A. Oatridge, J. M. Pennock, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1991, 15, 1. 4. P. Lanzer, E. H. Botvinick, N. B. Schiller, L. E. Crooks, M. Arakawa, L. Kaufman, P. L. Davis, R. Herfkens, M. J. Lipton, and C. B. Higgins, Radiology, 1984, 150, 121. 5. R. E. Wendt, R. Rokey, G. W. Vick, and D. L. Johnston, Magn. Reson. Imag., 1988, 6, 89. 6. T. A. Spraggins, Magn. Reson. Imag., 1990, 8, 675. 7. R. N. Dimick, L. W. Hedlund, R. J. Herfkens, E. K. Fram, and J. Utz, Invest. Radiol., 1987, 22, 17. 8. P. van Dijk, Diagn. Imag. Clin. Med., 1984, 53, 29. 9. G. H. Glover and N. J. Pelc, in `Magnetic Resonance Annual', ed. H. Y. Kressel, Raven Press, New York, 1988, p. 299.

6 CARDIAC GATING PRACTICE 10. G. W. Lenz, E. M. Haacke, and R. D. White, Magn. Reson. Imag., 1989, 7, 445. 11. D. E. Bohning, B. Carter, S. S. Liu, and G. M. Pohost, Magn. Reson. Med., 1990, 16, 303. 12. D. J. Bryant, J. A. Payne, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1984, 8, 588. 13. P. van Dijk, J. Comput. Assist. Tomogr., 1984, 8, 429. 14. G. L. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 15. S. R. Underwood, D. N. Firmin, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. Heart J., 1987, 57, 404. 16. M. H. Buonocore and H. Bogren, Magn. Reson. Med., 1992, 26, 141.

17. L. Sùndergaard, F. StaÊhlberg, C. Thomsen, T. A. Spraggins, E. Gymoese, L. Malmgren, E. MuÈller, and O. Henriksen, Magn. Reson. Imag., 1993, 11, 533.

Biographical Sketch David N. Firmin. b 1955. B.Sc., 1978, M.Phil., 1982, Ph.D., 1989, Magnetic Resonance Imaging of Blood Flow, University of London. Introduced to NMR by D. B. Longmore, National Heart and Chest Hospital, London. Senior Lecturer, National Heart and Lung Institute, University of London, 1982±present. Approx. 210 publications. Research specialties: rapid ¯ow, and cardiovascular imaging by use of magnetic resonance.

CARDIOVASCULAR NMR TO STUDY FUNCTION

Cardiovascular NMR to Study Function

1

means that virtually every cardiac diagnosis that previously required invasion can now be made noninvasively and costeffectively, without mortality or morbidity. Until recently the exception was coronary artery disease. The NMR alternative to the coronary angiogram is dealt with elsewhere.

Donald B. Longmore Royal Brompton Hospital, London, UK

1 INTRODUCTION Approximately half of those reading this volume will die of cardiovascular disease, a further quarter will die of cancer, and 12% of lung disease. The diagnosis of these diseases are all now within the capabilities of NMR. However, vast intellectual and ®nancial resources have been, and are being, applied to magnetic resonance technology, ignoring these major causes of death. Only a handful of the world's 9800 NMR machines are available to cardiologists, oncologists, and lung specialists. Presently NMR is used to study less than 3.5% of all disease. Open access rapid cardiovascular and lung imagers with facilities for intervention including laser and ultrasound surgery are technically feasible and needed urgently. Furthermore, the diagnostic power and, from the patient's point of view, the simplicity of a functional cardiovascular NMR examination, means that, for the ®rst time, the ultimate medical objective of preventing the commonest diseases could be realized. The potential of functional evaluation of the cardiovascular system is for population screening to enable an understanding of the natural history of atheromatous disease, the commonest cause of death in the western world. Detection of presymptomatic cardiovascular disease enables both secondary prevention monitoring the ef®cacy of therapeutic and preventive measures using repeated functional NMR studies. Demographic evidence reveals the changing incidence of the disease in the same population over time and the varying incidence from place to place in the same ethnic groups (see section 8). This, coupled with experimental evidence, shows atheroma to have causative factorsÐand therefore to be preventable.1±4 While the heart and circulation are being studied, it is only a small step to include screening for breast cancer5 (Breast MRI) and, with the new ultra-short (50 s to 500 s) TE sequences, to study the lung (Lung and Mediastinum MRI). Contemporary cardiovascular diagnosis relies on an expensive and time-consuming combination of clinical examination, chest radiography, electrocardiography (ECG), stress ECG, echo cardiography, nuclear medicine studies with and without pharmacological stress and invasive cardiac catherization, which carries a small but important mortality and morbidity. Unsuccessful attempts have been made to evaluate the cardiovascular system avoiding cardiac catherization using external recording devices such as apex cardiography, ballisto cardiography, ECG, impedance cardiography, etc. Now the balance has been profoundly altered following advances in nuclear medicine techniques, X-ray computed tomography, and echo cardiography. NMR is the most powerful diagnostic instrument yet conceived and this, combined with echo cardiography,

2

TECHNICAL REQUIREMENTS FOR CARDIOVASCULAR NMR

In order to study cardiovascular function, magnetic resonance techniques additional to those for imaging motionless parts of the body are required. Essential enhancements are cardiac gating, velocity mapping (Cardiac Gating Practice), rapid imaging, and velocity encoded rapid imaging (Whole Body MRI: Strategies for Improving Imaging Ef®ciency) in all planes.6±8 2.1

Imaging Strategy

A comprehensive heart NMR study is achieved in minimal time using a simple strategy. A transverse (transaxial) image of the chest is used to measure the angle of the heart to the left. A scan in the plane through the long axis of the left ventricle produces a vertical long axis (VLA) image through the left atrium and the left ventricle. From this, the downward angle of the heart is measured and a double oblique image from below produced to provide a horizontal long axis (HLA) view [Figure 1(a)] showing the ®ve functional chambers of the heart and the coronary arteries in cross-section. Image planes can be selected perpendicular to the HLA to produce the short axis (SA) image [Figure 1(b)].

Figure 1 (a) The spin echo HLA image shows the right atrium (RA), the right ventricle (RV), the left atrium (LA), the left ventricle (LV) and the left ventricular out¯ow tract (O). The tricuspid valve which is not indicated is at the same level as the mitral valve marked (MV). The right coronary artery (RCA) and the left coronary artery (LCA) are also seen in cross-section. Fat at the apex of the heart is prominent. (b) The spin echo SA image shows the `doughnut'-shaped left ventricular muscle (LV) with the right ventricle (RV) anterior to it. The left anterior descending coronary artery (LAD) is seen in the upper interventricular groove and the coronary sinus (CS) in the lower interventricular groove

2 CARDIOVASCULAR NMR TO STUDY FUNCTION area of diastolic VLA  area of diastolic HLA  0:85 minimum diastolic length apex to valve plane

2.2 Rapid Imaging applied to the Cardiovascular System Two rapid imaging techniques are currently available9,10 (dealt with in detail in Blood Flow: Quantitative Measurement by MRI). Orders of magnitude of reduction of acquisition times for NMR imaging were ®rst achieved by Mans®eld and Pykett8 with the one-shot echo planar `Snapshot' (EPI) technique. The advantage of EPI for the cardiovascular system is that data acquisition can follow preparation pulses required to study cardiovascular function for diffusion, chemical shift imaging, and velocity encoding. The alternative to EPI with velocity mapping is the fast low angle shot (FLASH) technique with velocity mapping. EPI has a better signal-to-noise ratio (SNR) and can freeze movement, which is vital in the neonate. Less than 50 ms acquisition time is needed to obtain a velocity encoded image. The FLASH technique uses a total acquisition time of over 300 ms. There is no stage in the cardiac cycle of an infant when the heart is relatively still for this long. Furthermore, at all stages throughout life the ®lling and contraction of the heart varies from beat to beat. This variability is never more marked than in the neonate with congenital disease. EPI is more suitable for the measurement of aperiodic movements (irregularly irregular) than FLASH, which relies on periodicity and the heart being in the same position and at the same stage of contraction for 16 or more heart beats. However, despite these disadvantages, because the gross appearance of a FLASH image is mainly determined by the data from a relatively few central phase encoding steps acquired in the ®rst 50 ms or so of a total acquisition of over 300 ms, acceptable FLASH velocity encoded imaging is possible; FLASH is also less demanding on the gradients, the rf, data collection, etc., and it can be implemented on standard machines with relatively simple modi®cation. Either technique can be used to ®nd a jet, measure its velocity and, from the velocity measurement, calculate the gradient and therefore the severity of the lesion using a modi®ed Bernoulli equation.11 Velocity mapping combined with good anatomical imaging usually reveals more useful functional information about congenital and acquired disease than can be obtained from a cardiac catheter angiogram.

and the end systolic volume from the formula: area of systolic VLA  area of systolic HLA  0:85 minimum length apex to valve plane

…2†

This method is only valid if the ventricles are contracting evenly. 2.3.1

Clinical Signi®cance of Volume Measurements

Subtraction of the systolic volume from the diastolic volume produces an accurate measure of stroke volume, which when multiplied by the heart rate gives the cardiac output. Continued life depends on the way the heart is functioning rather than on the output at a moment in time. An ef®cient heart retains a small residual volume of blood which when compared with the stroke volume gives the ejection fraction. A low ejection fraction carries a poor prognosis. 2.3.2

The Four Way Comparison

Volume measurements have been validated by comparing the stroke volumes of the right and left ventricles which in normal subjects is unity. The stroke volumes have also been compared with the aortic and pulmonary artery ¯ows. Three measurements (right ventricular output, pulmonary artery ¯ow, and left ventricular output) correspond to within 2%. Aortic ¯ow is 5±10% below the other measurements due to the `take off' of coronary ¯ow. Functional assessment of the cardiovascular system can be applied to all types of disease. In this article the classical `surgical sieve' is used, dividing disease into congenital and acquired, and further dividing acquired disease into traumatic, in¯ammatory, neoplastic, etc. There is a `gray area' where it is dif®cult to distinguish between diseases we are born with and those we acquire, possibly because we are genetically vulnerable.

3 2.3 Volume Measurements The volumes of all the cardiac chambers, notably the ventricles, can be measured accurately from two multislice sets of transverse, or HLA, or SA images, one set taken at end systole (when the volumes are least) and the other at end diastole (when the heart is fullest). Usually 10±12 slices are suf®cient to cover the whole of the heart. The areas of the cardiac cavities can be measured manually or by automatic edge detection or area growth techniques. Ventricular volume measurements are obtained by summing the areas of the images covering the ventricles in systole or diastole. The thickness of the slices is known and the volume of each slice calculated. Summation of the various slice volumes gives the total cavity volume.12 A `shorthand' method of measuring stroke volume uses area length calculations of VLA and HLA systolic and diastolic images assuming the ventricles to approximate to a cone. The end diastolic volume is calculated from the formula:

…1†

FUNCTIONAL ASSESSMENT OF CONGENITAL HEART DISEASE

Approximately three in every 1000 live births manifest one of the commoner congenital heart diseases, pulmonary artery stenosis, atrial septal defect, patent ductus arteriosis, and tetralogy of Fallot, though there is a signi®cant incidence of many other complex forms of congenital disease. NMR functional imaging can detect and quantify the threat to life and form a vital part of the management in all these conditions. Three discrete groups of patients require functional assessment. These are: neonates and infants; children, when the less catastrophic abnormalities begin to manifest themselves; and `grown up' congenital heart disease usually to follow up the results of surgical intervention. 3.1

Neonates and Infants

The small size of an infant and its inability to lie still, combined with the complexity of many congenital abnormalities,

CARDIOVASCULAR NMR TO STUDY FUNCTION

necessitates a rapid imaging technique and the use of special surface coils or a head coil. 3.2 Functional Imaging in Children Functional techniques have given an insight into the compensatory mechanisms which can allow an infant with gross cardiac abnormalities to survive to childhood when operative intervention is easier and safer. The ideal diagnostic techniques for children are a combination of rapid imaging used in neonates and infants and the standard adult functional imaging methods. 3.3 `Grown Up' Congenital Heart Disease (GUCH) NMR is the only available technique for the repeated longterm follow-up of adolescents and young adults who have undergone successful surgery. It is vital in postoperative cases to include velocity mapping searching for deterioration, because calci®ed obstructions are not visualized on anatomical images and can only be demonstrated by blood velocity mapping.13 NMR is uniquely able to detect the afferent and efferent vessels to congenital arteriovenous (AV) malformations, helping the surgeon to plan an operation. In cases with multiple AV malformations (Klippel, Tralawney syndrome) the measurement of the total blood ¯ow through all the afferent or efferent vessels to the AV malformations can predict the impending onset of high-output heart failure.14 3.4 The Late Effects of Congenital Heart Disease 3.4.1

Valve Disease

Valve disease may be acquired or present as a late manifestation of congenital heart disease. A small proportion of the population are born with a bicuspid aortic valve which will usually function normally until the patient reaches the fourth decade of life, by which time calci®cation causing stenosis and regurgitation results from the abnormal vibrations and stresses on the two cusps. NMR can be used to measure the jet velocity and to calculate the gradient and to measure the regurgitant fraction. This regurgitation can be quanti®ed directly using velocity mapping or by comparing the ratio of the stroke volumes of the right and left ventricles. 3.4.2

Septal Defects

Pulmonary valve stenosis and atrial septal defect are the commonest congenital deformities of the heart, but they may not prove to be signi®cant until adolescence or early adult life when the gradients across the pulmonary valve become signi®cant and can be measured from the jet velocity. The shunt across the septal defect can be measured by the volume method or by velocity mapping. It is important to determine the amount of blood shunting from the left (higher pressure side of the heart) to the right (lower pressure side). It is not uncommon for the shunt ratio to be as high as 4: 1, with four times as much blood ¯owing through the lungs as the body. This shunt gradually and irreversibly damages the lungs, increasing their resistance to ¯ow and eventually reversing the shunt and making closure of the defect irrelevant and dangerous. If only one

3

defect is present in the absence of valve regurgitation, the shunt can be measured by comparison of the outputs of the two sides of the heart. Velocity mapping measuring the shunt directly is more reliable. 4

FUNCTIONAL ASSESSMENT OF ACQUIRED HEART DISEASE

4.1

Traumatic Cardiovascular Disease

The commonest traumatic cardiovascular lesion due to road traf®c and other accidents is a partial tear of the aorta causing a dissecting aneurysm. These can be detected with a computed tomography (CT) scan or an anatomical MR image which reveal the presence of the dissection. A dissection usually travels down the length of the aorta and sometimes it extends back towards the origin of the coronary arteries. Before any repair can be attempted it is essential to use velocity mapping to determine where the blood is ¯owing, and which branches to vital organs arise from the true lumen and which from the false lumen.15 Tears and penetrating injuries such as stab wounds to the heart and the associated blood in the pericardium can be seen on spin echo and better on gradient echo images. Blood ¯owing in and out of false aneurysms indicates their size, and more signi®cantly, differentiates them from pericardial cysts. 4.2 4.2.1

In¯ammatory and Toxic Damage to the Heart Valve Disease

The mitral and aortic valves may be damaged by bacterial infection on an already damaged or abnormal valve, or by rheumatic fever which for the past decades has been uncommon in the western world. Hypersensitivity to Streptococcus spp. still commonly causes valve disease in developing countries and there is now a resurgence, notably in the USA. Thickened valve cusps can be seen on `black blood' images or as ®lling defects on white blood images. However, life depends on the normal function of valves, and appearance may not correlate with the severity of their malfunction. Turbulent and regurgitant ¯ow associated with stenosis and regurgitation due to rheumatic fever, etc. can be visualized as areas of signal loss on gradient echo images. If velocity mapping is not available, a remarkably accurate assessment of the volume of a regurgitant jet can be made from the proportion of the chamber occupied by the jet and the duration of the back ¯ow.16 Stenosis and regurgitation are best measured directly using velocity mapping with short echo times. A TE of 3 ms to the second echo can capture velocities of up to 6 m sÿ1: the highest velocity attained through a stenotic valve. Multivalve disease invalidates comparison of stroke volumes from the two sides of the heart. The same multislice technique can be used to measure ventricular mass, which is increased in congenital cardiomyopathies. 4.2.2

Heart Muscle Disease

Heart muscle can be affected directly by viral illnesses such as in¯uenza, the ME virus and glandular fever. Conditions such as post partum cardiomyopathy and in¯ammatory/allergic

4 CARDIOVASCULAR NMR TO STUDY FUNCTION in®ltrative processes such as rheumatic fever also impair muscle function. Alcohol, drugs, and industrial toxic substances can also cause generalized dysfunction of the myocardium. In ¯orid cases, cine NMR imaging is suf®cient to demonstrate the dilated poorly contracting heart with a low ejection fraction. In less obvious cases, measurement of global ventricular function, as described below, is required. 4.3 Neoplastic Disease Primary malignant tumors of the heart muscle are uncommon. They can be seen invading the muscle on multislice NMR images. Their effects on cardiac function can be determined by measuring global and regional function (see sections 4.4.2 and 4.4.3). Benign tumors and developmental cysts are commoner and do not usually have a signi®cant effect on heart function, but repeated studies are necessary to check them for malignant change. Tumors in the atria can ¯oat through the atrioventricular valves during artrial systole and be blown back into the atria during ventricular systole, causing cardiac dysfunction by obstructing the valves and by displacing blood in a pumping chamber. Anatomical images do not distinguish well between intracardiac blood clot and intracavity tumor on spin echo images and gradient echo, and velocity mapping techniques are needed to reveal the presence of clot. Tumors in the right atrium may appear to have an obvious connection to the atrial wall but are frequently extensions of tumors in other parts of the vascular system which have grown along the great veins into the atrium. An `NMR search' can locate the origin of the tumor. The hemodynamic signi®cance of intracavity space-occupying lesions can be determined by velocity mapping. More frequently the heart is invaded by a local tumor such as a lung cancer with hilar glandular involvement. In these cases cardiac function can be impaired in three ways: by stiffening of the wall, by reduction of chamber size due to the mass of tumor tissue, or directly by tumor replacing contractile elements of the heart. 4.4 Degenerative Heart Disease Most cardiovascular deaths, and hence about 40% of all deaths, are due to various forms of degenerative heart disease of which the commonest are hypertension and atherosclerotic disease. 4.4.1

Hypertension and Ventricular Mass

The reduction of ventricular mass following effective treatment of hypertension can be monitored using the multislice method (section 2.3). The muscle hypertrophy caused by high blood pressure can be quanti®ed using the same multislice technique as for ventricular volumes.17 Studies of the caval or pulmonary venous ¯ow which normally exhibit two similar ¯ow peaks, one in systole and one in diastole, can be used to distinguish between ventricular function which is abnormal due to valvar regurgitation which causes an enhanced second peak `constriction' of the heart by pericardial effusion or scar tissue around it. `Restriction' due to stiffness of the heart itself, usually as a result of an inadequate blood supply due to coronary artery disease, correlates with an enlarged ®rst peak.

Ventricular function can be studied globally or regionally. 4.4.2

Global Ventricular Function

Ventricular function measurements are clinically most relevant, but the heart has four chambers and the techniques described here are equally relevant to the atria. Global ventricular function measured in the resting state is frequently within normal limits, but the patient's disability indicates that there is inadequate cardiac output during exercise associated with normal activities. Pharmacological stress (section 4.4.4) can reveal occult abnormalties of global function in two ways: by changes in velocity and acceleration of blood in the aorta, or by changes of stroke volume and ejection fraction.18 In the normal heart, increasing doses of a pharmacological stress (section 4.4.4) agent such as dobutamine cause an increase in aortic blood ¯ow, velocity, and acceleration. In the abnormal heart, increasing the dose produces increased performance up to a much lower limit than in the normal heart, which has large reserves of power, and there is a `fall off' in performance at dose levels which in the normal heart would still increase velocity. The dose level at which the `fall off' occurs is an indication of the extent of the myocardial damage. When the resting global function is at the lower end of normal limits, and pharmacological stress (section 4.4.4) indicates myocardial dysfunction with exercise, this may be because regions of the ventricle are contracting suboptimally, not contracting, or even moving paradoxically, bulging during systole thereby further disadvantaging the remaining normal heart muscle. In these cases it is necessary to study regional ventricular function. 4.4.3

Regional Ventricular Function

Multislice VLA, HLA, or, most usefully, SA views of the heart can be used to assess regional thickening and movement of the heart wall during contraction. If these appear normal but the clinical history suggests otherwise, then pharmacological stress is required. The inner surface of the heart muscle is under maximum stress during cardiac contraction and is remote from the blood supply which enters the heart from its outer surface. The earliest sign that the heart is becoming injured is the failure of the subendocardial muscle ®bers to `actively relax' to enable ®lling at the beginning of diastole. Later in the disease process, contraction of these muscle ®bers is also impaired. Velocity mapping of the inner surface of the heart in the direction of the long axis of the heart provides a sensitive measure of the elasticity and capability of the heart to ®ll actively. Later in the disease process, the same velocity mapping technique can be used as an accurate measure of the local failure of contraction.19 4.4.4

Pharmacological Stress

To mimic normal exercise, pharmacological stress using inotropic agents to increase cardiac work are used. Intravenous dobutamine, which has a half-life of less than 100 s, is an example of a safer, effective agent than the long-acting coronary vasodilator dipyramidole (Persantin), which is more

CARDIOVASCULAR NMR TO STUDY FUNCTION

commonly used. Any angina caused by the dobutamine infusion passes off within 90 s of stopping the infusion.20 Pharmacological stress agents highlight areas which fail to contract normally, and indicate which coronary artery territory is receiving an inadequate blood supply. During simulated exercise there is good correlation between areas which do not thicken and move during systole and bolus tracking of a `pseudo blood pool' magnetic resonance contrast agent. Both correlate well with nuclear medicine techniques.21 Measurements such as global and regional ventricular function demonstrate the effect, rather than the cause, of reduced ventricular function, which is most commonly a result of a diminished blood supply caused by atheromatous disease partially blocking the coronary arteries.

4.5 Atheromatous Disease Atheromatous disease is progressive, and from demographic studies and animal experimentation is known to be reversible in its early presymptomatic stages (see section 8). The reader, hopefully a normal subject, requires answers to three fundamental questions about atheromatous disease. (a) Have I got it? (b) Are the atheromatous plaques of a type which can ulcerate and cause sudden death? (c) Can MR monitor the ef®cacy of preventive and therapeutic measures? An NMR machine capable of chemical shift imaging and velocity mapping is capable of answering all these questions. 4.5.1

Arterial Compliance

(a) Have I got arterial disease?: Atheromatous disease, mainly in the coronary arteries in the male and the carotid arteries in the female, appears not to coexist with a normal compliant arterial system. For this reason, measurements of aortic compliance, or its reciprocal the velocity of the onset of the arterial pulse wave along the aorta, can be used as a screening test to determine whether the subject is normal or an arteriopath.22 Aortic compliance can be determined by NMR directly by measuring the volume of the arch of the aorta at peak systole when it is most dilated, and at the end of diastole when its elastic recoil reduces it to its smallest volume. It can be derived indirectly with similar accuracy by measuring the onset of the arterial pulse wave in the ascending aorta and at the end of the arch of the aorta during peak systolic ¯ow. In a normal subject under the age of 55, the pulse wave takes an average of 32 ms to traverse the arch, whereas if the transit time of the leading edge of the pulse wave takes less than 12 ms diffuse arterial disease is present. This simple screening test divides subjects under the age of 55 into three groups: athletes who have remained ®t; the average normal; and arteriopaths. Over this age the normal age-related aortic dilatation and reduction of arterial compliance is such that all subjects appear to be arteriopaths. 4.5.2

5

con®rmed and artefacts excluded by detecting acceleration of blood upstream of the obstruction, increased velocity across it, and deceleration downstream of it. Water and fat images of the plaque can be acquired by using Dixon's selective reading of fat and water or Hinks' selective excitation of fat and water. A subtraction image indicates the percentage of lipid present.23 Figures 3(a) and (b) show an atheromatous plaque with low lipid content. The range of lipid content in a plaque varies from zero up to a maximum of 28%. Figures 4(a) and (b) show a plaque with a high lipid content. A ®brous plaque [Figures 5(a) and (b)] with no or very low lipid content, though it impairs the circulation, is unlikely to cause sudden death. A plaque with a high lipid content puts the subject at risk because the plaque may rupture and ulcerate, precipitating acute thrombosis on its surface and, frequently, sudden death. At the present stage of development, chemical shift imaging is only applicable to larger vessels, and the resolution is inadequate for NMR studies of coronary arteries based on single breathhold imaging acquiring 16 turbo-FLASH images. Furthermore, fat suppression techniques are needed in order to produce accurate images of the coronary arteries and velocity measurements in them, because phase changes due to the presence of fat surrounding the vessels invalidate the Dixon technique. Coronary velocity mapping is however adequate for the determination of the hemodynamic signi®cance of coronary plaques by measuring the increased velocity across the plaque taking 24 heart beats (a breathhold duration within the capabilities of most subjects). The way ahead for coronary imaging depends on the development of velocity encoded echo planar and spiral imaging, both techniques with potentially suf®cient resolution to enable analysis of coronary plaques. At present the inference has to be made that the lipid content of plaques in the coronary arteries is similar to that in plaques elsewhere in the circulation, as is usually the case.

Chemical Shift Imaging

(b) Are the atheromatous plaques of a type which can ulcerate and cause sudden death?: If the patient is shown to be an arteriopath by the screening tests outlined above, arterial plaques in the larger vessels can be found on spin echo images of the vessel walls (Figure 2). The presence of the plaque can be

Figure 2 Atheromatous plaques (P) in the aorta of a 59-year-old male seen on a black blood spin echo image

6 CARDIOVASCULAR NMR TO STUDY FUNCTION

Figure 3 (a) A spin echo image of an atheromatous plaque (P) in a cadaveric specimen of aorta. (b) The image produced by subtraction of the fat and water images. There is no signal in the plaque which contains no lipid, con®rmed by histology

4.5.3

Monitoring Atheroma using Functional NMR

(c) Can NMR monitor the ef®cacy of preventive and therapeutic measures?: Experiments with arti®cially induced atheroma in animals show that plaques regress with starvation (see section 8). Simple preventive measures including strict diet and exercise have also been shown to cause regression of atheromatous plaques. In subjects with very high cholesterol

Figure 4 (a) Spin echo cross-sectional images of a cadaveric specimen of human aorta with an atheromatous plaque (P) and a test tube with fat ¯oating on water (W + F). (b) The images produced by subtraction of fat and water images. The plaque (P) and the periaortic fat have high signals corresponding with the high lipid content in the histology specimen. There is a line of high signal on the interface between the fat and water in the test tube (arrowed)

and other lipid levels, lipid-lowering agents may be necessary to augment the simple preventive regime. The advantages of NMR are that functional studies can be repeated regularly to monitor the progress of preventive measures.

CARDIOVASCULAR NMR TO STUDY FUNCTION

7

be done using direct NMR vision. Ultrasound surgery with NMR vision has the advantage that the ultrasound energy can be applied at a low level using the changes to the NMR signal with increased temperature to con®rm that the ultrasound transducers are focused on the area of interest. The ultrasound energy can then be increased to ablate the lesion. The massive intellectual pool of knowledge of NMR will be switched from less common diseases to cardiology, oncology, and to the lung. Proper training in cardiovascular NMR is required, not so much for imagers, as for cardiologists and experts in vascular disease. 6

CONCLUSIONS

Functional NMR evaluation of the cardiovascular system using volume measurements, velocity encoded rapid imaging, and chemical shift imaging is capable of making the diagnosis in congenital and acquired cardiovascular disease which, between them, cause the largest number of deaths of any disease in the western world and massive morbidity and suffering. Furthermore, for the ®rst time in the history of medicine, there is the opportunity to apply preventive measures to eradicate the epidemic of preventable arterial disease. 7

RELATED ARTICLES

Blood Flow: Quantitative Measurement by MRI; Breast MRI; Echo-Planar Imaging; Heart: Clinical Applications of MRI; Lung and Mediastinum MRI; Whole Body Magnetic Resonance: Fast Low-Angle Acquisition Methods. 8

Figure 5 (a) A spin echo image of the abdominal aorta in a 61-yearold male showing a typical asymmetric plaque (P). (b) The image produced by subtraction of fat and water images. The plaque (P) has a lower signal than the marrow (M) in the vertebral body which contains fat in a similar proportion to a lipid-rich plaque; therefore this is a ®brous plaque which will not rupture and cause sudden death

5 THE FUTURE More versatile magnetic resonance machines capable of echo planar imaging and spiral velocity mapping will become available. These machines will use a new generation of openaccess magnets in which ultrasound and laser intervention will

BACKGROUND NOTE

Atheromatous disease was unknown in Central Europe after World War I. Autopsy studies at the London Hospital from 1908±40 showed a steady increase in coronary artery disease. Casualties in the starving population of the siege of Leningrad in World War II showed no arterial disease; likewise there was no atheromatous disease in the survivors of concentration camps. Young soldiers killed in the Korean and Vietnam Wars showed massive diffuse disease. Experimental atheroma in laboratory animals regresses with starvation. Atheromatous disease does not coexist with wasting diseases. It also varies not only from time to time but from place to place. There are wide variations in its incidence in apparently genetically similar populations in different countries. There is no relationship between the `spend' on health care and the incidence of disease. The USA spends the highest proportion of its GDP on health care and the UK nearly the lowest. Both countries are amongst those with the highest incidence. France, a high spender, and Japan, the lowest, share the lowest incidence of occlusive vascular disease.

9

REFERENCES

1. D. B. Longmore, Diagn. Imag., 1988, 399.

8 CARDIOVASCULAR NMR TO STUDY FUNCTION 2. Physiology and Treatment of Starvation: Experiences in Warstarved Europe, Report of Societies 818, Br. Med. J., June 9, 1945. 3. J. Brozek, S. Wells, and A. Keys, `Medical Aspects of Semistarvation in Leningrad (Seige 1941±42)'. 4. P. Mollison, Br. Med. J., January 1946. 5. B. A. Porter and J. P. Smith, Magn. Reson., 1993. 6. G. L. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 7. D. J. Bryant, J. A. Payne, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1984, 8, 588. 8. P. Mans®eld and I. L. Pykett, J. Magn. Reson., 1978, 29, 355. 9. J. Frahm, A. Haase, and D. Matthaei, Magn. Reson. Med., 1986, 3, 321. 10. D. N. Firmin, P. D. Gatehouse, and D. B. Longmore, Proc. XIth Annu. Mtg. Soc. Magn. Reson. Med., Berlin, 1992, 2915. 11. P. J. Kilner, C. C. Manzara, R. H. Mohiaddin, D. J. Pennell, M. G. St. J. Sutton, D. N. Firmin, S. R. Underwood, and D. B. Longmore, Circulation, 1993, 87, 1239. 12. D. B. Longmore, R. H. Klipsten, S. R. Underwood, D. N. Firmin, G. N. Houns®eld, M. Watanabe, C. Bland, K. Fox, P. A. PooleWilson. R. S. O. Rees, D. N. Denison, A. M. McNeilly, and E. D. Burman, Lancet, 1985, i, 1360. 13. P. J. Kilner, D. N. Firmin, J. Martinez, J. Somerville, S. R. Underwood, R. S. O. Rees, and D. B. Longmore, Eur. Heart J., 1991, 12(Suppl.), 284. 14. M. Huber, D. B. Longmore, D. N. Firmin, J. Assheuer, H. Bewermeyer, and W. D. Heiss, Digit. Bilddagnos., 1989, 9, 1.

15. H. G. Bogren, S. R. Underwood, D. N. Firmin, R. H. Mohiaddin, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. J. Radiol., 1988, 61, 456. 16. R. H. Mohiaddin and D. J. Pennell, in `Percutaneous Balloon Valvuloplasty', ed. T. O. Cheng, Igaku-Shoin Medical Publishers, New York, 1992, pp. 185±213. 17. S. M. Forbat, S. P. Karwatowski, P. D. Gatehouse, D. N. Firmin, S. R. Underwood, and D. B. Longmore, Br. J. Radiol, 1993, 66, 957. 18. D. J. Pennell, S. R. Underwood, and D. B. Longmore, J. Comput. Assist. Tomogr., 1990, 14, 167. 19. S. P. Karwatowski, R. H. Mohiaddin, G. Z. Yang, D. N. Firmin, M. G. St. J. Sutton, S. R. Underwood, and D. B. Longmore, JMRI, 1994, 4, 151. 20. D. J. Pennell, S. R. Underwood, C. C. Manzara, R. H. Swanton, J. M. Walker, P. J. Ell, and D. B. Longmore, Am. J. Cardiol., 1992, 70, 34. 21. D. J. Pennell, S. R. Underwood, P. J. Ell, R. H. Swanton, J. M. Walker, and D. B. Longmore, Br. Heart J., 1990, 64, 362. 22. R. H. Mohiaddin, S. R. Underwood, H. G. Bogren, D. N. Firmin, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. Heart J., 1989, 62, 90. 23. R. H. Mohiaddin, D. N. Firmin, S. R. Underwood, A. K. Abdulla, R. H. Klipstein, R. S. O. Rees, and D. B. Longmore, Br. Heart J., 1989, 62, 81.

Coronary Artery Disease Evaluated by MRI Robert R. Edelman and Warren J. Manning Beth Israel Hospital, Boston, MA, USA

1 Related Articles 2 References

3 3

Magnetic resonance imaging has had a profound clinical impact in several areas, including the brain, spine, and joints. A variety of technical advances have expanded the applications of MRI into areas where motion can be problematic, such as the abdomen. Fast imaging techniques also permit good quality images of the heart to be acquired, either for the demonstration of anatomy or for the measurement of function. Nonetheless, the clinical impact of MRI in the heart has only been modest. Aside from the technical challenges involved in imaging the beating heart, alternative, less expensive, and more readily available imaging modalities, such as ultrasound and radionuclide scintigraphy, can usually provide the desired information. There is one area where alternative noninvasive imaging tests fall short, i.e. in the assessment of myocardial ischemia. Cardiovascular disease remains the leading cause of death in the United States with an estimated 1.2 million myocardial infarctions and 600 000 deaths per year attributed to coronary artery disease.1 Minimally invasive tests for myocardial ischemia include single photon emission tomography (SPECT)2,3 and positron emission tomography (PET).4,5 These tests reveal perfusion abnormalities, but do not depict the coronary artery stenoses that cause them nor do they provide direct measurements of coronary artery blood flow. Transthoracic and transesophageal echocardiography can depict portions of the coronary arteries, but not sufficiently for routine clinical use.6 – 8 While transesophageal echocardiography does not require contrast injection, it does require esophageal intubation and has associated risks of death, esophageal perforation/laceration, aspiration/hypoxia, and arrhythmias.9 The ‘gold standard’ for the evaluation of the coronary arteries is contrast angiography with over 500 000 diagnostic cardiac catheterizations performed annually in the United States. Despite the development of multiple noninvasive tests for the detection of myocardial ischemia, up to 20% of these coronary angiograms reveal insignificant coronary artery disease.10 Information derived from such angiograms, however, is the standard by which mechanical interventions and many medical therapies are planned. In addition, prognostic information is also gained from data regarding coronary artery patency. Hospital charges alone for cardiac catheterizations have been estimated at over $1.8 billion annually. In addition, coronary angiography carries with it a low, though finite major complication rate—death (0.12–0.20%), cerebrovascular accident (0.03–0.20%), myocardial infarction (0.0–0.25%) and a

minor complication rate—vascular complication, local infection (0.57–1.6%), or arrhythmias (0.30–0.63%).11 – 13 The high cost and associated risk make routine coronary angiography inappropriate for use as a screening test. Furthermore, while semiquantitative techniques exist for estimating the flow restriction caused by a coronary artery stenosis based on the conventional angiogram,14 they do not provide a quantitative measure of coronary artery blood flow. Magnetic resonance (MR) imaging is well suited for evaluating the heart, with excellent soft-tissue contrast and the ability to image the heart in double-oblique tomographic sections. Complex cardiac anatomy15 and systolic function16 have already been elucidated using MR imaging techniques. Magnetic resonance angiography is currently being used in several clinical applications, including screening for intracranial aneurysms,17 detection of stenoses of the extracranial carotid arteries,18 vertebro-basilar system, and renal arteries.19 If the techniques of MR angiography could be extended successfully to the heart, then MR might be used to predetermine which patients should go on for coronary catheterization, to follow high-risk patients noninvasively, to quantify flow reserve so as to determine the physiological significance of a stenosis, or ultimately as a cost-effective substitute for catheterization. Human coronary arteries are small in caliber. The left main coronary artery is typically 4–6 mm in diameter20 while the left anterior descending and left circumflex coronary arteries are generally 3–4 mm at their origins and taper distally. The right coronary artery supplying less myocardium is smaller than the left main, and is typically 3–4 mm proximally. High-resolution MR angiography of arteries of similar caliber (e.g. circle of Willis) to epicardial coronary vessels has been successful, but reproducible coronary angiography has remained elusive due to the effects of significant cardiac and respiratory motion. Standard ECG-gated spin echo and gradient echo cine images only occasionally show portions of the coronary arteries, and these images are not adequate for detailed evaluation.21 There are several fast imaging methods that have been used for coronary artery imaging. Three-dimensional gradient echo methods22 permit thin sections and short TE , thereby minimizing partial volume averaging and flow-related dephasing. With 3D approaches, however, saturation effects are more severe than with 2D approaches, resulting in suboptimal flow contrast particularly in the setting of low flow velocities. Unfortunately, long scan time for 3D acquisitions (at least several minutes for ECG-gated studies) precludes breath-holding, so that significant blurring from respiratory motion will occur. Subtraction methods also have potential value for coronary artery imaging. One approach involves the interleaved acquisition of a pair of projection (thick section) images, of which one has a spatially localized inversion pulse applied to the aortic root to tag the coronary inflow.23 Since only the signal intensity of the coronary arteries is altered by the inversion pulse, background and cardiac chamber signals are eliminated by image subtraction. Potential drawbacks of this method are the need for prolonged breath-holding periods (e.g. 24 s or longer), inadvertent saturation of other structures such as blood within the left atrium, dependence of flow contrast on the R–R interval, and potential misregistration resulting in imperfect subtraction. The method also assumes that blood within the major epicardial coronary arteries is completely displaced

2 CORONARY ARTERY DISEASE EVALUATED BY MRI within one R–R interval by blood from the aortic root. The use of a thick section also has the drawback of increased intravoxel dephasing compared with the use of thin sections. Nonetheless, encouraging preliminary results have been demonstrated in healthy volunteers. Conventional gradient echo pulse sequences used for cine cardiac imaging consist of a radiofrequency pulse followed by a gradient refocused echo. The data acquisition is synchronized to the cardiac cycle. One line of data (one phase-encoding step) is acquired for each phase of the cardiac cycle within one R–R interval. As a result, the time required to acquire 128 lines of data (the amount required for a 128 × 256 matrix image) is 128 × R–R interval. For a 1-s interval, the imaging time is approximately 2 min, too long for a breath-hold. In order to acquire the images within a breath-hold interval, the sequence is modified so that up to 10 lines of data are acquired in rapid sequence within a 100 ms time period.24 This group of lines is called one segment, and the imaging technique is called segmented turboflash or breath-hold cine. In our experience, there is relatively little blurring of coronary artery detail for segment durations up to 100 ms, but the blurring worsens as the duration is further increased. Imaging is performed during diastole, when cardiac motion is least and coronary flow velocities are high. One segment is acquired during each heart

Figure 1 Left anterior descending coronary artery. (a) Axial segmented turboFLASH image shows the left anterior descending coronary artery (arrow) and a septal branch. (b) Oblique view in another subject shows the left anterior descending coronary artery (arrow) and a septal branch

beat, and a series of these segments, acquired over multiple heart beats, are then combined to create a whole image. For instance, 16 segments, each consisting of eight lines of data, would be combined so as to complete a 128 × 256 matrix. The imaging time would be 16 heart beats, on the order of 16 s or less. Cine imaging is also feasible by acquiring segments in rapid sequence (one segment per cardiac phase) within each breath-hold. Flow velocities in the coronary arteries of resting subjects are rather low, on the order of 10–30 cm s−1 . As a result, it is difficult to develop adequate flow contrast in the coronaries, compared to the adjoining myocardium, on the basis of inflow effects alone. This is a particular problem when the coronary arteries are imaged longitudinally. However, the epicardial portions of the coronary arteries are typically surrounded by fat. It is possible to apply fat suppression pulses (binomial or chemical shift selective) to diminish the signal intensity of the epicardial fat. By combining the segmented turboflash method with fat suppression techniques, the proximal and more distal portions of the epicardial coronary arteries are routinely depicted. We have begun clinical studies of MR coronary angiography (Figure 1 and 2).25 Sensitivity and specificity for coronary artery disease are in the 80–90% range, which is comparable with thallium scintigraphy. One problem has been poor flow contrast resulting from slow flow distal to a severe stenosis, making it difficult to distinguish a stenosis from an

Figure 2 Severe stenosis of the right coronary artery. (a) MR angiogram shows the lesion (arrow). (b) Coronary catheterization

CORONARY ARTERY DISEASE EVALUATED BY MRI

occlusion. This problem might be ameliorated by the use of contrast agents. Blood pool contrast agents are currently under development that could greatly improve the quality of coronary artery images. Compared with our original technique, we have recently obtained a greater than 50% improvement in signal-to-noise ratio by lowering the sampling bandwidth of the sequence from 195 Hz pixel−1 to 78 Hz pixel−1 , keeping the echo time at 7 ms. This was possible because of enhancements in the gradient capabilities of our MRI system. The signal-to-noise ratio improvement has allowed us to position the patient supine within the body coil, which is better tolerated than the prone positioning previously required with a flat surface coil. Further 2–3-fold signal-to-noise improvements will be possible with the introduction of circularly polarized whole-volume phasedarray coils. Flow velocity measurements can be done by using a phase contrast modification of segmented turboFLASH, in which two segmented turboFLASH images are acquired in a single breath-hold.26 In one sequence, the slice-selective gradient incorporates a three-lobed, velocity-compensated waveform that is insensitive to flow; in the other, a bilobed waveform is applied resulting in a measurable shift in the phase of the MR signal of flowing blood. The gradients were calibrated to permit flow velocity measurements up to 150 cm s−1 . A 96 × 256 acquisition matrix is used with pixel dimensions of 1.4 mm × 0.8 mm. Four images are produced by this sequence, comprising magnitude and phase reconstructions of the flow-insensitive and flow-sensitive data. The flowinsensitive magnitude image shows vessel anatomy, whereas subtraction of the flow-insensitive and flow-sensitive phase images produces a phase difference image in which the phase shift is proportional to the flow velocity along the direction of slice selection. Image subtraction eliminates background phase shifts unrelated to blood flow. The method was tested in 12 subjects studied at rest and four studied before and during pharmacologic stress using intravenous adenosine. Flow velocities at rest in the midportion of the right coronary artery were 9.9 ± 3.5 cm s−1 (n = 12); in the proximal left anterior descending artery they were significantly higher, measuring 20.5 ± 5.2 cm s−1 (n = 6). With adenosine, flow velocities typically increased at least fourfold. These results indicate that noninvasive measurement of coronary artery flow velocities is feasible using MR angiography. Table 1 summarizes the current performance of MR coronary angiography in our clinical practice. Finally, entirely new imaging techniques are likely to be developed that are advantageous compared with segmented turboFLASH. The fastest known method for creating an MR image is called echo planar imaging (EPI).27 The most commonly used EPI method, called blipped EPI, oscillates the read-out gradient so that a series of gradient echoes are generated. Each gradient echo is separately phase-encoded by a short duration gradient pulse (the blip); alternatively, a constant amplitude phase-encoding gradient can be applied for the entire read-out period. Using specially designed gradient coils, EPI can generate a series of 64 echoes in as little as 32 ms. Thus a 64 × 128 matrix image can be acquired in 1/30th of a second, which is sufficiently short to freeze any kind of physiologic motion, including that of the heart. We have had promising results imaging the coronary arteries recently using an echo planar MR angiography method that circumvents

3

Table 1 Sensitivity and Specificity of the MR Coronary Angiographic Technique to Correctly Classify a Vessel as having Significant Disease.a Sensitivity/specificity for Individual Vessels are as Listed.b

Left main LAD LCX RCA Patient

No. (%) with disease

Sensitivity (%)

Specificity (%)

PV (+)

PV (−)

2 23 7 20 29

100 87 71 100 97

100 92 90 78 70

1.00 0.95 0.63 0.83 0.90

1.00 0.80 0.93 1.00 0.88

(5) (64) (20) (53) (74)

a Sensitivity

= 90%; specificity = 90%. = left anterior descending coronary artery; LCX = left circumflex coronary artery; RCA = right coronary artery; PV (+) = predictive value positive; PV (−) = predictive value negative.

b LAD

the problem of vessel tortuousity by using thick slices and suppression of background tissue signal.

1 RELATED ARTICLES

Abdominal MRA; Whole Body Magnetic Resonance Angiography; Head and Neck Studies Using MRA; Phase Contrast MRA; Time-of-Flight Method of MRA.

2 REFERENCES 1. S. S. Yang, L. G. Bentivoglio, V. Maranhao, and H. Goldberg (eds.) Cardiac Catheterization Data to Hemodynamic Parameters, 3rd edn., F. A. Davis, Philadelphia, PA, 1988, p. 256. 2. B. L. Holman, S. C. Moore, P. M. Shulkin, C. M. Kirsch, R. J. English, and T. C. Hill, Invest. Radiol., 1983, 18, 322. 3. J. H. Cladwell, D. L. Williams, G. D. Harp, J. D. R. Stratton, and J. L. Ritchie, Circulation, 1984, 70, 1048. 4. R. C. Marshall, J. H. Tillisch, M. E. Phelps, S. C. Huang, R. Carson, E. Henze, and M. R. Schelbert, Circulation, 1983, 67, 766. 5. R. Brunken, J. H. Tillisch, M. Scwarger, J. S. Childs. R. Marshall, M. Mandelkern, M. E. Phelps, and H. R. Schelbert, Circulation, 1986, 73, 951. 6. C. C. Chen, J. Morganroth, S. Ogawa, and T. J. Mardelli, Circulation, 1980, 62, 288. 7. S. Ogawa, C. C. Chen, F. E. Hubbard, F. S. Pauletto, T. J. Mardelli, J. Morganroth, L. S. Dreifus, M. Akashi, and Y. Nakamura, Am. J. Cardiol., 1980, 45, 301. 8. F. B. Pearce, K. H. Sheikh, N. P. deBruijn, and J. Kisslo, J. Am. Soc. Echo, 1989, 2, 276. 9. W. B. Daniel, R. Erbel, W. Kasper, C. A. Visser, R. Engberding, O. R. Sutherland, E. Arribe, P. Hanroth, and B. Maisch et al., Circulation, 1991, 83, 817. 10. L. W. Johnson, E. C. Lozner, S. Johnson, R. Kron, A. D. Pichard, G. W. Vetrovec, and T. J. Nota, Cathet. Cardiovasc. Diagn., 1989, 17, 5. 11. K. Davis, J. W. Kennedy, H. G. Kemp, M. P. Judkins, A. J. Gosselin, and T. Killip, Circulation, 1979, 59, 1105. 12. J. W. Kennedy/ Registry Committee of the Society for Cardiac Angiography, Cathet. Cardiovasc. Diagn., 1982, 8, 5. 13. R. M. Wyman, R. D. Safian, V. Portway, S. J. Shilman, R. G. McKay, and D. S. Bain, J. Am. Coll. Cardiol., 1988, 12, 1400.

4 CORONARY ARTERY DISEASE EVALUATED BY MRI 14. K. L. Gould, Cardiovasc. Intervent. Radiol., 1990, 13, 5. 15. D. Didier, C. B. Higgins, M. R. Fisher, L. Osaki, N. H. Silverman, and M. D. Cheitlin, Radiology, 1986, 158, 227. 16. C. B. Higgins, W. Holt, P. Pflugfelder, and U. Sechtem, Magn. Reson. Med., 1988, 6, 121. 17. P. M. Ruggieri, G. A. Laub, T. J. Masaryk, and M. T. Modic, Radiology, 1989, 171, 785. 18. T. J. Masaryk, M. T. Modic, P. M. Ruggieri, J. S. Ross, G. Laub, G. W. Lenz, J. A. Tkaeh, E. M. Haacke, W. R. Selman, and S. I. Harik, Radiology, 1989, 171, 801. 19. K. C. Kent, R. R. Edelman, D. Kim, T. I. Steinman, D. H. Porter, and J. J. Skillman, J. Vasc. Surg., 1991, 13, 311. 20. S. Paulin, Acta Radiologica, 1964(Suppl.), 125. 21. S. Paulin, G. K. von Schulthess, E. Fossel, and H. P. Krayenbuehl, Am. J. Roentgen., 1987, 148, 665.

22. C. B. Paschal, E. M. Haacke, L. P. Adler, W. Lin, A. Sketty, and R. J. Alfidii, Proc. 9th Annu. Meet. Soc. Magn. Reson. Med., New York, Aug. 18–24, 1990 , p. 278. 23. S. J. Wang, B. S. Hu, A. Marcovski, and D. G. Nishimura, Magn. Reson. Med., 1991, 18, 417. 24. R. R. Edelman, W. J. Manning, D. Burstein, and S. Paulin, Radiology, 1991, 181, 641. 25. W. J. Manning, W. Li, and R. R. Edelman, N. Engl. J. Med., 1993, 328, 828. 26. R. R. Edelman, W. J. Manning, E. Gervino, and W. Li, J. Magn. Reson. Imag., 1993, 3, 699. 27. M. K. Stehling, R. Turner, and P. Mansfield, Science, 1991, 254, 43.

HEART: CLINICAL APPLICATIONS OF MRI

Heart: Clinical Applications of MRI

1

The ®rst electrocardiographically gated magnetic resonance images of the human heart were shown in 1984. In the decade that followed several important techniques have been developed which facilitate the use of MRI for the evaluation of cardiovascular morphology, contractile function, myocardial tissue characterization, and blood ¯ow. With these techniques, MRI has been found to be effective for the investigation of a variety of acquired and congenital cardiovascular diseases. This role of MRI in relation to preexisting cardiac imaging techniques continues to evolve. The clearly de®ned indications for the effective use of MRI in acquired heart disease include the evaluation of pericardial disease, right ventricular dysplasia, thoracic aortic disease, intra- and extracardiac masses, cardiomyopathies, valvular heart disease, ischemic myocardial disease, and the morphology of the coronary arteries. The indications for the application of MRI in congenital heart disease include the assessment of pulmonary artery morphology, coarctation of the aorta, vascular rings, complex congenital heart disease, and postoperative evaluation.

extracardiac structures (e.g. the great vessels). Anatomical information also can be obtained using electrocardiogram (ECG)gated cine gradient echo (cine GRE) and newer, fast gradient echo techniques that sample segmented k-space during a single breath-hold (breath-hold cine GRE). These latter two techniques are also useful for acquiring multiple, ECG referenced images of the heart, in either long or short axis projections similar to echocardiography, which may be electronically `looped' and viewed in a cinematic format to evaluate global ventricular function and segmental wall motion. Velocity-encoded cine MRI techniques can assess quantitatively blood ¯ow velocity and ¯ow volume within the heart to evaluate intracardiac shunts, and valvular pathology and function. This same technique may also be used in the great vessels to determine pulmonary and aortic blood ¯ow, including differential ¯ow to the right and left pulmonary arteries, as well as to estimate gradients across areas of stenosis such as in coarctation of the aorta. For multislice spin echo MRI, ECG gating is required to `freeze' the motion of the heart, and respiratory compensation further improves image quality (see also Cardiac Gating Practice and Motion Artifacts: Mechanism and Control). In adults, typical sequences include ECG gating T1-weighted images of 7±10 mm slice thickness obtained in the axial plane, with supplemental images acquired in the sagittal, and oblique sagittal, axial, and coronal planes, as dictated by the particular clinical concern. In adults and children, body coils are used most commonly. Flexible phased array surface coils are gaining in use for both general and special purposes, as they provide greater resolution and image contrast as a result of improved signalto-noise ratios. For some infants and small children a head coil may be practical, and will result in the greatest resolution and signal-to-noise ratio for the relatively small areas of interest.1

2 IMAGING TECHNIQUES

3

A multitude of sequences are available for cardiac imaging (Table 1). These include spin echo and multiplanar multiphasic techniques for morphological information of both cardiac and

Cardiovascular MRI is used primarily for detection and delineation of anatomical abnormalities according to a recent survey of members of the Society for Magnetic Resonance

Scott D. Flamm and Charles B. Higgins University of California at San Francisco, San Francisco, CA, USA

1 INTRODUCTION

LEVEL OF CURRENT CLINICAL ACTIVITY

Table 1 Techniques for MRI of the Heart Technique

Application

ECG-gated spin echo

Anatomy and morphology

Cine GRE

Anatomy; atrial and ventricular dimensions and volumes; global and regional ventricular function

Fast (turbo) GRE

Anatomy; myocardial perfusion; rapid imaging of great vessels

Breath-hold, segmented k-space GRE

Anatomy including coronary arteries; myocardial perfusion; global and regional ventricular function

Velocity encoded cine MRI

Valvular function (quanti®cation of regurgitation, and estimation of peak valvular gradient), measurement of stroke volume and cardiac output of each ventricle, quanti®cation of shunts, measurements of differential ¯ow in right and left pulmonary arteries, quanti®cation of gradients at stenoses of the pulmonary arteries or coarctation of the aorta

Echo planar

Global and regional function of left and right ventricles, quanti®cation of regional myocardial blood volume and perfusion

2 HEART: CLINICAL APPLICATIONS OF MRI Imaging. Of the responding centers, 46% reported exclusively clinical work, 3% exclusively research, and 51% both clinical and research. The majority of MRI studies are performed in the evaluation of thoracic aortic disease, congenital heart disease, and cardiac masses. While most studies are performed for morphological information, half of the centers also accomplish functional studies, usually for ventricular and/or valvular function, in conjunction with anatomical studies.2

4 ACQUIRED HEART DISEASE 4.1 Pericardial Disease Differentiating constrictive pericarditis from restrictive pericarditis is the most frequent indication for applying MRI in pericardial disease. This is an important distinction because both diseases cause similar hemodynamic abnormalities, and are usually indistinguishable on cardiac catheterization. Only constrictive pericarditis can be alleviated by surgery. Constrictive pericarditis is seen most commonly after cardiothoracic surgery, and radiation therapy. On spin echo sequences it is manifested by local or generalized thickening of the pericardium to 4 mm or greater. This thickening is more readily evident over the right ventricle and right atrium. Associated ®ndings include enlargement of the right atrium, inferior vena cava and hepatic veins, pericardial effusion, and tubular narrowing of the right ventricle. The reported diagnostic accuracy of MRI in establishing the diagnosis is 93%.3 MRI also can be used to examine the pericardium and its contents for evidence of effusion or hemorrhage, and mass lesions. Echocardiography is a fast, effective technique to determine the presence and general quantity of pericardial effusions, but can be less accurate in discerning hemorrhagic from nonhemorrhagic effusion, or determining the extent and location of loculated effusions, particularly in the posterior and basal portions of the heart. Spin echo MRI can easily determine the anatomical location and size of effusions, as well as the extent of loculations in all portions of the heart. It also has the advantage of more easily identifying the presence of pericardial masses, which may represent metastatic tumors responsible for hemorrhagic and/or recurrent effusion. The hemorrhagic nature of effusions can be identi®ed with T1weighted spin echo sequences as high signal intensity, in contrast to the very low signal intensity of nonhemorrhagic pericardial effusions.

4.2 Right Ventricular Dysplasia Right ventricular dysplasia is a relatively uncommon disorder generally occurring in young people, and manifesting as potentially life threatening arrhythmias originating from the right ventricle. The diagnosis is predicated on the identi®cation of fatty in®ltration or ®brosis replacing the right ventricular myocardium. The diagnosis of right ventricular dysplasia is dif®cult to establish even with electrophysiological studies and myocardial biopsy. The primary diagnostic feature on MRI is transmural fat within the free wall of the right ventricle, but the spectrum of

®ndings includes `islands' of fat within, or abnormal thinning of, the right ventricular free wall, regional or global contraction abnormalities of the right ventricle, and aneurysmal outpouching of the right ventricular out¯ow tract during systole.4 Axial spin echo images focused on the right ventricular free wall are used to demonstrate myocardial fat, and cine GRE sequences obtained in the axial or horizontal long axis are useful to evaluate regional right ventricular contraction abnormalities. The utility of MRI in establishing the diagnosis noninvasively was recently documented in a group of patients with biopsy proven right ventricular dysplasia. Those patients with inducible right ventricular arrhythmias were signi®cantly more likely to have identi®able fat within the myocardium, contraction abnormalities of the right ventricle, and/or a lower ejection fraction.5 The course of patients with right ventricular dysplasia typically has been followed with periodic right heart angiography and myocardial biopsy, which entails a signi®cant degree of risk. MRI appears to be an effective technique for identifying at least a subset of patients with right ventricular dysplasia, and may become the procedure of choice for follow-up. 4.3

Thoracic Aortic Disease

Aneurysm and dissection are the most common indications for MRI of the thoracic aorta. Spin echo sequences are used to demonstrate the morphology of thoracic aortic aneurysms, which may occur as a result of atherosclerosis, trauma, or Marfan's syndrome, or may be mycotic or syphilitic in origin. In aortic dissection the intimal ¯ap can often be identi®ed on spin echo images, but is more readily seen with GRE and cine GRE sequences (Figure 1). These techniques also are effective in correctly identifying the true and false lumens, and differentiating slow ¯ow versus thrombus within the false lumen. Additionally, MRI can identify the presence of hematoma within the wall of the aorta.6 This cannot be de®ned with Xray angiography. MRI is a highly sensitive and speci®c technique for the detection of aortic dissection that can identify both the intraand extravascular space, and the walls of the aorta. Conventional angiography has been considered the gold standard, but requires an invasive procedure, may incompletely identify the false lumen and its extent, and is unable to evaluate fully the walls of the aorta, and the extravascular space. MRI has proven to be superior to X-ray angiography and transthoracic echocardiography. MRI has been compared to transesophageal echocardiography, which is also a highly sensitive technique, and has the advantage of being portable, permitting its use in critical and unstable patients unable to leave the intensive care setting. Multiple studies have demonstrated a similar high sensitivity (98±100%) for both methods, but MRI has a signi®cantly higher speci®city (98±100%) than transesophageal echocardiography (68±77%) in high risk populations.7,8 Evaluation of the thoracic aorta usually requires approximately 20±30 min using standard cine GRE sequences. Newer sequences, particularly segmented k-space breath-hold cine GRE sequences, can acquire each slice in 12±16 s, providing images of the entire thoracic aorta in approximately 5 min. The study can be extended to the abdominal aorta to evaluate for potential involvement of the mesenteric vessels with the addition of a few minutes of imaging time.

HEART: CLINICAL APPLICATIONS OF MRI

3

Figure 1 Series of selected images from a patient with a type A aortic dissection involving the aortic arch, descending thoracic aorta, and abdominal aorta. These are images acquired using a breath-hold segmented k-space cine MRI sequence acquired at the level of (a) the aortic arch, (b) the left pulmonary artery, (c) the main and right pulmonary artery, and (d) the upper abdomen. Note the compression of the true lumen by the false lumen. The bright signal within the false lumen con®rms the presence of ¯ow. T, true lumen; F, false lumen

4.4 Intracardiac Masses Tumor, thrombus, and valvular vegetations (as occur in bacterial endocarditis) are the primary considerations when there is concern for an intracardiac mass. There are no studies demonstrating the bene®t of MRI for valvular vegetations; echocardiography remains the most effective noninvasive technique for this determination. MRI is effective for evaluating intracardiac masses, and in most instances can differentiate between tumor and thrombus.9 Spin echo sequences are usually adequate to identify the size and location of intracardiac masses. Both tumors and thrombus tend to be isointense to myocardium on spin echo images; however, some thrombi may have a low signal intensity and be indistinguishable from the ventricular cavity without the additional application of cine GRE sequences.10 Cine GRE sequences are particularly useful in that they permit identi®cation of the majority of masses, and typically are successful in differentiating thrombus from tumor.11 The most common tumors include myxomas (usually left atrial), rhabdomyomas (usually right ventricular), and metastases from primary carcinoma of the breast or lung, or melanoma. On cine GRE images tumors tend to be intermediate in signal intensity, with

myocardium being low±intermediate in intensity, while thrombus is frequently very low in signal intensity. In the case of relatively fresh thrombus, high signal intensity may be present on cine GRE images. These signal characteristics hold true in the majority of instances; however, some tumors, particularly myxomas which have bled internally, may contain increased amounts of hemosiderin, making them very low signal intensity on GRE images. 4.5

Cardiomyopathy

There are four recognized types of cardiomyopathyÐhypertrophic, congestive (or dilated), restrictive, and obliterativeÐeach of which has a distinct constellation of morphological and functional features. The etiology of the myocardial abnormality is varied. Most forms are idiopathic in nature, but chronic injury from ischemia or toxins, or in®ltrative disorders are responsible for a minority of cases. The hypertrophic and congestive forms have been studied extensively with MRI; less is known about the obliterative form as it is very rare, and encountered primarily in Africa. Hypertrophic cardiomyopathy usually manifests as global hypertrophy of the left ventricular myocardium, though many

4 HEART: CLINICAL APPLICATIONS OF MRI variants have been documented with MRI, including focal or diffuse septal hypertrophy, and midventricular and apical forms.12 Cine GRE, acquired in either the long or short axis, has been used to quantify left ventricular mass, volumes, and ejection fraction in this disease. MRI has shown high accuracy and interstudy reproducibility for quantifying left ventricular volumes and mass, providing a precise technique for serial assessment during treatment.13,14 Patients with hypertrophic cardiomyopathy also demonstrate abnormal diastolic and systolic function. The overall heart size remains normal, but the end-diastolic volume and ®lling rates are reduced in both the left and right ventricles.15 Systolic function is hyperdynamic, and near obliteration of the ventricular cavity, particularly at the apex, is a characteristic feature (Figure 2). In congestive cardiomyopathy the overall cardiac dimensions are increased, and there is left ventricular dilation. The

idiopathic forms have normal myocardial thickness, and as a result the myocardial mass is greater than normal. In patients with chronic injury or ischemic etiology the myocardium may be globally or focally thinned. As in the hypertrophic form, cine MRI in congestive cardiomyopathy can quantify the ventricular volumes and mass, and assess ventricular function and wall stress.16 Recent studies have demonstrated the effectiveness of using serial MRI examinations to follow patients with congestive cardiomyopathy undergoing treatment with angiotensin-converting enzyme inhibitor therapy.17,18 Restrictive cardiomyopathy may have either normal or thickened myocardium, and normal, increased, or decreased left ventricular end-diastolic volume. The salient characteristic functional feature is markedly reduced diastolic compliance, which is similar to that seen in constrictive pericarditis. Constrictive pericarditis, however, may be differentiated by the

Figure 2 Short axis images of the heart in a patient with hypertrophic cardiomyopathy. The upper row of images were acquired in diastole (the images on the left are from near the apex, and the images on the right from the mid-venticular region), and directly below each is the respective systolic image, demonstrating marked, global thickening of the left ventricular myocardium, and obliteration of the apical cavity during systole. The sequence used is a breath-hold segmented k-space cine MRI

HEART: CLINICAL APPLICATIONS OF MRI

abnormally thickened pericardium (>4 mm), usually noted over the right side of the heart. Patients who have restrictive cardiomyopathy with thickened myocardium (such as in amyloidosis) may be differentiated from those with hypertrophic cardiomyopathy based on the reduced systolic function in the former, compared with the hyperdynamic function in the latter.19 4.6 Valvular Heart Disease MRI using cine GRE sequences can identify the severity, extent, and direction of the turbulent jet of blood across stenoses of the aortic, pulmonic, mitral, and tricuspid valves. Turbulence, on cine GRE images is shown as a signal void against the background of bright signal intensity blood in the great vessels or ventricle. Using these images as a reference, the peak velocity of the turbulent jet may be determined with velocity-encoded cine MRI images obtained perpendicular to the long axis of the turbulent jet near the valve's ori®ce. Velocity-encoded cine MRI can determine the peak velocity of stenotic jets with accuracy up to 6 m sÿ1.20 By using the modi®ed Bernoulli equation [gradient (mmHg) = 4  (peak velocity (m sÿ1))2], the peak gradient across the stenosis can be calculated; this is one of the important clinical indicators used in determining the effectiveness of medical therapy, and, when necessary, timing of valve replacement surgery. Recent work has demonstrated the usefulness of this technique in both the aortic21 and mitral22 valves. Cine GRE sequences also have been used to measure the extent and direction of the regurgitant jet in aortic and mitral insuf®ciency. Quanti®cation of the volume of regurgitant ¯ow may be acquired using velocity-encoded cine MRI sequences where the cardiac cycle (each R±R interval) is divided into 16 phases, and ¯ow is measured at each phase of the cycle. In the case of aortic regurgitation, an imaging slice is acquired perpendicular to the long axis of the aorta proximal to the origin of the innominate artery. The data at each of the 16 phases can be plotted versus time, and the forward, reverse, and total ¯ow calculated.23 This method has the advantage of determining absolute ¯ow, and has high interstudy reproducibility.24 Regurgitant volume across the mitral valve has been calculated by measuring the amount of left ventricular in¯ow during diastole (velocity-encoded cine MRI slice across the mitral valve) and subtracting the left ventricular out¯ow during systole (imaging slice in the ascending aorta).25 4.7 Ischemic Heart Disease MRI has the potential to become an important technique to differentiate normal from ischemic or infarcted myocardium by combining information about morphology, function, and perfusion in a single study. Of these three the largest hurdle has been determining global and regional myocardial perfusion in a rapid, reproducible manner under conditions re¯ecting both the basal state, and stress. The earliest attempts at differentiating ischemic from normal myocardium focused on acute infarctions because of the edema that occurs secondary to vascular leak at the site of damage. Increased water in tissue results in increased signal on T2weighted scans in areas of acute infarction due to T2 prolongation.26 Subsequently, MRI contrast agents have been used to improve the distinction between acute and subacute infarc-

5

tions, and normal myocardium. Contrast enhancement (greater than the adjacent normal myocardium) has been noted uniformly in many reperfused areas of myocardium, or only peripherally in areas of infarction in which the central zones had necrosis, hemorrhage, or marked edema that limited or ceased blood ¯ow centrally.27,28 Detecting areas of stress induced ischemic myocardium has been approached using contrast-prepared fast imaging techniques. Recent studies have demonstrated the feasibility of monitoring the passage of a bolus of MRI contrast agent through the myocardium using rapid, sequential images.29±32 Myocardium supplied by coronary arteries with angiographically proven signi®cant stenoses demonstrates lower peak signal intensity after contrast administration, and a lower rate of signal increase than normal myocardium.29 Ischemic myocardium, induced with the administration of dipyridamole, a pharmacological stress agent, also demonstrates lower peak signal intensity as compared with normal myocardium.30 At present, imaging is limited to a few imaging planes for each study, but with the development of echo planar imaging, dynamic acquisition of the entire heart during the ®rst pass of MRI contrast media will be feasible (see Figure 3; also EchoPlanar Imaging). Ischemic myocardium also may be identi®ed by functional wall motion abnormalities observed on cine GRE sequences acquired in the basal state or during pharmacological stress. Areas of threatened myocardium demonstrate decreased contractility during dipyridamole infusion.33 One study found the sensitivity/speci®city (percent values) for localization of stenotic vessels at 78/100 for the left anterior descending, 73/100 for the left circum¯ex, and 88/87 for the right coronary artery.34 This technique has also been compared to radionuclide scanning using 99mTc±methoxyisobutyl isonitrile SPECT (MIBI SPECT), with no signi®cant differences detected in sensitivity for lesion localization.35 Dobutamine is another pharmacological stressor that has been used to assess wall motion abnormalities with cine GRE sequences. Dobutamine MRI studies have demonstrated close agreement with ®ndings on dobutamine thallium-201 SPECT,36 and a sensitivity and speci®city of 81 and 100%, respectively, using coronary arteriography as the gold standard.37 4.8

Coronary Arteries

MRA techniques have recently been used for imaging of the coronary arteries.38±40 The coronary arteries present a formidable challenge for MRI, because they are small structures whose motion within the thorax is complex, being in¯uenced by the spiral contraction of the heart, expansion of the chest, and descent of the diaphragm. The proximal and midportions of the coronary arteries have been demonstrated by MRA with a fat-suppressed, cine GRE technique using segmented k-space and breath-holding. In a group of 19 normal volunteers, and six patients with recent angiograms, the left anterior descending and right coronary arteries were identi®ed in 100%, the left main in 96%, and the left circum¯ex in 76%.41 A subsequent study using a similar MRA technique in 39 patients with angiographically identi®ed coronary stenoses demonstrated a sensitivity and speci®city of 90 and 92% for detection of signi®cant (550% narrowing) lesions.42

6 HEART: CLINICAL APPLICATIONS OF MRI

Figure 3 Series of short axis images of the heart acquired repetitively at the same level over a 90 s period during bolus administration of contrast media. The transit of contrast material increases signal intensity ®rst in the right ventricular blood pool, then left ventricular blood pool, and ®nally the myocardium of the right and left ventricles. The patient has had an infarction in the anterior and anteroseptal distribution (arrowheads); note the delayed rise in signal intensity compared to the adjacent normal myocardium

The use of complex, oblique planes has permitted visualization of more than 8 cm lengths of the right coronary artery and left anterior descending coronary artery (Figure 4).38 Other recent advances include non-breath-hold three-dimensional

imaging using fat saturation and magnetization transfer contrast techniques. Initial work has demonstrated the ®rst 3±10 cm of the left and right coronary arteries from a single 7±10 min acquisition.43 Further improvements in scanning time, image

Figure 4 MRA of the coronary arteries using a breath-hold segmented k-space cine sequence. (a) An oblique coronal image demonstrating the proximal and midportions of the right coronary artery (arrowheads). (b) An axial image showing the left main and proximal left anterior descending coronary arteries (arrowheads)

HEART: CLINICAL APPLICATIONS OF MRI

resolution, and surface coil technology may result in a complete magnetic resonance angiogram of the coronary arteries.

5 CONGENITAL HEART DISEASE 5.1 Coarctation of the Aorta MRI can provide excellent anatomic visualization of the aorta. Images are acquired ®rst in the axial plane using a spin echo sequence, and then using these images as a reference a set of oblique sagittal images are acquired parallel to the plane of, and including, the ascending and descending aorta. These two sets of images can adequately describe the anatomical location and severity of the stenosis, including involvement of arch vessels, and degree of poststenotic dilation. A cine GRE sequence in the same oblique sagittal plane can demonstrate the jet of turbulent ¯ow along the descending thoracic aorta just distal to the site of coarctation.44 The gradient across the stenosis can be estimated from the peak velocity (using the modi®ed Bernoulli equation described previously); measurements performed with velocity-encoded cine MRI reveal a close correlation (r = 0.95) with values obtained by continuous wave Doppler ultrasound.45 The length of the turbulent jet has been correlated with the severity of stenosis, a factor important in gauging the appropriate timing of initial corrective surgical procedure, or subsequent repair in recoarctation. At present, determining the optimal time for corrective surgery has been based on a combination of clinical ®ndings and the echocardiographically determined gradient across the coarctation, both of which are only loosely correlated with the severity of stenosis. Recent work has revealed a unique ability of MRI in the evaluation of coarctation of the aorta. Velocityencoded cine MRI was used to measure blood ¯ow in the descending thoracic aorta just distal to the coarctation and at the level of the diaphragm, with the difference in ¯ow representing the contribution from collateral vessels (Figure 5). While normal individuals have an average 8% decrease in blood ¯ow along the descending thoracic aorta, patients with signi®cant coarctations have an average 80% increase in ¯ow.46 This technique represents the only currently available noninvasive method to measure the extent of collateral development, and their contribution in supplying blood to the abdominal aorta.

5.2 Pulmonary Arteries The evaluation of pulmonary arterial anatomy, orientation, size, and ¯ow is critical to the management of children with lesions affecting the pulmonary arteries such as tetralogy of Fallot, and pulmonary atresia and stenosis.47 The pulmonary arteries are best visualized using 3±5 mm ECG-gated spin echo images in the axial plane, followed by oblique sagittal images through the pulmonary infundibulum, and/or right and left pulmonary arteries. MRI can de®ne accurately the location and severity of narrowing in the right ventricular out¯ow tract and main pulmonary artery, as well as associated poststenotic dilation.48,49 In a series of 12 patients who had undergone the arterial switch operation MRI correctly identi®ed

7

92% of the pulmonary artery stenoses, while echocardiography identi®ed only 58%. Moreover, MRI was superior in identifying peripheral pulmonary stenoses, as echocardiography visualized only seven of the 17 peripheral stenoses seen by MRI.50 Pulmonary atresia is identi®ed by a ®brous and/or muscular band at the expected junction of the main pulmonary artery and the right ventricular out¯ow tract, on both axial and sagittal images. Determining the presence or absence of the main pulmonary artery and continuity of the right and left pulmonary arteries is crucial information for surgical planning. At least two studies, using angiography as the gold standard, have con®rmed the ability of MRI to determine correctly the status of main, left and right pulmonary arteries, without the need for an invasive procedure.51,52 Quantitative, physiological information regarding ¯ow within the pulmonary arteries can be obtained using velocityencoded cine MRI images. This technique can accurately quantify both total, and right and left lung differential blood ¯ow,53 and correlates well with radionuclide lung scanning for the evaluation of relative blood ¯ow to the left and right lungs.54 This information is useful for determining the severity of stenoses and their effect on total and differential pulmonary blood ¯ow (Figure 6). In addition, it may prove particularly valuable in serially assessing ¯ow at sites of stenosis after balloon valvuloplasty, or surgical repair.

5.3

Vascular Rings

MRI is ideal to de®ne the aortic arch and great vessel relationships because of its multiplanar capabilities and ability to image the adjacent air-®lled trachea and esophagus, both of which may be compressed by the vascular structures. Spin echo sequences are typically employed in the axial and sagittal planes to demonstrate the commonly encountered abnormalities which include double aortic arch, right aortic arch with mirror image branching, right aortic arch with aberrant left subclavian artery, left aortic arch with aberrant left subclavian artery, and pulmonary sling. The sagittal images often best demonstrate the compression on both the trachea and esophagus. Coronal images may provide further information regarding double aortic arches. Imaging may also be performed using magnetic resonance angiography, with either time-of-¯ight or phase contrast techniques.55,56 Both techniques produce signal primarily from moving blood resulting in images that re¯ect the intravascular space, similar to conventional contrast angiograms, but without the need for contrast or an invasive procedure. These sequences have a further advantage in that the images may be reconstructed into a three-dimensional format, and viewed from any desired angle. Further postprocessing permits close examination of a particular area of interest for detailed analysis.

5.4

Complex Congenital Heart Disease

Complex congenital heart disease requires de®nition of the complicated anatomical relationships of the heart. MRI permits segmental analysis of complex congenital heart disease by providing tomograms encompassing the great vessels, heart, and infradiaphragmatic structures. Axial spin echo sequences are

8 HEART: CLINICAL APPLICATIONS OF MRI

900

(e)

Flow volume (ml min–1)

800 700

4.8 l min–1

600

Post coarctation Level of diaphragm

500 400 300 200

2.5 l min–1

100 0 1 2 3 4 5 6 7

8 9 10 11 12 13 14 15 16

Time frames

Figure 5 Patient with coarctation of the aorta. (a, b) ECG-gated spin echo images acquired in the axial and oblique sagittal projections, respectively. (c, d) Magnitude and phase images, respectively, from a VEC sequence acquired at the level of the diaphragm. Regions-of-interest have been placed over the aorta. (e) A ¯ow curve can be derived, and blood ¯ow at that level calculated (as described in the text). Similar information is obtained from a slice positioned just distal to the site of coarctation. The difference in blood ¯ow between the sites in the descending thoracic aorta is a direct measure of the contribution of collateral blood ¯ow to the abdominal aorta. A, ascending aorta; D, descending aorta; arrow, coarctation

usually suf®cient to de®ne the anatomy, with sagittal, coronal, and oblique imaging planes used to further complement the anatomical information. A segmental approach is used to de®ne the anatomy beginning with determination of the abdominal situs, and the relationship of the inferior vena cava and aorta. After this, the

pattern of blood ¯ow through the heart and great vessels is determined starting with the systemic and pulmonary venous connections, and then identifying the atrial situs, and presence and position of the atrioventricular valves. Next, the ventricular anatomy, and ventriculoarterial connections and great artery relationship are determined.

HEART: CLINICAL APPLICATIONS OF MRI (b)

9

Flow volume (ml min–1)

1400 MPA flow = 5.5 l min–1 LPA flow = 4.6 l min–1 RPA flow = 1.0 l min–1

1200 1000 800 600 400 200 0 –200

2 3 4 5 6 7

8 9 10 11 12 13 14 15 16

Time frames

Figure 6 (a) Axial spin echo image of a patient with stenosis (between arrows) of the proximal right pulmonary artery. VEC MRI can be performed perpendicular to the main, and proximal right and left pulmonary arteries, and the total blood ¯ow to the lungs, and differential blood ¯ow to each lung calculated. (b) The different ¯ows calculated for each of the respective arteries are demonstrated here. R, right pulmonary artery; L, left pulmonary artery; M, main pulmonary artery

The absolute and relative sizes of the ventricular system including quanti®cation of ventricular mass, and the quantity and spatial extent of septal tissue are all well de®ned by MRI.57±59 Many complex lesions have been evaluated correctly with MRI including double-outlet right ventricle,60,61 doubleoutlet left ventricle,62 double-inlet ventricle,63 single ventricle and atrioventricular canal,64 partial and total anomalous pulmonary venous connection,65,66 and Ebstein's anomaly.67 In a study of 29 patients with complex lesions, MRI was shown to be more informative for de®ning complex congenital heart disease than conventional angiography.64 Receiver operating characteristic curves have been derived for MRI at a speci®city level of 90%, demonstrating 100% sensitivity for evaluating great vessel relationships, ventricular septal defects, visceroatrial situs, and ventricular loop, 95% for right ventricular out¯ow tract obstructions, 94% for thoracic aortic anomalies, and 91% for atrial septal defects.68 5.5 Postoperative Evaluation The metallic clips, wires, stents, shunts, and prosthetic valves used in interventional and surgical correction of congenital heart disease are sometimes problematic for MRI. On magnetic resonance images these devices may result in paramagnetic artifacts that degrade image quality, but fortunately, in most instances, they do not affect signi®cantly the diagnostic utility of the study. The characteristics of noninvasiveness, lack of ionizing radiation, and good interstudy reproducibility are all distinct advantages for an imaging study intended to be used serially in children. Congenital obstruction at the level of the main pulmonary artery, right ventricular out¯ow tract, or tricuspid valve requires a corrective procedure that involves construction of an atriopulmonary connection (Fontan procedure) or ventriculopulmonary conduit (Rastelli procedure) using native tissue and/or homograft material. These procedures can be complicated by insidious stenosis of the conduit, or the anastomotic site(s), and will bene®t from periodic imaging examinations in addition to clinical follow up. The Fontan procedure has been evaluated

with the spin echo technique alone.69 Addition of velocityencoded cine MRI can provide further information regarding quanti®cation of blood ¯ow through the connection into the pulmonary arteries. This technique has demonstrated the normal biphasic pattern of blood ¯ow, similar to venous ¯ow, seen in atriopulmonary Fontans, or the normal monophasic pattern, seen in atrioventricular Fontans, thus con®rming the success of the operation.70 Rastelli conduits may also be complicated by stenosis, which is often dif®cult to evaluate with echocardiography because of the anterior location of the conduit in the chest. The velocity-encoded cine MRI technique has been applied in these patients to measure the gradient and to localize accurately the site of obstruction, providing a measure of the severity of the lesion which was superior to echocardiography.71 The Norwood72 and Jatene50,73 procedures have been evaluated similarly, and MRI has been shown to be equivalent to angiography, and superior to echocardiography, respectively. Extracardiac conduits such as the Glenn (superior vena cava to the right pulmonary artery) and Blalock±Taussig (subclavian artery to the ipsilateral pulmonary artery) shunts also can be evaluated easily with spin echo techniques, often in a single imaging plane. Serial imaging studies are valuable to evaluate for size, course, and patency of these shunts, as well as growth of the pulmonary arteries.63,74

6

THE FUTURE

The technology relevant to MRI of the heart continues to evolve. Consequently, the application of MRI in cardiovascular disease can be expected to increase further in the next several years.

7

RELATED ARTICLES

Cardiac Gating Practice; Cardiovascular NMR to Study Function; Echo-Planar Imaging; Lung and Mediastinum MRI;

10 HEART: CLINICAL APPLICATIONS OF MRI Motion Artifacts: Mechanism and Control; NMR Spectroscopy of the Human Heart. 8 REFERENCES 1. M. Erlichman, Health Technol. Assess. Report, 1990, 4, 1. 2. R. D. White, R. L. Ehman, and J. C. Weinreb, JMRI, 1992, 2, 365. 3. T. Masui, G. R. Caputo, J. C. Bowersox, and C. B. Higgins, J. Comput. Assist. Tomogr., 1995. 4. L. M. Blake, M. M. Scheinman, and C. B. Higgins, Am. J. Roentgenol., 1994, 162, 809. 5. W. Auffermann, T. Wichter, G. Breithardt, K. Joachimsen, and P. E. Peters, Am. J. Roentgenol., 1993, 161, 549. 6. E. K. Yucel, F. L. Steinberg, T. K. Egglin, S. C. Geller, A. C. Waltman, and C. A. Athanasoulis, Radiology, 1990, 177, 779. 7. C. A. Nienaber, Y. von Kodolitsch, V. Nicolas, V. Siglow, A. Piepho, C. Brockhoff, D. H. Koschyk, and R. P. Spielmann, N. Engl. J. Med., 1993, 328, 1. 8. C. A. Nienaber, R. P. Spielmann, Y. von Kodolitsch, V. Siglow, A. Piepho, T. Jaup, V. Nicolas, P. Weber, H. J. Triebel, and W. Bleifeld, Circulation, 1992, 85, 434. 9. N. Fujita, G. R. Caputo, and C. B. Higgins, Am. J. Card. Imaging, 1994, 23, 959. 10. M. Jungehulsing, U. Sechtem, P. Theissen, H. H. Hilger, and H. Schicha, Radiology, 1992, 182, 225. 11. K. C. Seelos, G. R. Caputo, C. L. Carrol, H. Hricak, and C. B. Higgins, J. Comput. Assist. Tomogr., 1992, 16, 169. 12. J. H. Park, Y. M. Kim, J. W. Chung, Y. B. Park, J. K. Han, and M. C. Han, Radiology, 1992, 185, 441. 13. J. D. Allison, F. W. Flickinger, J. C. Wright, D. G. Falls, L. M. Prisant, T. W. VonDohlen, and M. J. Frank, Magn. Reson. Imag., 1993, 11, 329. 14. R. C. Semelka, E. Tomei, S. Wagner, J. Mayo, G. Caputo, M. O'Sullivan, W. W. Parmley, K. Chatterjee, C. Wolfe, and C. B. Higgins, Am. Heart J., 1990, 119, 1367. 15. J. Suzuki, J. M. Chang, G. R. Caputo, and C. B. Higgins, J. Am. Coll. Cardiol., 1991, 18, 120. 16. J. Suzuki, G. R. Caputo, T. Masui, J. M. Chang, M. O'Sullivan, and C. B. Higgins, Am. Heart J., 1991, 122, 1035. 17. N. E. Doherty, K. C. Seelos, J. Suzuki, G. R. Caputo, M. O'Sullivan, S. M. Sobol, P. Cavero, K. Chatterjee, W. W. Parmley, and C. B. Higgins, J. Am. Coll. Cardiol., 1992, 19, 1294. 18. N. Fujita, J. Hartiala, M. O'Sullivan, D. Steiman, K. Chatterjee, W. W. Parmley, and C. B. Higgins, Am. Heart J., 1993, 125, 171. 19. T. Masui, S. Finck, and C. B. Higgins, Radiology, 1992, 182, 369. 20. P. J. Kilner, D. N. Firmin, R. S. Rees, J. Martinez, D. J. Pennell, R. H. Mohiaddin, S. R. Underwood, and D. B. Longmore, Radiology, 1991, 178, 229. 21. A. C. Eichenberger, R. Jenni, and G. K. von Schulthess, Am. J. Roentgenol., 1993, 160, 971. 22. P. A. Heidenreich, J. C. Steffens, N. Fujita, M. O'Sullivan, G. R. Caputo, E. Foster, and C. B. Higgins, Am. J. Cardiol., 1994, 75, 365. 23. N. Honda, K. Machida, M. Hashimoto, T. Mamiya, T. Takahashi, T. Kamano, A. Kashimada, Y. Inoue, S. Tanaka, and N. Yoshimoto, Radiology, 1993, 186, 189. 24. M. C. Dulce, G. H. Mostbeck, M. O'Sullivan, M. Cheitlin, G. R. Caputo, and C. B. Higgins, Radiology, 1992, 185, 235. 25. N. Fujita, A. F. Chazouilleres, J. J. Hartiala, M. O'Sullivan, P. Heidenreich, J. D. Kaplan, H. Sakuma, E. Foster, G. R. Caputo, and C. B. Higgins, J. Am. Coll. Cardiol., 1994, 23, 951. 26. M. T. McNamara, C. B. Higgins, N. Schechtmann, E. Botvinick, M. J. Lipton, K. Chatterjee, and E. G. Amparo, Circulation, 1985, 71, 717.

27. A. de Roos, A. C. van Rossum, E. van der Wall, S. Postema, J. Doornbos, N. Matheijssen, P. R. van Dijkman, F. C. Visser, and A. E. van Voorthuisen, Radiology, 1989, 172, 717. 28. M. C. Dulce, A. J. Duerinckx, J. Hartiala, G. R. Caputo, M. O'Sullivan, M. D. Cheitlin, and C. B. Higgins, Am. J. Roentgenol., 1993, 160, 963. 29. W. J. Manning, D. J. Atkinson, W. Grossman, S. Paulin, and R. R. Edelman, J. Am. Coll. Cardiol., 1991, 18, 959. 30. S. Schaefer, R. van Tyen, and D. Saloner, Radiology, 1992, 185, 795. 31. M. A. Klein, B. D. Collier, R. S. Hellman, and V. S. Bamrah, Am. J. Roentgenol., 1993, 161, 257. 32. F. P. van Rugge, J. J. Boreel, E. E. van der Wall, P. R. van Dijkman, A. van der Laarse, J. Doornbos, A. de Roos, J. A. den Boer, A. V. Bruschke, and A. E. van Voorthuisen, J. Comput. Assist. Tomogr., 1991, 15, 959. 33. D. J. Pennell, S. R. Underwood, P. J. Ell, R. H. Swanton, J. M. Walker, and D. B. Longmore, Br. Heart J., 1990, 64, 362. 34. F. M. Baer, K. Smolarz, M. Jungehulsing, P. Theissen, U. Sechtem, H. Schicha, and H. H. Hilger, Am. J. Cardiol., 1992, 69, 51. 35. F. M. Baer, K. Smolarz, P. Theissen, E. Voth, H. Schicha, and U. Sechtem, Int. J. Card. Imaging, 1993, 9, 133. 36. D. J. Pennell, S. R. Underwood, C. C. Manzara, R. H. Swanton, J. M. Walker, P. J. Ell, and D. B. Longmore, Am. J. Cardiol., 1992, 70, 34. 37. F. P. van Rugge, E. E. van der Wall, A. de Roos, and A. V. Bruschke, J. Am. Coll. Cardiol., 1993, 22, 431. 38. H. Sakuma, G. R. Caputo, J. C. Steffens, A. Shimakawa, T. K. F. Foo, and C. B. Higgins, Radiology, 1993, 189(P), 278 (abstract). 39. R. R. Edelman, W. J. Manning, D. Burstein, and S. Paulin, Radiology, 1991, 181, 641. 40. R. R. Edelman, W. J. Manning, J. Pearlman, and W. Li, Radiology, 1993, 187, 719. 41. W. J. Manning, W. Li, N. G. Boyle, and R. R. Edelman, Circulation, 1993, 87, 94. 42. W. J. Manning, W. Li, and R. R. Edelman, N. Engl. J. Med., 1993, 328, 828. 43. D. Li, C. B. Paschal, E. M. Haacke, and L. P. Adler, Radiology, 1993, 187, 401. 44. S. Rees, J. Somerville, C. Ward, J. Martinez, R. H. Mohiaddin, R. Underwood, and D. B. Longmore, Radiology, 1989, 173, 499. 45. R. H. Mohiaddin, P. J. Kilner, S. Rees, and D. B. Longmore, J. Am. Coll. Cardiol., 1993, 22, 1515. 46. J. C. Steffens, M. W. Bourne, H. Sakuma, M. O'Sullivan, G. R. Caputo, and C. B. Higgins, Radiology, 1993, 189(P), 302 (abstract). 47. A. S. Gomes, Radiol. Clin. North Am., 1989, 27, 1171. 48. P. R. Julsrud, R. L. Ehman, D. J. Hagler, and D. M. Ilstrup, Radiology, 1989, 173, 503. 49. A. S. Gomes, J. F. Lois, and R. G. Williams, Radiology, 1990, 174, 51. 50. F. Blankenberg, J. Rhee, C. Hardy, G. Helton, S. S. Higgins, and C. B. Higgins, J. Comput. Assist. Tomogr., 1994, 18, 749. 51. D. A. Lynch and C. B. Higgins, J. Comput. Assist. Tomogr., 1990, 14, 187. 52. J. M. Parsons, E. J. Baker, A. Hayes, E. J. Ladusans, S. A. Qureshi, R. H. Anderson, M. N. Maisey, and M. Tynan, Int. J. Cardiol., 1990, 28, 73. 53. G. R. Caputo, C. Kondo, T. Masui, S. J. Geraci, E. Foster, M. M. O'Sullivan, and C. B. Higgins, Radiology, 1991, 180, 693. 54. J. M. Silverman, P. J. Julien, R. J. Herfkens, and N. J. Pelc, Radiology, 1993, 189, 699. 55. K. S. Azarow, R. H. Pearl, M. A. Hoffman, R. Zurcher, F. H. Edwards, and A. J. Cohen, Ann. Thorac. Surg., 1992, 53, 882. 56. R. B. Jaffe, Semin. Ultrasound. CT MR, 1990, 11, 206.

HEART: CLINICAL APPLICATIONS OF MRI 57. E. J. Baker, V. Ayton, M. A. Smith, J. M. Parsons, E. J. Ladusans, R. H. Anderson, M. N. Maisey, M. Tynan, N. L. Fagg, and P. B. Deverall, Br. Heart J., 1989, 62, 305. 58. G. R. Caputo, J. Suzuki, C. Kondo, H. Cho, R. A. Quaife, C. B. Higgins, and D. L. Parker, Radiology, 1990, 177, 773. 59. N. E. Doherty, N. Fujita, G. R. Caputo, and C. B. Higgins, Am. J. Cardiol., 1992, 69, 1223. 60. J. R. Mayo, D. Roberson, B. Sommerhoff, and C. B. Higgins, J. Comput. Assist. Tomogr., 1990, 14, 336. 61. J. M. Parsons, E. J. Baker, R. H. Anderson, E. J. Ladusans, A. Hayes, N. Fagg, A. Cook, S. A. Qureshi, P. B. Deverall, and M. N. Maisey, J. Am. Coll. Cardiol., 1991, 18, 168. 62. S. A. Rebergen, G. L. Guit, and A. de Roos, Br. Heart J., 1991, 66, 381. 63. I. C. Huggon, E. J. Baker, M. N. Maisey, A. P. Kakadekar, P. Graves, S. A. Qureshi, and M. Tynan, Br. Heart J., 1992, 68, 313. 64. B. A. Kersting-Sommerhoff, L. Diethelm, P. Stanger, R. Dery, S. M. Higashino, S. S. Higgins, and C. B. Higgins, Am. Heart J., 1990, 120, 133. 65. T. M. Vesely, P. R. Julsrud, J. J. Brown, and D. J. Hagler, J. Comput. Assist. Tomogr., 1991, 15, 752. 66. B. Kastler, A. Livolsi, P. Germain, A. Gangi, A. Klinkert, J. L. Dietemann, D. Willard, and A. Wackenheim, Pediatr. Radiol., 1992, 22, 262. 67. B. Kastler, A. Livolsi, H. Zhu, E. Roy, G. Zollner, and J. L. Dietemann, J. Comput. Assist. Tomogr., 1990, 14, 825. 68. B. A. Kersting-Sommerhoff, L. Diethelm, D. F. Teitel, C. P. Sommerhoff, S. S. Higgins, S. S. Higashino, and C. B. Higgins, Am. Heart J., 1989, 118, 155. 69. B. A. Kersting-Sommerhoff, K. C. Seelos, C. Hardy, C. Kondo, S. S. Higgins, and C. B. Higgins, Am. J. Roentgenol., 1990, 155, 259. 70. S. A. Rebergen, J. Ottenkamp, J. Doornbos, E. E. van der Wall, J. G. Chin, and A. de Roos, J. Am. Coll. Cardiol., 1993, 21, 123.

11

71. J. E. Martinez, R. H. Mohiaddin, P. J. Kilner, K. Khaw, S. Rees, J. Somerville, and D. B. Longmore, J. Am. Coll. Cardiol., 1992, 20, 338. 72. C. Kondo, C. Hardy, S. S. Higgins, J. N. Young, and C. B. Higgins, J. Am. Coll. Cardiol., 1991, 18, 817. 73. C. E. Hardy, G. J. Helton, C. Kondo, S. S. Higgins, N. J. Young, and C. B. Higgins, Am. Heart J., 1994, 128, 326. 74. B. Kastler, A. Livolsi, P. Germain, G. Zollner, and J. L. Dietemann, Int. J. Card. Imaging, 1991, 7, 1.

Biographical Sketches Scott D. Flamm. b 1960. B.A., 1982, University of California, Berkeley. M.D., 1988, George Washington University, USA. Residency training in diagnostic radiology, University of California, Los Angeles. Fellowship training in cardiovascular imaging and MRI (supervisor Charles B. Higgins), University of California, San Francisco. Currently clinical instructor in cardiovascular imaging/MRI, University of California, San Francisco, Approx. 10 publications. Research specialties: MRI of acquired and congenital cardiac disease, MRA, and quantitative vascular ¯ow measurements. Charles B. Higgins. b 1944. B.S., 1963, Villanova University. M.D., 1967, Jefferson Medical College, USA. Residency training in surgery, University of California, Los Angeles, and in diagnostic radiology, University of California, San Diego. Fellowship training in cardiovascular radiology, Stanford University. Research training at the Cardiovascular Research Laboratory (supervisor Eugene Braunwald), University of California, San Diego. Currently Professor of Radiology, and Chief, Magnetic Resonance Imaging and Cardiac Imaging Sections, University of California, San Francisco. Approx. 450 publications. Research specialties: MRI of acquired and congenital cardiac disease, and strategies for detection of myocardial ischemia using MR contrast media.

LUNG AND MEDIASTINUM MRI

Lung and Mediastinum MRI Robert J. Herfkens Stanford University School of Medicine, CA, USA

The utilization of MRI techniques for visualization of lung and mediastinal abnormalities has been greeted with great enthusiasm due to the excellent contrast resolution of MRI. However, the complex motions associated with those respiratory and cardiac events have posed signi®cant challenges to the utilization of MRI for visualizing pulmonary and mediastinal abnormalities. The utilization of motion compensation techniques have greatly improved the image quality of MRI, and its intrinsic qualities have indeed provided unique diagnostic information about thoracic disease. In order to compensate for the complex motions associated with the heart motion and blood ¯ow, methods for coordinating the acquisition of the images in concert with the cardiac cycle have been developed.1 Gating or triggering of the acquisition to the cardiac cycle has been obtained through a number of methods; primarily, electrocardiographic (ECG) gating methods have been employed.2,3 In addition, methods of plethysmographic or pulse oximeter methods have been utilized, but these are somewhat less precise in coordinating the acquisition with the cardiac cycle. The ECG signal triggers the magnetic resonance (MR) scanner as providing a consistent point in the cardiac cycle for initiating multislice or cine imaging. This, however, provides a limitation in imaging of the chest, as the R±R interval determines the repetition time of the scan. This provides a signi®cant limitation to the ¯exibility of presenting imaging contrast. For an average heart rate of 60 beats per minute, the minimum repetition time would be 1000 ms. Depending upon the patient's physiologic status, the heart rate may vary quite greatly, thus having a signi®cant effect on the overall image contrast, due to varying TR times. Respiratory motion, which is also periodic, causes signi®cant artifacts in MRI. These have a strong impact on chest images. The simplest solution to respiratory artifacts originally was to provide multiple signal averages, which blur the respiratory motion and, in general, improve the signal-to-noise ratio. These increases in imaging time, associated with respiratory averaging, are somewhat inappropriate on modern systems. Techniques that coordinate the phase encoding of the image acquisition with the respiratory cycle, so-called `respiratory ordered phase encoding' or `respiratory compensation', can signi®cantly reduce the amount of respiratory artifacts without unduly increasing the imaging time. Recent fast imaging techniques with fast repetition times as short as 10 ms have allowed breath-held images of the chest. Although these images may introduce some blurring of the cardiac silhouette, improved de®nition of mediastinal and diaphragmatic structures have been realized. Basic spin echo techniques gated to the cardiac cycle will provide relatively unique contrast within the mediastinum.

1

The basic contrast associated with relative T1-weighted imaging shows fat as a relatively high signal intensity, with normal mediastinal structures, such as the esophagus and heart, with intermediate signal intensity.4 The intrinsic ¯ow sensitivity spin echo images show a relative decrease in signal intensity of the vascular structures. In gated spin echo imaging, however, the signal intensity of major vessels may vary, depending upon when the speci®c image is obtained in the cardiac cycle. Those images obtained during systole have a relatively strong signal void, and those obtained during diastole may have signi®cant intravascular signal, because of the relatively slow ¯ow. Utilization of ¯ow sensitive sequences, such as gradient-recalled imaging techniques, have improved the ability to differentiate intravascular slow ¯ow signal from soft-tissue signals. Gradient-recalled images show an intrinsic sensitivity to ¯owing blood due to ¯ow related enhancement, thus demonstrating an increase in signal intensity associated with ¯ow. By utilizing gating with gradient-recalled sequences, phase-encoding signals can be advanced with each cardiac cycle, thus providing a cine image throughout the cardiac cycle, generating sensitivity to ¯ow. In addition, phase contrast techniques have been applied to the cine technique, allowing the demonstration of velocity changes throughout the cardiac cycle (Figure 1). These phase contrast techniques have been used to characterize ¯ow in major vascular structures and may be important in characterizing pulmonary artery ¯ow abnormalities5 (Figure 2). When using spin echo sequences, the repetition time is determined by the R±R interval. In order to obtain T2weighted images in the chest, techniques which would allow for the gating of the acquisition to multiple heart beats, which is every third or every fourth heart beat, allow for the acquisition of T2-weighted data. In general, the utilization of T1-weighted data has been superior for anatomic de®nition. The intrinsic contrast, as previously noted, where fat is bright and muscle and esophagus are dark, leave any other intermediate signal intensities as being pathologic. The relatively high contrast between ¯owing blood in spin echo images and other soft tissues also allows for excellent contrast to characterize other pathologic entities related to vascular involvement or improved de®nition of the spatial relationship associated with the relative signal void of vascular structures (Figure 3). The ability of MR to generate any plane has been a signi®cant advantage over conventional techniques, such as computerized tomography (CT). The use of coronal planes has signi®cantly enhanced the visualization of abnormalities and the apices of the lungs, and has improved the visualization of lesions associated with the diaphragm. The ability to utilize oblique planes has been extremely helpful in characterizing pathologies associated with the aorta or other structures.6 Another major element which degrades images within the thorax is the susceptibility associated with the multiple air± water interfaces in the pulmonary parenchyma. This marked change in susceptibility has resulted in a very poor delineation of other abnormalities within the lungs themselves. A number of recent techniques have provided signi®cant improvement in the visualization of pulmonary abnormalities; however, they still appear to be signi®cantly less sensitive than CT for detecting pulmonary pathologies.7±9 The development of fast spin

2 LUNG AND MEDIASTINUM MRI

Figure 1 (a) A gated T1-weighted spin echo image showing prominent soft tissue density along the lateral wall of the heart. This mass in a 32year-old female was identi®ed on chest X-ray. (b) Eight frames of a cine gradient-recalled imaging sequence from diastole through systole. The mass noted in (a) is shown with signal variations suggesting ¯ow. (c) A cine phase contrast imaging sequence encoded in the superior to inferior direction corresponding to the image shown in (b). Note the signi®cant change in signal intensity during systolic contraction in both the superior and inferior directions in the region of the mass, suggesting that this mass is a pulmonary arterial±venous malformation. (d) A pulmonary angiogram con®rming the presence of a pulmonary arterial±venous malformation

LUNG AND MEDIASTINUM MRI

3

Figure 2 Flow analysis of the right pulmonary artery ¯ow pattern in a patient after lung transplant. This is the native lung and shows a pulsatile ¯ow pro®le going from positive to negative throughout the cardiac cycle, showing a low net total ¯ow to the lung. Such pulmonary ¯ow analysis may be helpful in determining vascular compliance of the lung, as well as in relative ratios in blood ¯ow in postoperative patients

echo imaging techniques has recently been shown to increase the sensitivity to pulmonary parenchyma abnormalities, such as metastasis. However, CT is signi®cantly superior. Projection± reconstruction techniques can utilize extremely short echo times (as short as 100 s), and thus allow imaging of the pulmonary parenchyma or lung water associated with the pulmonary parenchyma10 (Figure 4). These techniques remain somewhat experimental but will allow for the demonstration of pulmonary parenchymal signals, which are less sensitive to the susceptibility generated within the lung. These signals represent excellent maps of lung water and show exceptional promise in improving MR's ability for characterizing subtle intrapulmonary abnormalities. The utilization of paramagnetic contrast agents has also advanced the abilities of thoracic MRI to characterize a number of lesions.11 The T1 shortening effects of paramagnetic materials have provided the ability to enhance the differentiation of atelectatic from existing tumors as well as characterizing nonperfused cystic structures. MRI has been valuable in characterizing a number of signal intensities related to tissues. The speci®c signal characteristics of hemorrhage have allowed the identi®cation of short T1 changes associated with methemoglobin in order to characterize hemorrhagic processes from pulmonary parenchymal diseases. Pulmonary hemorrhage appears as a relatively short T1 of bright signal intensity on T1-weighted images. Hemorrhage into the mediastinum or associated with aortic dissections can be characterized due to the signal intensity changes associated with hemoglobin as it changes from oxyhemoglobin to deoxyhemoglobin to methemoglobin.

Figure 3 (a) A coronal-gated spin echo image showing the intimate relationship of a bronchiogenic carcinoma with the right upper lobe bronchus. (b) A gated spin echo image posterior to (a), showing a larger portion of the mass and possible association with the chest wall

The mediastinum contains relatively simple contrast relationships for MRI. The esophagus and heart remain intermediate in signal intensity, contrasted to the relatively high signal intensity on T1-weighted images of thoracic fat. The normal thymus appears as a signal of intermediate intensity; it generally decreases in size with age, and increases in signal intensity with fatty replacement. Of note is the fact that the thymus may rebound in adulthood due to some physiologic stress, thus providing an intermediate anterior mediastinal sig-

4 LUNG AND MEDIASTINUM MRI

Figure 4 Comparison of images obtained with (a) conventional T1weighted gated (1 R±R) multislice spinwarp (TR & 750 ms, TE = 15 ms, 256  2 NEX, scan time 7.6 min, using respiratory compensation and spatial presaturation) and (b) multislice Hadamard projection reconstruction (TR = 200 ms, TE = 0.25 ms, 512 views  8 NEX, scan time 4 min). The images have been windowed to show the lung parenchyma, which causes the ¯ow artifacts in (a), to appear abnormally intense. Note that the projection±reconstruction acquisition has intrinsically lower respiratory and ¯ow artifacts, even without the compensation methods employed in (a), and despite the reduced scan time. The lack of respiratory motion artifacts with the projection± reconstruction sequence is critical to the successful visualization of lung parenchyma

nal intensity. This, in general, can be differentiated from most pathologic processes on the basis of morphology. Although MRI provides excellent tissue characterization, the ability of MRI in differentiating between benign and malignant changes within the thymus itself is somewhat limited. CT and MRI both provide an excellent technique for evaluating anatomic changes associated with thymic disorders.12±14 MRI provides an excellent method for identifying and characterizing abnormalities in the pulmonary hyla. The sensitivity to ¯ow associated with spin echo imaging and the increase in signal intensity from gradient-recalled images provide an excellent format for identifying soft tissue abnormalities in the hyla. MRI appears to be signi®cantly more sensitive than CT for the detection of these abnormalities; however, the characterization of these may be somewhat dif®cult.15±17 The pathologic nature of hilar and mediastinal lymph nodes remains largely based on the size of the individual lesions. MRI provides an excellent format for visualizing lymphadenopathy. The ability to image in the coronal plane has increased the sensitivity for the detection of aortapulmonary window adenopathy. The major advantage of MRI over CT is that the intrinsic contrast relationships allow identi®cation of the lymph nodes and their separation from major vascular structures without the need to administer an iodinated contrast agent. Characterization of pathologies by T1- and T2-weighted images, however, has been somewhat disappointing. Both in¯ammatory and neoplastic adenopathy appear to have prolonged T2 times. The primary mechanism for distinguishing between benign and malignant adenopathy remains that of lymph node morphology and size.18,19 Staging of bronchiogenic carcinoma with MRI has provided an excellent means of detecting mediastinal lymph nodes on an anatomic basis. Characterization of lymphomas has also bene®ted signi®cantly from the detection of mediastinal lymphadenopathy, due to the excellent contrast characteristics of MRI. The relative T2 prolongation associated with lymphomas has been identi®ed, but its use remains limited due to the lack of ability to characterize changes other than

those based on size (Figure 5). The evaluation of patients with recurrent lymphoma has been suggested to be improved by the ability of MRI to characterize increased water content of nodes. In most treated lymphomas, there is a generalized decrease in size and signal intensity related to therapy. The recurrence of lymphoma can be characterized by either an increase in size or an increase in signal intensity. The increase in signal intensity in T2-weighted images has been suggested to be grounds for suspecting recurrence, and individual patients with changes in signal intensity or size should be assessed for recurrent disease.20 The evaluation of bronchiogenic carcinoma is important in determining therapy and prognosis. MRI and CT both provide excellent anatomic images of the primary tumor as well as the ability to characterize the presence of lymphadenopathy. A speci®c problem for which MRI may provide improved speci®city is in the associated changes in atelectasis associated with proximal bronchiogenic lesions.21,22 MRI has been suggested to be an excellent method for differentiation of obstructed atelectasis from the primary mass. The utilization of paramagnetic contrast material appears to provide signi®cantly improved information over CT, and aids in differentiating mass from atelectasis23,24 (Figure 6). Mediastinal cysts can often be confused with neoplastic lesions. The speci®c characteristics of very long T2 and very long T1 associated with mediastinal cysts enable these lesions to be characterized by MRI. The relative homogeneity associated with these benign congenital cysts and their anatomic location provides speci®c clues as to their etiology. Mediastinal cysts can be divided into duplication cysts, neurenteric cysts, and pleuroparenchymal cysts. In general, they are speci®cally characterized by their long T1 and long T2 and relatively sharp margins. The additional use of an intravenous paramagnetic contrast material will characterize these lesions as being avascular. Esophageal carcinoma represents a continuing diagnostic dilemma. The presence of the primary mass can easily be identi®ed by the relative increase in soft tissue density associated with the esophagus. MRI provides an excellent method for the staging of esophageal cancer and characterizing the association with vascular structures. The obliteration of fat planes about the aorta and other adjacent vascular structures, such as the left atrium, can be extremely helpful in planning therapy. These changes, however, do not appear to have signi®cantly impacted on the ultimate outcome of esophageal cancer (Figure 7). Approximately 10% of parathyroid glands are ectopic. Two thirds of these appear to arise in the anterior mediastinum. In the presence of hyperparathyroidism unresponsive to resection of the cervical components of the thyroid glands, the mediastinum becomes the primary source. The behavior of hyperplastic parathyroid glands on MRI is characterized by a markedly prolonged T2 relaxation time. This makes the identi®cation of T1and T2-weighted images extremely important. It should be noted that the identi®cation of a markedly prolonged T2 signal in the mediastinum should be separated from relatively slow ¯ow in thoracic veins by the utilization of some form of gradient-recalled imaging sequence. The excellent contrast characteristics of MRI provide an excellent means of staging of thoracic neoplasms. The sensitivity to direct invasion of vascular structures by obliteration of

LUNG AND MEDIASTINUM MRI

5

Figure 5 (a) A gated T1-weighted spin echo image through the upper mediastinum. (b) A spin density gated image through the upper mediastinum. (c) A T2-weighted gated image through the upper mediastinum. (d) A fat-suppressed, fast spin echo image through the upper mediastinum. Note the bulky adenopathy in the anterior mediastinum and the right paratracheal space. In addition, note the large lymph node in the right axilla. There are bilateral pleural effusions present, which give signi®cantly variable signal intensities, as do all these images, to the relatively long T1 and T2 of the effusion, which is affected strongly by ¯ow secondary to cardiac and respiratory motion. Note the difference in the degree of conspicuousness of the axillary node in the fat-suppressed image (d) and the signi®cant difference in the signal characteristics of the mediastinal adenopathy. The fast spin echo image associated with both fat suppression and magnetization transfer effects will have signi®cantly different contrast in this mediastinal adenopathy. The ability to characterize adenopathy, due to signal changes, remains a complex issue

associated short T1 fat planes provides an excellent method (Figure 8). The ability of MRI to characterize changes along the thoracic wall provides a unique opportunity to demonstrate thoracic wall invasion. The short T1 associated with subpleural fat and the relatively long T1 of thoracic neoplasms give MRI great sensitivity in detecting thoracic wall invasion (Figure 9). These changes in thoracic wall invasion observed in both lymphoma and bronchiogenic carcinomas may signi®cantly change therapy.25 Either identi®cation of the lesion as being nonresectable or the provision of information about thoracic wall invasions may signi®cantly alter the radiation therapy planning. Similarly, the ability to provide coronal images of Pancoast's tumors may provide important information about resectability and therapy.26,27 Pleural effusions are easily identi®ed in MRI scans. The signal intensities were originally thought to provide important

information about transudative versus exudative behavior of pleural effusions. However, the strong effect of ¯ow within the pleural effusions provides the dominant mechanism for signal changes in spin echo imaging. Free ¯owing pleural effusion may even appear as a signal void on spin echo images, and can easily be identi®ed as areas of increased signal intensity and gradient-recalled images due to their ¯ow related enhancement.28 Pulmonary artery abnormalities, such as the presence of pulmonary emboli, have been identi®ed as areas of increased signal intensity on spin echo images. MRI represents an excellent method for characterizing the presence of central pulmonary emboli. In general, chronic pulmonary emboli have intermediate signal intensity on both T1- and T2-weighted images. However, the presence of acute pulmonary emboli may be characterized by the presence of an extremely high

6 LUNG AND MEDIASTINUM MRI intensity signal on T2- and T1-weighted images due to the presence of methemoglobin in the thrombi (Figure 10). The ¯ow sensitivity of MRI has been utilized to characterize pulmonary arteriovenous malformations. The relatively intermediate signal intensity seen on T1-weighted images may be present due to relatively slow ¯ow. The sensitivity of gradient-recalled images or phase contrast methods has allowed the characterization of these pulmonary parenchyma abnormalities by demonstrating the ¯ow within them (Figure 1). In summary, the excellent contrast characteristics of MRI, associated with its sensitivity to ¯ow, make it an excellent method for identifying and characterizing pulmonary and med-

Figure 6 (a)±(d)

iastinal pathologies. The relatively limited spatial resolution within the pulmonary parenchyma is a signi®cant limitation; however, the superb contrast sensitivities have ensured the place of MRI in the identi®cation and characterization of thoracic pathologies.

1

RELATED ARTICLES

Breast MRI; Heart: Clinical Applications of MRI; Whole Body Magnetic Resonance Angiography.

LUNG AND MEDIASTINUM MRI

7

Figure 6 (a) A CT scan obtained from a patient presenting with a right-sided mass and large pleural effusion. (b) A gated coronal image through the same patient as in (a) following a thoracentesis. This image still shows a rather prominent mass, with occlusion of the right upper lobe bronchus. The question of primary mass versus atelectasis is often raised. (c) A gated transverse spin echo image through the mass. Note the relatively high signal intensity of this `T1-weighted image' in the periphery of this mass, the ®nding which is described in atelectatic lung. (d) A fatsaturated fast spin echo image showing marked heterogeneity in the mass but relatively clear delineation between the peripheral components and more central low intensity mass. (e) A T1-weighted fat-saturated gated fast spin echo image through the same region as in (d) showing marked enhancement of the mass, as well as the adjacent pleura on this side

Figure 7 A gated transverse spin echo image through the midthorax shows a subcarinal mass associated with relatively subtle esophageal carcinoma. There is a fat plane between the aorta and mediastinal soft tissue, suggesting no de®nite invasion. Of more importance are the soft tissue densities in each lung ®eld. These relatively hazy, ill-de®ned nodules are characteristic of metastatic disease on MRI. CT remains the primary method of screening for pulmonary metastases

Figure 8 A gated spin echo image through the upper mediastinum shows prominent anterior mediastinal adenopathy as well as a prominent para-aortic node. On the right-hand side there is pulmonary parenchymal consolidation and ®brotic changes secondary to prior radiation therapy. Shoddy mediastinal adenopathy is present. Of note is the absence of a signi®cant signal void in the superior vena cava, due to vena caval occlusion

8 LUNG AND MEDIASTINUM MRI

Figure 9 (a) A gated spin echo image through a bronchiogenic carcinoma of the right upper lobe. (b) A gated spin-density-weighted image through the carcinoma. (c) A STIR image obtained through the upper mediastinum. Note the excellent delineation on all images of the primary mass in relation to the pulmonary parenchyma. Paratracheal adenopathy is clearly visible on the spin echo images, but is somewhat obscured on the STIR images. The STIR image does, however, provide much better detail of the mass and its intimate association with the chest wall, suggesting chest wall invasion. In addition, the signi®cant motion artifacts present on the STIR image are due to the relative dif®culty of gating inversion±recovery images. The STIR images themselves are useful in investigating chest wall invasion but, due to the relative temporal inef®ciency, may not be appropriate for all patients

LUNG AND MEDIASTINUM MRI

Figure 10 A gated T1-weighted transverse image through the aortic arch, showing a very high signal intensity area on the left upper lung corresponding to a subacute pulmonary embolus

2 REFERENCES 1. L. Axel, R. M. Summers, H. Y. Kressel, and C. Charles, Radiology, 1986, 160, 795. 2. L. Hedlund, R. Herfkens, J. Deitz, R. Blinder, R. Nassar, E. Coleman, E. Effman, and C. Putman, Proc. IVth Ann Mtg. Soc. Magn. Reson. Med., London, 1985, 1, 583. 3. C. B. Higgins, E. H. Botvinick, P. Lanzer, R. J. Herfkens, M. J. Lipton, L. E. Goorn, and L. Kaufman, Cardiol. Clin., 1983, 1, 527. 4. G. Gamus and D. Sostman, Am. Rev. Respir. Dis., 1989, 139, 254. 5. N. J. Pelc, R. J. Herfkens, A. Shimakawa, and D. R. Enzmann, Magn. Reson. Q., 1991, 7, 229. 6. P. Y. Poon, M. J. Bronskill, R. M. Henkelman, D. F. Redeout, M. S. Sholman, G. L. Weisbrod, M. I. Steinhardt, M. J. Dunlap, A. R. J. Ginsberg, and R. Feld, Radiology, 1987, 196, 651. 7. S. L. Primack, J. R. Mayo, T. E. Hartman, R. R. Miller, and N. L. MuÈller, J. Comput. Assist. Tomogr., 1994, 18, 233. 8. N. L. MuÈller, J. R. Mayo, and C. V. Zwirewich, Am. J. Roentgenol., 1992, 158, 1205. 9. E. H. Moore, W. R. Webb, N. L. MuÈller, and R. Sollitto, Am. J. Roentgenol., 1986, 146, 1123. 10. C. J. Bergin, J. M. Pauly, and A. Macovski, Radiology, 1991, 179, 777.

9

11. C. J. Herold, J. E. Kuhlman, and E. A. Zerhouni, Radiology, 1991, 178, 715. 12. G. M. Glazer, M. B. Orringer, T. L. Chenevert, J. A. Borrello, M. W. Penner, L. E. Quint, K. C. Li, and A. M. Arsers, Radiology, 1988, 168, 429. 13. G. M. Glazer, Chest, 1989, 96 (Suppl. 1), 44S. 14. W. R. Webb, M. Sarin, E. A. Zerhouni, R. T. Heelan, G. M. Glazer, and C. Gatsonis, J. Comput. Assist. Tomogr., 1993, 17, 841. 15. W. R. Webb, C. Gatsonis, E. A. Zerhouni, et al., Radiology, 1991, 178, 705. 16. W. B. Gefter, Semin. Roentgenol., 1990, 25, 73. 17. O. Musset, P. Grenier, M. F. Carette, G. Frija, M. P. Hauuy, H. T. Desbleds, P. Giraird, J. M. Bigot, and D. Lallemand, Radiology, 1986, 160, 607. 18. G. M. Glazer, B. H. Gross, A. M. Aisen, L. E. Quint, I. R. Francis, and M. B. Orringer, Am. J. Roentgenol., 1985, 145, 245. 19. H. S. Glazer, R. G. Levitt, J. K. T. Lee, B. Emami, S. Gronemeyer, and W. A. Murphy, Am. J. Roentgenol., 1984, 143, 729. 20. R. S. Nyman, S. M. Rehn, B. L. Glinelius, M. E. Hagberg, A. L. Memmingtson, and C. J. Sundstrom, Radiology, 1989, 170, 435. 21. J. A. Verschakelen, P. Demaerel, J. Coolen, M. Demeols, G. Marshal, and A. L. Baert, Am. J. Roentgenol., 1989, 152, 965. 22. C. J. Herold, J. E. Kuhlman, and E. A. Zerhouni, Radiology, 1991, 178, 715. 23. P. M. Bourgouin, T. C. McLoud, J. F. Fitzgibbon, E. J. Mark, J. A. Shepard, E. M. Moore, E. Rummeny, and T. J. Brady, J. Thorac. Imag., 1991, 6, 22. 24. J. Tobler, R. C. Levitt, H. S. Glazer, J. Moran, E. Crouch, and R. G. Evens, Invest. Radiol., 1987, 22, 538. 25. R. T. Heelan, B. E. Demas, J. F. Caravelli, N. Martini, M. S. Bains, P. M. McCormack, M. Burt, D. M. Panicek, and A. Mitzner, Radiology, 1989, 170, 637. 26. D. R. Pennes, G. M. Glazer, K. J. Wimbish, B. H. Gross, R. W. Long, and M. B. Orringer, Am. J. Roentgenol., 1985, 144, 507. 27. A. M. Haggar, J. L. Pearlberg, J. W. Froelich, D. O. Hearshen, G. H. Beure, J. W. Lewis, Jr., G. W. Schkuder, C. Wood, and P. Gniewck, Am. J. Roentgenol., 1987, 148, 1075. 28. P. Vock, L. W. Hedlund, R. J. Herfkens, E. L. Effmann, M. A. Brown, and C. E. Putman, Invest. Radiol., 1987, 22, 382.

Biographical Sketch Robert J. Herfkens. b 1949. M.D., 1974, Loyola University, Stritch School of Medicine. Associate Professor and Director of MRI, Stanford University School of Medicine, 1989. Principal investigator on research grants sponsored by Government and private industry. Approx. 100 scienti®c papers, presented at over 100 national meetings; several textbooks. Main research interest: the development of applications and new techniques directed towards body imaging, most speci®cally those related to cardiovascular diseases.

MARKER GRIDS FOR OBSERVING MOTION IN MRI

Marker Grids for Observing Motion in MRI Leon Axel University of Pennsylvania, Philadelphia, PA, USA

1 INTRODUCTION Early in the development of NMR, it was noticed that motion could affect the NMR signal.1 It was quickly understood that this re¯ected the fact that moving material carries its magnetization with it. If the magnetization is perturbed locally, that perturbed magnetization is carried with the material as it moves. This principle was applied in some early applications of NMR to the measurement of ¯uid ¯ow.2±4 With the development of MRI, selective excitation was applied to the region to be imaged. The resulting local perturbation of magnetization within the slice being imaged led to intensity changes in the images of ¯owing blood or cerebrospinal ¯uid, due to `time-of-¯ight' (TOF) effects.5,6 (See also articles on Flow in this Encyclopedia, listed in the Related Articles below.) These TOF effects due to selective slice excitation can be used for qualitative or quantitative evaluation of ¯ow through the plane of the image. We will review here some aspects of using regional perturbation of the magnetization in planes perpendicular to the image to create magnetic tags. Subsequent imaging allows us to display directly the motion (of ¯uid or tissue) in the plane of the image during the time between tag production and image data acquisition, as a corresponding displacement of the tags in the image. We will review some aspects of tag production, approaches to tagged image analysis, and initial applications. While the phase shifts acquired by excited spins moving along magnetic ®eld gradients can also affect their signal, we will not consider these effects here.

2 TAG PRODUCTION When perturbed, the altered magnetization (excitation or saturation) in a tag will persist for times of the order of the relaxation time. An excitation tag can be produced with initial excitation of a plane perpendicular to the imaging plane and subsequent conventional position encoding in the imaging plane. These excited spins will be seen as a bright stripe in the image, representing the intersection of the initially excited plane with the imaging plane. Any displacement of the spins in the plane of the image between the times of initial excitation and subsequent position encoding will be seen as a corresponding displacement of the tagged stripe. This approach has been used to demonstrate ¯uid ¯ow, with the excited spins detected either as a conventional spin echo7 or as a gradient echo.8 If conventional spin echo detection is used, the refocusing rf

1

pulse can be used to select the plane to be imaged. With gradient echo excitation tagging, a two-dimensionally selective `spike' excitation could be used to restrict the tagging plane. The motion sensitivity of tagging depends on the time between excitation and readout. This, in turn, is limited by transverse relaxation to be of the order of the T2 relaxation time, or T2* for gradient echo readout. An alternative way to create tags is with local production of saturation or inversion in a plane or planes perpendicular to the image plane. The altered longitudinal magnetization will result in stripes with a correspondingly altered image intensity in a subsequent image acquisition, representing the intersections of the tagged planes with the imaging plane. Any displacement in the image plane between the times of tagging and imaging will be seen as a corresponding displacement of the tagged stripe in the image. In this case, tag persistence is limited by longitudinal relaxation so that displacement can be tracked for times of the order of the T1 relaxation time (with initial inversion leading to longer tag persistence than initial saturation). The local perturbation of magnetization can be created selectively. Selective tagging is analogous to slice selection excitation, with an rf excitation pulse with tailored frequency content applied in the presence of a magnetic ®eld gradient, in order to excite only a desired plane perpendicular to the gradient. If only longitudinal magnetization tagging is desired, any remnant transverse magnetization from the tagging can be `spoiled' by applying a further magnetic ®eld gradient prior to excitation of the imaging plane. Multiple tagging planes can be created by applying the tagging process repeatedly at several locations or by tailoring the rf pulse to excite several locations simultaneously.9 The latter approach can allow more rapid production of tagging patterns, but may require more rf power. By following selective inversion tagging with a nonselective inversion pulse, we can create a tag pattern with bright stripes (where the magnetization has been driven back up toward equilibrium) against a darker background.10 Separate images of different tagging patterns can be combined to give the effect of having used a more complex tagging pattern.11 The tagging pattern can also be created nonselectively. In this case, the magnetic ®eld gradient is applied between nonselective rf excitation pulses.12,13 The resulting periodic phase variation along the direction of the gradient leads to a net periodic modulation of the magnetization along the direction of the gradient, or spatial modulation of magnetization (SPAMM). This is precisely analogous to the use of binomial pulses14 and related techniques in NMR spectroscopy for solvent suppression; in spectroscopy the corresponding periodicity of the suppression is usually chosen to be greater than the relevant spectral bandwidth. This spatially modulated magnetization, in turn, will be seen in a ®nal image as a periodic stripe pattern along the direction of the tagging gradient. The spacing of the stripes will be determined by the strength and duration of the gradient pulses; the orientation of the stripes will be determined by the orientation of the tagging gradient. The relative position of the tag pattern along the gradient direction can be shifted by shifting the relative phases of the rf pulses. By using multiple rf pulses with suitable relative intensities, we can control the intensity pro®le of the resulting tagging pattern.

2 MARKER GRIDS FOR OBSERVING MOTION IN MRI The result of one application of SPAMM is a set of parallel tagging stripes. Creating a second set at right angles to the ®rst can produce a tagging grid.13 Images with tagging patterns of opposite polarity can be combined to increase the effective amplitude of the tags.15 Tags can also be created with tagging planes that are not perpendicular to the imaging plane, although the stripes may be blurred due to partial volume effects with the oblique intersection of the tagging plane with the slice being imaged.16,17 The SPAMM principle can be used with a continuous gradient if the rf pulses have suf®ciently broad spectral content (are `hard' enough) to excite the region to be tagged in the presence of the gradient.18 The SPAMM principle can also be applied in combination with selective excitation tagging to create periodic amplitude modulation along the tag.19 Saturation or inversion tags can be used as a preconditioning sequence incorporated into any imaging pulse sequence, much as a conventional inversion pulse can be used to add T1 relaxation time weighting to images. For example, with cardiac-gated imaging, the cardiac-derived trigger can be used to control initiation of the tagging, and then imaging data can be acquired at subsequent desired delays into the cycle (Figure 1). The tag pattern will re¯ect any motion of the underlying tissue between the times of tagging and imaging. If used in conjunction with a rapid enough imaging technique, images can be acquired with suspended respiration, removing image degradation due to respiratory motion.11,20,21 With a further decrease in imaging time, motion of soft tissue structures can be evaluated with a voluntary motion initiated at the time of tagging (e.g. in response to the audible sound of the gradients used in the tagging sequence)10 and a `one-shot' image acquired when the new position has stabilized but before the tags have faded too much.

3

TAGGED IMAGE ANALYSIS

Much qualitative motion information can be gained from visual inspection of the deformation of the tag pattern (Figure 2), particularly with a dynamic `cine' display of images acquired with progressive delays after the tagging. However, there is the potential for further extraction of quantitative measures of motion from analysis of the displacement and deformation of the tags. Motion can be considered as composed of rigid body motion (displacement and rotation) and deformation (`strain'). The rigid body displacement is a property of material points. It is de®ned as the motion of the points relative to an external reference system, for example, the centroid of a ventricle (at a ®xed time or moving with the ventricle), and will be affected by bulk motion of the structure of which a particular point is a part. In contrast, the `strain' is a property of the material in a small region about a given point. It is de®ned as the relative change in length of material line segments passing through the point. As it depends only on the deformation of this region, it will be independent of the choice of external reference system. Another important difference between displacement and strain, as properties describing motion, is that whereas displacement is a vector quantity described by a magnitude and a direction, strain is a higher-dimensional tensor quantity that is orientation dependent. While a line segment oriented in one direction may be stretched a given fractional amount, other line segments passing through the same point in different directions will, in general, be stretched different fractional amounts, or even shrunk. Although it would seem dif®cult or impossible to describe ef®ciently the full strain tensor, the strain in a small region around a point can be conveniently summarized by the `princi-

Figure 1 Tagged short axis images of the heart, (a) immediately after tagging at end diastole and (b) at end systole: chest wall (CW), right ventricle (RV), left ventricle (LV)

MARKER GRIDS FOR OBSERVING MOTION IN MRI

Figure 2

3

Superimposed positions of tag lines measured from images in Figure 1

pal strains' at the point. In general, a hypothetical unit sphere (circle in two dimensions) about the point in an initial state will be transformed by motion into an ellipsoid (ellipse in two dimensions). The three major axes of the ellipsoid (or two axes of the ellipse) will suf®ce to de®ne the ®nal length of any initial line segment. The lengths of these axes for an initial unit length are the `eigenvalues' of the strain; their directions give the `eigenvectors'. For homogeneous deformation within a given region de®ned by a two-dimensional tagging grid, the rigid body motion and deformation motion variables can be calculated from the motion of the grid points. In particular, for a twodimensional homogeneous deformation, the motion of the corners of a triangular region is suf®cient to de®ne the rigid body motion and strain within the triangle.22,23 An alternative approach to calculating the motion variables within a tagged region is to use ®nite element methods to ®t the motion between the tags to a smoothly varying function of position; this model of the motion permits calculating the rigid body motion and strain at each point without requiring the assumption of homogeneous strain within each portion of the tagging grid. The motion data from two orthogonal sets of tagged images can be combined into one uni®ed three-dimensional (3D) ®nite-element model and used to calculate 3D strains.24 Note that in calculating motion from tagging grid points, it is generally not wise to use the coordinates of the intersections of tag lines with the boundaries of a moving object as distinct tagged landmarks, because through-plane motion of an obliquely oriented or tapered object can result in artifactual motion due to the change of the intersection of the boundary of the object with the ®xed imaging plane.

When the amplitude of the motion is of the order of the spacing between tags, care must be taken to avoid confusing one tag for another in the deformed state. If this is a problem, additional images with a more widely spaced grid or at intermediate times can help resolve the ambiguity. Tracking the sequential positions of multiple points in a tagging grid through multiple imaging slice locations can be a time-consuming and tedious process. The use of computer programs speci®cally designed for tag analysis can signi®cantly speed up the analysis process25. In particular, techniques to make the process of extraction of tag position from the images more automated can be very useful. These approaches can include means to match the tags to a `template' of a model tag or tag grid intersection.26±28 Alternatively, various edge or contour tracking techniques can be used to track the tags.29,30 Having derived a large amount of regional motion data from tagged images, we are faced with the problem of how to present it in a readily assimilated form. Visual display of the local values of motion variables as `functional images' is an ef®cient and appealing approach to this problem, made practical by the continuously improving performance/cost ratio of computers with graphical displays. Scalar quantities, such as the magnitude of the displacement, can be relatively easily displayed, for example, as pseudocolor overlays on the images. Even here, however, the task of displaying the time evolution of motion in a three-dimensional object may require interactive dynamic display on a computer screen to convey fully the nature of the motion. Vector motion quantities, such as directed displacement, may require overlaying the display with symbols such as arrays of directed arrows. Higher order motion quan-

4 MARKER GRIDS FOR OBSERVING MOTION IN MRI tities, such as strain eigenvectors, are intrinsically dif®cult to display fully on a ¯at screen or page, particularly for threedimensional motions. Tensor properties may require display as lower order components such as the strain eigenvalues. The reliability of tagged imaging studies as markers of the underlying motion has been validated with imaging studies of phantoms with independently known motions. This includes both phantoms undergoing only rigid body motion31 and phantoms undergoing deformation.32 As expected, the tags do move with the underlying material. 4 APPLICATIONS The development of tagged imaging methods and of methods for the analysis of the resulting images is still ongoing. However, experience is being gained with initial applications to a variety of motion studies, including ¯ow imaging, regional heart wall motion, and other soft tissue deformation. A simple, but very useful, application of tagged imaging to ¯ow studies is simply to detect the presence of ¯ow. Although the usual TOF and phase shift effects of motion seen in conventional MRI are generally suf®cient to distinguish ¯owing blood from static thrombus in MR images, it is not uncommon to have dif®culty distinguishing slow ¯ow from thombus. Tagging methods provide a straightforward way to increase the sensitivity of MRI to slow ¯ows. If grid saturation tags are created in the region of question as part of a cardiac-gated imaging sequence, and a delay of 200 ms or so is allowed to elapse between tagging and image data acquisition, any ¯uid motion of the order of a pixel or more during this interval will result in a visible shift and distortion of the tag pattern, or even complete smearing out of the tag pattern due to shearing

Figure 3

motions of the ¯uid.12 For example, in pulmonary artery embolization, the resulting slow ¯ow can be dif®cult to distinguish from the emboli; SPAMM can be very helpful in de®ning the emboli.33 The problem of visualizing patterns of ¯uid ¯ow is a long standing and important one, with many different solutions such as the injection of opaque tracers into transparent ¯owing ¯uids. The use of tagged MRI provides a new and potentially powerful tool for the study of ¯ow patterns. The tagging process does not disturb the ¯ow in any way, MR tagging can be employed with opaque ¯uids, there is no distortion due to refraction, and images are obtained in registered 3D sectional planes. In addition to biological ¯uid ¯ow applications, there may be useful engineering and industrial applications.34,35 An important area of potential application is to the study of regional heart-wall motion. Traditional imaging methods of studying cardiac function have been invasive or limited in the kinds of motion data they provide. The lack of trackable landmarks within the heart wall, and the few surface landmarks, have largely limited heart-wall motion analysis to the study of the motion of the endocardial surface or of wall thickening. This analysis has been further limited by the uncertainty about what is the appropriate reference system against which to measure the surface motion, and by the potential error due to through-plane motion of the curved and tapering heart wall. Within-wall and nonradial components of heart-wall motion can be evaluated at selected points through biplane imaging of radioopaque markers embedded in the heart wall, but the dif®culty of tracking many markers and the invasive nature of the embedding process has limited such studies to relatively few markers and precluded routine clinical application of these methods. In contrast, tagged MRI is completely noninvasive and permits a full tensor analysis of motion patterns within the heart wall.

As in Figure 1, but for a patient with diffuse hypertrophic cardiomyopathy

MARKER GRIDS FOR OBSERVING MOTION IN MRI

As the cardiac motion data provided by MRI are new, initial applications of tagged MRI to the heart have studied the normal patterns of regional wall motion.36,37 Application to the study of abnormal motion patterns in hypertrophy due to pressure overload38 or hypertrophic cardiomyopathy39 is attractive both because of their clinical importance and because of the relative ease of studying motion within thicker heart walls (Figure 3). Ischemic heart disease is another area of potential signi®cance, as MRI-derived strain measures may provide a more precise and sensitive way to characterize regional contractile dysfunction than conventional wall motion studies. This may lead to a better understanding of phenomena such as the impairment of function seen in the perinfarct zone. Other soft tissue structures, such as skeletal muscle, may be studied in a manner similar to the heart wall. They can be imaged either with repetitive motions using a modi®ed, tagged, cardiac imaging sequence and an appropriate trigger signal,17 or with a `one-shot' tagged imaging sequence.10 While it is still early in the development of these areas of application, the power of tagged MRI to provide a noninvasive means to analyze other `hidden' motions should lead to a new understanding of normal and abnormal function, and hopefully to a useful new diagnostic tool.

5 RELATED ARTICLES Heart: Clinical Applications of MRI; Selective Excitation in MRI and MR Spectroscopy; Water Suppression in Proton MRS of Humans and Animals; Whole Body Magnetic Resonance Angiography.

6 REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

G. Suryan, Proc. Indian Acad. Sci. Sect. A, 1951, 33, 107. J. R. Singer, Science, 1959, 130, 1652. A. N. Garroway, J. Phys. D, 1974, 7, L159. J. H. Battocletti, R. E. Halbach, S. X. Salles-Cunha, and A. Scances, Med. Phys., 1981, 8, 435. I. R. Young, M. Burl, G. J. Clarke, A. S. Hall, T. Pasmore, A. G. Collori, D. T. Smith, J. S. Orr, G. M. Bydeler, F. H. Doyle, R. H. Greenshan, and R. E. Steiner, Am. J. Roentgenol., 1981, 137, 895. L. Axel, Am. J. Radiol., 1984, 143, 1157. L. Axel, A. Shimakawa, and J. MacFall, Magn. Reson. Imag., 1986, 4, 199. T. Matsuda, K. Shimizu, T. Sakurai, T. Matsuda, K. Shimizu, T. Sakurai, A. Fujita, H. Ohara, S. Okamura, S. Hashimoto, S. Tomaki, and C. Kowai, Radiology, 1987, 162, 857. E. A. Zerhouni, D. M. Parish, W. J. Rogers, A. Yang, E. P. Shapiro, Radiology, 1988, 168, 59. M. Niitsu, N. G. Campeau, A. E. Holsinger-Bampton, S. J. Riederer, and R. L. Ehman, J. Magn. Reson. Imag., 1992, 2, 155. E. R. McVeigh and E. Atalar, Magn. Reson. Med., 1992, 28, 318. L. Axel and L. Dougherty, Radiology, 1989, 171, 841. L. Axel and L. Dougherty, Radiology, 1989, 172, 349. P. J. Hore, Magn. Reson., 1983, 55, 283.

5

15. S. E. Fischer, G. C. McKinnon, S. E. Maier, and P. Boesiger, Magn. Reson. Med., 1993, 30, 191. 16. L. Axel and L. Dougherty, Radiology, 1989; 173(P), 223. 17. J. G. Pipe, J. L. Boes, and T. L. Chenevert, Radiology, 1991, 181, 591. 18. T. J. Mosher and M. B. Smith, Magn. Reson. Med., 1990, 15, 334. 19. B. Bolster, Jr., E. R. McVeigh, and E. A. Zerhouni, Radiology, 1990, 177, 769. 20. L. Axel, Z. A. Fayad, and L. Dougherty, Radiology, 1992, 185(P), 223. 21. Z. A. Fayad, D. L. Kraitchman, and L. Axel, `Magnetic resonance tagging: Conventional spin echo versus fast breath-hold gradientecho imaging using phased-array coils', in Proc. 12th Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 1217. 22. G. D. Meier, M. C. Ziskin, W. P. Santamore, and A. A. Bove IEEE Trans. Biomed. Eng., 1980, 27, 319. 23. L. Axel, R. GoncËalves, and D. Bloomgarden, Radiology, 1992, 183(3), 745. 24. A. A. Young and L. Axel, Radiology, 1992, 185, 241. 25. L. Axel, D. Bloomgarden, C.-N. Chang, D. Kraitchman, and A. A. Young, `SPAMMVU: a program for the analysis of dynamic tagged MRI', Proc. 12th Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 724 26. D. Fisher, `Automated tracking of cardiac wall motion using magnetic resonance markers', Dissertation, University of Iowa, 1990. 27. D. L. Kraitchman, L. Axel, and A. A. Young, `Springs: a fast method for detection and correspondance of cardiac magnetic resonance tags', Proc. 12th Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 725. 28. M. A. Guttman, J. L. Prince, and E. R. McVeigh, IEEE Trans. Med. Imag., in press. 29. M. Kass, A. Witkin, and D. Terzopoulos, Int. J. Comput. Vis., 1987, 1, 321. 30. D. Kraitchman, A. A. Young, and L. Axel, `Active contour models for tracking magnetic resonance tags', Proc. of the 15th Annu. Int. Conf. of the IEEE Eng. in Med. and Biol. Soc., San Diego, CA, 1993, pp. 166, 167. 31. E. R. McVeigh and E. A. Zerhouni, Radiology, 1991, 180, 677. 32. A. A. Young, L. Axel, L. Dougherty, D. K. Bogen, and C. S. Parenteau, Radiology, 1993, 188, 101. 33. H. Hatabu, W. B. Gefter, L. Axel, H. I. Palevsky, C. Cope, N. Reichek, L. Dougherty, J. Listerud, and H. Y. Kressel, Radiology, 1994, 190, 791. 34. M. V. Icenogle, A. Caprihan, and E. Fukushima, J. Magn. Reson., 1992, 100, 376. 35. M. Tyszka, R. C. Hawkes, and L. D. Hall, J. Magn. Reson., 1992, 97, 391. 36. N. R. Clark, N. Reichek, P. Bergey, E. A. Hoffman, D. Brownson, L. Palmon, and L. Axel, Circulation, 1991, 84, 67. 37. A. A. Young, H. Imai, C.-N. Chang, and L. Axel, Circulation, 1994, 89(2), 740. 38. L. C. Palmon, N. Reichek, S. B. Yeon, N. R. Clark, D. Brownson, E. Hoffman, and L. Axel, Circulation, 1994, 89(1), 122. 39. A. A. Young, C. M. Kramer, V. A. Ferrari, L. Axel, and N. Reichek, Circulation, 1994, 90, 854.

Biographical Sketch L. Axel. b 1947. B.S. Syracuse University, 1967; Ph.D. Princeton University, 1971; M.D., University of California at San Francisco, 1976. Department of Radiology, Hospital University of Pennsylvania, 1981± present. Approx. 110 publications. Research interests include development of NMR imaging methods to study cardiovascular function.

NMR SPECTROSCOPY OF THE HUMAN HEART

NMR Spectroscopy of the Human Heart Paul A. Bottomley Johns Hopkins University, Baltimore, MD, USA

1 INTRODUCTION The heart is the largest consumer of energy per gram of tissue, and defects involving energy metabolism, including energy supply and demand, are central to much disease of the heart. It is inevitable, therefore, that human cardiac spectroscopy has focused on the energy metabolites that are NMR detectable. Phosphorus (31P) NMR spectroscopy (MRS) can see adenosine triphosphate (ATP), the fundamental energy currency of the body, and phosphocreatine (PCr), a reservoir of cellular energy, in the heart, as well as the metabolic by-product inorganic phosphate (Pi). The signal-to-noise ratio is about 1  10ÿ5 or 2  10ÿ5 of that of the proton (1H) NMR signal from muscle water (from the same volume), at the same magnetic ®eld strength.1,2 PCr and ATP cannot therefore be imaged with the same spatial resolution, signal-to-noise ratio, and scan time as tissue water protons in the same magnetic ®eld. However, with compromises in spatial resolution and scan time, PCr and ATP can be seen in anterior wall volume elements (voxels) as small as 8 mL in the anterior myocardium.3 PCr and ATP are linked via the creatine kinase (CK) reaction: PCr ‡ ADP „ ATP ‡ creatine

…1†

whereupon inorganic phosphate (Pi) may be formed as ATP ! ADP ‡ Pi ‡ energy

…2†

The total creatine pool (CR) comprising unphosphorylated (Cr) plus PCr is also detectable in the heart via its N-methyl resonance at 3.0 parts per million (p.p.m.) using 1H MRS, with a current resolution of 3±9 ml in scan times of <10 min.4 Consequently, a combination of 1H and 31P MRS could potentially provide a near-complete picture of CK metabolism in the heart. Under anaerobic conditions, such as may occur in ischemic heart disease in regions of the heart supplied by blocked arteries,5 a transient mismatch between oxygen supply and demand can cause excess PCr consumption to maintain adequate ATP, along with possible elevations in Pi. This is manifested as reductions in PCr/ATP, PCr/Pi, and/or ATP/Pi in the corresponding 31P NMS during the ischemic episode. The rapid depletion of ATP and PCr levels assayed after coronary occlusion or low-¯ow ischemia has been demonstrated in animal models.6 Biochemical assays of biopsies taken at surgery have also demonstrated chronic reductions in the levels of PCr, Cr, and creatine kinase activity in patients with dilated cardiomyopathy (DCM), hypertrophy, and coronary artery disease (CAD).7,8 These ®ndings link changes in energy metabolism to

1

contractile dysfunction, but in the absence of evidence of active ischemia.9 Thus, the levels and ratios of PCr, ATP, CR, Cr, and Pi measured by 1H and 31P MRS may be useful indices of supply-side myocardial energy metabolism. To date, 31P NMR has been applied extensively in studies of patients with cardiomyopathies, including DCM, hypertrophic cardiomyopathy (HCM), and pressure overload left ventricular hypertrophy (LVH); patients with transplanted hearts, in an attempt to assess histological rejection; and patients with CAD and myocardial infarction, including exercise stress testing of patients with reversible ischemia. The literature on 31P NMR of the heart has recently been reviewed in detail.10 Abnormalities of varying extents in the phosphate metabolites have been reported in all of these disease states. In particular, the variability of ®ndings in cardiomyopathy, and the results from transplanted hearts, suggest the existence of confounding variables that are as yet incompletely understood. However, quantitative studies are approaching a consensus on the true PCr/ATP ratio for human heart, and there are preliminary values reported for the PCr and ATP concentrations, the intracellular pH, and the forward CK reaction rate, and ¯ux. Quanti®cation of Pi has proved dif®cult because of the presence of neighbouring blood 2,3-diphosphoglycerate (DPG) resonances. Partial saturation effects and contamination from blood and chest muscles are probably the main sources of discrepancy amongst reported myocardial 31P NMR PCr/ATP data. 1 H NMR spectra from the human heart are dominated by water and fat resonances, but metabolites such as CR can be measured either relative to water, or as concentrations.4 CR appears to be depleted in myocardial infarction,4 and alterations are also anticipated in cardiomyopathy and heart failure.9 NMR spectroscopy studies of the human heart with nuclei other than 31P or 1H are scarce. Carbon (13C) NMR potentially provides access to glycolytic and citric acid cycle metabolites such as lactate and glutamate but has the disadvantage of low NMR sensitivity and a 1.1% natural isotopic abundance. This effectively prevents their detection at natural abundance with current technology without resorting to 13C-enriched substrates.2 Fatty acid resonances are the main contributors to the natural abundance 13C human heart spectrum, and pericardial fat is probably a dominant contaminant, but glycogen may be discernible with 1H decoupling and NOE.11 While strictly not spectroscopy since there is usually only a single resonance, sodium (23Na) NMR has been performed in human heart12,13 and has potential for detecting sodium increases in reperfused infarction where sodium±potassium pump function is lost.14

2 2.1

METHODS Localization

Localized 31P NMR spectroscopy of the human heart was ®rst reported in 1985.1 That study, and all those since, employed surface detection coils placed on the chest closest to the anterior myocardium. Myocardium was distinguished from chest wall using MRI slice selection (the depth resolved surface coil spectroscopy, or DRESS, technique), and 1H MRI was

2 NMR SPECTROSCOPY OF THE HUMAN HEART used for tissue identi®cation. While an unlocalized surface coil study of the chest of an infant with congenital cardiomegaly was also reported in 1985,15 the ®rst studies of heart patients with localized spectroscopy protocols were not published until 1987.16,17 These were limited to two to four patients with hypertrophic cardiomyopathy or recent myocardial infarction, and localization was afforded by rotating frame zeugmatography (RFZ), which uses B1 gradients, or by the DRESS technique. Full three-dimensional MRI gradient localization of 31P NMR spectra to single voxels in the heart via the image selected in vivo spectroscopy (ISIS) technique appeared in 1988.18 Multiple voxel, chemical shift imaging (CSI) studies of the heart using MRI phase-encoding gradients in one or more

dimensions were published around 1990.5,18,19 These spectroscopy localization techniques are reviewed in greater detail elsewhere (see Chemical Shift Imaging, Selective Excitation in MRI and MR Spectroscopy, and Single Voxel Whole Body Phosphorus MRS), and have various advantages and disadvantages. For example, a problem with three-dimensional CSI is that the large number of phase-encoding steps required to encode the entire sample volume sets a minimum scan time, which may not be compatible with clinical study protocols of limited duration such as in stress testing for myocardial ischemia. One-dimensional CSI and RFZ can be completed faster, but at the expense of the fuzzy localization afforded by the surface coil in the two dimensions that are not gradient encoded. An example of a one-dimensional CSI data set and the corre-

DPG

PCr

(b)

16

PDE DPG ATP

PCr

Ref.

15 PME ·20

14 Pi

13 ·10

Depth 12

9 20

0

ppm

Figure 1 (a) Typical axial 1H surface coil image of the chest and heart of a normal volunteer, annotated to show the location of eight 1 cm thick coronal slices localized in the spectroscopy exam. (b) Cardiac-gated one-dimensional CSI surface coil 31P NMR spectra from the slices show a reference peak from a vial embedded in the coil (slices 9, 10), chest muscle (slices 12, 13), and myocardium (slices 14±16). Scan time for the spectra was about 12 min, in a clinical 1.5 T scanner. (Reproduced by permission of the Radiological Society of North America from Bottomley et al.20)

NMR SPECTROSCOPY OF THE HUMAN HEART

sponding annotated cardiac image is shown in Figure 1.20 Use of single voxel methods such as DRESS or ISIS requires accurate selection of a suspected region of abnormality a priori: incorrect voxel selection for a stress test, for example, could result in a missed ®nding for reversible ischemia. 1 H NMR spectroscopy studies of myocardial CR in patients to date have been performed using the single voxel pointresolved surface coil spectroscopy method (PRESS)21 and the stimulated echo method (STEAM)22 with water suppression. Each method localizes a spin-echo or stimulated echo NMR signal to a voxel de®ned in all three dimensions by the use of three spatially selective NMR pulses (90 ±180 ±180 for PRESS; 90 ±90 ±90 for STEAM). 2.2 Coils and NOE A typical experimental cardiac 31P NMR surface coil probe is pictured in Figure 2.23 When the sample is the dominant source of the noise detected by the surface coil, the coil diameter should be chosen roughly equal to the depth of the tissue of interest. This leads to heart 31P detectors with typical diameters of 6±12 cm for the anterior wall to mid-wall. Preferably, the signal is excited with a separate, larger rf coil to minimize spatially dependent distortions due to B1 inhomogeneity.23 Additional 1H MRI coils ensure correct 31P detection coil placement, which is critical because of the limited range of sensitivity of the surface coil, and possible avoidable contamination from nearby tissue, especially chest muscle. Sensitivity improvements of up to about H2 compared with the best positioned surface coil, and the need for careful positioning is relaxed if phased-array surface detection coils are

3

used.24 This technology is in use in MRI,25 and has been implemented in 31P studies of normal volunteers,24 but not yet in patient studies. Other factors offering potential sensitivity gains are the use of a prone, as opposed to supine, patient orientation, to bring the heart closer to the detection coil(s);23 synchronization of NMR acquisitions to the cardiac cycle (see Cardiac Gating Practice); and 1H NOE.26 At 1.5 T, NOEs of  = 0.61 ‹ 0.25 for PCr,  = 0.6 ‹ 0.3 for -ATP, and  = 0.3 ‹ 0.2 for ATP have been reported for the human heart.26 Because of the differences in  for PCr and ATP, the NOE may distort the PCr/ATP ratio by up to 20%, necessitating a correction for quantitative intersite comparisons.26 3 3.1

QUANTIFICATION Myocardial PCr/ATP Ratio

If the PCr/ATP ratio is the most important metabolic index that 31P NMR can reliably detect in the heart, it is important to establish its normal range for inter- and intralaboratory comparisons. However, published values from healthy volunteers range from 0.9 ‹ 0.322 to 2.1 ‹ 0.4.28 Why the difference? There are two main causes. First, data are distorted to varying extents by partial saturation effects that result from the use of pulse sequence repetition periods, TR, that are comparable to or less than the T1 values of PCr and ATP. Differences in the T1 values of the two moieties produce differential distortion of PCr/ATP.29 The 95% con®dence intervals for published T1 values are given in Table 1.17,19,26,30±35 With a ratio of the T1

Figure 2 The cardiac 31P NMR surface coil probe used to acquire the localized 31P and 1H NMR data in Figure 1.23 The patient lies prone on the padded coil set (left), which is positioned on the patient table in the center of the whole body NMR magnet. The coil set comprises a square base with a 0.40  0.40 m 31P transmit coil (lower left), and a smaller probe containing a 0.065 m mean diameter spiral 31P detector coil, and a 0.08  0.13 m ®gure-8 1H receiver coil, used for imaging and shimming (right). Black strips (right) are Velcro, permitting the smaller probe to be positioned at will. Passive diode circuitry minimizes coil interactions23

4 NMR SPECTROSCOPY OF THE HUMAN HEART Table 1 Phosphorus-31 NMR Measurements of Phosphate Metabolites in Normal Volunteers Parameter T1(PCr) T1( -ATP) T1( -ATP) [PCr]

Value

Reference a

4.37 ‹ 0.48 2.52 ‹ 0.45a 2.28 ‹ 0.54a 11 ‹ 3 mol (g wet wt)ÿ1 12 ‹ 4.3 mol (g wet wt)ÿ1 10 ‹ 2 mol (g wet wt)ÿ1 [ATP] 6.9 ‹ 1.6 mol (g wet wt)ÿ1 7.7 ‹ 3.0 mol (g wet wt)ÿ1 5.8 ‹ 1.6 mol (g wet wt)ÿ1 Pi/PCr <0.25 0.14 ‹ 0.06 pH 7.15 ‹ 0.2 7.15 ‹ 0.03 CK forward rate 0.5 ‹ 0.2 sÿ1 CK forward ¯ux 6 ‹ 3 mol (g wet wt s)ÿ1 NOE of PCr at 1.5 T 0.61 ‹ 0.25 0.6 ‹ 0.1 NOE of ATP at 1.5 T 0.6 ‹ 0.3 ( -ATP) 0.3 ‹ 0.2 ( -ATP) 0.4 ‹ 0.2 ( - and -ATP)

30 30 30 19 31 32 19 31 32 17 33 17 33 34 34 26 35 26 26 35

CK, creatine kinase; -ATP and -ATP, - and - phosphates of ATP. a 95% con®dence intervals.

values of PCr and ATP of about 1.9, the distortion or saturation factor, F, by which the PCr/ATP ratio measured in the short TR experiment must be multiplied by to yield the true myocardial PCr/ATP ratio, will vary between 1 and ~2, depending on TR and the NMR pulse ¯ip angles.29 Thus, measurements that are corrected for partial saturation can differ from those that have not, by up to a factor of about two. Correction of experimental PCr/ATP values for partial saturation can be done by two methods. First, F can be calculated using known myocardial T1 values and a measurement of the experimental NMR pulse ¯ip angle (e.g. with equation (2) of Bottomley et al.).29 Due to time constraints, T1 values are usually either measured on a separate group of individuals, or else published values are adopted. Second, F may be measured directly from the ratio of spectra acquired under partially saturated conditions and fully relaxed conditions. The unlocalized, uniformly excited signal detected by a surface coil on the chest is suf®ciently sensitive to permit the measurement of saturation factors in about 6 min, which can easily be incorporated into the study protocol.5,29 This use of an unlocalized saturation factor depends on the assumption that T1(PCr)/T1(ATP) is essentially the same in the chest and heart tissue that contribute to the unlocalized spectrum.29 The assumption is consistent with the available data,30 and errors resulting from differences in the ratio can be minimized by using small or Ernst ¯ip angles for excitation. The second, lesser source of scatter in the reported myocardial PCr/ATP values is that of blood which contains ATP but no PCr. This reduces the apparent PCr/ATP ratio in voxels that intersect the ventricles.36 The PCr/ATP ratio can be corrected for blood ATP contamination by measuring the amount of blood using the blood DPG signal present in the spectrum

as a marker. DPG has a characteristic doublet at 5.4 and 6.3 ppm, near Pi17 and other phosphomonoesters (PMs). The amount of ATP added by blood is estimated from corrected ATP ˆ contaminated ATP ÿ 0:5  …‰ATPŠ=‰DPGŠ†  total DPG signal

…3†

where [ATP]/[DPG] & 0.30,33,37±39 and remembering that DPG has two phosphates. Blood ATP corrections typically increase PCr/ATP values by a relatively small amount, 13 ‹ 6%.20,33,36,39,40 Potential problems with the correction may arise from differences between the T1 of ATP and DPG for short TR protocols (possibly offset, however, by the effect of in¯owing unsaturated blood), variations in the blood ATP/DPG ratio with localization method and patients,41 and possible contamination of DPG by Pi or PMEs, which would tend to cause overestimation of the amount of blood contamination. Considering only those human studies in which both saturation and blood corrections were performed20,31,33,36,40,42±48 or in which blood contamination was ruled out,3 a consensus is emerging on the normal myocardial PCr/ATP value, as evident in Table 2. The mean value is 1.81 ‹ 0.14, omitting the highest and lowest values. This should represent the best current estimate for the true PCr/ATP ratio in the normal human heart, since tissue ATP and PCr appear to be 100% NMR visible.49,50

3.2

PCr and ATP Concentrations

Tissue [PCr] and [ATP] can be determined by combining the fully corrected PCr/ATP ratio from a voxel, with a measurement of the NMR signal from a concentration reference and a measurement of the volume of myocardium present in the voxel. The reference may lie outside of the subject, which requires that corrections be made for differences in ®eld strength and sensitivity of the NMR coils at the two locations.16 Alternatively, a reference located at the same position as the heart may be observed in a separate experiment in

Table 2 Some Corrected Normal Human Myocardial PCr/ATP Ratios Value

Field (T)

Reference

1.80 ‹ 0.21 1.93 ‹ 0.21 1.8 ‹ 0.1 1.95 ‹ 0.45 1.65 ‹ 0.26 1.85 ‹ 0.28 1.8 ‹ 1 1.71 ‹ 0.13 1.98 ‹ 0.07 2.02 ‹ 0.11 1.42 ‹ 0.18 1.61 ‹ 0.18 2.46 ‹ 0.53 1.6 ‹ 0.3

1.5 1.5 4.0 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 2.0

36 20 3 40 33 42 31 39 43 44 45 46 47 48

NMR SPECTROSCOPY OF THE HUMAN HEART

which differences in coil loading are taken into account.31 The tissue volumes must again be estimated by MRI, which may introduce weighting errors when the coil sensitivity is nonuniform across the localized volume. A quanti®cation method that does not require tissue volumetry employs measurements of the 1H signal from water, sW, as a concentration reference, using the identical localization sequence.32 The water signal is recorded with the same 31P coil used to measure the PCr and ATP metabolite signals, sP; as a result, both 31P and 1H spectra are acquired with essentially the same B1 pro®le. This is possible because of the enormous concentration advantage of the water signal compared with the metabolites. The method requires calibration with a phosphate reference to determine the ratio of the 31P signal per phosphate to the 1H signal per proton, CPH, as well as knowledge of the myocardial tissue water content [W], which appears to be relatively constant.4 The metabolite concentration in a voxel is calculated from: 

‰PŠ ˆ

2sP ‰WŠCPH FP EP  sW FW EW

 …4†

where FPFW are the saturation factors and E the factors accounting for signal decay during TE (e.g. E = exp{TE/T2*} where T2* is the transverse signal decay time) with subscripts W for water, P for the metabolite (subscript P). The factor of 2 accounts for the two protons on water. The value of CPH is determined in a separate experiment using a standard solution of phosphate.32 The tissue water signal can be corrected for contamination from water in ventricular blood and in pericardial fat via a 31P DPG measurement and a measurement of the fat resonance in the 1H spectrum, respectively. The contaminating water signal from ventricular blood is: S…blood† ˆ

  S…DPG†WB CPH FDPG EDPG  ‰DPGŠ EB

…5†

The contaminating water signal from fat is LS(CH2), where S(CH2) is the lipid 1H signal, and L is the water content of fat, which is about 15%.32 The corrected water signal to use in equation (4) is: SW ˆ …measured water signal† ÿ S…blood† ÿ LS…CH2 †:

…6†

Concentrations reported to date for the human heart are in fair agreement and are summarized in Table 1. 3.3 Total Myocardial Creatine The detection of the N-methyl 1H resonance of CR at 3.0 ppm provides a means of quantifying myocardial CR.4 Unlike PCr/ATP ratios, CR cannot be reliably measured as a ratio from a single 1H acquisition relative either to water, which is normally suppressed, or to fat resonances, which are often variable because of contamination from pericardial fat. A solution is to acquire two consecutive 1H spectra under the same conditions, with and without water suppression. The CR peak in the water-suppressed spectrum is measured relative to the water peak in the unsuppressed spectrum. Because 1H sensitivity is high and because the 1H T1 values are shorter than for 31P, the experiment can essentially be done fully relaxed. The CR/water ratios measured with the same STEAM or PRESS timing par-

5

ameters can be used for patient comparative studies.4 The normal CR/water ratio is about 1.3  10-3. Such ratios can be translated into absolute concentration measurements using the known myocardial tissue water content and correcting for the signal loss caused by the STEAM or PRESS echo delay, represented by the E factors as in equation (4):4 ‰CRŠ ˆ

  2SCR ‰WŠ FCR ECR  3SW FW EW

…7†

with the extra factor of 3 for the N-methyl protons. The initial 1H NMR estimate for normal human heart [CR] is 28 ‹ 6 mol (g wet wt)-1, which is consistent with canine and human necropsy studies.4 Concentrations reported to date for the human heart are in agreement, and summarized in Table 1. 3.4

Pi, pH, and Creatine Kinase Reaction Kinetics

Intracellular pH can be measured from the chemical shift of the Pi resonance, which varies from 3.7 to 5.2 ppm relative to PCr over a pH range of 6.0±7.3.51 In the normal heart, Pi is small and dif®cult to resolve unambiguously from blood DPG. It is probably better described by inequalities17 (Table 1). Even with 1H decoupling, which may reduce the linewidths of the DPG signals and permit a less ambiguous determination of any neighboring Pi, Pi still appears undetectable in about half of the normal subjects studied.33 When detectable, the normal myocardial pH is about 7.15 (Table 1). It is also possible that Pi is partially NMR invisible.49,50 It is unlikely, however, that Pi and pH measurements are contaminated by Pi in blood,52 since Pi is imperceptible in 31P spectra from human blood39,41 and the [Pi] in whole human blood is listed as only 0.08 mM.53 The ¯ux of PCr through the creatine kinase reaction [see equation (1)] in the heart can be measured using the saturation transfer NMR experiment.34 The experiment involves acquisition of 31P spectra while the -phosphate resonance of ATP is being saturated, for example, by application of a long rf pulse tuned to its resonant frequency. When ATP is saturated, the PCr signal declines as PCr is converted to ATP via the forward reaction, in the absence of any refreshment by unsaturated phosphates produced via the reverse reaction. The ratio of PCr signals measured with and without saturation is directly proportional to the ®rst-order rate constant, in units of the T1 of PCr in the presence of the saturating radiation. Studies in a 4 T whole body instrument indicate that about half of the PCr is turned over per second (Table 1), consistent with data from animals.34

4

PATIENT STUDIES

Some data from human heart 31P NMR studies of patients are summarized in Table 3.5,17,20,28,31,33,36,39,40,42±44,47,48,54,55 4.1

Cardiomyopathy

Evidence anticipating reductions in myocardial high energy phosphates in human cardiomyopathy and hypertrophic disease comes from 31P NMR studies of animal models,56±61 and

6 NMR SPECTROSCOPY OF THE HUMAN HEART Table 3

Summary of Some Findings for Myocardial PCr/ATP in Patients

Patients

PCr/ATP

Controls

Comments

Reference

Cardiomyopathy studies showing signi®cant changes (see text) 1.3 ‹ 0.3* 2.1 ‹ 0.4 1.4 ‹ 0.4* 2.1 ‹ 0.4 1.3 ‹ 0.3*** 1.7 ‹ 0.3 1.1 ‹ 0.4** 1.7 ‹ 0.3 1.1 ‹ 0.3** 1.6 ‹ 0.2a 1.5 ‹ 0.31* 1.8 ‹ 0.2 1.4 ‹ 0.5*** 1.9 ‹ 0.4a 1.3 ‹ 0.5 1.94 ‹ 0.11 1.5 ‹ 0.1* 2.0 ‹ 0.1 1.3 ‹ 0.3** 1.6 ‹ 0.3 2 ‹ 0.4*** 2.4 ‹ 0.5

MD, BB, CA patients HCM patients NYHA class 0±II HCM patients HCM patients NYHA class II±III LVH patients NYHA class II±IV DCM patients NYHA class 5III DCM patients DCM with NYHA class II±III, high mortality Aortic valve disease, NYHA class III Patients with severe mitral valve disease HCM patients

39 54 36 40 43 44 48 47

Heart transplant patients 1.6 ‹ 0.5** 1.6 ‹ 0.5***

1.9 ‹ 0.2 1.9 ‹ 0.4a

All transplant patients Patients with rejection (myocyte necrosis)

20 20

1.6 ‹ 0.4 2.0 ‹ 0.5 1.85 ‹ 0.28 1.8 ‹ 1.0

Pi/ATP elevated 5±9 days-post-MI [PCr]b, [ATP]d reduced in old MI Patients studied 0.5±24 months post-MI Patients with ®xed 201Tl defects Fixed 201Tl defects; [PCr], [ATP] reduced***

17 55 40 42 31

1.5 ‹ 0.3a 1.6 ‹ 0.2a

During isometric exercise stress testing Isometric exercise, reversible 201Tl defects

5 42

Myocardial infarction 1.7 ‹ 0.4 (normal) normal 1.8 ‹ 0.5 1.2 ‹ 0.3** 0.9 ‹ 0.4*** Ischemia 0.9 ‹ 0.2* 1.0 ‹ 0.3*

28 28 .

Values are mean ‹SD. Data from abstracts are omitted. MD, muscular dystrophy; BB, cardiac beri-beri; CA, cardiac amyloidosis; MI, myocardial infarction. Controls were normal, or LVH patients not in heart failure (indicated by a),54 patients with dilated cardiomyopathy NYHA class
biochemical analyses of patient biopsies taken at surgery.7,8 Human in vivo 31P NMR studies have demonstrated statistically signi®cant reductions in anterior myocardial PCr/ATP ratios in a group of patients with cardiomyopathies due to speci®c disease (muscular dystrophy, beri-beri, amyloidosis);28 in patients with HCM;28,33,39,47 in patients with valve disease who were undergoing treatment for heart failure;44,48,54 and in patients with DCM who were in heart failure [New York Heart Association (NYHA) clinical classi®cation for heart failure 5II].36,40,43 One report found a signi®cant negative correlation between the NYHA classi®cation for failure, and the myocardial PCr/ATP ratio.40 Also, in patients whose NYHA classi®cation was improved by drug therapy, myocardial PCr/ ATP values recovered,40 while patients with signi®cantly reduced PCr/ATP ratios had a poorer survival rate compared with patients with normal PCr/ATP ratios, suggesting that the ratio may be of prognostic value.43 A number of other studies have reported no statistically signi®cant changes in myocardial PCr/ATP in LVH,27,28 or DCM,27,28,33,62,63 although possible correlations between reduced myocardial PCr/ATP ratio and the NYHA heart failure classi®cation40 were not explicitly ruled out by these studies. The ®ndings are reviewed in detail elsewhere.10 Except for the NYHA classi®cation, PCr/ATP ratios have generally correlated weakly with etiology and functional indices of disease severity

such as the left ventricular ejection fraction or fractional shortening although such links may be proved with ongoing studies.48 The factors responsible for the differences in the signi®cant ®ndings for PCr/ATP ratios in cardiomyopathies are probably (i) variations in the range and severity of heart failure in patient study populations, and (ii) variations in statistical sensitivity. Regarding the former, it appears that averaging results from patients in mild and severe failure, or from a group of patients in predominantly mild failure, can mask signi®cant PCr/ATP changes,40 and that abnormalities might only be seen in subsets of patients in more advanced stages of heart failure.54 With respect to the second factor, it should be noted that in all studies of HCM, DCM, and LVH so far, the mean myocardial PCr/ATP ratios were lower by 2±54% relative to the corresponding normal control groups with varying statistical signi®cance.10 The 31P NMR data linking PCr/ATP changes to heart failure would indicate that abnormal myocardial PCr/ATP ratios may ®rst become detectable in DCM and LVH between NYHA classes II and III,40,54 which is the point at which physical activity is limited by fatigue, and accompanying symptoms such as palpitation, or dyspnea. This is consistent with studies that conclude that the ability of the cell to sustain adequate levels of ATP is compromised only in the more advanced stages of

NMR SPECTROSCOPY OF THE HUMAN HEART

failure.64 One hypothesis is that the reduced energy reserve, as demonstrated by the reduction in myocardial PCr/ATP ratios seen by 31P NMR, and by the decreases in myocardial creatine and creatine kinase activity evident in surgical biopsies, may limit the ability of the heart to do work, and lead to contractile dysfunction.9 On the other hand, in at least two studies of HCM,15,33 the PCr/ATP ratio was reduced in the absence of a link to heart failure, so it is possible that there are differences between HCM, and DCM and LVH. In the clinic, 31P NMR might help identify patients in heart failure where physical activity or diagnosis is complicated by other conditions, including age54 and lung disease. In patients with valve disease, the detection of metabolic abnormalities might ®nd use in surgical planning to minimize permanent injury and maximize bene®t.65 Further work is needed to de®ne where, speci®cally, the bene®ts lie. 4.2 Heart Transplant Patients The idea that changes in myocardial metabolite ratios might predict histological rejection in human heart transplants stems from animal 31P NMR studies, usually of nonimmunosuppressed allografts, which showed metabolic changes prior to the occurrence of histological evidence for acute rejection in the ®rst week or so posttransplantation.66±70 In the management of heart transplant patients, the standard criterion for assessing the existence of signi®cant allograft rejection of severity suf®cient to warrant augmentation of immunosuppressive therapy is histological evidence for myocyte necrosis in endomyocardial biopsies acquired during regular cardiac catheterization procedures. While some earlier conference reports on 31P NMR studies in transplant patients dating from 1988 show reduced myocardial PCr/ATP and PCr/Pi ratios in transplanted hearts, the success with which 31P NMR can reliably predict the outcome of histological evaluations is mixed. The ®rst published paper on 19 31P NMR examinations of patients studied up to 5.5 years posttransplantation did ®nd signi®cantly lower resting anterior myocardial PCr/ATP ratios relative to normal controls, consistent with the animal studies.10,20 However, the results showed agreement between 31P NMR abnormalities and histological evidence for necrosis in only about 60±70% of examinations,20 suggesting that 31P NMR is not a precise predictor of signi®cant histological rejection in many transplant patients. To date there is no ®rm evidence linking the ®nding of signi®cant reductions in the myocardial PCr/ATP (and possibly PCr/Pi) ratio in transplanted hearts to other factors such as hypertrophy, or CAD involving the major vessels. However, one conference report of 13 transplant recipients on 71 occasions found reduced PCr/ATP ratios in the ®rst few weeks after transplantation, suggesting that lower creatine levels or injury or edema following surgery may be possible causes.71 The observation that PCr/ATP ratios may not precisely predict histological rejection is likely to re¯ect fundamental differences between histological and metabolic indices. In particular, myocyte necrosis would not cause altered PCr/ATP ratios because dead cells can contribute no high energy phosphates, whereas necrosis is important in histological evaluation. If the PCr/ATP reduction is associated with rejection, it may re¯ect an earlier phase of the process, which would not be inconsistent with the observed PCr/ATP reduction in the early weeks postsurgery

7

because rejection episodes tend to be more frequent during this period. A much more frequent schedule of NMR examinations and biopsies would be needed to test this hypothesis. Nevertheless, the occurrence of reductions in the high energy phosphate metabolism of a majority of transplant patients is a concern meriting further study. 4.3 4.3.1

Coronary Artery Disease Myocardial Infarction

Published papers to date show no signi®cant alterations in resting myocardial PCr/ATP ratios in myocardial infarction.17,40,46,55 Two reports do show signi®cantly reduced resting PCr/ATP in patients with infarction and irreversible 201 Tl defects by radionuclide imaging.31,42 It is unclear whether other factors such as heart failure and/or cardiomopathy may play a role in these patients and could explain the different ®ndings.10 In recent anterior myocardial infarction, signi®cant elevations in Pi levels can be detected within the ®rst week or so after onset,17 which is consistent with canine 31P NMR studies showing elevated Pi levels persisting for several days postcoronary occlusion, as a byproduct of PCr and ATP consumption.72 Biochemical analyses of animal hearts also show that essentially all the ATP and PCr is depleted within the ®rst few hours of ischemic injury that results in cell death.6 Thus again, as dead cells can contribute no high energy phosphates, it is likely that the normal resting PCr/ATP ratios derive from a mixture of metabolically normal tissue, scar tissue and, possibly, jeopardized myocardium adjacent to, or interspersed with, the infarction. Accordingly the infarction itself may best be characterized by the absence of any contribution to the spectrum. Evidence for the possible use of 31P NMR for assessing myocardial viability, is provided by a study of 29 patients with recent myocardial infarction.46 This revealed no changes in PCr/ATP ratios in patients with myocardial stunning after reperfusion, suggesting that relative levels of high-energy phosphates are not depleted in stunned human myocardium. There is now 31P NMR evidence that the metabolite concentrations are reduced in infarction itself.31,55 This includes the observation of a signi®cant negative correlation between ATP levels and the size of perfusion de®cits in the heart, as quanti®ed by thallium 201Tl radionuclide imaging,55 and reductions in [PCr] and [ATP] in patients with ®xed 201Tl defects compared with those with reversible defects.31 Concentration measurements made by 31P NMR must be treated cautiously, however, since the tissue volume present in the voxel may be reduced by any wall thinning that occurs after infarction. This may partially offset the signal loss due to infarction, depending on how tissue volume is accounted for in the concentration measurements. Nevertheless, the observations31,55 are consistent with a model for infarction wherein the myocardial PCr and ATP levels are reduced in the heart spectrum in proportion to the volume of infarcted tissue that intersects the voxel.10 Reductions of about 60% in myocardial [CR] can also be detected with water-referenced 1H NMR in patients with myocardial infarction.4 The results are consistent with animal studies and the concept of metabolic depletion in infarction. Because the resolution and sensitivity are better with 1H NMR than with 31P NMR (and because the CR resonance has three

8 NMR SPECTROSCOPY OF THE HUMAN HEART protons compared with the single phosphate group of PCr), 1H NMR could provide a useful metabolic means for distinguishing healthy from infarcted nonviable myocardium.

(a)

2.5

2.0

In patients with ischemic heart disease involving severe stenosis of the anterior vessels the resting anterior myocardial PCr/ATP ratio is normal,40 or nearly so.5 To observe metabolic change corresponding to reversible ischemia, safe and effective stress protocols that can be performed in the NMR magnet for the duration of a localized 31P NMR examination are necessary. Three types of stress have been tried: aerobic exercise involving the lifting of weights with the legs;73 an isometric exercise, performed, for example, with a handgrip dynamometer;5,42 and stress induced by pharmaceutical agents such as dobutamine.45,63 The increase in cardiac work-load indexed by the heart rate±blood pressure product in these protocols is about 70% for an aerobic leg exercise lifting 5 kg weights;73 about 30±35% with the isometric hand-grip exercise at 30% of the subjects maximum force;5 and up to 130% (2.3-fold) with dobutamine infusion.45,63 While producing a lesser increase in cardiac work, the isometric exercise protocol can produce re¯ex coronary vasoconstriction in stenosed vessels. It also minimizes motion problems during NMR acquisition compared with aerobic exercise, and, in addition, produces vasoconstriction in patients with critical coronary disease.5 It is also well tolerated by CAD patients, and can be immediately terminated should complications arise. Phosphorus-31 NMR exercise stress testing of healthy volunteers who are free of signi®cant CAD appears to elicit no signi®cant alterations in anterior myocardial PCr/ATP ratios, with a 30±70% increase in rate±pressure product.5,73 Dobutamine stress testing at 230% rate±pressure product produced no signi®cant change in seven normal controls, although a recent study of 20 healthy volunteers elicited a statistically signi®cant 14% reduction in myocardial PCr/ATP at a 300% rate±pressure product.45 There are two papers reporting 31P ®ndings from isometric hand-grip stress testing of patients with documented anterior wall ischemia.5,42 In the ®rst hand-grip exercise, the PCr/ATP ratio decreased by 37% in a group of 16 patients with severe anterior stenoses (Figure 3).5 After exercise, metabolite ratios recovered to near-normal, pre-exercise values. Eleven healthy control subjects, and nine patients with nonischemic heart disease (cardiomyopathy or valve disease) exhibited no PCr/ATP changes during the same exercise, suggesting that the stressinduced changes in PCr/ATP are speci®c to CAD. Five CAD patients underwent repeat 31P NMR stress-testing after successful revascularization therapy.5 Prior to therapy, stress provoked a 33% decrease in PCr/ATP in these patients, as in the larger group. Exercise stress testing performed posttherapy produced no change, indicating that the metabolic abnormality resolves with successful clinical outcome. The second report found a similar 40% decrease in the PCr/ ATP ratio in patients with reversible anterior wall ischemia that was con®rmed by exercise 201Tl radionuclide imaging.42 Twelve patients with ®xed 201Tl defects indicative of myocardial infarction, as well as normal controls, exhibited no exercise induced PCr/ATP changes. This, and the observation that dobutamine stress testing induced no signi®cant reduction in myocardial PCr/ATP, on average, in patients with DCM,63

PCr/ATP

Myocardial Ischemia

1.5

1.0

0.5

(b)

Rest

Exercise

Recovery

Rest

Exercise

Recovery

2.5

2.0 PCr/ATP

4.3.2

1.5

1.0

0.5

Figure 3 Transient changes in the anterior myocardial PCr/ATP ratio in (a) control subjects free of CAD, and (b) patients with CAD involving the anterior wall, in response to continuous isometric handgrip exercise at 30% of the subjects maximum force.5 Error bars show means ‹SD. (Reproduced by permission of New England Journal of Medicine from Weiss et al.5)

is further evidence that the stress-induced changes may be speci®c for ischemia. Present 31P NMR studies are limited to the anterior wall. Whether 31P NMR stress testing might play a role in the evaluation of ischemia in the clinic may depend in part on extending 31P NMR to other regions of the heart, which will necessitate further improvements in sensitivity such as those afforded by NOE and phased array detection coils. Comparative studies of sensitivity relative to existing clinical modalities would then be needed. 4.3.3

Cardiac Model

In summary, the 31P studies of myocardial infarction and reversible ischemia suggest a model in which the cardiac spectrum from a voxel comprises of up to four components: (i) myocardium either with an essentially normal resting PCr/ATP ratio or, possibly, a reduced resting PCr/ATP ratio due to complications arising from chronic infarction such as heart failure or cardiomyopathy; (ii) jeopardized myocardium whose PCr/ ATP ratio decreases with stress testing; (iii) scar tissue with lower ATP and PCr concentrations; and (iv) infarcted myocar-

NMR SPECTROSCOPY OF THE HUMAN HEART

,,,,, ,,,,, ,,,,, DPG

Nonjeopardized and jeopardized tissue

Blood

+

=

or

Pi?

+

9

Infarction

Observed

(Jeopardized tissue)

Stress

Figure 4 Model for the observed cardiac spectrum as the integral of four potentially distinguishable components: (i) myocardium either with an essentially normal resting PCr/ATP or, possibly, a reduced resting PCr/ATP ratio associated with heart failure; (ii) jeopardized myocardium, possibly only distinguishable from normal or failing heart by stress testing; (iii) a mixture of infarction with no PCr or ATP but perhaps some residual Pi shortly after infarction, and scar tissue, presumably with lower ATP and PCr concentrations; and (iv) contaminating signal from blood10

dium with no PCr, ATP, or CR but possibly some residual Pi persisting several days after infarction. Appropriate choice of the 31P or 1H NMR protocols may resolve these components: measurements of the metabolite concentrations at rest may index the fraction of scar tissue and infarction present in a selected volume, while jeopardized myocardium may be indexed by stress testing (Figure 4).10

5 RELATED ARTICLES Cardiac Gating Practice; Cardiovascular NMR to Study Function; Chemical Shift Imaging; Heart: Clinical Applications of MRI; Magnetization Transfer between Water and Macromolecules in Proton MRI; Proton Decoupling During In Vivo Whole Body Phosphorus MRS; Proton Decoupling in Whole Body Carbon-13 MRS; Rotating Frame Methods for Spectroscopic Localization; Selective Excitation in MRI and MR Spectroscopy; Single Voxel Whole Body Phosphorus MRS.

6 REFERENCES 1. P. A. Bottomley, Science, 1985, 229, 769. 2. P. A. Bottomley, Radiol. 1989, 170, 1. 3. R. S. Menon, K. Hendrich, X. Hu, and K. Ugurbil, Magn. Reson. Med., 1992, 26, 368. 4. P. A. Bottomley and R. G. Weiss, Lancet, 1998, 351, 714. 5. R. G. Weiss, P. A. Bottomley, C. J. Hardy, and G. Gerstenblith, N. Engl. J. Med., 1990, 323, 1593. 6. R. B. Jennings and K. A. Reimer, Am. J. Pathol., 1981, 102, 241. 7. J. L. Swain, R. L. Sabina, R. B. Peyton, R. N. Jones, A. S. Wechsler, and E. W. Holmes, Proc. Natl. Acad. Sci. USA, 1982, 79, 655.

8. J. S. Ingwall, M. F. Kramer, M. A. Fifer, B. H. Lorell, R. Shemin, W. Grossman, and P. D. Allen, N. Engl. J. Med., 1985, 313, 1050. 9. J. S. Ingwall, Circulation, 1993, 87(Suppl. VII), VII-58. 10. P. A. Bottomley, Radiology, 1994, 191, 593. 11. P. A. Bottomley, C. J. Hardy, P. B. Roemer, and O. M. Mueller, Magn. Reson. Med., 1989, 12, 348. 12. J. B. Ra, S. K. Hilal, C. H. Oh, and J. K. Mun, Magn. Reson. Med., 1988, 7, 11. 13. T. B. Parish, D. S. Fieno, S. W. Fitzgerald, and R. M. Judd, Magn. Reson. Med., 1997, 38, 653. 14. R. J. Kim, J. O. A. C. Lima, E. L. Chen, S. B. Reeder, F. J. Klocke, E. A. Zerhouni, and R. M. Judd, Circulation, 1997, 95, 1877. 15. J. R. Whitman, B. Chance, H. Bode, J. Maris, J. Haselgrove, R. Kelley, B. J. Clark, and A. H. Harken, J. Am. Coll. Cardiol., 1985, 5, 745. 16. B. Rajagopalan, M. J. Blackledge, W. J. McKenna, N. Bolas, and G. K. Radda, Ann. N.Y. Acad. Sci., 1987, 508, 321. 17. P. A. Bottomley, R. J. Herfkens, L. S. Smith, and T. M. Bashore, Radiology, 1987, 165, 703. 18. S. Schaefer, J. Gober, M. Valenza, G. S. Karczmar, G. B. Matson, S. A. Camacho, E. H. Botvinick, B. Massie, and M. W. Weiner, J. Am. Coll. Cardiol., 1988, 12, 1449. 19. P. A. Bottomley, C. J. Hardy, and P. B. Roemer, Magn. Reson. Med., 1990, 14, 425. 20. P. A. Bottomley, R. G. Weiss, C. J. Hardy, and W. A. Baumgartner, Radiology, 1991, 181, 67. 21. P. A. Bottomley, US patent, 4 480 228, Oct. 30, 1984. 22. J. Frahm, H. Bruhn, M. L. Gyngell, K. D. Merbold, W. Hanicke, and R. Sauter, Magn. Reson. Med., 1989, 9, 79. 23. P. A. Bottomley and C. J. Hardy, Philos. Trans. R. Soc. London, Ser. A, 1990, 333, 531. 24. C. J. Hardy, P. A. Bottomley, K. W. Rohling, and P. B. Roemer, Magn. Reson. Med., 1992, 28, 54. 25. P. B. Roemer, W. A. Edelstein, C. E. Hayes, S. P. Souza, and O. M. Mueller, Magn. Reson. Med., 1990, 16, 192.

10 NMR SPECTROSCOPY OF THE HUMAN HEART 26. P. A. Bottomley and C. J. Hardy, Magn. Reson. Med., 1992, 24, 384. 27. S. Schaefer, J. R. Gober, G. G. Schwartz, D. B. Twieg, M. W. Weiner, and B. Massie, Am. J. Cardiol. 1990, 65, 1154. 28. Y. Masuda, Y. Tateno, H. Ikehira, T. Hashimoto, F. Shishido, M. Sekiya, Y. Imazeki, H. Imai, S. Watanabe, and Y. Inagaki, Jap. Circ. J., 1992, 56, 620. 29. P. A. Bottomley, C. J. Hardy, and R. G. Weiss, J. Magn. Reson., 1991, 95, 341. 30. P. A. Bottomley and R. Ouwerkerk, Magn. Reson. Med., 1994, 32, 137. 31. T. Yabe, K. Mitsunami, T. Inubushi, and M. Kinoshita, Circulation, 1995, 92, 15. 32. P. A. Bottomley, E. A. Atalar, and R. G. Weiss, Magn. Reson. Med., 1996, 35, 664. 33. A. de Roos, J. Doornbos, P. R. Luyten, L. J. M. P. Oosterwaal, E. E. van der Wall, and J. A. den Hollander, J. Magn. Reson. Imaging, 1992, 2, 711. 34. P. A. Bottomley and C. J. Hardy, J. Magn. Reson., 1992, 99, 443. 35. H. Kolem, R. Sauter, M. Friedrich, M. Schneider, and K. Wicklow, in `Cardiovascular Applications of Magnetic Resonance', ed. G. M. Pohost, Futura, Mt. Kisco, NY, 1993, p. 417. 36. C. J. Hardy, R. G. Weiss, P. A. Bottomley, and G. Gerstenblith, Am. Heart J., 1991, 122, 795. 37. S. Minakami, C. Suzuki, T. Saito, and H. Yoshikawa, J. Biochem., 1965, 58, 543. 38. E. Beutler, in `Hematology', 3rd edn., eds. W. J. Williams, E. Beutler, A. J. Erslev, and M. A. Lichtman, McGraw-Hill, New York, 1983, p. 283. 39. H. Sakuma, K. Takeda, T. Tagami, T. Nakagawa, S. Okamoto, T. Konishi, and T. Nakano, Am. Heart J., 1993, 125, 1323. 40. S. Neubauer, T. Krahe, R. Schindler, M. Horn, H. Hillenbrand, C. Entzeroth, H. Mader, E. P. Kromer, G. A. J. Riegger, K. Lackner, and G. Ertl, Circulation, 1992, 86, 1810. 41. M. Horn, S. Neubauer, M. Bomhard, M. Kadgien, K. Schnackerz, and G. Ertl, Magma, 1993, 1, 55±60. 42. T. Yabe, K. Mitsunami, M. Okada, S. Morikawa, T. Inubushi, and M. Kinoshita, Circulation, 1994, 89, 1709. 43. S. Neubauer, M. Horn, M. Cramer, K. Harre, J. B. Newell, W. Peters, T. Pabst, G. Ertl, D. Hahn, J. S. Ingwall, and K. Kochsiek, Circulation, 1997, 96, 2190. 44. S. Neubauer, M. Horn, T. Pabst, K. Harre, H. Stromer, G. Bertsch, J. Sandstede, G. Ertl, D. Hahn, and K. Kochsiek, J. Invest. Med., 1997, 45, 453. 45. H. J. Lamb, H. P. Beyerbacht, R. Ouwerkerk, J. Doornbos, B. M. Pluim, E. E. van der Wall, A. van der Laarse, and A. de Roos, Circulation, 1997, 96, 2969. 46. R. Kalil-Filho, C. P. de Albuquerque, R. G. Weiss, A. Mocelim, G. Bellotti, G. Cerri, and F. Pileggi, J. Am. Coll. Cardiol., 1997, 130, 1228. 47. W. Jung, L. Sieverding, J. Breuer, T. Hoess, S. Widmaier, O. Schmidt, M. Bunse, F. van Erckelens, J. Apitz, O. Lutz, and G. J. Dietze, Circulation, 1998, 97, 2536. 48. M. A. Conway, P. A. Bottomley, R. Ouwerkerk, G. K. Radda, and B. Rajagopalan, Circulation, 1998, 97, 1716. 49. S. M. Humphrey and P. B. Garlick, Am. J. Physiol., 1991, 260, H6. 50. P. B. Garlick and R. M. Townsend, Am. J. Physiol., 1992, 263, H497. 51. D. G. Gadian, `Nuclear Magnetic Resonance and its Applications to Living Systems', Oxford University Press, Oxford, 1982, p. 30. 52. K. M. Brindle, B. Rajagopalan, N. M. Bolas, and G. K. Radda, J. Magn. Reson., 1987, 74, 356. 53. D. L. Altman and D. A. Dittmer, FASEB, 1961, 21, 29. 54. M. A. Conway, J. Allis, R. Ouwerkerk, T. Niioka, B. Rajagopalan, and G. K. Radda GK, Lancet, 1991, 338, 973.

55. K. Mitsunami, M. Okada, T. Inoue, M. Hachisuka, M. Kinoshita, and T. Inubishi, Jap. Circ. J., 1992, 56, 614. 56. W. Markiewicz, S. Wu, W. W. Parmley, C. B. Higgins, R. Sievers, T. L. James, J. Wikman-Coffelt, and G. Jasmin, Circ. Res., 1986, 59, 597. 57. S. A. Camacho, J. Wikman-Coffelt, S. T. Wu, T. A. Watters, E. H. Botvinick, R. Sievers, T. L. James, G. Jasmin, and W. W. Parmley, Circulation, 1988, 77, 712. 58. S. Wu, R. White, J. Wikman-Coffelt, R. Sievers, M. Wendland, J. Garrett, C. B. Higgins, T. James, and W. W. Parmley, Circulation, 1987, 75, 1058. 59. K. Nicolay, W. P. Aue, J. Seelig, C. J. A. van Echteld, T. J. C. Ruigrok, and B. de Kruijff, Biochim. Biophys. Acta, 1987, 929, 5. 60. S. J. Kopp, L. M. Klevay, and J. M. Feliksik, Am. J. Physiol., 1983, 245, H855. 61. N. Afzal, P. K. Ganguly, K. S. Dhalla, G. N. Pierce, P. K. Singal, and N. S. Dhalla, Diabetes, 1988, 37, 936. 62. W. Auffermann, W. M. Chew, C. L. Wolfe, N. J. Tavares, W. W. Parmley, R. C. Semelka, T. Donnelly, K. Chatterjee, and C. B. Higgins, Radiology, 1991, 179, 253. 63. S. Schaefer, G. G. Schwartz, S. K. Steinman, D. J. Meyerhoff, B. M. Massie, and W. M. Weiner, Magn. Reson. Med., 1992, 25, 260. 64. S. M. Krause, Heart Failure, 1988, 267. 65. Editorial, Lancet, 1991, 338, 981. 66. R. C. Canby, W. T. Evanochko, L. V. Barrett, J. K. Kirklin, D. C. McGiffen, T. T. Sakai, M. E. Brown, R. E. Foster, R. C. Reeves, and G. M. Pohost, J. Am. Coll. Cardiol., 1987, 9, 1067. 67. C. E. Haug, J. L. Shapiro, L. Chan, and R. Weil, Transplantation, 1987, 44, 175. 68. C. D. Fraser, V. P. Chacko, W. E. Jacobus, R. L. Soulen, G. M. Hutchins, B. A. Reitz, and W. A. Baumgartner, Transplantation, 1988, 46, 346. 69. C. D. Fraser, V. P. Chacko, W. E. Jacobus, P. Mueller, R. L. Soulen, G. M. Hutchins, B. A. Reitz, and W. A. Baumgartner, J. Heart Transplant., 1990, 9, 197. 70. C. D. Fraser, V. P. Chacko, W. E. Jacobus, G. M. Hutchins, J. Glickson, B. A. Reitz, and W. A. Baumgartner, Transplantation, 1989, 48, 1068. 71. J. O. van Dobbenburgh, N. de Jonge, C. Klopping, J. R. Lahpor, S. R. Woolley, and C. J. A. van Echteld, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, 3, 1093. 72. P. A. Bottomley, L. S. Smith, S. Brazzamano, L. W. Hedlund, R. W. Redington, and R. J. Herfkens, Magn. Reson. Med., 1987, 5, 129. 73. M. A. Conway, J. D. Bristow, M. J. Blackledge, B. Rajagopalan, and G. K. Radda, Br. Heart J., 1991, 65, 25.

Acknowledgements I would like to thank R. G. Weiss, M. Conway, and B. Rajagopalan for many helpful discussions.

Biographical Sketch Paul A. Bottomley. b 1953. B.Sc. Hon., 1974, Ph.D. 1978, University of Nottingham, UK, with E. Raymond Andrew and Waldo Hinshaw. Research Associate at Johns Hopkins Medical Institutions, 1978±1980. Physicist, G.E. Research and Development Center, 1980±1994. Presently Russell H. Morgan Professor and Director, Division of MR Research, Dept of Radiology, Johns Hopkins University, Baltimore, U.S.A. Approx. 120 publications, 130 conference reports, 27 patents. Gold Medal, Society of Magnetic Resonance in Medicine, 1989; G.E. Coolidge Medal and Fellow, 1990. Research specialties: in vivo NMR imaging, localized spectroscopy, tissue relaxation times, MRI, human cardiac NMR spectroscopy.

BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS

Brain Infection and Degenerative Disease Studied by Proton MRS Robert E. Lenkinski, Dolores LoÂpez-Villegas, and Supoch Tunlayadechanont

1

the observation of lactate to indicate the presence of an infarct. 1.3

Intracranial Tuberculomas

Gupta et al.5 have reported the results of localized proton MRS carried out on two patients with intracranial tuberculo-

University of Pennsylvania, Philadelphia, PA, USA NAA

1 INFECTIONS Cho Cr

1.1 Creutzfeldt±Jakob Disease (CJD) Bruhn et al.1 showed in an initial case report of CJD that there was signi®cant loss of N-acetylaspartate (NAA) in both white matter (40%) and gray matter (30%). The level of inositol, presumably myo-inositol (MI), was higher than normal (30%) in white matter. The conventional magnetic resonance (MRI) images of this patient showed only mild cortical atrophy and hyperintensities in the lentiform nuclei. These authors suggested that proton MRS might be useful in the early detection of CJD. Graham et al.2 have studied two patients with biopsy proven CJD using in vivo MRS. High-resolution NMR spectroscopy was also carried out on an extract of a biopsy specimen of a third patient with CJD.2 These NMR results were compared with neuronal cell counts. Marked decreases in the levels of NAA (detected in vivo) were observed at later stages of disease in two patients. However in the early stages of CJD smaller decreases were observed (15% and 27%) in the level of NAA. The NMR spectrum obtained from the biopsy sample taken from a third patient showed little or no reduction in the levels of metabolites. In contrast with the previous case report of Bruhn et al.1 Graham et al. suggested that there should be little or no change in the levels of metabolites in the early stages of CJD. This suggestion was supported by referring to the known pathophysiology of CJD which indicates that neurons are lost relatively late in the disease process.

3.5

3.0

(a)

2.5 2.0 1.5 Chemical shift (ppm)

1.0

0.5

Cho

Cr NAA Lac

1.2 Herpes Simplex Encephalitis (HSE) Menon et al.3 showed in an early case report that the ratio of NAA to choline (NAA/Cho) was reduced in an 11 year old boy. This reduction observed at eight weeks after the onset of symptoms was found to be unaltered on follow-up examination (16 weeks), indicating that there was no progressive neuronal loss. This observation was consistent with the clinical evaluation of the patient. As part of an MRI study of HSE involving eight patients, Demaerel et al.4 obtained proton spectra from two patients with HSE. The proton spectra from one of the HSE subjects are shown in Figure 1. Note the reduction in NAA observed in the affected region as compared with the contralateral temporal lobe. There is also a signi®cant amount of lactate observed. The authors interpret

3.5 (b)

3.0

2.5 2.0 1.5 Chemical shift (ppm)

1.0

0.5

Figure 1 Solvent-suppressed proton spectra obtained using the STEAM sequence at 1.5 T (TR = 3000 ms, TE = 270 ms) from a 27 cm3 voxel located at (a) the left and (b) the right temporal lobe. Each spectrum is the sum of 256 acquisitions. Reproduced with permission from Demaerel et al.4

2 BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS Lipid 4.5 4.0 3.5 3.0

Signal

2.5 2.0 1.5 1.0 0.5 0.0 3.2

2.8 3.0

2.4 2.6

2.0 2.2

1.6 1.8

1.2 1.4

0.8 1.0

0.4 0.6

0.2

Chemical shift (ppm)

(a)

A

2.2 2.0 1.8 B

1.6 Signal

1.4

C

1.2 1.0 D

0.8 0.6 0.4 0.2 0.0 3.8 (b)

3.4

3.0

2.6 2.2 1.8 1.4 1.0 Chemical shift (ppm)

0.6

0.2

Figure 2 Solvent-suppressed proton spectra obtained using the STEAM sequence at 2.0 T (TR = 5000 ms, TE = 20 ms) from an 8 cm3 voxel located (a) in a tuberculoma and (b) in a contralateral normal brain. Each spectrum is the sum of 128 acquisitions. Reproduced with permission from Gupta et al.5

mas. A large resonance was observed between 0.7±1.6 ppm (see Figure 2) which was assigned to the (CH2)n group of saturated fatty acids. The presence of this peak was attributed to the large lipid fraction present in the tubercule bacillus. 1.4 Human Immunode®ciency Virus (HIV) Infection Menon et al.6 reported the ®rst spectroscopic study of the brains in patients infected by Human Immunode®ciency Virus

(HIV). The authors examined two patients with AIDS [Centers for Disease Control classi®cation (CDC) Group IV] and evidence of focal CNS pathology. A multiple-echo spectroscopy acquisition (MESA) sequence with a TR of 2000 ms and TE of 270 ms was employed. The spectra were collected from a 64 cm3 voxel in the right parietal region. Although these patients had abnormal MRI, the spectra were taken in areas of normal appearing white matter. In both cases the spectra showed a marked reduction in the ratio of NAA/Cho and NAA/creatine (NAA/Cr) when compared with control subjects. The authors proposed that the spectral changes in these regions which appear normal on MRI may be attributed to other causes but is most likely to be due to primary HIV infection. The authors ®rst demonstrated the ability of proton MRS to detect brain abnormalities in HIV-infected patients at a stage when it is undetectable by imaging and proposed a possible role for the early selection of patients for treatment with antiviral drugs. The same group examined 11 patients with HIV infection and varying stages of AIDS dementia complex (ADC) (actually known as HIV-1-associated cognitive/motor complex).7 In this study, patients with pathology seen on MR imaging that could not be directly attributed to HIV infection of the brain were excluded from the study. Spectra were obtained from 27 to 64 cm3 voxels in the parieto±temporal regions using the same sequence as in the previous study. Spectra from patients with moderate (Stage 2) to severe ADC (Stage 3), when compared with spectra from normal volunteers, exhibited signi®cant reductions in NAA/Cr ratio and a tendency to increased Cho/Cr ratio, although this last trend did not reach statistical signi®cance. Spectra from patients with no ADC (Stage 0) or early ADC (Stage 1) were not signi®cantly different from normal volunteers (see Figures 3 and 4). Many of the patients in this study exhibited abnormalities on MRI, but apart from the presence of atrophy, which was seen in all of the patients with moderate to severe ADC, MRI did not seem to discriminate between patients with and without ADC. The authors concluded that although the NAA/Cr ratio may not be an early or sensitive marker of ADC, it may be relatively speci®c since all of the patients with signi®cantly low values of this ratio had a clinical diagnosis of ADC. Meyerhoff et al.8 examined 14 HIV seropositive patients, 10 with varying degrees of cognitive impairment and four who were cognitively asymptomatic. Spectra were obtained from nine 2.5 cm3 volumes in the centrum semiovale and the mesial cortex in each patient. Signi®cantly reduced NAA/Cho and NAA/Cr ratios were observed in cognitively impaired subjects versus normal controls, without signi®cant regional differences between the voxels studied. No signi®cant differences were found between groups with cognitive impairment and asymptomatic groups or between asymtomatic and control groups. This study reported diffuse reductions in NAA in individuals with cognitive impairment due to HIV. Contrary to previous studies, most of the patients in this study had normal appearing MRI (80%) suggesting that proton MRS is more sensitive than imaging in assessing the effects of HIV infection on the brain. Jarvik et al.9 examined 11 HIV seropositive patients without clinical, radiological, or laboratory evidence of CNS infections other than HIV. Proton spectra were acquired with the stimulated echo acquisition mode (STEAM), TR of 2000 ms, TE of

BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS

3

NAA

(a)

(b)

(c)

(d)

Cr Cho

3.0

2.0

3.0

ppm

2.0 ppm

3.0

2.0 ppm

3.0

2.0 ppm

Figure 3 Solvent-suppressed proton spectra obtained from a 27 cm3 voxel located in the parietal lobe from a healthy volunteer and three patients with HIV infection. The spectra were obtained with a double-echo sequence, TR = 2000 ms, TE = 270 ms. Each spectrum is the sum of 128 acquisitions. The spectra are displayed with creatine scaled similarly. The spectra are from (a) a healthy volunteer, (b) a patient with ADC Stage 0, (c) a patient with ADC Stage 1, and (d) a patient with ADC Stage 3. Reproduced with permission from Menon et al.7

combined these three ratios. This aggregate score proved to be a good discriminator between the patient and control populations (P = 0.001) (see Figure 5). The aggregate scores were abnormal (>2 SDs from the mean of the control subjects) in 13 out of 15 patient spectra (87%), while 8 out of 11 patients (73%) had abnormal MR images. Moreover, only one out of 10 control spectra (10%) was abnormal while four out of 11 controls (36%) had abnormal imaging. These results suggested that MRS may be more sensitive and speci®c than MR imaging in detecting CNS involvement in HIV-infected patients. Chong et al.10 reported the largest study in which proton MRS was performed in 103 HIV seropositive patients and 23 control subjects. Spectra were collected from an 8 cm3 voxel placed in a normal parieto±occipital region of the brain using a 90 , 180 , 180 spin echo sequence (PRESS), TR of 1600 ms and TE of 135 ms. In the ®rst part of the study, the spectra of HIV seropositive patients were compared and correlated with clinical, inmunologic, and radiologic measures of HIV infection. A signi®cant reduction in the NAA/Cr ratio was seen in patients with late-stage disease (CDC Group IV). The NAA/Cr and NAA/Cho ratios were also reduced in patients with CD4 counts <200 mm±3 and in patients with neurologic signs. Signi®cant increases in Cho/Cr ratios were seen in patients with low CD4 counts and abnormal MR images. Reduced NAA ratios correlated with diffuse but not focal MR imaging abnormalities. In the second part of the study, the authors evaluated the utility of combining the results of MR imaging and spectroscopy, ®nding that the combination of both modalities provides closer relationships to clinical and immunologic measures of disease than either modality alone. Moreover, abnormal spectra correlated more with abnormal neurologic ®nding than did abnormal MR images, suggesting that spec-

,   

,
,
,   ,

,
,
, ,  

19 ms, and TM of 10.6 ms. Voxels of 3.4±8 cm3 were chosen to cover areas of abnormal white matter signal intensity if present, or centrum semiovale if the white matter appeared normal at imaging. Analysis of the images showed that there was a signi®cant difference between the patients and control subjects with respect to atrophy, although no signi®cant difference was found between the appearance of the white matter in patients and control subjects. Analysis of the spectra showed that the NAA/Cr ratio was signi®cantly lower and Cho/Cr and marker peak/Cr ratios signi®cantly higher in patients as compared with control subject. The authors calculated an aggregate score that

4

3.5 3

*

2.5 2

1.5 1

0.5 0

Normal

ADC ST 0

ADC ST 1

ADC ST 2/3

Figure 4 A comparison of the NAA/Cr ratios in normals and patients with different stages of ADC. The bars represent 1 SD. The NAA/Cr ratio is signi®cantly reduced in the ADC Stage 2 and 3 as compared with any of the other groups (P < 0.05). Reproduced with permission from Menon et al.7

4 BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS 8 7 Aggregate

6

+sd

5

m

4 3

–sd

2 1 0 Observations (a) Normal volunteers 8 7 Aggregate

6 5 4 3

+sd m –sd

2 1 0 Observations (b) Patients

Figure 5 Aggregate spectral scores (Cr/NAA + Cho/Cr + marker/Cr) obtained from normal volunteers and patients with HIV infection. The mean and 1 SD lines are indicated. Note the almost complete separation between the two groups. See Jarvik et al. for experimental details9

troscopy may be sensitive to changes in cerebral chemistry that are clinically relevant and are not apparent on MR images. The same authors11 studied 43 HIV seropositive patients, including 26 who had clinical or radiologic evidence of latestage disease (CDC Group IV) and 17 in the early stages of infection (CDC Groups II and III). Using the same sequence as described above, spectra were obtained from a 8 cm3 voxel placed in a normal parieto-occipital region of the brain. When patients grouped by different criteria were compared, a signi®cant reduction in the NAA/(NAA + Cho + Cr) ratio was found in patients in the late stage of the disease as compared with those in early stage disease. This ratio was lower in patients with HIV-1associated cognitive/motor complex as compared with patients who were neurologically healthy. Also this ratio was decreased in patients with abnormal MR images (diffuse white matter abnormalities) as compared with those with normal appearing white matter. When patients were compared with healthy control subjects, the NAA/(NAA + Cho + Cr) ratios were signi®cantly higher in control subjects than in CDC Group II and III patients and also signi®cantly higher than in all seropositive patients with normal MR imaging. The authors noted, however, that the control group was younger and so these last results may be in¯uenced by the expected decrease in NAA levels which parallel neuronal loss with age. The authors reported the ®rst followup spectroscopic study performed in 15 patients between three and eight months after their initial studies. A signi®cant reduction in the NAA/(NAA + Cho + Cr) ratio was observed at follow-up study when compared with the initial examinations.

The results of all of the MRS studies of HIV infection are summarized in Table 1. A common ®nding in all of the studies on HIV infection of the brain is the reduced level of NAA. Although NAA, which resonates at 2.0 ppm, is the most prominent resonance in the brain proton spectrum, its precise biochemical function is not clear. It has been suggested that NAA is involved in the regulation of lipid synthesis, the regulation of protein synthesis, or as a storage buffer for aspartate. Since NAA is largely con®ned to neurons it has been proposed as a neuronal marker. Hence, decreases in the level of NAA may be interpreted as a sign of neuronal loss or injury. This ®nding is consistent with the pathologic evidence of neuronal loss reported in HIV infected patients. Another common ®nding, although not present in all of the studies, is the increase in Cho/Cr ratio. Choline, which resonates at 3.2 ppm, is an important precursor of cell membrane synthesis and is often found to be elevated in tissues that are rapidly regenerating or undergoing membrane disruption. The signi®cance of the elevation in Cho/Cr ratio remains uncertain. A possible interpretation for this increase in HIV infected patients is that the increased Cho may not directly arise from changes in neurons but may result from metabolic alterations in glial or in¯ammatory cells.7 An alternative interpretation is that the increase in Cho/Cr ratio combined with the increase in marker/Cr ratio, also observed in these patients, may re¯ect myelin damage.9 Choline phosphoglycerides contribute 11.2% of myelin lipids, with phosphatidylcholine being the most abundant. Therefore, as myelin damage occurs, free choline may be released increasing the choline resonance detected by MRS. The `marker peak' region between 2.1±2.6 ppm may be a combination of nonspeci®c amino acids and possibly myelin catabolites which may increase with myelin breakdown. Both explanations for the increase in Cho/Cr ratio are consistent with the neuropathologic ®nding of in¯ammatory in®ltrates and white matter abnormalities in the brain of HIV infected patients. An alternative explanation proposed for the increase in this ratio is the decrease in Cr resonance.10 This resonance at 3.0 ppm re¯ects the concentration of the total creatine pool (PCr and Cr). The level of this peak remains stable under many conditions in the brain. The decrease in creatine pool may indicate impairment of cellular metabolism in the brain of these patients and is supported by the decrease in PCr peak observed in phosphorus MRS studies. Abnormal proton spectra may be found in normal-appearing white matter on imaging, suggesting that MRS may be more sensitive than MRI in detecting CNS involvement in HIV infected patients. Moreover, MRS seems to be a better discriminator between the patient and control populations than MRI, suggesting that MRS may be also more speci®c. When individuals displaying different manifestations of illness were compared, the abnormalities in the spectra correlated with the presence and the severity of the cognitive impairment and with clinical and immunologic measures of late-stage disease. These ®ndings suggest that MRS may serve as an indicator of the degree of CNS involvement. Therefore, MRS may serve to monitor the course of disease progression and may have a prognostic role regarding CNS involvement. This modality may provide a sensitive method for the early detection of HIV migration to the CNS. Finally, MRS can be employed in longitudinal studies to monitor the response to therapy and thus may lead to individual optimized treatment effectiveness.

Table 1 A Summary of the Proton MRS Findings in HIV infection Study

Patients (p) Stagea Controls (c)

Menon et al. (1990)

2p/6c

late-stage

2

Menon et al. (1992)

11p/8c

late-stage

10

early-stage Meyerhoff et al. (1993) 14p/7c

Chong et al. (1993)d

Chong et al. (1994)d

a

11p/8c

103p/23c

43p/8c

MRS sequence and parameters

MRS ®ndings

Groups statistically distinguished by MRS

cogn. imp. + focal signs

1 1

focals lesions not attributed to HIV

2

MESA TR 2000 ms TE 270 ms V = 64 cm3 normal parietal region

;NAA/Cho ;NAA/Cr

patients/controls

cogn. imp. (ADC)

7

abnormal

9

MESA TR 2000 ms TE 270 ms

;NAA/Cr

1

no cogn. imp

4

normal

2

V = 27±64 cm3 parieto-temporal region

:Cho/Cr (NS)

ADC Stage controls

late-stage

4

cogn. imp.

abnormal

3

CSI

;NAA/Cr

early-stage

10

late-stage

10

no cogn. imp

4

normal

4

cogn. imp.

9

abnormal

8

early-stage

7

no cogn. imp.

2

normal

3

late-stage

70

neurologic signs

early-stage

22

no neurologic signs 31

late-stage

26

early-stage

17

cogn. imp. (HIV-1-associated cognitive/motor complex) no cogn. imp.

19

14

V = 2.5 cm3 centrum semiovale and ;NAA/Cho mesial cortex

patients/ controls

late-stage/earlystage CD4 count neurologic signs/no neurologic signs abnormal MRI/ normal MRI

34e

PRESS TR 1600 ms TE 135 ms

;NAA/Cr

normal

36

V = 8 cm3 normal parieto-occipital

;NAA/Cho

region

:Cho/Cr

PRESS TR 1600 ms TE 135 ms

;NAA/NAA + Cho + Cr

normal

9 34

cogn. impair/ controls

STEAM TR 2000 ms TE 19 ms TM;NAA/Cr 10.6 ms V = 3.4±8 cm3abnormal white matter :Cho/Cr or centrum semiovale : Marker peak/Cr

abnormal

abnormal 6

11

2/3/

V = 8 cm3 normal parietooccipital region

late-stage/early stage cogn imp/no cogn imp abnormal MRI/ normal MRI early-stage/controls normal MRI patients/controls

Stage: late stage refers to CDC Group IV, early stage to CDC Groups II and III. bADC refers to the aids dementia complex, actually known as HIV-1-associated cognitive/motor complex. Cogn. imp. is an abbreviation for cognitive impairment. cAbnormal MRI: abnormalities were either GM atrophy or WM signal changes. dNot all of the subjects had complete neurological examinations. e The remaining patients (33) were excluded because they exhibited focal lesions on MRI.

BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS

Jarvik et al. (1993)

MRI ®ndingsc

Clinical evidence of CNS involvement

5

6 BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS 60

Mole % GLU

50 (b) 40

NAA

30 Cr Glx

Glx

Lipids

20 0

10

20 Mole % NAA

30

40

MI

(a)

Cho

Figure 6 The variation in the mol% of glutamate with the mol% of NAA in the NMR spectra of extracts of brain samples from patients with AD. See Klunk et al. for experimental details12 4

2 DEGENERATIVE DISEASES 2.1 Alzheimer's Disease (AD) In Vitro Studies of AD

Klunk et al.12 determined the mole ratios of several amino acid metabolites in perchloric extracts of 12 AD and ®ve control brain samples from either the junction of the superior and middle frontal cortex or the superior temporal cortex. NAA and -aminobutyric acid (GABA) were found to be lower in AD. The NAA levels also had a negative correlation to the numbers of senile placques (SP) and neuro®brillary tangles (NFT). As shown in Figure 6, glutamate levels were greater in AD and had an inverse correlation with NAA levels. The authors suggested that these ®ndings applied to in vivo studies and re¯ected neuronal loss, while the remaining neurons were exposed to the potentially excitatotoxic effect of glutamate. 2.1.2

1

2 Chemical shift (ppm)

Figure 7 Representative summed solvent-suppressed proton spectra obtained using the STEAM sequence (TR = 1500 ms, TE = 30 ms) from (a) normal volunteers (n = 8) and (b) patients with AD (n = 8). Reproduced with permission from Miller et al.13

with no signi®cant brain atrophy or reduction in regional blood ¯ow detected by SPECT. There was no reduction of NAA/Cr in PH. These authors concluded that, in the appropriate clinical setting, proton MRS was a useful measure for the early detection and study of PDD (see Figure 8).

Normal NPH PDD

In Vivo Studies of AD

Miller et al.13 used the STEAM sequence, at ®eld strength of 1.5 T and TE 30 ms, to acquire proton spectra from 10±15 ml voxels in white matter located in the parietal area (WM) and gray matter in the occipital cortex (GM). Summed spectra are shown in Figure 7. They reported an increased myoinositol/ creatine (MI/Cr) ratio, which suggested abnormalities in the inositol polyphosphate messenger pathway, and reduced NAA/ Cr in AD compared to controls. The difference in NAA/Cr level was small but statistically signi®cant in both WM and GM voxels. The increase in MI/Cr was more prominent, especially in GM. However, there was no correlation between the severity of the disease and spectroscopic ®ndings. Shiino et al.14 studied a patient group (n = 9) with primary degenerative dementia (PDD) which included seven patients with probable AD, three normal pressure hydrocephalus (NPH), and healthy controls. The DRY STEAM technique was applied with a TR of 2500 ms, TE of 19 ms, and TM of 5.7 ms. The NAA/Cr was signi®cantly reduced in patients with PDD,

2.0 1.8 1.6 NAA/Cr

2.1.1

3

1.4 1.2 1.0 0.8 0.6 20

30

40

50

60

70

80

90

Age (years)

Figure 8 A correlation plot of NAA/Cr versus age. Normals are designated by ®lled circles. Patients with primary degenerative dementia (PDD) are designated with open circles and patients with normal pressure hydrocephalus are designated by triangles. See Shiino et al.14 for experimental details

BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS

2.2 Parkinson's Disease (PD) Shiino et al.14 examined two patients with PD using proton MRS. Spectra were acquired from a 27 cm3 voxel located in the insular area of the brain. A decrease in the NAA/Cr ratio was observed in the two patients with PD when compared with control subjects. Even though the mean age in the control group was lower, the authors observed that there were no agerelated changes in the mean area ratio of NAA/Cr in this group. Both PD patients examined in this study exhibited marked atrophy on MRI, and since the spectra were obtained from voxels placed in the insular area these results should be interpreted with caution. 2.3 Huntington's Disease (HD) Jenkins et al.15 reported the results of a proton MRS study in a series of HD patients. Sixteen patients with clinical signs of de®nite HD and con®rmed family history and two individuals without neurologic affectation but with linked DNA markers, indicating a high probability to have inherited the HD gene, were examined. Proton spectroscopy was performed using a STEAM sequence with TR of 2000 ms, TE of 272 ms, and TM of 10 ms. Spectra were collected from a 15.6 cm3 voxel placed over the visual cortex and from a 5.4±20 cm3 voxel placed in basal ganglia. A 5 inch surface coil was used to study the occipital cortex and the quadrature head coil was used for studies of the basal ganglia. Elevated lactate levels were observed in the occipital cortex in all of the symptomatic patients when compared with normal controls. The lactate level correlated with the duration of illness. Lactate levels were normal in the two asymptomatic subjects with high probability of having the HD gene. In the basal ganglia, the levels of NAA were decreased and the levels of Cho elevated relative to creatine. Several patients also showed elevated lactate levels in the basal ganglia. The authors suggested that the increase in lactate may be explained by a defect in oxidative phosphorylation in HD. This possibility is supported by biochemical studies that have reported reduced mitochondrial enzyme activity in HD patients, and ultrastructural studies that have shown abnormalities in the mitochondria of these patients. The reduced NAA and increased Cho were interpreted in these patients as resulting from neuronal loss and gliosis. The authors suggested that lactate likely precedes neuronal death and that this may be the explanation for the more variable lactate elevation in the basal ganglia where evidence of more neuronal loss and gliosis was found. On this basis, they proposed that the elevated lactate may provide a simple marker to monitor the progression of the disease and possible therapies for HD patients. High-resolution proton MRS was employed by Nicoli et al.16 to study CSF and serum metabolic samples obtained from patients with HD. Serum and CSF samples were collected from 11 patients suffering HD and 12 reference patients suffering miscellaneous neurological diseases. A signi®cant increase in the pyruvate concentration in CSF was found in patients with HD. This ®nding may be explained by the decrease of pyruvate-dehydrogenase and Kreb's cycle enzyme activity observed in HD patients. This observation is consistent, as was the ®nding of elevated lactate reported by Jenkins et al.,15 with

7

the hypothesis of the existence of a defect in oxidative phosphorylation present in this disease.

3

RELATED ARTICLES

Brain MRS of Human Subjects; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Systemically Induced Encephalopathies: Newer Clinical Applications of MRS.

4

REFERENCES

1. H. Bruhn, T. Weber, V. Thorwirth, and J. Frahm, Lancet, 1991, 337, 1610. 2. G. D. Graham, O. A. Petroff, A. M. Blamire, G. Rajkowska, P. Goldman-Rakic, and J. W. Prichard, Neurology, 1993, 43, 2065. 3. D. K. Menon, J. Sargentoni, C. J. Peden, J. D. Bell, I. J. Cox, G. A. Coutts, C. Baudouin, and C. G. Newman, J. Comput. Assist. Tomogr., 1990, 14, 449. 4. Ph. Demaerel, G. Wilms, W. Robberecht, K. Johannik, P. Van Hecke, H. Carlton, and A. L. Baert, Neuroradiology, 1992, 34, 490. 5. R. K. Gupta, R. Pandey, E. M. Kahn, P. Mittal, R. B. Gujral, and D. K. Chhabra, Magn. Reson. Imag. 1993, 11, 443. 6. D. K. Menon, C. J. Baudouin, D. Tomlinson, and C. Hoyle, J. Comput. Assist. Tomogr., 1990, 14, 882. 7. D. K. Menon, J. G. Ainsworth, I. J. Cox, R. C. Coker, J. Sargentoni, G. A. Coutts, C. J. Baudouin, A. E. Kocsis, and J. R. Harris, J. Comput. Assist. Tomogr., 1992, 16, 538. 8. D. J. Meyerhoff, S. Mackay, L. Bachman, N. Poole, W. P. Dillon, M. W. Weiner, and G. Fein, Neurology, 1993, 43, 509. 9. J. G. Jarvik, R. E. Lenkinski, R. I. Grossman, J. M. Gomori, M. D. Schnall, and I. Frank, Radiology, 1993, 186, 739. 10. W. K. Chong, B. Sweeney, I. D. Wilkinson, M. Paley, M. A. Hall-Craggs, B. E. Kendall, J. K. Shepard, M. Beecham, R. F. Miller, I. V. D. Weller, S. P. Newman, and M. J. G. Harrison, Radiology, 1993, 188, 119. 11. W. K. Chong, M. Paley, I. D. Wilkinson, M. A. Hall-Craggs, B. Sweeney, M. J. G. Harrison, R. F. Miller, and B. E. Kendall, AJNR, 1994, 15, 21. 12. W. E. Klunk, K. Panchalingam, J. Moossy, R. J. McClure, J. W. Pettigrew, Neurology, 1992, 42, 1578. 13. B. L. Miller, R. A. Moats, T. Shonk, T. Ernst, S. Woolley, and B. D. Ross, Radiology, 1993, 187, 433. 14. A. Shiino, M. Matsuda, S. Morikawa, T. Inubushi, I. Akiguchi, and J. Handa, Surg. Neurol., 1993, 39, 143. 15. B. G. Jenkins, W. J. Koroshetz, M. F. Beal, and B. R. Rosen, Neurology, 1993, 43, 2689. 16. F. Nicoli, J. Vion-Dury, J. M. Maloteaux, C. Delwaide, S. Confort-Gouny, M. Sciaky, and P. Cozzone, Neurosci. Lett., 1993, 154, 47.

Acknowledgements Supported in part by NIH grant NS 31464 (Robert E. Lenkinski). Dr. LoÂpez-Villegas would like to acknowledge support by the Ministry of Health of Spain (FIS: BAE 94/5033). Dr. Tunlayadechanont would like to acknowledge support from Ramathibodi Hospital in Thailand.

8 BRAIN INFECTION AND DEGENERATIVE DISEASE STUDIED BY PROTON MRS Biographical Sketches Robert E. Lenkinski. b 1947. B.Sc., 1969, University of Toronto, Ph.D., 1973, University of Houston. Postdoctoral Fellow, Weizmann Institute of Science, Isotope Department, Rehovot, Israel, 1973±75. Faculty at University of Houston, 1975±76; Comprehensive Cancer Center, University of Alabama in Birmingham, 1976±80; Department of Chemistry, University of Guelph, Ontario, Canada, 1980±86; University of Pennsylvania, 1986±present. Approx. 92 publications. Research interests include clinical MRS particularly in the diagnosis and staging of disease. Dolores LoÂpez-Villegas. b 1964. M.D., 1988, University of Barcelona, Medical School. Intern in Medicine, Hopital Clinic, University of Barcelona, 1988. Resident in Medicine, Hopital de la Santa Creu i Sant Pau, Autonomous University of Barcelona, 1989. Resident in Neurol-

ogy in the same hospital, 1990±92. Research Fellow in MRS (supported by the Ministry of Health of Spain), University of Pennsylvania, Department of Radiology, 1993±present. Approx. 10 publications. Research interests include application of MRS to investigation of neurological diseases. Supoch Tunlayadechanont. b 1962. M.D., 1986, Ramathibodi Hospital, Mahidol University. Residency training in Neurology, Ramathibodi Hospital, 1989±92; Graduate Diplomate in Clinical Science, 1990, Mahidol University; Diplomate the Thai Board of Neurology, 1992, Thai Medical Council. Training in MRI and MRS, Institute of Magnetic Resonance, University of Innsbruck, Austria, 1993. Research Fellow in MRS, University of Pennsylvania, 1993±present. Approx. 10 publications. Research interests include clinical MRS particularly in neurological disease.

BRAIN MRS OF INFANTS AND CHILDREN

Brain MRS of Infants and Children Ernest B. Cady University College London Hospitals, UK

and E. Osmund R. Reynolds University College London Medical School, UK

1 INTRODUCTION Magnetic resonance spectroscopy (MRS) was ®rst introduced into pediatrics in the early 1980s as a method for the noninvasive investigation of perinatal brain injury.1 This injury is often due to cerebral periventricular hemorrhage (PVH) in preterm infants (born before 37 weeks of gestation) or to hypoxia-ischemia in both preterm and term infants. It is responsible for permanent neurodevelopmental impairments in about 10% of disabled children of school age. In the 1970s, Xray computerized tomography (CT) and, more especially, ultrasound imaging using portable equipment at the cotside, were found to be useful for investigating the incidence, pathogenesis, and prognostic signi®cance of PVH, but long-term follow-up studies indicated that hypoxia-ischemia, often causing periventricular leukomalacia, was a more important cause than PVH of permanent disabilities in preterm infants.2 In term infants, hypoxia-ischemia was the mechanism by which `birth asphyxia' (critically impaired gas exchange during labor) damaged the brain. Overall, hypoxia-ischemia turned out to be the most prevalent cause of serious perinatal brain injury, so there was an obvious need for noninvasive investigation of cerebral oxidative metabolism, and hence of the pathogenesis and evolution of hypoxic-ischemic brain injury. MRS was thought to provide a suitable, noninvasive, method for obtaining biochemical information from the otherwise inaccessible neonatal brain.3±5 In the early 1980s, 31P MRS had already been used to investigate oxidative metabolism in human skeletal muscle in vivo and the ®rst in vivo spectra had been obtained from mammalian brain.6 However, studies of newborn infants had to await the development of superconducting magnets with a suf®ciently wide bore (approximately 20 cm) and a high enough ®eld strength for spectroscopic work. These magnets (intended primarily for studies of human limb muscles), became available in the early 1980s, and investigations of the brain in newborn infants started shortly thereafter.1,7 The ®rst in vivo spectra from the human brain were obtained from a preterm baby at University College London in 1982 using the 31 P nucleus.1 It rapidly became clear that metabolites involved in energy metabolism [e.g. adenosine triphosphate (ATP), phosphocreatine (PCr), and inorganic phosphate (Pi)] as well as those related to more ®xed cellular components, namely phosphomonoesters (PMEs) and phosphodiesters (PDEs), were easily detected. Intracellular pH (pHi) could

1

also be measured. Obvious spectral abnormalities were soon found in the brains of infants with hypoxic ischemic injury.1,7 31 P spectra have also been obtained from the brains of older children and developmental changes have been elucidated. In addition, 1H spectra have demonstrated products of anerobic glycolysis such as lactate (Lac), various amino acids (e.g. glutamate and glutamine), and other brain metabolites, including those containing choline (Cho; related to membrane metabolism) and N-acetylaspartate (NAA; a putative neuronal marker). Localization of spectra to speci®c regions of the brain was initially crude, relying solely on the sensitive volume of a surface coil. Recently, techniques have been developed that enable spectra to be acquired from well de®ned volumes of tissue, so deeper and focal lesions can be investigated. The purposes of this article are to summarize the methods used for MRS of the brain in infants and children, to describe what has been learnt about normal brain development, and then to consider the role of MRS, particularly in the investigation of hypoxic ischemic brain injury.

2

PATIENT MANAGEMENT

MRS studies of the brains of infants and children are safe provided that current recommendations are observed,8 though the requirements for physiological support and monitoring are often much greater than for adults. Newborn infants, who may be in an unstable condition, have to be conveyed and studied in a specially designed transport incubator incorporating a nonmagnetic, cylindrical pod that encloses the baby and is suitable for insertion into the magnet. Conventional facilities for temperature control, monitoring, and mechanical ventilation are provided.9,10 Monitoring of heart rate [by electrocardiogram (ECG) or ultrasound Doppler ¯ow], blood pressure, arterial oxygen saturation (by pulse oximetry), end-tidal CO2, transcutaneous pO2 and pCO2, and body temperature (by skin and/or rectal sensors) should be available. Intravenous infusions may have to be provided. Newborn infants can often be studied satisfactorily whilst naturally asleep following a feed. However, sedation with, for example, chloral hydrate is often required for studies of older children. Especially for the newborn, the head must be gently immobilized during studies using, for example, a polythene bag containing expanded polystyrene beads and which can be evacuated. Clinical apparatus must be carefully designed to take into account the large range of size of infants and children. It is important that the data acquisition protocol is such that useful information can be obtained even if small movements take place. Interactions between magnetic ®elds (both static and transient) and the pulsed rf, and the monitoring equipment needed for patient care, require consideration. Some sensors contain ferromagnetic components that can impair the homogeneity of the magnetic ®eld. Many manufacturers of monitoring equipment can supply special magnetic resonance compatible devices. Alternatively, it may be possible to position ferromagnetic components so that their effects are unimportant. Some sensors may pick up rf interference, thereby decreasing the signal-to-noise ratio and transient rf and pulsed magnetic ®elds may overload ECG and other devices. The accidental formation

2 BRAIN MRS OF INFANTS AND CHILDREN of conductive loops by sensor cables must be avoided. This is in order to eliminate the risk of burns that could be caused if currents are induced by transient rf and pulsed gradient ®elds.11 Microprocessors, included in many newer monitors, emit rf interference and thus special rf Faraday shields and ®lters are required. Also, the spectrometer must be safe in the presence of the high inspired oxygen concentrations needed by some patients. Probes (rf) must be tested to ensure that transmitter pulses do not cause arcing; they must be in sealed containers which, as a further precaution, can be ¯ushed with a relatively inert gas such as N2. 3 DATA ACQUISITION Using surface coils with diameters in the range 5±7 cm and without further localization techniques, good quality 31P spectra can be obtained from 10±50 mL of the cerebral cortex of the newborn infant. These spectra have very little contamination from other tissues since skin and fat contain little 31P, cranial muscle is sparse, and the cranial bones are poorly mineralized and have relatively immobile 31P nuclei. For basic studies of the normal brain and abnormalities associated with nonfocal brain injury, valuable information has been obtained without better spectroscopic localization. However, the sensitive volume provided solely by a surface coil is not well de®ned. For investigations of focal metabolism, particularly deep in the brain, and in older infants and children who have better developed cranial musculature, and also for 1H studies, in which fat contamination can be a problem, spectroscopic localization techniques are essential. Many volume localization methods exist which have been used successfully for studies of infants and children. These techniques all employ pulsed magnetic ®eld gradients to obtain low-contamination spectra from a well de®ned voxel positioned with reference to MRI scout images. In these multipulse techniques, rf power is higher than for `single pulse' surface coil methods and consequently greater attention to safety is mandatory. Image-selected in vivo spectroscopy (ISIS; see Single Voxel Whole Body Phosphorus MRS)12 has been used for 31 P studies of the developing brain13 and, in order to combine MRI with localized 31P MRS, double-tunable probes dedicated to the examination of children's brains have been developed.14 For neonatal applications, standard probeheads are usually too large and give poor signal-to-noise ratio with the tiny voxels (about 1 mL) often necessary when studying the smaller anatomy. Localized 1H spectra from infants and older children have been acquired using both the point resolved spectroscopy (PRESS) and stimulated echo amplitude mode (STEAM) techniques (see Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo).15 MR spectroscopic imaging16 (see Chemical Shift Imaging) has not been as widely applied as the single voxel methods. Absolute concentrations (i.e. in millimoles per kilogram wet weight, or millimoles per liter) of 31P metabolites in developing brain have been estimated both by PRESS with brain water as an internal reference,17 and by ISIS with an external reference.18 The absolute concentrations of 1H metabolites are more dif®cult to estimate, due largely to the dependences of peak areas on T2 relaxation and, in the case of multiplets, on phase modulation. However, some information about developmental changes has been acquired using

an external reference19 or brain water as an internal reference.17,20

4

NORMAL BRAIN DEVELOPMENT IN INFANTS AND CHILDREN

Information about the relative and absolute concentrations of metabolites detectable by 31P and 1H MRS in the developing human brain has been acquired at several centers worldwide with good agreement of results. 4.1

The Neonatal Brain

The major 31P energy related resonances in spectra from skeletal muscle had been securely assigned when the ®rst studies of the neonatal brain were carried out. Peaks attributable to -, -, and -nucleotide triphosphate (NTP), PCr, and Pi, all of which are involved in energy metabolism, were also detected in the brains of newborn infants (see Figure 1). Magnesium complexed ATP is the main contributor to cerebral NTP, with guanosine, uridine, and cytidine triphosphates making up about a quarter of the remaining NTP pool;21 the - and -NTP resonances have small contributions from nucleotide diphosphates; and an often unresolved nicotinamide adenine dinucleotide (NAD + NADH) quartet resonates on the righthand shoulder of -NTP.22 Within the physiological range, the PCr resonant frequency is independent of both pHi (it only shifts signi®cantly at very low pH) and metal ion concentration; it is often used as a chemical shift reference (assigned as 0 ppm). Only a single Pi resonance (peak 2) is visible in the 31P spectra shown in Figure 1. Its chemical shift relative to PCr (Pi) depends strongly on the concentration ratio of its acidic and basic species (pKa & 6.8) and hence pHi can be estimated from the Henderson±Hasselbalch equation:23 pHi ˆ 6:77 ‡ log10 ‰…Pi ÿ 3:29†=…5:68 ÿ Pi †Š

…1†

Estimation of pHi is accurate to about 0.1 pH units. At high ®eld strengths (54.7 T), the Pi peak can often be resolved into extra- and intracellular24 and perhaps also mitochondrial components,25 implying the presence of compartments of different pH. When the ®rst 31P spectra were acquired from the neonatal brain, problems arose with the interpretation and assignment of prominent resonances other than those particularly involved in energy metabolism. The PME and the moderately broad, unresolved PDE peaks were both surprisingly large, and surface coil spectra were superimposed on a broad hump [see Figure 1(c)]. Chromatography and 31P and 1H MRS of extracts of neonatal mammalian brain have identi®ed phosphoethanolamine (PEt) as the main contributor to the PME peak26 which also includes phosphocholine (PC) and several smaller resonances, due for example to phosphoserine and phosphoinositol.22 These PMEs are involved in phospholipid metabolism, including the synthesis of myelin. The moderately broad resonance at about 2.9 ppm [see Figures 1(a) and 1(b), peak 3] is largely attributable to cross linked PDEs in membrane bilayer phospholipids and also to a small pool of mobile phospholipid breakdown products

BRAIN MRS OF INFANTS AND CHILDREN

3

(a)

(c)

1 (b)

4 2

5

6

3

7

40

30

20

10

0

–10

–20

–30

(ppm) 15

10

5

0

–5 –10 (ppm)

–15

–20

–25

Figure 1 31P spectra acquired from the temporo-parietal cortex of (a) a preterm infant of gestational plus postnatal age (GPA) 31 weeks; (b) an infant born and studied at full term. Pi is relatively smaller and PCr larger in spectrum (b) as compared with spectrum (a). Spectrum (c) is from the same infant as (b) but without removal by postacquisition processing of the broad underlying hump which was due to relatively immobile 31P in bone and membrane phospholipids. Peak identi®cations: (1) PME; (2) Pi; (3) PDE; (4) PCr; (5) -NTP; (6) -NTP; (7) -NTP. (Reproduced by permission of Plenum Press from E. B. Cady, `Clinical Magnetic Resonance Spectroscopy', New York, 1990, p. 86)

[glycerolphosphorylethanolamine (GPE) and glycerolphosphorylcholine (GPC)]. The membrane bilayer phospholipids exhibit chemical shift anisotropy leading to a reduced signalto-noise ratio at higher ®eld strengths (e.g. 4.7±7 T). The very broad underlying signal, shown in Figure 1(c), is due to more rigidly bound 31P nuclei mainly in membrane phospholipids and myelin. Well resolved 1H spectra can be acquired from the brains of newborn infants if long echo times (TE) are used (e.g. 270 ms). In Figure 2 an in vivo spectrum from a normal newborn infant is compared with an in vitro high-resolution spectrum of adult rat brain extract.27 The major in vivo peaks originate from Cho, Cr, and NAA; a smaller but clearly detectable Lac methyl resonance is also present. In the extract spectrum, Cho is seen to include GPC, PC, and taurine, and NAA is adjacent to glutamate, glutamine, and -aminobutyric acid (GABA). At shorter TE times (e.g. about 25 ms), T2 relaxation and phase modulation are less important and numerous peaks are easily detected; however, 1H spectra become more dif®cult to analyze. This problem is further complicated by the presence of

many broad overlapping resonances from fatty acids (from extravoxel contamination and perhaps also free fatty acids) and macromolecules (e.g. proteins). Unlike 31P metabolites, whose biochemical pathways are largely known, many 1H resonances other than Lac still pose problems of interpretation.28 For example, although NAA is widely regarded as an intraneuronal marker, its biochemical role remains unclear and NAA has also been found in oligodendrocyte progenitor cells.29 4.2

Changes with Age

Age-dependent changes in the concentration ratios and absolute concentrations of 31P and 1H metabolites have been de®ned in several studies. A major reason for doing these investigations has been to establish normal age-related values so that metabolic abnormalities may be securely identi®ed. Attention has focused on the rapid changes taking place in the last 3 months of gestation as determined in preterm and term infants studied during the ®rst days of life. Investigations of development during later childhood have also been undertaken. Figure

4 BRAIN MRS OF INFANTS AND CHILDREN

Cho

(a)

NAA Cr

Lac

Cr H-a, H-a′

(b)

PCr CH3–

N–Ac–Asp CH3–

Cr CH3–

Gly (H-a)

Glu H-a Gln H-a

Ac Tau H-a, H-a′ Tau H-b, H-b′

N–Ac–Asp H-a –glucose

Asp H-a

Lac H-a

Gln Glu H-g, H-g1′ H-g, H-g1′

GPC

PCr H-a, H-a′

Lac CH3–

N–Ac–Asp PC

Thr CH3–

H-b′

GABA H-b H-g, H-g′ Asp

Asp

H-b

H-b′

GABA H-a, H-a′

GABA H-b, H-b′ BHBA Val CH3– Ala CH3– Leu CH3–

4.0

3.0

2.0

1.0

(ppm from TSP)

Figure 2 An in vivo 1H PRESS spectrum from the thalamic region of a normal newborn infant of GPA 35 weeks (a) compared with an in vitro spectrum of an extract of adult rat brain (b). Spectrum (a) was collected at 100 MHz and the acquisition conditions were: TE 270 ms; repetition time (TR) 2 s, 128 averaged echoes; and an 8 mL voxel. Resonance identi®cations: Cho, choline containing compounds; Cr, creatine + phosphocreatine; NAA, N-acetylaspartate; Lac, lactate. Spectrum (b) was obtained at 360 MHz with TR 3 s and a 90 ¯ip angle (water-suppressed single pulse sequence). [Spectrum (a) was obtained in collaboration with A. Lorek, J. Penrice, J. S. Wyatt, R. Aldridge, and M. Wylezinska. Spectrum (b) is reproduced by permission of Elsevier Science Publishers from S. Cerdan, R. Parrilla, J. Santoro, and M. Rico, FEBS Lett., 1985, 187, 167]

BRAIN MRS OF INFANTS AND CHILDREN (b) 0.40

1.4

0.35

0.28

1.2

0.30

0.24

1.0

0.25

0.20

0.8

(c)

PDE/Ptot

1.6

PME/Ptot

PCr/Pi

(a)

0.20

0.32

0.16

0.6

0.15

0.12

0.4

0.10

0.08

0.2

0.05

0.04 0.00

0.00

0.0 24

28

32 36 (Weeks)

40

44

5

24

28

32 36 (Weeks)

40

44

24

28

32 36 (Weeks)

40

44

Figure 3 Relations between 31P metabolite concentration ratios in the temporo-parietal cortex and GPA in 30 infants of appropriate weight for gestational age. Regression lines and 95% con®dence limits are shown. (Reproduced by permission of International Pediatric Research Foundation Inc. from D. Azzopardi, J. S. Wyatt, P. A. Hamilton, E. B. Cady, D. T. Delpy, P. L. Hope, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 440)

3 and Table 1 give 31P surface coil data from the temporoparietal cortex of 30 infants of appropriate weight for gestational age (AGA) studied at a median postnatal age of 4 days.30 It can be seen that, with increasing gestational plus postnatal age (GPA), [PCr]/[Pi] and [PDE]/[total mobile phosphorus (Ptot)] increased, whereas [PME]/[Ptot] fell. These alterations, together with an increase in [PCr]/[NTP], continue during the ®rst months of life, but then the changes slow down as adulthood is approached (see Figures 4 and 5).13 Adult values are attained by about 4 years. Values from normal, small-for-gestational-age (SGA; i.e. with birthweights below the 3rd centile for gestation) infants were similar to those from AGA infants.30 [PCr]/[Pi] is directly related to the phosphoryl-

ation potential and to the free energy change of hydrolysis of ATP. The increase in [PCr]/[Pi] with age may be attributable to a greater need for an energy reserve in the mature brain and perhaps also to less ef®cient oxidative phosphorylation in the immature one. In the newborn human infant, [PCr]/[Pi] is similar to that in other altricious species, such as the newborn rat,31 but smaller than in precocious animals such as the lamb.32 The large amplitude of the PME resonance in the neonatal period is due to high [PEt].33,34 This metabolite may have an important role as a phospholipid precursor related to increased synthesis of membrane phospholipids and myelin at this time. Of the 31P resonances only PME shows a change in relaxation with development; T1 decreases from a neonatal value of about

Table 1 Relationships between Metabolite Concentration Ratios and pHi, in the temporo-parietal cortex and Gestational plus Postnatal age in 30 Infants of Appropriate Weight for Gestational Agea x PCr/Pi NTP/Ptot PCr/Ptot Pi/Ptot PME/Ptot PDE/Ptot PCr/NTP Pi/NTP PME/NTP PDE/NTP Hump/NTPb pHic a

0.024 0.001 0.001 ÿ0.001 ÿ0.004 0.002 0.009 ÿ0.013 ÿ0.050 0.016 1.011 ÿ0.003

c 0.190 0.075 0.051 0.128 0.429 0.131 0.717 1.482 4.887 1.658 ÿ7.790 7.233

r 0.58 0.19 0.56 ÿ0.40 ÿ0.68 0.47 0.20 ÿ0.24 ÿ0.37 0.15 0.49 0.12

p

28 weeks

42 weeks

<0.001 NS <0.01 <0.05 <0.001 <0.01 NS NS <0.05 NS <0.05 NS

0.85 0.09 0.09 0.10 0.32 0.19 0.79 1.13 3.49 2.11 20.5 7.14

1.18 0.10 0.11 0.09 0.26 0.22 1.09 0.95 2.79 2.34 34.7 7.09

‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹

0.33 0.03 0.02 0.02 0.04 0.04 0.42 0.50 1.23 1.03 20.5 0.28

‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹ ‹

0.33 0.03 0.02 0.02 0.04 0.04 0.42 0.50 1.25 1.05 21.6 0.28

x, Slope; c, ordinate intercept from the regression; p, signi®cance of the correlation coef®cient r. Data for 28 and 42 weeks are mean values ‹95% con®dence intervals from the regressions. bn = 24. cn = 23; values for the remaining infants were not calculated, because spectra were not well resolved. (Reproduced by permission of International Pediatric Research Foundation Inc. from D. Azzopardi, J. S. Wyatt, P. A. Hamilton, E. B. Cady, D. T. Delpy, P. L. Hope, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 440)

6 BRAIN MRS OF INFANTS AND CHILDREN (a) 150

(b) 1000 PCr

125 100

600

50

n (Hz)

PDE γ–ATB α–ATP Pi β–ATP

75 n (Hz)

800

PME

25

400 200

0 0

–25

–200

–50 20 (c)

15

10

5

0 –5 (ppm)

–10 –15 –20

15 (d)

150

10

5

0 –5 (ppm)

–10

–15

–20

–15

–20

175 150

125

125

100 n (Hz)

n (Hz)

100 75 50

75 50

25

25

0

0

–25

–25 20

15

10

5

0 –5 –10 –15 –20 (ppm)

15

10

5

0 –5 (ppm)

–10

Figure 4 31P ISIS spectra from the paraventricular region of the human brain at postnatal ages of 1 month (a), 4 months (b), 30 months (c), and 15 years (d). The spectra were obtained at 1.5 T using a sensitive volume varying from 54 mL in infants to 72 mL in older children. TR was 3.75 s and 256 FIDs were averaged. Signi®cant developmental changes occurred within the ®rst few years of life. (Reproduced by permission of The Radiological Society of North America from M. S. Van der Knaap, J. van der Grond, P. C. van Rijen, J. A. J. Faber, J. Valk, and K. Willemse, Radiology, 1990, 176, 509)

1.8 ms to about 1.2 ms in adulthood.18 This fall is probably due to alterations in the constituents of the peak, notably the reduction in PEt. No change has been detected in pHi during maturation (pHi ' 7.10).13,30 Measurements of the absolute concentrations of 31P metabolites in the newborn human brain obtained by PRESS17 and ISIS18 techniques have shown disagreements. For example, with the ISIS technique, the apparent [NTP] was much less than estimated with the PRESS method. These discrepancies could be due to a number of factors, including the acquisition of data from different parts of the brain. Furthermore, ISIS and PRESS estimates of [PDE] may be affected by relaxation during the delays and the pulses of the acquisition sequence; the cross-linked membrane bilayer fraction of PDE has a T2 of about 3 ms.35 As with 31P metabolites, the relative peak areas13,19,36 and absolute concentrations17,19 of 1H metabolites change with increasing age (see Figures 6±8). NAA/Cr increases and Cho/ Cr decreases up to about 3 years when the rate of change declines. [NAA] and [Cr] both appear to increase with age,

whereas both [Cho] and [myoinositol] decrease.17,19,37 The relatively low NAA signal seen in younger subjects and the increase with age are supported by biochemical studies, indicating that [NAA] increases greatly during mammalian brain development.38 Since NAA is mainly located in neurons, this increase may re¯ect neuronal maturation. The Cho peak contains signals from PCh, GPC and other metabolites; changes in its amplitude probably relate to alterations in the PME and PDE peaks in 31P spectra. Recent evidence suggests that Lac is detectable in the brains of normal newborn infants [see Figure 2(a)] and is higher in preterm and SGA infants. The implied greater reliance on anerobic glycolysis in preterm and SGA infants may relate to their reduced values for [PCr]/[Pi].17,36 5

CEREBRAL PATHOLOGY 31

P MRS has proved particularly valuable for the investigation of hypoxic-ischemic brain injury, especially in newborn infants. Most of the data have been acquired from infants who

BRAIN MRS OF INFANTS AND CHILDREN (a)

(c)

(b)

2.5

7

2.5

1.5 1.0

PDE/b–NTP

2.0 PME/PCr

PME/b–NTP

4.0 2.0

1.5

3.0

1.0 2.0

0.5

0.5

0 0

50

100 150 Age (months)

1.0

0

200

0

(d)

50

100 150 Age (months)

0

200

(e)

50

100 150 Age (months)

200

(f)

2.5 1.5

1.0

2.0 PCr/Pi

PCr/b–NTP

Pi/b–NTP

1.5

1.5

1.0

1.0

0.5

0.5 0.5

0

0 0

50

100 150 Age (months)

200

0 0

50

100 150 Age (months)

200

0

50

100 150 Age (months)

200

Figure 5 Relationships between age and PME/ -NTP (a), PME/PCr (b), PDE/ -NTP (c), Pi/ -NTP (d), PCr/ -NTP (e), and PCr/Pi (f) in the paraventricular region. Regression lines and 2.5 and 97.5 percentile lines are shown. (Reproduced by permission of The Radiological Society of North America from M. S. Van der Knaap, J. van der Grond, P. C. van Rijen, J. A. J. Faber, J. Valk, and K. Willemse, Radiology, 1990, 176, 509)

have suffered birth asphyxia, but similar results have been obtained in those with periventricular leukomalacia and other forms of hypoxic ischemic injury. More recently, 1H MRS has been used for monitoring abnormal glycolytic pathway activity (via increased Lac) and neuronal degeneration (via decreased NAA).

5.1 Perinatal-Hypoxia Ischemia Studies of babies with suspected cerebral hypoxic-ischemic injury were initiated on the premise that in conditions where the supply of oxygen to the brain was curtailed, or the mitochondrial mechanisms for its consumption were damaged, [ATP] would tend to fall. It was expected that the decrease in [ATP] would initially be very small because, unless [PCr] is grossly depleted, [ATP] is buffered by the creatine kinase reaction39 which maintains it close to normal, but at the expense of a fall in [PCr] and a reciprocal rise in [Pi]. It was hypothesized that [PCr]/[Pi] in the brains of infants with hypoxic ischemic injury would be reduced and that, in extreme circumstances, [ATP] might also be low. Surprisingly, studies of severely birth-asphyxiated babies have shown that cerebral 31P spectra

were often normal on the ®rst day of life.40 Subsequently [PCr]/[Pi] gradually fell, with a nadir at 2±4 days of age (see Figure 9 and Table 2), in spite of normal arterial oxygen saturation, blood pressure, and blood glucose. In the most affected infants, [NTP] then fell and death ensued. In contrast to the profound intracellular acidosis seen in experimental hypoxiaischemia,32 pHi tended to be slightly alkaline. (It is of interest that PCr and NTP were almost undetectable in two infants with inborn metabolic errors, propionic acidemia and arginosuccinic aciduria.41 However, cerebral pHi was profoundly acidic in these infants.) In survivors, the spectra usually returned to normal within about 2 weeks, but the overall 31P signal amplitude was often reduced, indicating loss of brain cells. Since the lowest observed [PCr]/[Pi] values were strongly related to long-term prognosis (see below) and because of the long timescale of the postnatal biochemical changes, in order not to miss useful information it is important that spectra are obtained on several occasions during the ®rst 2±4 days of life during which metabolic abnormalities are at their greatest. An explanation for the observed sequence of spectral changes is that the brain suffered an acute episode of hypoxia-ischemia before birth causing `primary' energy failure, which was reversed by resuscitation at delivery, so that the energy state of

8 BRAIN MRS OF INFANTS AND CHILDREN (a)

(b)

4 days

5 months

4 years

NAA

Cr

Adult

Ch Cr ML Glx

4

NAA Glx

3

2 (ppm)

Lac CH2 CH3

1

0

4

3

2 (ppm)

1

0

Figure 6 1H STEAM spectra obtained from (a) the parietooccipital cortex (predominantly white matter) and (b) the midline occipital cortex (mostly gray matter) of human brain at the ages of 4 days, 5 months, 4 years and from an adult. Amplitude scales have been adjusted so that absolute differences depending on both development and location can be visualized directly. Acquisition conditions: TR 2.87 s; mixing time (TM) 13 ms; TE 272 ms. Peak assignments: Ch, choline containing compounds; Cr, creatine + phosphocreatine; NAA, N-acetylaspartate; Glx, glutamate and glutamine; MI, myoinositol; Lac, lactate; CH2/CH3, lipid. (Reproduced by permission of Williams and Wilkins from R. Kreis, T. Ernst, and B. D. Ross, Magn. Reson. Med., 1993, 30, 424)

the brain returned to normal. The subsequent changes in 31P spectra have been attributed to a `secondary' impairment or failure of energy generation set off by the primary phase or events associated with it.41 Studies using near-infrared spectroscopy (NIRS) in infants developing secondary energy failure have shown that both cerebral blood ¯ow and blood volume are increased; continuing inadequate oxygen or substrate supply is therefore not likely to be causal. Many mechanisms have been suggested to account for the progression from primary to secondary energy failure, which is associated with delayed neuronal death. For example, one possibility is that excitatory neurotransmitters, particularly glutamate, which are released at the synapses in response to acute hypoxia ischemia may, by stimulating N-methyl-D-aspartate (NMDA) and other receptors, cause massive calcium entry to

cells and damage to the mitochondrial electron transport chain. Other possible mechanisms include those involving prostanoids, nitric oxide, free radicals, immune mechanisms, phagocytes, growth factors, and impaired protein synthesis. Because of the relation between secondary energy failure and long-term prognosis, much effort is being made to explore the mechanisms involved and how to interrupt them, with a view to the cerebroprotective treatment of asphyxiated infants. The same mechanisms are likely to be involved in stroke in adults. Some evidence is emerging from animal studies of the feasibility of cerebroprotection. The development of secondary energy failure has been modeled in the newborn piglet, giving 31P MRS results closely resembling those seen in birth-asphyxiated human infants.42 This model is suitable

BRAIN MRS OF INFANTS AND CHILDREN (a)

9

(b) 1.4 NAA/Cr

NAA/Cr

1.4 1.0 0.6 0.2 200

300

400

500

600

1.4

1.4

1.0

1.0

Ch/Cr

Ch/Cr

100

0.6 0.2

0

100

200

300

400

500

600

0

100

200

300

400

500

600

0

100

200

300

400

500

600

0

100

200

300 400 GA (weeks)

500

600

0.6 0.2

0

100

200

300

400

500

600

1.4

1.4

1.0

1.0

MI/Cr

MI/Cr

0.6 0.2

0

0.6 0.2

0.6 0.2

0

100

200

300

400

500

600

0.8

0.8

0.6

0.6

Glx/Cr

Glx/Cr

1.0

0.4 0.2

0.4 0.2

0

100

200

300 400 GA (weeks)

500

600

Figure 7 Relations between gestational age (GA) and cerebral peak-amplitude ratios at the same anatomical locations as in Figure 6. Acquisition conditions and resonance identi®cations are as in Figure 6. Cho/Cr and MI have been ®tted to a monoexponential function, whereas the NAA/Cr ®t is biexponential. (Reproduced by permission of Williams and Wilkins from R. Kreis, T. Ernst, and B. D. Ross, Magn. Reson. Med., 1993, 30, 424)

for testing clinically feasible cerebroprotective strategies, with the amelioration of secondary energy failure, as measured by 31 P and 1H MRS, providing an initial end-point. Recently, mild hypothermia, applied after a severe cerebral hypoxic± ischemic insult, was shown to be markedly cerebroprotective in this model,43 whereas intravenous magnesium sulfate was ineffective.44 1 H MRS studies of infants with hypoxic-ischemia brain injury36,45,46 show that increased Lac (due to failure of oxidative phosphorylation and consequent anerobic glycolysis), and occasionally also reduced NAA (due to neuronal death), may be detected (see Figure 10). The Lac/NAA peak area ratio is often high and this index may prove a sensitive marker of

cerebral injury. Reduced [NAA] has also been seen in newborn infants with a variety of nonfocal pathologies,19 and in children with generalized demyelination.47 Abnormal 1H spectra have been found in older children with a variety of inborn errors of metabolism, including increased NAA in Canavan's disease (see Figure 11), and increased Lac in Leigh, Schilder, and Cockayne disease, and also in neuroaxonal dystrophy.47 5.2

Prognosis

The prognostic signi®cance of secondary energy failure has been investigated by Azzopardi et al.40 They studied 61 infants recruited because of evidence or suspicion of hypoxic

[Cr] (mmol kg–1)

[NAA] (mmol kg–1)

10 BRAIN MRS OF INFANTS AND CHILDREN 10.0 8.0 6.0 4.0 2.0 0.0 0

100

200

300

400

500

600

0

100

200

300

400

500

600

0

100

200

300

400

500

600

0

100

200

300 400 GA (weeks)

500

600

8.0 6.0 4.0 2.0

[Ch] (mmol kg–1)

0.0

2.5 2.0 1.5 1.0 0.5

[MI] (mmol kg–1)

0.0

14.0 10.0 6.0 2.0 0.0

Figure 8 Absolute concentrations (mmol/kg brain tissue) of the main cerebral metabolites detectable by in vivo 1H MRS. Un®lled symbols are data from parietal white matter [location (a) in Figure 6]; ®lled symbols are data from occipital gray matter [location (b) in Figure 6]. (&, &) Data points from both normal infants and subjects with no primary cerebral pathology; (^, ^) data points from infants with essentially normal neurological status for whom the eventual diagnosis was considered immaterial to the cerebral metabolite pro®le. The solid lines are best ®ts to the data; biexponential for [NAA] and monoexponential for [Cr], [Cho], and [MI]. (The dotted curve in the [Cr] plot is triexponential and appears to ®t the data better.) The vertical dashed lines indicate term at 40.5 weeks gestational age. The horizontal dashed lines show the mean adult values measured at location (a) in Figure 6. (Reproduced by permission of Williams and Wilkins from R. Kreis, T. Ernst, and B. D. Ross, Magn. Reson. Med., 1993, 30, 424)

ischemic brain injury, many of whom had sustained birth asphyxia. Clear evidence was found that the chances of survival and the severity of neurodevelopmental impairments at 1 year of age were directly related to the maximum detected extent of energy failure during the ®rst days of life, quanti®ed as the minimum observed value for [PCr]/[Pi] (see Table 2 and

Figure 12). Furthermore, if [NTP]/[Ptot] fell, death was almost inevitable. As a continuation of the same study, Roth et al.48 recruited a further group of birth-asphyxiated infants, so that data from a total of 52 such infants were available for analysis. These infants had been in very poor condition at birth; their mean arterial base excess in cord blood shortly after delivery was ÿ20 mmol Lÿ1, 38 had ®ts, and 26 required mechanical ventilation. The relation between minimum recorded [PCr]/[Pi] in the ®rst days of life and neurodevelopmental outcome is illustrated in Figure 13. These data extend those of Azzopardi et al. and con®rm the bad prognosis of secondary energy failure. The sensitivity, speci®city and positive predictive value for death or multiple disabling impairments of values of [PCr]/[Pi] more than 2 SD below normal were 72%, 92%, and 91%, respectively. Roth et al. also found a relationship between minimum observed [PCr]/[Pi] and head growth. Although the infants had normal head circumferences at birth, head growth was slowest, leading to microcephaly, in those whose values for [PCr]/[Pi] fell the most. Further follow-up indicated that the severities of adverse outcomes at age 4 years were also closely related to the extent of cerebral energy impairment in the ®rst week of life.49 The results of these studies may be summed up as showing that secondary energy failure leads both to neurodevelopmental disabilities and to microcephaly. The worse the energy failure, the worse the outcome is likely to be. 31 P spectra from the brains of infants with increased echo densities on ultrasound scans are also often abnormal, showing reduced values for [PCr]/[Pi] and sometimes [NTP]/[Ptot] (see Table 2).40,50 Such echo densities are usually due to hypoxic ischemic injury or to hemorrhage. 31P spectroscopy can be useful for segregating infants whose echo densities carry a relatively good, from those implying a bad, prognosis. Data are becoming available which indicate that abnormal 1 H spectra also imply a bad prognosis.36,45 Observation of raised Lac and of reduced NAA appear predictive of an unfavorable outcome. Further investigations, including long-term follow-up, are required before de®nitive statements can be made.

6

CONCLUSIONS

The ®rst in vivo MRS studies of the human brain were in newborn infants. Since these studies were carried out, investigations of older children and adults have been undertaken and developmental changes have been de®ned for 31P and 1H metabolites. For example, [PCr]/[Pi] rises with age, indicating increased phosphorylation potential, and [PME]/[PDE] falls. 31 P MRS has proved particularly useful for investigating the cerebral metabolic consequences of severe birth-asphyxia. Spectra are often normal shortly after birth, but then [PCr]/[Pi] and, in severe cases, [NTP]/[Ptot] gradually falls, reaching minimum values at 2±4 days of age. The extent of the falls in [PCr]/[Pi] and [NTP]/[Ptot] are strongly related to the likelihood of long-term neurodevelopmental impairments or death. These changes in 31P metabolite concentrations have been termed `secondary' cerebral energy failure, on the hypothesis that they were initiated by an episode of `primary' energy failure taking

BRAIN MRS OF INFANTS AND CHILDREN

11

55 h

36 h

31 h

1 23 4 5

6

7 8h

15 10

5

0

–5 –10 –15 –20 –25 (ppm)

Figure 9 31P spectra from the temporo-parietal cortex of a birth-asphyxiated infant born at 37 weeks gestation. The postnatal ages at the times of study are given. Peak assignments are as in Figure 1. At age 8 h, [PCr]/[Pi] was 0.99, [NTP]/[Ptot] was 0.09 and pHi was 7.06; pHi increased to a maximum of 7.28 at 36 h. The minimum [PCr]/[Pi] was 0.32 at 55 h, when [NTP]/[Ptot] was 0.04 and pHi was 6.99. The infant died aged 60 h. (Reproduced by permission of International Pediatric Research Foundation Inc. from D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delpy, J. Baudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 445)

Table 2 Age-dependent Standard Deviation Scores (SDS) for Phosphorus Metabolite Concentration Ratios and pHi in the temporo-parietal cortex of Newborn Infants Suspected of Hypoxic±ischemic Brain Injurya All infants

Birth asphyxia

(n = 61) PCr/Pi NTP/Ptot PCr/Ptot Pi/Ptot PME/Ptot PDE/Ptot PCr/NTP Pi/NTP PME/NTP PDE/NTP pHi a

Postnatal asphyxia

(n = 40) b

ÿ2.14 ‹ 2.10 ÿ0.98 ‹ 1.86c ÿ1.72 ‹ 2.58b 5.40 ‹ 8.77b 0.42 ‹ 1.69 0.44 ‹ 1.43 0.52 ‹ 2.98 7.21 ‹ 16.41b 1.94 ‹ 4.40d 1.02 ‹ 3.84 0.05 ‹ 1.47 (n = 55)

(n = 5) b

ÿ2.14 ‹ 2.19 ÿ1.17 ‹ 1.82c ÿ1.79 ‹ 2.59b 5.97 ‹ 8.91b 0.63 ‹ 1.58 ÿ0.60 ‹ 1.44d 0.65 ‹ 3.14 8.96 ‹ 17.94b 2.49 ‹ 4.71c 1.43 ‹ 4.48 0.23 ‹ 0.97 (n = 36)

ÿ0.03 ÿ0.01 0.33 0.69 ÿ0.79 0.39 0.15 0.35 ÿ0.26 0.17 ÿ0.28 (n =

‹ 0.76 ‹ 1.03 ‹ 1.04 ‹ 0.99 ‹ 1.26 ‹ 0.50 ‹ 0.66 ‹ 1.06 ‹ 1.11 ‹ 0.77 ‹ 0.22 5)

Increased cerebral echodensities (n = 16) ÿ2.04 ‹ 1.78b ÿ0.83 ‹ 1.99 ÿ2.19 ‹ 2.52b 5.51 ‹ 9.06c 0.30 ‹ 1.83 ÿ0.32 ‹ 1.47 0.31 ‹ 2.91 4.94 ‹ 13.27c 1.25 ‹ 3.72 0.23 ‹ 1.81 0.00 ‹ 2.21 (n = 14)

Values are those obtained when PCr/Pi was at its lowest. Mean values for SDS ‹ SD vs the normal control infants (see Table 1) are given. p < 0.001. cp < 0.01. dp < 0.05. (Reproduced by permission of International Pediatric Research Foundation Inc. from D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delpy, J. Baudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 445)

b

12 BRAIN MRS OF INFANTS AND CHILDREN Cho

Cr NAA

(d) Lac 14 days

(c) 7 days

(b) 3 days

–PD

(a)

2 days

3

2 (ppm)

1

Figure 10 1H PRESS spectra acquired at 100 MHz from the thalamus of a severely birth-asphyxiated term infant. The ages at which the spectra were acquired are shown. Acquisition conditions: TE 270 ms; TR 2 s; 128 averaged echoes; 8 mL voxel. The lactate peak at about 1.3 ppm is greatly increased when compared with normal infants [see Figure 2(a)], although this may be partly due to raised free fatty acids. The peak at about 1.1 ppm is due to propan-1,2-diol (PD; the injection medium for phenobarbitone and phenytoin preparations). (Reproduced by permission of Williams and Wilkins from E. B. Cady, A. Lorek, J. Penrice, E. O. R. Reynolds, R. A. Iles, S. P. Burns, G. A. Coutts, and F. M. Cowan, Magn. Reson. Med., 1994, 32, 764)

place shortly before or during delivery. The prognostic information acquired by 31P MRS is valuable for guiding the clinical care of infants suspected of hypoxic ischemic brain injury. In the future, the prevention of secondary energy failure, as measured by MRS, may be used to test the ef®cacy of cerebroprotective strategies. 1H spectroscopy has demonstrated increased Lac/NAA peak area ratios in the brains of asphyxiated infants and this may also have prognostic signi®cance.

Recent technological developments will enhance the value of MRS to clinical pediatrics. Five major areas have potential for immediate improvement: the development of special neonatal head probes to increase the signal-to-noise ratio; optimized acquisition methods to enhance the quality and variety of spectroscopic information; spectroscopic quanti®cation; and spectral editing to detect resonances that would otherwise be dif®cult to resolve. Spectroscopic imaging for metabolite

BRAIN MRS OF INFANTS AND CHILDREN (b)

(a) 5

6 TE = 135 ms

TE = 135 ms

NAA

5

4

4

3

3

2

2

Signal (mV)

Signal (mV)

0 TE = 270 ms

3

Ch

NAA Cr/PCr

1

Cr/PCr

1

4

13

Cr/PCr

0 4 TE = 270 ms

3 2

2 1

1

0

0

4.0

3.5

3.0 2.5 2.0 1.5 Chemical shift (ppm)

1.0

0.5

4.0

3.5

3.0 2.5 2.0 1.5 Chemical shift (ppm)

1.0

0.5

Figure 11 1H PRESS spectra from occipital voxels in the brain of a 7-month-old child with Canavan disease (a) and in that of an age-matched, normal control (b). Spectrum (a) shows increased N-acetylaspartate (NAA) signal, and the resonance from choline-containing compounds (Cho) is almost undetectable. Canavan disease has been linked with high body ¯uid levels of NAA caused by aspartoacylase de®ciency. The disease also affects myelin sheaths due to a defect of myelin formation and/or glial metabolism, and this may be the reason for the low Cho signal. (Reproduced by permission of The Radiological Society of North America from W. Grodd, I. Krageloh-Mann, U. Klose, and R. Sauter, Radiology, 1991, 181, 173)

3.0

1.5

–1.5 PCr/Pi SDS

PCr/Pi, SDS

0.0

–3.0

–4.5

–6.0 45

65

85

105

125

145

GQ

Figure 12 The relationship between minimum observed PCr/Pi in the temporo-parietal cortex expressed as standard deviation score (SDS) vs normal infants and the Grif®ths general quotient (GQ) assessed at age 1 year in 38 infants surviving after hypoxic ischemic brain injury. Scores below 50 are recorded as 50 (r = 0.67, p < 0.001). (Reproduced by permission of International Pediatric Research Foundation Inc. from D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delpy, J. Baudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 445)

4 3 2 1 0 –1 –2 –3 –4 –5 –6 –7 –8 Normal

Minor

Major

Multiple

Dead

Impairments

Figure 13 Minimum observed age-related PCr/Pi standard deviation score (SDS) in the temporo-parietal cortex of birth-asphyxiated newborn infants and neurodevelopmental outcome at age 1 year: (*) term AGA infants; (*) term SGA infants; (!) preterm AGA infants; (&) preterm SEA infants. (Reproduced by permission of Mac Keith Press from S. C. Roth, D. Azzopardi, A. D. Edwards, J. Baudin, E. B. Cady, J. Townsend, D. T. Delpy, A. L. Stewart, J. S. Wyatt, and E. O. R. Reynolds, Dev. Med. Child Neurol., 1992, 34, 285)

14 BRAIN MRS OF INFANTS AND CHILDREN mapping and direct comparison of MRI and spatial MRS data should help to improve the quality of the information available and further develop MRS as an acceptable and valuable clinical tool.

7 RELATED ARTICLES Brain Infection and Degenerative Disease Studied by Proton MRS; Brain MRS of Human Subjects; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; MRI in Clinical Medicine; Patient Life Support and Monitoring Facilities for Whole Body MRI; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Single Voxel Whole Body Phosphorus MRS; Tissue and Cell Extracts MRS.

8 REFERENCES 1. E. B. Cady, A. M. de L. Costello, M. J. Dawson, D. T. Delpy, P. L. Hope, E. O. R. Reynolds, P. S. Tofts, and D. R. Wilkie, Lancet, 1983 i, 1059. 2. A. L. Stewart, R. J. Thorburn, P. L. Hope, M. Goldsmith, A. P. Lipscomb, and E. O. R. Reynolds, Arch. Dis. Child., 1983, 58, 598. 3. E. B. Cady, J. Hennig, and E. Martin, in `Imaging Techniques of the CNS of Neonates', ed. J. Haddad, D. Christmann, and J. Messer, Springer, Berlin, 1991, Chap. 6. 4. E. B. Cady, in `Magnetic Resonance Spectroscopy in Biology and Medicine', ed. J. D. de Certaines, W. M. M. J. Bovee, and F. Podo, Pergamon, Oxford, 1992, Chap. 23. 5. E. B. Cady, C. Boesch, and E. Martin, in `Perinatal Asphyxia', ed. J. Haddad and E. Saliba, Springer, Berlin, 1993, Chap. 13. 6. A review of this early in vivo work is given in: E. B. Cady, `Clinical Magnetic Resonance Spectroscopy', Plenum, New York, 1990. 7. D. P. Younkin, M. Delivoria-Papadopoulos, J. C. Leonard, V. H. Subramanian, S. Eleff, J. S. Leigh, and B. Chance, Ann. Neurol., 1984, 16, 581. 8. Department of Health (United Kingdom), `Guidelines for Magnetic Resonance Diagnostic Equipment in Clinical Use', HMSO, Norwich, 1993. 9. A. Chu, D. T. Delpy, and S. Thalayasingam, in `Fetal and Neonatal Physiological Measurements', ed. P. Rolfe, Butterworths, London, 1986, Chap. 59. 10. C. Boesch and E. Martin, Radiology, 1988, 168, 481. 11. F. G. Shellock and G. L. Slimp, Am. J. Roentgenol., 1989, 153, 1105. 12. R. J. Ordidge, A. Connelly, and J. A. B. Lohman, J. Magn. Reson., 1986, 66, 283. 13. M. S. Van der Knaap, J. van der Grond, P. C. van Rijen, J. A. J. Faber, J. Valk, and K. Willemse, Radiology, 1990, 176, 509. 14. R. Gruetter, C. Boesch, M. Muri, E. Martin, and K. Wuthrich, Magn. Reson. Med., 1990, 15, 128. 15. C. T. W. Moonen, M. von Kienlin, P. C. M. van Zijl, J. Cohen, J. Gillen, P. Daly, and G. Wolf, NMR Biomed., 1989, 2, 201. 16. J. H. Duyn and C. T. W. Moonen, Magn. Reson. Med., 1993, 30, 409. 17. E. B. Cady, J. Penrice, P. N. Amess, A. Lorek, M. Wylezinska, R. F. Aldridge, F. Franconi, J. S. Wyatt, and E. O. R. Reynolds, Magn. Reson. Med., 1996, 36, 878. 18. R. Buchli, E. Martin, P. Boesiger, and H. Rumpel, Pediatr. Res., 1994, 35, 431.

19. R. Kreis, T. Ernst, and B. D. Ross, Magn. Reson. Med., 1993, 30, 424. 20. P. Christiansen, O. Henriksen, M. Stubgaard, P. Gideon, and H. B. W. Larsson, Magn. Reson. Imaging, 1993, 11, 107. 21. A. G. Chapman, E. Westerberg, and B. K. SiesjoÈ, J. Neurochem., 1981, 36, 179. 22. T. Glonek, S. J. Kopp, E. Kot, J. W. Pettegrew, W. H. Harrison, and M. M. Cohen, J. Neurochem., 1982, 39, 1210. 23. O. A. C. Petroff, J. W. Prichard, K. L. Behar, J. R. Alger, J. A. den Hollander, and R. G. Shulman, Neurology, 1985, 35, 781. 24. R. C. Robbins, R. S. Balaban, and J. A. Swain, J. Thorac. Cardiovasc. Surg., 1990, 99, 878. 25. P. B. Garlick, S. Soboll, and G. R. Bullock, NMR Biomed., 1992, 5, 29. 26. J. W. Pettegrew, S. J. Kopp, J. Dadok, N. J. Minshew, J. M. Feliksik, T. Glonek, and M. M. Cohen, J. Magn. Reson., 1986, 67, 443. 27. S. Cerdan, R. Parrilla, J. Santoro, and M. Rico, FEBS Lett., 1985, 187, 167. 28. Reviews of biochemical aspects of the resonances seen in 1H brain spectra are given, in: B. L. Miller and B. D. Ross, NMR. Biomed., 1991, 4, 47, 59. 29. J. Urenjak, S. R. Williams, D. G. Gadian, and M. Noble, J. Neurosci., 1993, 13, 981. 30. D. Azzopardi, J. S. Wyatt, P. A. Hamilton, E. B. Cady, D. T. Delpy, P. L. Hope, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 440. 31. P. Tofts and S. Wray, J. Physiol. (London), 1985, 359, 417. 32. P. L. Hope, E. B. Cady, A. Chu, D. T. Delpy, R. M. Gardiner, and E. O. R. Reynolds, J. Neurochem., 1987, 49, 75. 33. H. C. Agrawal and W. A. Himwich, in `Developmental Neurobiology', ed. W. A. Himwich, C. C. Thomas, Spring®eld, IL, 1970, Chap. 9. 34. N. Okumura, S. Otsuki, and A. Kameyama, J. Biochem., 1960, 47, 315. 35. W.-I. Jung, S. Widmaier, M. Bunse, U. Seeger, K. Straubinger, F. Schick, K. Kuper, G. Dietze, and O. Lutz, Magn. Reson. Med., 1993, 30, 741. 36. J. Penrice, E. B. Cady, A. Lorek, M. Wylezinska, P. N. Amess, R. F. Aldridge, A. L. Stewart, J. S. Wyatt, and E. O. R. Reynolds, Pediatr. Res., 1996, 40, 6. 37. P. S. Huppi, S. Posse, F. Lazeyras, R. Burri, E. Bossi, and N. Herschkowitz, Pediatr. Res., 1991, 30, 574. 38. H. H. Tallan, J. Biol. Chem., 1957, 223, 41. 39. B. K. SiesjoÈ, `Brain Energy Metabolism', Wiley, New York, 1978. 40. D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delpy, J. Baudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. R. Reynolds, Pediatr. Res., 1989, 25, 445. 41. P. L. Hope, A. M. de L. Costello, E. B. Cady, D. T. Delpy, P. S. Tofts, A. Chu, P. A. Hamilton, E. O. R. Reynolds, and D. R. Wilkie, Lancet, 1984, ii, 366. 42. A. Lorek, Y. Takei, E. B. Cady, J. S. Wyatt, J. Penrice, A. D. Edwards, D. Peebles, M. Wylezinska, H. Owen-Reece, V. Kirkbride, C. Cooper, R. F. Aldridge, S. C. Roth, G. Brown, D. T. Delpy, and E. O. R. Reynolds, Pediatr. Res., 1994, 36, 699. 43. M. Thoresen, J. Penrice, A. Lorek, E. B. Cady, M. Wylezinska, V. Kirkbride, C. E. Cooper, G. C. Brown, A. D. Edwards, J. S. Wyatt, and E. O. R. Reynolds, Pediatr. Res, 1995, 37, 667. 44. J. Penrice, P. N. Amess, S. Punwani, M. Wylezinska, L. Tyszczuk, P. D'Souza, A. D. Edwards, E. B. Cady, J. S. Wyatt, and E. O. R. Reynolds, Pediatr. Res., 1997, 41, 1. 45. E. B. Cady, Neurochem. Res., 1966, 21, 1043. 46. E. B. Cady, P. Amess, J. Penrice, M. Wylezinska, V. Sams, and J. S. Wyatt, Magn. Reson. Imaging, 1997, 15, 605. 47. W. Grodd, I. Krageloh-Mann, U. Klose, and R. Sauter, Radiology, 1991, 181, 173.

BRAIN MRS OF INFANTS AND CHILDREN 48. S. C. Roth, A. D. Edwards, E. B. Cady, D. T. Delpy, J. S. Wyatt, D. Azzopardi, J. Baudin, J. Townsend, A. L. Stewart, and E. O. R. Reynolds, Dev. Med. Child Neurol., 1992, 34, 285. 49. S. Roth, J. Baudin, E. B. Cady, K. Johal, J. Townsend, J. S. Wyatt, E. O. R. Reynolds, and A. L. Stewart, Dev. Med. Child Neurol., 1997, 39, 718. 50. P. A. Hamilton, P. L. Hope, E. B. Cady, D. T. Delpy, J. S. Wyatt, and E. O. R. Reynolds, Lancet, 1986, i, 1242.

Biographical Sketches Ernest B. Cady. b 1952. B.Sc. (astronomy), University College London, 1973; Dip. Adv. Sc. Studies (radio astronomy), Manchester Uni-

15

versity, 1974. Introduced to NMR by Prof. D. R. Wilkie FRS. Medical Physicist, University College London Hospitals, 1978±present. Approx. 100 publications, including the book Clinical Magnetic Resonance Spectroscopy. Research interests: probe design, data analysis, absolute quanti®cation, and neonatal brain metabolism. E. Osmund R. Reynolds. b 1933. B.Sc. (physiology), 1955, M.B., B.S., 1958, M.D., 1965, London University; M.R.C.P., 1966, F.R.C.P., 1975, F.R.C.O.G. (ad eundem), 1983; F.R.S., 1993; C.B.E., 1995; Hon. F.R.C.P.C.H., 1997; Professor of Neonatal Pediatrics, University College London Medical School, 1976±96 (emeritus). Approx. 300 publications. Research interests: noninvasive investigation of perinatal brain injury, in particular employing ultrasound imaging, NMR spectroscopy, and near-infrared spectroscopy.

CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI

Central Nervous System Degenerative Disease Observed by MRI Frank J. Lexa Leonard Davis Institute of Health Economics, University of Pennsylvania, Philadelphia, PA, USA

1 INTRODUCTION The degenerative diseases encompass some of the most interesting and important pathological entities in clinical medicine. In the past decade, MR imaging (MRI) has emerged as the premier radiological method for examining the brain and spinal cord. MRI has many advantages over computerized tomography (CT) for the detection of subtle changes in white and gray matter structures. This has led to the ability ®nally to detect and characterize some of these diseases early in their course without resorting to invasive techniques. This section will discuss the use of MRI to detect and characterize neural degeneration, beginning with the general case of the processes of Wallerian degeneration, which are common to many types of brain injury, and then move on to discuss the salient clinical and imaging features of the most important and best characterized of the neurodegenerative disorders. In the interest of brevity, only the most important clinical and imaging facets of the diseases will be covered. The reader is referred to two standard texts which are able to provide a more comprehensive discussion and bibliography.1,2

2 WALLERIAN DEGENERATION: THE NEURONAL REACTION TO INJURY Waller is credited for making the ®rst histological description of the series of events which occur in the distal axonal segment following a proximal injury.3±5 This can occur after a wide variety of insults to the brain or spinal cord and thus represents an important marker of signi®cant central nervous system (CNS) damage. The earliest phase is cessation of axonal transport, and collapse of the axonal segment.6 The myelin begins to fragment, and the Schmidt±Lanterman incisures open up. This is followed by reaction of the surrounding tissue with cellular proliferation and edema, during which the myelin is chemically degraded and removed. This eventually leads to formation of a gliotic scar. Until the advent of MRI, it was dif®cult or impossible to detect degeneration of axonal pathways using antemortem imaging techniques. CT is usually only capable of detecting signi®cant volume loss in end-stage cases. Early reports of the MRI appearance of Wallerian degeneration described T1 and T2 prolongation along corticofugal pathways in patients with acquired lesions such as strokes and Schilder's disease.7±13 In these reports, the lesions were at least several months old when

1

they had these signal characteristics, although these changes have been reported to be observable earlier.14 Observations from the ®rst month were reported in which no abnormality could be detected.15 High signal intensity in the dentatorubral± thalamic tract has been reported after dentate nucleus resection.16 Other reports described a more complex pattern with no detectable signal changes in approximately the ®rst month, a transient period of hypointensity on long TR images at approximately 1 month, followed by normalization of that hypointensity and later development of the more familiar T1 and T2 prolongation effects at approximately the 3±4 month mark.17±18 Recent work suggests that fast spin echo (FSE) imaging shows this early hypointense phase of Wallerian degeneration more conspicuously than conventional proton density imaging.19 In clinical medicine, this has important implications for the diagnosis and treatment of brain injury. Animal models have been used to address the controversies raised by the above studies. Magnetic resonance spectroscopy of peripheral nerves showed signi®cant increases in T1 and T2 relaxation times 15 days after sectioning the sciatic nerve.20 Imaging of the tibial nerve con®rmed that the T2 signal intensity was elevated on long TR/TE images at day 15 in a crush injury model.21 Histological con®rmation of Wallerian degeneration was obtained in a feline model of radiation injury. These areas showed high signal intensity on long TR/TE images over 200 days after radiation injury.22 Newer techniques utilizing magnetization transfer techniques appear very promising for the early detection and separation of some of these changes. Using a cortical ablation model, Lexa et al. were able to demonstrate changes as early as the ®rst week after injury (Figure 1).23 Magnetization transfer images detected degenerating tracts at a distance from the primary injury site before conventional spin echo images or even routine light microscopy showed signi®cant evidence of injury. Electron microscopy con®rmed that the ®rst phase of Wallerian degeneration was underway (Figure 2). Myelin degeneration and cellular degeneration were not detectable by light-microscopic techniques until signi®cantly later. Moreover, the biphasic nature of the early changes in magnetization transfer suggests that it may be possible to separate some of the processes occurring in early axonal injury with magnetization transfer (Figure 3).

3

DEGENERATIVE DEMENTING ILLNESSES: ALZHEIMER'S AND PICK'S DISEASES

Alzheimer's disease is both the commonest cause of dementing illness as well as one of the most common causes of death in the industrialized countries. Both pathological and radiological studies demonstrate extensive atrophy, particularly of the hippocampus.24,25 Pick's disease is a much rarer dementing illness with similar cognitive de®cits to Alzheimer's disease; however, abnormal behaviors, apathy, abulia, and the KluÈver±Bucy syndrome are more common.26 The disease appears to be transmitted in a dominant, although modi®ed, fashion, and is more common in females. Neuroradiological studies and gross pathology show ®ndings of marked atrophy affecting the frontal and anterior

2 CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI

MTR(MTR affected –MTR control%)

10

5

0

+

+

+

+

+

–5

–10

–15

0

5

10

15

20 25 30 35 Days postoperation

40

45

50

Figure 1 Demonstration of changes in magnetization transfer in a feline model of Wallerian degeneration at 1.5 T. The x axis shows the number of days postplacement of a cortical lesion. The y axis shows the change in the magnetization transfer ratio (MTR) of the affected side relative to a control in three structures: the degenerating superolateral white matter (~), remote white matter (^), and the ipsilateral lateral geniculate nucleus (+). The ipsilateral white matter tracts show an early rise and later fall in magnetization transfer as early degeneration progresses. (Reproduced by permission of The American Society of Neuroradiology from Lexa et al.23)

temporal lobes with relative sparing of the parietal lobes and sparing of the posterior portion of the superior temporal gyrus and the occipital regions.26 Alzheimer's disease appears to have associations with other degenerative diseases of the CNS±

Figure 3 Long TR FSE images in a woman 4 weeks after a middle cerebral artery stroke showing evidence of Wallerian degeneration in the ipsilateral corticospinal tract with marked signal loss (arrow) relative to the unaffected side

most notably Parkinson's disease (see below). In some reports, close to half of those patients with Alzheimer's disease have evidence of the degenerative changes associated with Parkinson's disease.27,28

4

Figure 2 Electron micrograph of white matter immediately superior to the lateral geniculate nucleus 8 days after ablation of the cortical sites of visual processing. Arrows mark examples of early degeneration with axonal collapse and increased cytoplasmic staining. (Reproduced by permission of The American Society of Neuroradiology from Lexa et al.23)

NORMAL PRESSURE HYDROCEPHALUS

Hydrocephalus can occur secondary to a wide variety of etiologies. This discussion will be limited to the `degenerative' form of normal pressure hydrocephalus (NPH). The syndrome of NPH is seen predominately in older patients, and includes a triad of clinical ®ndingsÐdementia, gait ataxia, and urinary incontinence. The underlying mechanism which leads to NPH is controversial and includes theories related to diminished absorption from prior hemorrhage or in¯ammatory disease, changes in the physical properties of the ventricular walls, or, rarely, a preexisting congenital form of hydrocephalus. Patients may respond to shunting, and favorable factors include response to a preliminary trial of cerebrospinal ¯uid removal, presentation with gait disturbance, and short duration of symptoms.29 NPH and white matter ischemic disease appear to be signi®cantly related, including a direct correlation of the severity of the diseases.30 On cross-sectional imaging techniques, NPH may overlap signi®cantly in appearance with other diseases in this age group. In particular, the overlap with atrophy from other causes is problematic. The most useful features are those found in

CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI

other forms of hydrocephalus: thinning and elevation of the corpus callosum, and distention of the third ventricle.31 MRI may provide an additional clue by allowing detection and quanti®cation of ¯ow in the cerebral aqueduct. The magnitude of ¯ow in this structure is increased in the setting of NPH, and is probably secondary to the diminished compliance of the ventricular system.32 5 DEGENERATION OF THE CENTRAL GRAY MATTER STRUCTURES 5.1 Huntington's Chorea Huntington's chorea is an autosomal dominant degenerative disease which has been localized to chromosome 4.33 Presentation is in the fourth and ®fth decades, with choreathetosis and dementia. Neuroradiological studies con®rm characteristic atrophy of the caudate and putamen as well as more generalized atrophy.34 On MRI, abnormal signal intensity on long TR/ TE images may be present in the striatum. In addition to reductions in the caudate and putamen, detailed volumetric studies show volume loss in the thalamus and mesial temporal lobes.35 In mildly affected individuals, the putaminal measurement appears to be sensitive and speci®c.36 5.2 Wilson's Disease (Hepatolenticular Degeneration) Wilson's disease is a genetic de®ciency of ceruloplasmin, which leads to toxic copper deposition particularly in the liver and the lenticular nuclei of the brain. It is an autosomal recessive disorder which localizes to chromosome 13.37 Clinical manifestations include liver disease, movement disorders of tremor, ataxia, dysarthria and rigidity, and psychiatric symptoms. The Kayser±Flesicher ring of copper deposition in Descemet's membrane is characteristic, while the de®nitive diagnosis is made by low serum ceruloplasmin, increased urinary copper, and elevated liver copper levels. Atrophy, particularly of the gray matter, is a hallmark of Wilson's disease. Focal white matter lesions may occur which are hypointense on short TR, but of variable intensity on long TR.38 These lesions appear to have a reversible component during appropriate therapy.38±40 Clinical neurological symptoms appear to correlate with the presence of radiological abnormalities. The location of lesions can be useful in predicting the type of neurological manifestation.41 5.3 Parkinson's Disease Parkinson's syndrome, or Parkinsonism, is a broad clinical term referring to a cluster of symptoms due to loss of the majority of the dopaminergic efferent projections of the substantia nigra. The predominant clinical symptom is an involuntary tremor with rigidity and akinesia. The syndrome can be caused by a wide variety of etiologies, including stroke, trauma, in¯ammatory processes (particularly an epidemic which occurred after World War I, von Economo's encephalitis, now rarely with viral illnesses such as Cocksackie B, Japanese B and St. Louis viral encephalitides), drug effects, and associations with other degenerative diseases such as progressive supranuclear palsy, striatonigral degeneration, and Parkinson's disease. The

3

®nal common pathway is loss of the dopaminergic effects of these nigral projections on the striatal structures. Parkinson's disease is a true degenerative disease of the CNS with idiopathic loss of nigral±striatal projections. This entity is common, with an age of onset in the 50±60 years range, affecting approximately 1 in 1000 of people over the age of 50 years.27 The disease is progressive, probably due in part to ongoing loss of nigrostrial function. Early in the course of the disease, patients may respond well to a variety of medications, often with progressive resistance as the degeneration advances. On pathological examination, there are characteristic cell inclusions called Lewy bodies in the neurons of the substantia nigra, as well as in several related regions of the brain. MRI of Parkinson's disease patients reveals a decrement in the width of the pars compacta of the substantia nigra. This ®nding is shared with diseases leading to Parkinsonism such as striatonigral degeneration and progressive supranuclear palsy. Atrophy of the brain is also a characteristic that Parkinson's disease shares with other degenerative disorders, although sometimes this can be of a relatively small magnitude. 5.4

Striatonigral Degeneration

This degenerative disease presents with similar symptoms to Parkinson's disease. The primary differences rest with the much greater early involvement of the striatum (the putamen more than the caudate), which may in part explain the poor response to traditional Parkinsonism pharmacotherapy. This putaminal involvement may be detected on MRI with greater than normal hypointensity of the putamen on T2-weighted sequences. The magnitude of this ®nding has been reported to correlate with the severity of clinical involvement.42 A second area of detectable involvement occurs in the substantia nigra, with loss of width of the pars compacta. 5.5

Progressive Supranuclear Palsy

Named for the presence of supranuclear ophthalmoplegia of vertical gaze, this syndrome includes manifestations of pseudobulbar palsy, rigidity, extrapyramidal signs, and dementia, presenting in late middle age and in the elderly population. This degenerative disease localizes predominantly to the midbrain. In particular, the superior colliculi are atrophic, constituting a relatively speci®c MRI sign of the disease. Neuro®brillary tangles can be detected in the periaqueductal gray matter of the mesencephalon with gliosis±similar to the ®ndings in Alzheimer's disease.43,44 Changes in this region can be seen on MRI, with high signal intensity on relatively T2weighted sequences.43 As in the other degenerative diseases related to Parkinson's disease, there is diminished size of the pars compacta.

6

DEGENERATION OF THE CEREBELLUM AND BRAIN STEM

There are several degenerative diseases which have a propensity to attack the structures of the posterior fossa (Figure 4).

4 CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI occurs from a combination of congenital hypoplasia and perhaps ongoing degenerationÐparticularly in the cerebellum. The cerebellar loss is controversial, and other groups have shown greater effects on loss of gray matter volume. 6.3

Autism

Autism has been reported to be associated with ®ndings of focal loss of volume in the superior vermian lobules (declive, folium, and tuber), as well as more generalized volume loss in other parts of the brain.48 This ®nding has been contested by others.49 6.4

Figure 4 Sagittal T1-weighted MRI scan demonstrating idiopathic vermian atrophy in a patient with marked cerebellar symptoms

Olivopontocerebellar Degeneration

This rare neurodegenerative disorder can be inherited or sporadic. Ataxia is the presenting sign, and can occur at any age. Because of the nonspeci®c nature of the presentation, MRI may be particularly useful in providing a speci®c diagnosis. Pathological and imaging ®ndings include marked loss of volume in the body of the pons, middle cerebellar peduncles, cerebellum, and inferior olive. The pathways which interconnect these structures may show abnormal increased signal intensity on long TR sequences.50 This probably depends on the stage of axonal degeneration.

6.1 Down's Syndrome

6.5

Down's syndrome is a very important cause of inherited mental retardation. Supranormal amounts of the genetic material on chromosome 21 usually from nondisjunction are the underlying cause. MRI measurements show a global reduction in both gray and white matter. Interestingly, Down's syndrome is a very important risk factor for the ultimate development of Alzheimer's disease, with close to 100% of Down's patients developing a pathology similar to Alzheimer's disease in later life. The gene for the production of the amyloid- protein which has been linked to some of the pathology in Alzheimer's disease also maps to chromosome 21. Reports of older patients with Down's syndrome (up to 64 years old) show signi®cant widening of the temporal horns on MRI±similar to what has been reported in patients affected by Alzheimer's disease.45 Finally, NMR spectroscopic analysis has shown a rapid decline in N-acetylaspartate (NAA)/choline in older Down's patients, which correlates well with onset of clinical symptoms and precedes atrophic changes.46

Cerebellar cortical degeneration can present either with an early onset familial form or a late-onset sporadic form. There is a slowly progressive ataxia affecting the limbs, the trunk, and speech. Pathological examination and imaging studies both show marked cerebellar atrophy, particularly in the anterior vermis.

6.2 Fragile X and Rett's Syndrome The fragile X syndrome is a rare genetic cause of mental retardation. The syndrome is characterized by autism, repetitive movements, and hyperactive, self-destructive behaviors. The most profound expression is in males, but female carriers also show mild clinical and neuroradiological features of the disease. MRI reports have shown atrophy of the cerebellar vermis with concomitant enlargement of the fourth ventricle.47 In contrast, Rett's syndrome is a disease of girls which manifests also with autism and stereotyped movements with loss of language and motor skills. The brain appears to have a globally decreased volume, particularly in the frontal lobes. This probably

6.6

Cerebellar Cortical Degeneration

Acetazolamide-Responsive Familial Paroxysmal Ataxia

This unusual disorder is characterized by recurrent attacks of cerebellar symptoms which are associated with anterior vermian atrophy in the cerebellum. 6.7

Friedreich's Ataxia

Presenting in the second decade of life, this genetic degenerative disorder (localizing to chromosome 9 with multiple inheritance patterns) leads to a progressive ataxia. This leads to gait ataxia and dif®culty with speech, as well as kyphoscoliosis, probably on a muscular basis. In addition, to the degeneration of the spinocerebellar tracts, there is loss of the posterior columns of the spinal cord. The spinal cord is atrophic, particularly in the posterior and lateral columns. The majority of subjects also have cerebellar atrophy; however, this is less pronounced than other processes which affect the cerebellum.

7

CONCLUSIONS

MR imaging has revolutionized the detection and diagnosis of many of the neurodegenerative diseases. The patterns of

CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI

atrophy and Wallerian degeneration in these entities often allow a speci®c diagnosis to be made noninvasively. Moreover, in the near future, advances in MRI such as magnetization transfer techniques may allow much earlier detection and characterization of the degenerative processes within the neuron which lead to these tragic diseases.

8 RELATED ARTICLES Brain Neoplasms Studied by MRI; Hemorrhage in the Brain and Neck Observed by MRI; Intracranial Infections; Magnetic Resonance Imaging of White Matter Disease; Structural and Functional MR in Epilepsy.

9 REFERENCES 1. R. D. Adams and M. Victor, `Principles of Neurology', McGraw Hill, New York, 1989. 2. S. W. Atlas, `Magnetic Resonance Imaging of the Brain and Spine', Raven Press, 1991. 3. A. V. Waller, Philos. Trans. R. Soc. London, 1850, 140, 423. 4. R. J. Rossiter, `Chemical Pathology of the Nervous System', ed. J. Folch-Pi, Pergamon Press, New York, 1961. 5. P. M. Daniel and S. J. Strich, Acta Neuropathol. (Berlin), 1969, 12, 314. 6. H. Webster and F. De, Ann. N.Y. Acad. Sci., 1964, 122, 29. 7. L. D. DeWitt, J. P. Kistler, D. C. Miller, E. P. Richardson Jr., and F. S. Buonanno, Stroke, 1987, 18, 342. 8. S. R. Cobb and C. M. Mehringer, Radiology, 1987, 162, 521. 9. M. J. Kuhn, K. A. Johnson, and K. R. Davis, Radiology, 1988, 168, 199. 10. M. Bouchareb, T. Moulin, F. Cattin, J. L. Dretemann, A. Rack, H. Verdot, and J. F. Bonneville, J. Neuroradiol., 1988, 1988, 238. 11. A. Uchino, K. Onomura, and M. Ohno, Radiat. Med., 1989, 7, 74. 12. J. E. Pujol, J. L. Mari-Vilata, C. Junque, P. Vendrell, J. Fernandez, and A. Capdevila, Stroke, 1990, 21, 404. 13. T. Orita, T. Tsurutani, A. Izumihara, and T. Matsunaga, J. Comput. Assist. Tomogr., 1991, 15, 802. 14. Y. Inoue, Y. Matsumura, T. Fukuda, Y. Nemoto, N. Shirahata, T. Suzuki, M. Shakudo, S. Yawata, S. Tanaka, and K. Takemoto, Am. J. Roentgenol., 1990, 11, 897. 15. A. Danek, M. Bauer, and W. Fries, Eur. J. Neurosci., 1990, 2, 112. 16. N. P. Bontozoglou, D. W. Chakeres, G. F. Martin, M. A. Brogen, and R. B. McGhee, Radiology, 1991, 180, 223. 17. M. J. Kuhn, D. J. Mikulis, D. M. Ayoub, B. E. Kosofsky, K. R. Davis, and J. M. Taveras, Radiology, 1989, 172, 179. 18. A. Uchino, H. Imada, and M. Ohno, Neuroradiology, 1990, 32, 191. 19. N. Kumar, and F. J. Lexa, in preparation. 20. F. A. Jolesz, J. F. Polak, P. W. Ruenzel, and D. F. Adams, Radiology, 1984, 152, 85. 21. D. S. Titelbaum, J. L. Frazier, R. I. Grossman, P. M. Joseph, L. T. Yu, E. A. Kassab, W. F. Mickey, D. laRossa, and M. J. Brown, Am. J. Roentgenol., 1989, 10, 741. 22. R. I. Grossman, C. M. Hecht-Leavitt, S. M. Evans, R. E. Lenkinski, G. A. Holland, J. J. van-Winkle, J. T. McGrath, W. J. Curran, A. Shetty, and P. M. Joseph, Radiology, 1988, 169, 305. 23. F. J. Lexa, R. I. Grossman, and A. C. Rosenquist, Am. J. Roentgenol., 1994, 15, 201.

5

24. M. L. Schmidt, V. M.-Y. Lee, and J. Q. Trojanowski, Lab. Invest., 1989, 60, 513. 25. J. W. Dahlbeck, K. W. McCluney, J. W. Yeakley, M. J. Fenstermacher, C. Bonmati, G. van-Horn, and J. Aldeg, Am. J. Roentgenol., 1991, 12, 931. 26. J. L. Cummings, and L. W. Duchen, Neurology, 1981, 31, 1415. 27. A. Barbeau, `Handbook of Clinical Neurology', Elsevier, Amsterdam, 1986, Vol. 49, p. 87. 28. E. B. Larson, B. V. Rei¯er, S. M. Sumi, C. G. Con®eld, and N. M. Chinn, Arch. Intern. Med., 1986, 146, 1917. 29. C. Wikkelso, H. Andersson, C. Blomstrand, M. Matousek, and P. Svenelson, Neuroradiology, 1989, 31, 160. 30. W. G. Bradley Jr., A. R. Whittemore, A. S. Watanabe, S. T. Davis, L. M. Teresi, and M. Komyak, Am. J. Roentgenol., 1991, 12, 31. 31. T. El Gammal, M. B. Allen, B. S. Brooks, and E. K. Mark, Am. J. Roentgenol., 1987, 8, 591. 32. W. G. Bradley Jr., K. E. Kortman, and B. Burgoyne, Radiology, 1986, 159, 611. 33. J. F. Gusella, N. S. Wexler, P. M. Conneally, S. L. Naylor, M. A. Anderson, R. E. Tanzi, P. C. Watkins, K. Ottina, M. R. Wallace, and A. Y. Sakaguchi, Nature, 1983, 306, 234. 34. S. T. Grafton, J. C. Mazziotta, J. J. Pahl, P. St. George Hyslop, J. L. Haines, J. Gusella, J. M. Hoffman, L. R. Baxter, and M. E. Phelps, Arch. Neurol., 1992, 49, 1161. 35. T. L. Jernigan, D. P. Salmon, N. Butters, and J. R. Hesselink, Biol. Psychol, 1991, 29, 68. 36. G. J. Harris, G. D. Pearlson, C. E. Peyser, E. H. Aylward, J. Roberts, P. E. Barta, G. A. Chose, and S. E. Folstein, Ann. Neurol., 1992, 31, 69. 37. S. Starosta-Rubinstein, A. B. Young, K. Kluin, G. Hill, A. M. Aisen, T. Gabrielsem, and G. J. Brewer, Arch. Neurol., 1984, 44, 365. 38. L. Prayer, D. Wimberger, J. Kramer, G. Grimm, K. Oder, and H. Imhof, Neuroradiology, 1990, 32, 211. 39. H. Nazer, J. Brismar, M. Z. Al-Kawi, T. S. Gunasekaran, and K. H. Jorulf, Neuroradiology, 1993, 35, 130. 40. K. A. Thuomas, S. M. Aquilonius, K. Bergstrom, and K. Westermark, Neuroradiology, 1993, 35, 134. 41. W. Oder, L. Prayer, G. Grimm, J. Spatt, P. Ferenci, K. Holleger, B. Schneider, A. Gangl, and L. Deeke, Neurology, 1993, 43, 120. 42. R. Brown, R. J. Polinsky, G. Di Chiro, B. Postakia, L. Wener, and J. J. Simmons, J. Neurol. Neurosurg. Psychiatry, 1987, 50, 913. 43. M. Savoiardo, L. Strada, F. Girotti, L. D'Incerti, M. Sherna, P. Soliveri, and A. Bolzarini, J. Comput. Assist. Tomogr., 1989, 13, 555. 44. M. L. Schmidt, V. M.-Y. Lee, J. Q. Trojanowski, and H. Hurtig, Lab. Invest., 1988, 59, 45. M. LeMay and N. Alvarez, Neuroradiology, 1990, 32, 104. 46. T. Murata, Y. Koshino, M. Omori, I. Murati, M. Nishio, T. Horie, Y. Umezawa, K. Isaki, H. Kimura, and S. Itoh, Biol. Psychol., 1993, 34, 290. 47. E. H. Aylward and A. Reiss, J. Psychiatr. Res., 1991, 25, 159. 48. E. Courchesne, G. Press, and R. Yeung-Courchesne, Am. J. Roentgenol., 1993, 160, 387. 49. M. A. Nowell, D. B. Hackney, A. S. Muraki, and M. Coleman, JMRI, 1990, 8, 811. 50. M. Savoiardo, L. Strada, F. Girotti, R. A. Zimmerman, M. Grisoli, D. Testa, and R. Petrillo, Radiology, 1990, 174, 693.

Biographical Sketch Frank J. Lexa. b 1958. A.B. (Biology), Harvard University, USA, 1980. M.S. (Physiology), 1982, M.D. 1985, Stanford University School of Medicine, USA. Postgraduate medical training in internal medicine, Brigham and Women's Hospital, Harvard Medical School,

6 CENTRAL NERVOUS SYSTEM DEGENERATIVE DISEASE OBSERVED BY MRI and diagnostic radiology/neuroradiology, Hospital of the University of Pennsylvania, USA. Currently Fellow, Leonard Davis Institute of Health Care Economics and Graduate MBA Candidate, Class of '99, The Wharton School Director, Medical Acquisitions, BTG Interna-

tional. Approx. 30 publications. Current research specialties: early detection and characterization of axonal degeneration using magnetization transfer techniques, MRI contrast agent development for advanced medical applications.

CRANIAL NERVES INVESTIGATED BY MRI

Cranial Nerves Investigated by MRI Anton N. Hasso University of California, Irvine, CA, USA

and Peggy J. Fritzsche Loma Linda University, Loma Linda, CA, USA

1 INTRODUCTION The reader of this article will gain an understanding of the normal pathways of the cranial nerves and be able to correlate the function with the course of each nerve. The functional anatomy will also be addressed in conjunction with the clinical manifestations of pathologic entities. Tumors, infections, and in¯ammations of the cranial nerves will be discussed separately. 2 NORMAL ANATOMY 2.1 Cranial Nerve I (Olfactory) Smell is consciously perceived in the gray matter covering the medial and lateral striae of the olfactory gyri. The lateral striae travel from the inferior medial aspect of the temporal lobe (pyriform area) and the medial striae from the medial and inferior aspects of the frontal lobe (subcallosal region) to the olfactory trigone located at the anterior perforated substance. The intermediate stria is not signi®cant in humans, but joins the lateral and medial striae at the trigone to extend anteriorly as the olfactory tract. The olfactory tract terminates in the olfactory bulb. The olfactory bulbs are located inferior to the olfactory sulci (Figure 1). The olfactory sulci separate the gyri rectus from the medial orbitofrontal gyri. Sensory nerve bundles exit the olfactory bulbs via the cribriform plate of the skull to separate into multiple neurosensory cells which reside in the nasal epithelium. The function of the olfactory nerve is to provide olfaction. Neuropathy of the ®rst cranial nerve manifests as anosmia, or acute loss of olfaction. Anosmia is usually unilateral, and typically occurs secondary to pathologic processes arising in the nasal cavities. Intracranial neoplasms, infections, arterial disease, or congenital anomalies also can cause anosmia.1

1

the lateral geniculate body of the thalamus [Figure 2(a)]. Bilateral optic tracts then exit the lateral geniculate bodies and meet in the midline to form the optic chiasm. From the optic chiasm ventrally, the optic ®bers then divide into the bilateral optic nerves coursing initially into the cranial openings of the optic canals. The optic nerves exit the orbital openings of the optic canals and extend forward in a sinusoidal manner to enter the globes at the optic nerve heads. Lesions involving the intraorbital and prechiasmal portions of the optic nerve result in total monocular blindness. The sudden onset of a monocular de®cit suggests a possible vascular or demyelinating process. The slow onset of a monocular optic neuropathy may be caused by a lesion intrinsic to the optic nerve sheath or secondary to extrinsic compression by an adjacent mass. The ®bers from the outer (temporal) portion of both visual ®elds cross at the optic chiasm and thus are carried via the optic tracts to the contralateral calcarine cortex of the occipital lobes. The ®bers from the medial (nasal) portion of the visual ®elds do not cross at the optic chiasm and are projected to the ipsilateral visual cortex. Lesions involving the central portion of the chiasm particularly produce bitemporal hemianopsias. Lesions along the optic tracts result in loss of vision in the ipsilateral (uncrossed) nasal ®eld and contralateral (crossed) temporal ®eld. Since the retrochiasmal segment of the visual pathway recognizes the opposite visual ®eld, a retrochiasmal lesion leads to a homonomous hemianopsia contralateral to the lesion. Contralateral hemianopsias may be produced by lesions involving any portion of the retrochiasmal optic pathways including the lateral geniculate bodies, the optic radiations, or the occipital cortices.1±3 2.3

Cranial Nerve III (Oculomotor)

The motor component of cranial nerve III originates in the motor cortex of the cerebrum. The ®bers then travel centrally

2.2 Cranial Nerve II (Optic) The optic ®bers originate in the calcarine portion of the occipital cortex and travel superiorly, inferiorly, and laterally to the occipital horn of the ventricle. The optic ®bers then become the optic radiations as they course over the temporal horn of the lateral ventricle and the tail of the caudate nucleus to enter

Figure 1 Normal anatomy of the olfactory bulbs. The olfactory bulbs (arrows) are located inferior to the olfactory sulci which separate the gyrus rectus from the medial orbitofrontal gyrus

2 CRANIAL NERVES INVESTIGATED BY MRI

Figure 2 (a, b)

CRANIAL NERVES INVESTIGATED BY MRI

Figure 2 (c, d)

3

4 CRANIAL NERVES INVESTIGATED BY MRI

Figure 2 Illustrations of normal anatomy of brainstem and cranial nerves. (a) Lateral view of brainstem with motor nuclei. (b) Posterior view of brainstem with sensory and motor nuclei. (c) Normal axial anatomy of the brainstem at the level of the midbrain. (d) Normal axial anatomy of the brainstem at the level of the pontine isthmus. (e) Normal axial anatomy of the brainstem at the pons. (f) Normal axial anatomy of the brainstem at the upper medulla. (g) Normal axial anatomy of the brainstem at the middle medulla. CN, cranial nerve. Reproduced with permission from W. G. Bradley, Jr, Radiology, 1991, 179, 319

CRANIAL NERVES INVESTIGATED BY MRI

5

Figure 3 Lymphomatous meningitis in a patient with AIDS. Axial enhanced images with fat saturation pulses. (a) Note the focal nodule in the right side of the interpeduncular cistern (arrow). (b) The characteristic V-shape of the oculomotor nerves in the prepontine cistern is well shown (arrows)

through the internal capsule to the ipsilateral superior colliculus. The ®bers cross between the red nuclei in the dorsal tegmental decussation and continue distally in the paramedian raphe to synapse in the pontine reticular formation. The ®bers then ascend in the medial longitudinal fasciculus to the oculomotor nucleus which is situated in the paramedian midbrain tegmentum ventral to the aqueduct of Sylvius at the level of the superior colliculus. At this level, additional ®bers from the parasympathetic nucleus, paramedian nucleus, Perlia's nucleus, and Edinger±Westphal nucleus join the motor ®bers. These combined ®bers bow laterally to extend through the medial aspect of the red nucleus and the cerebral peduncle (Figure 2). Fibers then exit from the brainstem anteriorly at the pontomedullary junction near the midline where the two nerves form parallel structures in a V-con®guration as they extend through the interpeduncular fossa (Figure 3). The oculomotor nerves then travel between the superior cerebellar and posterior cerebral arteries through the prepontine cistern. Each cranial nerve III then penetrates the dura of the cavernous sinus and rises superiorly along the lateral wall of the cavernous sinus adjacent to the fourth cranial nerve and above the sixth cranial nerve and ophthalmic division of the ®fth cranial nerve. The parasympathetic and motor ®bers are in close proximity, and altogether the nerves penetrate the superior orbital ®ssure to reach the orbital structures. The motor ®bers provide the somatic motor function to the medial rectus, inferior rectus, superior rectus, and inferior oblique extraocular muscles. Lesions of this portion of the oculomotor nerve lead to a downward abducted eye, due to the unopposed action of the superior oblique and lateral rectus muscles. The oculomotor nerve also supplies the levator palpebrae superioris, which elevates the eyelid. Lesions of this portion of cranial nerve III also result in lid droop or ptosis. The visceral motor ®bers from the Edinger±Westphal nucleus

provide parasympathetic motor function to the pupillary constrictor and ciliary muscles. Lesions of this portion of the oculomotor nerve result in pupillary dilatation due to unopposed sympathetic input and failure to accommodate. Isolated third-nerve palsies are regarded as either complete (including parasympathetic involvement) or incomplete (due to lack of parasympathetic involvement).1±3 2.4

Cranial Nerve IV (Trochlear)

The motor impulses of cranial nerve IV originate in the motor cortex and then extend centrally via the internal capsule to the ipsilateral superior colliculus. The ®bers cross between the red nuclei in the dorsal tegmental decussation and travel caudally in the paramedial raphe to synapse in the pontine reticular formation. The ®bers then ascend in the medial longitudinal fasciculus to reach the motor nucleus. The fourth cranial nerve nucleus is located in the tegmentum of the mesencephalon at the level of the inferior colliculus immediately caudal to the third nerve nucleus. Fibers from the nucleus loop posteriorly to decussate in the tectum of the lower midbrain beneath the inferior colliculi. This pathway forms a sickle-shaped arch around the aqueduct of Sylvius (Figure 2). The trochlear nerve exits from the dorsal aspect of the brainstem, emerging from the superior medullary vellum just beneath the inferior colliculus. The nerve extends around the cerebral peduncle between the superior cerebellar and posterior cerebral arteries, just lateral to the third cranial nerve (Figure 4). The nerve then pierces the dura of the cavernous sinus coursing along the superior margin adjacent to the oculomotor nerve and the ophthalmic portion of the trigeminal nerve en route to the superior orbital ®ssure.

6 CRANIAL NERVES INVESTIGATED BY MRI

Figure 4 Tectal and right trochlear nerve tumor in a patient with neuro®bromatosis I. (a) Axial image at the level of the inferior colliculus. There is a hyperintense mass in the right tectum encompassing the nucleus of cranial nerve IV. Note the incidental dysplasia of the left sphenoid wing which is commonly seen in patients with NF-1. (b and c) Sagittal unenhanced images. There is a hypointense tumor expanding the inferior colliculus [arrow, (b)]. Note the expansion of the right trochlear nerve in the prepontine cistern [broad arrow, (c)]. (d) Axial enhanced image. The tumor foci in the tectal plate (white arrow) and in the right prepontine cistern (black arrow) do not enhance signi®cantly. A hamartoma or low grade glioma is suspected. Note the uninvolved oculomotor nerves in the prepontine cistern

The fourth cranial nerve innervates the superior oblique muscle in the orbit. Lesions of the trochlear nerve or its nucleus result in outward rotation of the affected eye. Because the trochlear nerve has the longest intracranial course (7.5 cm), it has the greatest chance of injury from trauma or surgery in the region of the midbrain.1,3,4 2.5 Cranial Nerve V (Sensory Trigeminal) The sensory impulses from the upper, middle, and lower part of the face extend to the gasserian ganglion via the ophthalmic (V1), maxillary (V2), and mandibular (V3) branches of cranial nerve V. These branches enter the skull via

the superior orbital ®ssure, foramen rotundum and foramen ovale, respectively. The ophthalmic and maxillary divisions extend through the cavernous sinus, while the mandibular branch bypasses the cavernous sinus and directly enters the trigeminal (gasserian) ganglion. This ganglion lies in the inferior portion of Meckel's cavity and contains the cell bodies of numerous afferent sensory ®bers. Facial numbness and burning (tic douloureux) represent signs and symptoms caused by sensory trigeminal neuropathies. Sensory impulses from the face then leave the trigeminal ganglion to enter the midlateral pons through the prepontine cistern. As they enter the brainstem, branches go to the three sensory nuclei: the principal sensory nucleus, the mesencepha-

CRANIAL NERVES INVESTIGATED BY MRI

7

Figure 5 Plexiform neuro®broma involving the left trigeminal nerve. Adult patient with neuro®bromatosis I. Enhanced images with fat saturation pulses. (a and b) Coronal images demonstrate the expansion and enhancement in the left maxillary division [arrow, (a)] and mandibular division [arrow, (b)] of the trigeminal nerve. (c and d) Axial images demonstrate the course of the left axillary nerve from the trigeminal ganglion through the foramen rotundum [arrows, (c)]. Note the expansion of the tumor into the left foramen ovale and pterygopalatine fossa [arrows, (d)]

lic nucleus, and the spinal nucleus. Pain and temperature ®bers descend to the C2±C3 cord level where they synapse in the spinal nucleus of cranial nerve V. These ®bers then cross the midline ascending in the ventral trigeminal lemmiscus to reach the ventral posteromedial nucleus of the thalamus. The thalamocortical tract then carries the ®bers via the internal capsule to the sensory cortex. The sensory impulses that mediate proprioception from the face, particularly of the mandible, synapse in the mesencephalic nucleus of cranial nerve V which extends superiorly into the periaqueductal gray matter of the midbrain. This portion of the sensory nucleus is composed of the sensory neurons that travel with the mandibular nerve. Sensory nerves that mediate touch and pressure from the face synapse in the principal sensory nucleus of cranial nerve V and travel to the ventral posteromedial nucleus of the thalamus via the contralateral dorsal trigeminal lemmiscus. The extension to the sensory cortex is via the thalamocortical tract which passes through the internal capsule (Figure 2).1±3,5

2.6 Cranial Nerve V (Motor Trigeminal) The impulses for the motor components of cranial nerve V originate in the motor cortex of the cerebral hemisphere. The pathway then extends through the genu of the internal capsule and the cerebral peduncle (along the corticobulbar tract) to the motor nucleus which is situated medial to the principal sensory nucleus. The fused motor and sensory components of the ®fth nerve emerge from the ventrolateral pons, coursing through the anterior portion of the cerebellopontine angle cistern near the apex of the medial petrous ridge. The nerve ®bers then pierce the dura of Meckel's cavity and enter the trigeminal ganglion. At the distal aspect of Meckel's cavity, the motor ®bers of the mandibular nerve exit at the inferolateral surface and extend through the foramen ovale (Figure 5). The motor division of the mandibular nerve supplies the muscles of mastication plus other small muscles (tensor villi palatini, mylohyoid, anterior belly of the digastric and tensor tympani).

8 CRANIAL NERVES INVESTIGATED BY MRI 2.7

Cranial Nerve VI (Abducens)

The impulses of cranial nerve VI originate in the motor cortex of the cerebral hemisphere. The ®bers extend through the internal capsule and the corticobulbar tract to the ipsilateral superior colliculus. At this point, the ®bers cross between the red nuclei in the dorsal tegmental decussation. They then extend caudally in the paramedian raphe of the brainstem to synapse in the pontine reticular formation. Both ipsilateral and contralateral ®bers ascend in the medial longitudinal fasciculus to reach the brainstem nucleus of the abducens nerve which lies in the pons immediately anterior to the fourth ventricle. These ®bers pass through the pontine tegmentum and exits anteriorly from the brainstem at the pontomedullary junction. The abducens nerve then courses superiorly and slightly laterally along the clivus. At the base of the dorsum sella, the sixth cranial nerve courses anteriorly piercing the dura through Dorello's canal to enter the cavernous sinus. The nerve then extends obliquely through the cavernous sinus just inferior and medial to cranial nerves III and IV, immediately adjacent to the cavernous portion of the internal carotid artery. Cranial nerve VI then exits anteriorly through the cavernous sinus to enter the superior oblique ®ssure en route to the lateral rectus muscle (Figure 2). An isolated lateral rectus palsy is the most common lesion of the cranial nerves supplying the extraocular muscles. Injury to the abducens nerve produces paralysis of the ipsilateral lateral rectus muscle, with resultant horizontal diplopia. The unopposed pull of the medial rectus muscle causes the eye to turn inward (adduct), thereby producing internal strabismus.1±3 2.8

Figure 6 Perineural extension of squamous cell carcinoma. Adult patient with previous known facial cancer. Coronal enhanced images. (a) Note the lateral deviation of the cavernous sinus wall and enhancement along the right trigeminal ganglion (arrow). (b) There is marked denervation atrophy of the right muscles of mastication. The fat which replaces the muscle shows moderate diffuse enhancement

The exit point of the motor trigeminal nerve from the brainstem is also the entry point of the sensory trigeminal nerve. This area is termed the `root entry zone' of the sensory trigeminal nerve and is subject to vascular compression by dolichoectatic vessels or aneurysms which may cause tic douloureux or other trigeminal neuropathies. Alternatively, an ipsilateral or contralateral posterior fossa tumor or other mass lesion can torque the brainstem and cause a distention of this root entry zone. If the motor trigeminal portion is affected by a pathologic process, weakness of the muscles of mastication (with or without denervation atrophy) can occur (Figure 6).1±3,5

Cranial Nerve VII (Facial)

The motor impulses of cranial nerve VII originate in the motor cortex of the cerebral hemisphere and extend via the genu of the internal capsule and cerebral peduncle to the brainstem nucleus. The motor nucleus of cranial nerve VII is situated in the caudal one-third of the ventral pontine tegmentum. Fibers from the ipsilateral and contralateral sides meet in this brainstem nucleus. From the nucleus, the ®bers extend posteriorly toward the ¯oor of the fourth ventricle and loop around the nucleus of cranial nerve VI (abducens). This portion of the facial nerve contributes to the facial colliculus, a mound of tissue that protrudes internally towards the ¯oor of the fourth ventricle. Extending from the loop, cranial nerve VII emerges from the brainstem at the anterolateral aspect of the pontomedullary junction. The nerve then extends laterally in the cerebellopontine angle cistern reaching the anterior superior quadrant of the internal auditory canal. The facial nerve extends through the temporal bone to the fundus of the internal auditory canal. It then exits to join the geniculate ganglion within the roof of the petrous bone (Figure 7). At this point, there is a synapse with the sensory neurons for taste to the anterior two-thirds of the tongue and the cutaneous ®bers for light touch in the region of the external ear. The superior salivatory nucleus in the pons is the site of origin of the parasympathetic ®bers that terminate in and stimulate the lacrimal, sublingual, and submandibular glands. The nucleus solitarius represents the end point of the ®bers that convey taste sensation from the anterior two-thirds (via cranial

CRANIAL NERVES INVESTIGATED BY MRI

9

Lesions of the facial nerve result in a facial palsy, which can be either peripheral or central. Central paralysis refers to a supranuclear injury that becomes manifest as paralysis of the contralateral muscles of facial expression, with sparing of the muscles of the forehead. Peripheral facial nerve function implies injury to the facial nerve from the level of the brainstem nucleus to the end of its motor ®bers. All the ipsilateral muscles of facial expression are paralyzed in a peripheral injury. Dysfunction of the facial nerve branches within the greater super®cial petrosal nerve is manifested as an impairment of ipsilateral lacrimation. Dysfunction of the stapedius branch of the facial nerve is manifested by hyperacusis. Dysfunction of the chorda tympani nerve causes loss of taste from the anterior two-thirds of the tongue. An impression on the `root exit zone' of the facial nerve from the brainstem by a vascular anomaly may lead to spasmodic contractures of the ipsilateral muscles of facial expression (facial tic) (Figure 8).1±3,6 2.9

Figure 7 Axial enhanced images of the geniculate ganglion and proximal facial nerve schwannoma. (a) Note the tumor enhancement extending through the fundus of the left internal auditory canal to the geniculate ganglion (arrow). (b) The expansion of the tumor along the surface of the left petrous bone into the middle cranial fossa can be clearly seen (arrow)

nerve VII) and from the posterior third (via cranial nerve IX) of the tongue. The cell bodies of these taste ®bers are found in the geniculate ganglion. The impulses from these nuclei form the intermediate nerve ®bers which join the motor root of the facial nerve as it exits the brainstem (Figure 2). From the region of the geniculate ganglion, the secretomotor ®bers project forward as the greater super®cial petrosal nerve which joins some parasympathetic ®bers from the carotid artery to form the nerve of the pterygoid canal. The pterygoid canal will transmit the nerve to the pterygopalatine fossa from where the parasympathetic components synapse and ®bers extend to affect lacrimation and salivation. From the geniculate ganglion, the facial motor ®bers and taste ®bers form a 180 turn and are directed posteriorly, at the anterior genu of the facial nerve. Extending posteriorly, these ®bers lie at the medial aspect of the tympanic cavity medial to the ossicular chain and below the lateral semicircular canal. At the posterior aspect of the middle ear, the motor and taste ®bers extend inferiorly at the posterior genu of the facial nerve to enter the stylomastoid foramen. At this point, the stapedial nerve exits to supply the stapedial muscle which controls the mobility of the ossicles. The chorda tympani ®bers for taste also separate from the seventh nerve within the stylomastoid foramen extending via the petrotympanic ®ssure and eventually joining the lingual branch of the mandibular nerve to participate in the sensory ®bers for taste to the anterior two-thirds of the tongue. The seventh nerve then exits the stylomastoid foramen into the parotid gland and extends anteriorly to innervate the muscles of facial expression as well as the posterior belly of the digastric, the stylohyoid, and the platysma muscles.

Cranial Nerve VIII (Vestibular)

Movement impulses are modulated by the hair cells within the utricle, saccule, and three semicircular canals and transmitted through the superior and inferior vestibular nerves (Figure 9). The combined vestibular nerve courses along with the cochlear nerve and facial nerve within the internal auditory canal. The superior and inferior vestibular divisions enter the brainstem at the pontomedullary junction after crossing the cerebellopontine angle cistern. These ®bers extend into the vestibular nuclear complex with postganglionic ®bers carrying information regarding equilibrium to many coordination sites including the folliculonodular lobe of the cerebellum, the vestibulospinal tract, both medial longitudinal fasciculi and other pathways which affect the control of the eyes and the muscles of balance (Figure 2).1±3 2.10 Cranial Nerve VIII (Cochlear) Sound waves from the external environment are converted to energy in the inner ear in the form of vibratory impulses at the oval window of the vestibule. The kinetic impulses are propagated as ¯uid waves in the cochlea. These impulses are then converted to electrical impulses along the spiral organ of Corti. The auditory signals are then transmitted via the spiral ganglion in the cochlear nerve through the internal auditory canal and cerebellopontine angle to the origin of cranial nerve VIII at the ventrolateral pontomedullary junction. There are two cochlear nuclei (dorsal and ventral) located lateral to the inferior cerebellar peduncles in the upper medulla. The dorsal cochlear nucleus carries high frequencies, while the ventral cochlear nucleus connects low frequencies. Somewhat more than half of the ®bers from these nuclei cross to the contralateral superior olivary nucleus to form the trapezoid body. From the trapezoid body, the ®bers enter the posteriorly located lateral lemniscus to ascend to the inferior colliculus in the midbrain. From the inferior colliculus, the ®bers travel to the medial geniculate body of the thalamus and then, via the auditory radiations to the superior temporal gyrus (Figure 2). Clinical manifestations of acoustic pathway injury depend on the level of the injury. Lesions of the cochlear portion of the eighth nerve result in hearing loss and tinnitus (ringing or roaring in the ear). If unilateral sensorineural hearing loss is

10 CRANIAL NERVES INVESTIGATED BY MRI low the corticobulbar tract through the midbrain to the nucleus ambiguus. The ®bers extend to the posterior olivary sulcus leaving the medulla at this point to extend anterolaterally to the superior and inferior glossopharyngeal ganglia situated in the jugular foramen. The ®bers then travel inferiorly to supply the stylopharyngeus and constrictor muscles of the pharynx. In addition to this motor component to the muscles, there is a solitary nucleus for sensory ®bers and a salivatory nucleus for parasympathetic ®bers. These nuclei represent the primary sensory neurons for general visceral and special sensation, including taste, for the posterior third of the tongue (Figure 2). The sensory portion of the glossopharyngeal nerve joins the motor portion which then extend together into the pars nervosa of the jugular foramen (Figure 10). There is a small tympanic branch (Jacobson's nerve) which supplies sensation to the middle ear and eustachian tube as well as parasympathetic ®bers to the parotid gland. A small visceral branch supplies the carotid sinus and carotid body which controls the pressor and chemoreceptor functions. Pharyngeal sensory branches supply the posterior oropharynx and soft palate. The lingual branch includes ®bers for both sensation and taste for the posterior third of the tongue. Clinical manifestations of ninth nerve dysfunction include otalgia (referred pain along the tympanic branch to the ear), dysphagia (stylopharyngeus muscle dysfunction), and tachycardia or bradycardia with hypotension (sinus nerve dysfunction). Other symptoms de®nitely related to ninth nerve injury are loss of the afferent limb of the gag re¯ex and loss of taste to the posterior third of the tongue.1±3

Figure 8 Enhanced images of dolichoectasia of the basilar artery compressing the root exit zone of the left facial nerve. (a) This axial view shows an indentation by the basilar artery on the brainstem at the origin of the left facial nerve (open arrow). (b) This coronal view documents the tortuous course of the left basilar artery, including the brainstem compression at the origin of the facial nerve (arrow)

present, the injury has occurred somewhere between the cochlea and the cochlear nuclei of the brainstem. Unilateral involvement of the auditory pathway above the cochlear nuclei usually causes bilateral hearing loss due to the multiple crossings through the acoustic pathways. The hearing loss is greater in the ear contralateral to the site of the lesion. Cortical acoustic pathway lesions, which rarely cause disruption of auditory function, may result in auditory agnosia.1±3,7 2.11 Cranial Nerve IX (Glossopharyngeal) Impulses from the motor cortex of the cerebral hemisphere extend to the genu of the internal capsule and subsequently fol-

Figure 9 Vestibular schwannoma of the left utricle and saccule. (a) Unenhanced axial image. The small tumor of the membranous labyrinth is dif®cult to identify within the left petrous temporal bone (arrow). (b) Enhanced axial image. The bulbous tumor in the left vestibule is clearly shown (arrow)

CRANIAL NERVES INVESTIGATED BY MRI

11

jugular foramen. As previously discussed, the ninth cranial nerve enters the pars nervosa of the jugular foramen. After exiting the jugular foramen, the vagus nerve plunges like a plumb line along the posterolateral aspect of the carotid artery to the aortopulmonic window of the mediastinum on the left side and to the clavicle on the right side. The right recurrent laryngeal branch turns cephalad around the right subclavian artery, while the left recurrent laryngeal branch turns cephalad by looping through the aortopulmonic window. Both recurrent laryngeal nerves reach the larynx via the tracheoesophageal groove. Sensory ®bers of the vagus nerve begin in the thoracoabdominal viscera and chemoreceptors at the aortic arch. These then synapse in the solitary nucleus (pure sensory ®bers) and the dorsal motor nucleus (parasympathetic secretomotor ®bers) (Figure 2). Injury to the vagus nerve below the hyoid bone causes manifestations of recurrent laryngeal nerve malfunction and include hoarseness, aspiration, and cervical dysphagia. Injury to the vagus nerve above the hyoid bone results in the above symptoms in combination with oropharyngeal symptoms of uvular deviation and loss of the efferent limb of the gag re¯ex secondary to pharyngeal plexus malfunction. Proximal vagal neuropathy usually involves some combination of cranial nerves IX, X, XI, and XII.1±3,8 2.13 Cranial Nerve XI (Spinal Accessory)

Figure 10 Schwannoma of the right glossopharyngeal nerve in the jugular foramen. (a) Coronal image. The homogeneously hyperintense tumor smoothly expands the right jugular foramen. The apex of the tumor is along the brainstem with its inferior portion extending through the right side of the skull base. (b) Axial image. The homogeneously hyperintense tumor mass in the right side of the skull base is clearly seen on this T2-weighted image

2.12 Cranial Nerve X (Vagus) Impulses from the motor cortex of the cerebral hemisphere travel through the genu of the internal capsule and cerebral peduncle via the corticobulbar tract to the nucleus ambiguus. The ®bers extend to the posterior olivary sulcus slightly more lateral than cranial nerve IX exiting the medulla in the groove between the inferior peduncle and olive. Cranial nerve X continues an anterolateral course with its companion nerves (cranial nerve IX and the bulbar portion of cranial nerve XI) through the perimedullary cistern. The vagus nerve and bulbar portion of the eleventh cranial nerve enter the anterior aspect of the pars vascularis of the

Impulses from the motor cortex of the cerebral hemisphere extend through the genu of the internal capsule and cerebral peduncle via the corticobulbar tract to the nucleus ambiguus. The motor ®bers of the bulbar portion of the spinal accessory nerve are joined by additional ®bers which arise from the anterior horn cells of the ®rst ®ve cervical cord segments. The ®bers from the spinal nucleus have ascended through the foramen magnum. Together, the combined ®bers from the bulbar and spinal portions of the spinal accessory nerve pass anteriorly to exit from the medulla at the posterior olivary sulcus. The eleventh cranial nerve joins the companion cranial nerves IX and X to pass through the perimesencephalic cistern. The spinal accessory nerve then leaves the skull via the posterior portion of the pars vascularis of the jugular foramen. From the skull base, the eleventh cranial nerve enters the carotid space and then quickly diverges posterolaterally to descend along the medial aspect of the sternocleidomastoid muscle. It then continues its course across the posterior triangle of the neck to terminate in the trapezius muscle. Clinical manifestations of spinal accessory nerve injury are often found in conjunction with ninth, tenth, and twelfth nerve symptoms. When isolated, the symptoms will consist of shoulder droop and an inability to lift the arm.1±3 2.14 Cranial Nerve XII (Hypoglossal) Impulses from the motor cortex travel via the corona radiata through the genu of the internal capsule and cerebral peduncle via the corticobulbar tract to reach the hypoglossal nucleus, which is found in a paramedian location in the ¯oor of the fourth ventricle. Hypoglossal nerve root ®bers as well as ®bers from the nucleus ambiguus exit forward from the medulla passing lateral to the medial lemniscus. The ®bers exit from the

12 CRANIAL NERVES INVESTIGATED BY MRI cases, schwannomas can be removed by surgery without transecting the nerve. Neuro®bromas, however, are part of the nerve which has to be excised with the tumor. 3.2

Figure 11 Multiple neuro®bromas in a patient with neuro®bromatosis I. There are bilateral neuro®bromas in the carotid spaces which surround the carotid arteries. The lesions are readily identi®ed by their high signal intensity (arrows). Additional smaller neuro®bromas are seen in the posterior scalp region

brainstem in the sulcus between the pyramid and olive as multiple small rootlets. The rootlets extend anteriorly to form the hypoglossal nerve extending anterolaterally through the hypoglossal canal. Once exited from the skull base, the ®bers descend inferiorly in the carotid space lying medial to cranial nerves IX, X, and XI and passing lateral to the carotid bifurcation. The hypoglossal nerve then continues forward to enter the posterior portion of the sublingual space of the oral cavity. From here it curves upward medially to the submandibular salivary glands to supply the intrinsic and extrinsic muscles of the tongue and the infrahyoid strap muscles. Clinical manifestations of hypoglossal nerve injury include deviation of the tongue toward the side of the lesion on protrusion. Tongue atrophy, including both the intrinsic and extrinsic muscles will be evident in chronic lesions. Fasciculations of the tongue muscles may also be seen.1±3 3 TUMORS OF CRANIAL NERVES 3.1 Neural Tumors Schwannoma and neuro®broma are the two most common tumors of the cranial and peripheral nerves. Both tumors are derived from Schwann cells, but present quite differently. Schwannomas are usually solitary masses arising from more proximal nerves, while neuro®bromas may be multiple, and usually arise in more distal nerves (Figure 11). Histologically, schwannomas show areas of high and low cellularity known as Antoni-A and Antoni-B areas, respectively, while neuro®bromas show spindle cells with wavy nuclei not seen in schwannomas. Another differentiation between these two tumors is the presence of a capsule around schwannomas, whereas neuro®bromas are usually not encapsulated. In many

Schwannoma (Neurilemoma)

Schwannoma is a benign tumor which arises from the Schwann cells surrounding the cranial nerves, peripheral nerves and sympathetic plexus. The Schwann cells represent structures that both encircle the axons of the nerves and provide the nerves with mechanical support and their myelin sheaths. Histologically, a schwannoma contains a cellular component (formed from the Antoni-A cells) and a myxoid component (formed from the Antoni-B cells). The presence of these components and the absence of nerve ®bers in the body of the tumor allow the pathologist to differentiate a schwannoma from a neuro®broma. Schwannomas of the olfactory system and visual pathway are rare, since cranial nerves I and II are not true cranial nerves, but rather are embryologic invaginations of ®ber tracts from the telencephalon and diencephalon. The remaining cranial nerves and the upper spinal nerves are the sites of origin of the majority of schwannomas. These tumors are more common in women aged 30±60 years. The signs and symptoms vary according to the speci®c site of origin. Schwannomas of cranial nerves III, IV and VI may be seen along the cisternal, cavernous or intraorbital course of these nerves. Symptoms include diplopia, ophthalmoplegia or proptosis, depending upon the site of greatest mass effect. These tumors may be visualized on MRI as they course through the cranial canals. Bulbous expansion of the canals at the site of a lesion is considered pathognomonic; however, differentiation of schwannomas from various malignancies extending along nerve roots may be dif®cult. Schwannomas of the trigeminal ganglion and trigeminal nerve usually cause progressive facial numbness, pain, and paresthesias. Some trigeminal tumors can attain a considerable size in the absence of facial pain or numbness, since they may arise from the nerve sheath and secondarily compress the nerve ®bers. Schwannomas involving the trigeminal nerve are divided into three anatomical divisions: preganglionic, ganglionic, and postganglionic. Lesions that originate from the trigeminal ganglion typically straddle the incisura with a component extending into the cavernous sinus (postganglionic) and another portion extending into the prepontine cistern (preganglionic) (Figure 12).9±12 Facial nerve schwannomas may present with facial nerve palsy and/or hearing loss, or may be relatively asymptomatic. The development of facial nerve schwannomas from the nerve sheaths allows some of these tumors to enlarge and decompress into the adjacent air-containing cavities of the temporal bone without producing symptoms. Schwannomas originating from the geniculate ganglion typically become quite large without producing facial nerve dysfunction. These lesions may expand into the superior surface of the temporal bone and present as middle fossa masses (Figures 7 and 13). Acoustic nerve schwannomas originate from the vestibulocochlear nerve along its course from the vestibule and cochlea, internal auditory canal, and cerebellopontine angle cistern. These tumors represent 5±10% of all primary intracranial neoplasms and approximately 85±90% of all cerebellopontine

CRANIAL NERVES INVESTIGATED BY MRI

Figure 12 Schwannoma of the right trigeminal ganglion. (a) There is a primarily hypointense lesion in the right Meckel's cave and trigeminal ganglion region (arrow). (b) This enhanced view documents homogeneous enhancement of the tumor mass (arrow)

angle tumors. The clinical symptoms of acoustic nerve schwannoma are typically unilateral or asymmetric neurosensory hearing loss. These symptoms are directly related to the size of the tumor. As the tumor enlarges and ®lls the internal auditory canal, it compresses the cochlear nerve, causing neurosensory hearing loss and tinnitus, which are the most common initial symptoms (Figure 13). The early hearing loss is of the highfrequency variety, which may be related to the anatomic arrangement of the high frequency ®bers around the periphery of the nerve. The hearing loss is typically gradual and progressive, but may have an acute onset. Vestibular dysfunction, such as dizziness and disequilibrium, are relatively uncommon presenting symptoms, even though 85% of acoustic nerve schwannomas arise from the vestibular nerves. Small intracana-

13

licular lesions may be either relatively asymptomatic or may cause progressive symptoms. Some larger neoplasms may develop intramural or extramural hemorrhage with the sudden onset of headache, nausea, and vomiting. True central vertigo may be associated with brainstem involvement. Facial nerve dysfunction may be related to compression of the seventh nerve. Jugular foramen schwannomas can originate from different sites along cranial nerves IX, X or XI. Early symptoms consist of loss of taste in the posterior third of the tongue (glossopharyngeal nerve), paralysis of the vocal cord and palate (vagus nerve), and paresis of the trapezius and sternocleidomastoid muscles (spinal accessory nerve). Proximal schwannomas may extend into the inferior portion of the posterior fossa, producing hearing loss, vertigo, ataxia, hoarseness, and swallowing dif®culties (Figure 10). Schwannomas of the twelfth nerve usually cause hemiatrophy and paresis of the tongue. Some large tumors of the jugular foramen can extend into the hypoglossal canal and large hypoglossal schwannomas may expand into the adjacent jugular foramen. Concomitant involvement of cranial nerves IX, X, XI, and XII is rare, but may be seen with bulky tumors. Additional disturbances of posterior fossa function may include gait abnormalities, swallowing dif®culties, dizziness, and vomiting. In the neck, schwannomas are the most common neoplasm of the carotid space and the second most common lesion of the parapharyngeal space. Clinical presentation is a painless, slow growing mass in the anterolateral neck or posterolateral oropharynx. These lesions may become painful with associated neuropathies if they continue to grow and compress the nerves.11±13 The MRI appearance of all schwannomas is similar. The key differentiating point is the location of the central portion of the tumor within or near the nerve of origin. Schwannomas are fairly fusiform in appearance and well circumscribed because of their associated capsule. There may be smooth and uniform enlargement of the cranial canal through which the nerve is transmitted. The signal intensities, however, differ dramatically with respect to the size of the tumor. Smaller tumors usually appear homogeneous (since they are predominantly of the Antoni-A cell type), while larger tumors demonstrate heterogeneity and variable signal intensities owing to the presence of hemorrhage, necrosis, and cystic degeneration (Figure 13). Larger tumors typically contain many Antoni-B type cells. These schwannomas can have foci of high and low signal intensities on both the T1- and T2-weighted images. The presence of intramural cysts will characteristically produce foci of high signal intensity on the T2-weighted images (Figure 14).9,12 One MRI study suggested that cystic changes in acoustic nerve schwannomas may be related to either intramural cysts (Figures 13 and 14) or extramural arachnoid cysts (Figure 15). The intramural cysts showed high signal intensity on the T2weighted images, which was thought to be due to necrotic material, blood, or colloid-rich ¯uid. The extramural cysts showed high signal intensity related to higher protein and/or colloid contents secreted by the tumor. The extramural cysts were thought to originate from peritumoral adhesions which caused a pseudoduplication by the trapping effect of ¯uid between the leptomeninges and the schwannoma.14 Enhancement following gadolinium contrast administration is also variable, although it is typically rapid due to extravasation of contrast into the extracellular compartment of the often

14 CRANIAL NERVES INVESTIGATED BY MRI

Figure 13 Large schwannomas of the right seventh and eighth cranial nerves extending from the geniculate ganglion to the cerebellopontine angle (CPA). The patient had marked hearing loss, but only minimal facial nerve dysfunction. Enhanced images with fat saturation pulses. (a) This coronal image shows the tumor in the right CPA and internal auditory canal (arrow). (b and c) Axial images. The portion of the tumor in the right CPA shows both solid and cystic components (arrows). The solid component extends through the right internal auditory canal where it merges with a companion tumor in the right geniculate ganglion [arrow, (c)]

vascular schwannoma. In nearly all cases, there is a sharp demarcation between the margin of the tumor and the adjacent structures. Enhancement of the cranial nerves or their ganglia without an associated soft tissue mass is consistent with cranial nerve neuritis and/or ganglionitis rather than a neoplasm. Sarcomas, hemangiomas, or paragangliomas may have similar MRI enhancement characteristics and need to be considered in the differential diagnosis of schwannomas.12,15 3.3 Neuro®broma A neuro®broma contains neural elements with a ®brous core. This tumor may arise from the cranial, spinal, or peripheral nerves. The patient's symptoms depend upon the nerve of involvement and may be similar to the symptoms associated

with a schwannoma. In the extracranial head and neck, neuro®bromas may present as subcutaneous soft tissue nodules in the skin or as a mass lesion in the carotid or parapharyngeal spaces (Figure 11). In such cases, the clinical symptoms result from tumor bulk and may include dysphagia or dif®culty turning the neck. Some neuro®bromas cause pain or other focal neurologic de®cits. Most neuro®bromas are seen in young adult patients. Approximately 10% of patients with neuro®bromas have neuro®bromatosis type I (NF-1 or von Recklinghausen's disease). Two types of neuro®broma are known to be associated with NF-1. Multiple or single solid lesions may be seen along the cranial, spinal, or peripheral nerves (Figure 11). Another type of lesion presents as an in®ltrating soft tissue mass with extension along the cranial nerves through the orbits, skull base, or

CRANIAL NERVES INVESTIGATED BY MRI

15

than the margins of a schwannoma, because neuro®bromas lack a capsule. In many cases, it may not be possible to differentiate a neuro®broma from a schwannoma on MRI characteristics.12,15 3.4

Figure 14 Acoustic nerve schwannoma with solid and intramural cystic components. (a) This axial image shows the heterogeneous mass in the left internal auditory canal and cerebellopontine angle cistern. The cystic component of the tumor merges with the bright signal intensity of the cerebrospinal ¯uid. (b) This axial enhanced image shows the solid and cystic components. Note that the cystic portion shows peripheral enhancement con®rming an intramural cyst (arrow)

posterior fossa. This so-called `plexiform' neuro®broma encases the muscles and soft tissues along the course of the nerves (Figure 5).16 The MRI characteristics of neuro®bromas are quite varied and simulate schwannomas in many cases. Neuro®bromas project an intermediate signal intensity on the T1- and protondensity-weighted images. They sometimes have a `salt and pepper' appearance whenever there is intense vascularity within the stroma of the tumor. The T2-weighted images are variable, depending on whether the tumor is cystic or solid in nature. Variable enhancement is seen following gadolinium administration due to the presence of cystic areas, calci®cations, or vascularity. The margins of a neuro®broma appear less distinct

Malignant Schwannoma

Many terms have been used to describe a malignant schwannoma, including malignant neuro®broma, neuro®brosarcoma, neurosarcoma, neurogenic sarcoma, malignant neurilemoma, and malignant nerve sheath tumor. Many malignant schwannomas contain interlacing fascicles of spindle cells in a herring-bone pattern, as typically seen in ®brosarcomas. Unlike ®brosarcomas, however, on electron microscopy malignant schwannomas are seen to contain basement membranes. The light microscopic ®ndings include high cellularity, pleomorphism, high incidence of mitotic ®gures, presence of necrosis, and presence of histologic invasion. There is no capsule around a malignant ®brosarcoma.16 Malignant schwannomas are highly malignant lesions that recur following local excision in 50±80% of patients. The histologic grade is the most important factor in determining prognosis and is the primary factor considered in staging. Five-year survival rates in one study of 17 patients over the course of 37 years was 47%. Wide and aggressive excision with or without adjuvant radiation therapy is considered the best treatment alternative. Generally, patients with malignant schwannomas of the head and neck have a poorer prognosis than do those with similar tumors of the extremities, which can be treated with amputation. One factor contributing to the poor prognosis of this tumor is its resistance to radiation therapy. The incidence of malignant schwannomas is approximately 5±10% of all sarcomas and 2±12% of all nerve sheath tumors. Patients with NF-1 have a 5±30% chance of developing a de novo malignant neurogenic tumor. The typical NF-1 patient with a malignant neural tumor is 20±50 years old. The 5-year survival rate in patients with NF-1 is 15±30%. Recurrences and metastases may occur as late as 5±10 years following treatment. The tumor initially causes a fusiform, nodular swelling of the involved nerve and ultimately diffusely in®ltrates the nerve. It then spreads into the surrounding soft tissues and often contains areas of hemorrhage or necrosis (Figure 16). The tumor may cause neurologic symptoms such as pain, paresthesias, atrophy, and muscle weakness. Occasionally, a symptomless mass along a nerve distribution is seen.12,15,16 The MRI appearance is of a tubular mass along the involved nerve with regional extension into the skull base (cranial nerve) or cervical soft tissues. The signal intensity is varied, but is typically hypointense on the T1-weighted images and hyperintense on the T2-weighted images. Gadolinium contrast administration helps to outline the margins of the tumor including extension beyond the site of origin into the spinal canal or cranial cavity (Figure 17)12,15 3.5

Cranial Nerve Seeding

Both metastatic neoplasms and primary malignant tumors of the central nervous system can disseminate via the subarachnoid cisternal pathways to distant locations. This type of

16 CRANIAL NERVES INVESTIGATED BY MRI

Figure 15 Acoustic nerve schwannoma with a small arachnoid cyst in the left cerebellopontine angle. (a and b) The large tumor is highly heterogeneous and deforms the left side of the brainstem and cerebellum. Note the homogeneously bright signal in the small left arachnoid cyst (arrows). (c) This image shows enhancement of the tumor. The arachnoid cyst, however, is extramural and does not enhance (arrows)

Figure 16 Malignant schwannoma of the right trigeminal and inferior alveolar nerves. Coronal enhanced images with fat saturation pulses. (a) This composite image documents the extensive in®ltration of the tumor in the right muscles of mastication. There is a pathologic fracture of the right ramus of the mandible (arrows). (b) This composite view documents the extension of the tumor in a tubular manner through the right foramen ovale into the right trigeminal ganglion (arrows)

CRANIAL NERVES INVESTIGATED BY MRI

17

phomatous foci may be deposited adjacent to the cranial nerves along their courses through the posterior and middle cranial fossae (Figures 3 and 18).16,17 Intracranial leptomeningeal metastases tend to be diffuse along the subarachnoid spaces and cranial nerves. Microscopically, the tumor cells in®ltrate the leptomeninges as a single layer or as thicker, multilayered aggregates. The earliest in®ltration is along cortical vessels and is limited to the perivascular Virchow±Robin spaces. Clinically, patients most often present with a variety of symptoms, which are usually multifocal. Headache, change of mental status, cranial nerve de®cits, and gait disturbances are the most common. These metastatic deposits are visualized best on the T1weighted postcontrast images, particularly if fat saturation

Figure 17 Coronal enhanced images of a malignant schwannoma of the right brachial plexus with intraspinal extension. (a and b) The tumor ®lls the region of the scalene muscles and brachial plexus. There is downward extrapleural extension [arrow, (a)]. Note a small intraspinal portion along the cervical nerve roots [arrow, (b)]

metastatic spread may be manifested as either a diffuse spread along the leptomeningeal surfaces of the brain or loculated deposits of tumor nodules along speci®c sites within the subarachnoid spaces such as the cranial nerves. The most common dissemination in children is from primitive neuroectodermal tumors and ependymomas. The most common cause of seeding in adults is from a glioblastoma multiforme. The incidence of seeding from a glioblastoma is less than 5% in all age groups. Other tumors that can seed to the leptomeninges or cranial nerves are sarcomas, carcinomas, and adenocarcinomas. Leptomeningeal metastases from various lymphoproliferative tumors are also common in both children and adults. Metastatic lym-

Figure 18 Lymphomatous meningitis of the brainstem and cranial nerves. Axial enhanced images. There is a focal tumor nodule in the left cerebellopontine angle region [arrow, (a)]. There is diffuse in®ltration of the midbrain and enhancement along the left trochlear nerve [arrow, (b)]

18 CRANIAL NERVES INVESTIGATED BY MRI 4

Figure 19 Axial enhanced image of seeding of metastatic adenocarcinoma to the subarachnoid spaces and cranial nerves. There are sheets and nodules of tumor surrounding the midpons and cerebellar folia. A large focal nodule occupies the origin and preganglionic portions of the left trigeminal nerve (arrow)

techniques are utilized. The areas of nodular or linear enhancement follow the contours of the meningeal surfaces or cranial nerves as they traverse through the cerebrospinal ¯uid (Figure 19). The T2-weighted images are less sensitive because of the surrounding high signal intensity of the cerebrospinal ¯uid.12,15

INFECTIONS AND INFLAMMATIONS OF CRANIAL NERVES

Infections of the cranial nerves commonly are caused by a latent virus such as herpes simplex I or varicella zoster. These neurotropic viruses tend to affect the cell bodies of the sensory neurons, in which they may reside for many years. Reactivation of the dormant virus results in a swelling of the involved ganglion or of the involved nerve. After contrast medium administration, there may be enhancement in the ganglion or along the course of an infected nerve.15,20 The most common viral mononeuritis is Bell's palsy, which is thought to be caused by reactivation of a herpes simplex virus in the geniculate ganglion. Following gadolinium administration, there may be enhancement along the entire course of the facial nerve from the internal auditory canal through the mastoid foramen. More commonly, the portion of the facial nerve in the region adjacent to the geniculate ganglion is enhanced as a result of entrapment of the seventh cranial nerve in the fallopian canal (Figure 21). Herpes simplex trigeminal neuritis and ganglionitis have been described, revealing involvement of the ®fth cranial nerve on contrast enhanced MRI (Figure 22).21,22 The clinical prevalence of herpetic involvement of the cranial nerves is most likely underestimated. However, in patients infected with the human immunode®ciency virus (HIV), it is fairly common to recognize postgadolinium enhancement along one, two or more cranial nerves. The mechanism is similar to that of herpes simplex virus, since the HIV virus is also neurotropic and directly invades sensory ganglia and their associated nerves.20,21

3.6 Perineural Spread of Malignant Neoplasms

5

OPTIC NERVE AND SHEATH DISORDERS

Perineural tumor extension is a form of metastatic disease in which primary tumors spread along neural pathways and gain access to noncontiguous regions. Since the treatment and prognosis are altered when perineural spread is identi®ed, proper evaluation is critical. The in®ltration of the nerves by tumor is best seen on gadolinium enhanced MRI.18,19 Perineural tumor spread is visible as smooth thickening of the cranial or spinal nerves and/or concentric expansion of the foramina and ®ssures through which the nerves traverse. Involvement of the maxillary and mandibular divisions of the trigeminal nerve is common because of the large size of these branches and their strategic passage through the pterygopalatine fossa. In cases of perineural spread to the trigeminal ganglion, there may be replacement of the normal trigeminal cistern hypointensity by an isointense or enhancing soft tissue mass (Figure 20). Other signs of trigeminal ganglion involvement include lateral bulging of the cavernous sinus dural membranes and atrophy of the masticator muscles supplied by the trigeminal nerve branches (Figure 6). Perineural metastatic disease may occur along the course of the facial nerve, most commonly resulting from neoplasms of the parotid gland, particularly adenoid cystic carcinoma.15,18,19

Primary disorders of the optic nerve and optic nerve sheath typically cause visual disturbances and occasionally proptosis in cases of large neoplasms. The most common primary neoplasms of the optic nerve include gliomas (astrocytomas and hamartomas) or meningiomas. There may be a paraoptic component of optic nerve tumors, which is characteristically seen in patients with neuro®bromatosis type I. The paraoptic component results from perineural arachnoidal gliomatosis, consisting of proteinaceous material that seeds and spreads within the optic nerve subarachnoid space and is a portion of the tumor. This process may result in optic nerve elongation and resultant kinking of the nerve just posterior to the posterior portion of the globe. There is no correlation between the size of the lesion and visual loss. Extension of tumors into the optic chiasm, optic tracts, and lateral geniculate bodies of the thalami is more accurately depicted on MRI than computed tomography (CT). The size and shape of the optic canals are best assessed in the axial projection, while the size and shape of the optic nerves are best appreciated on coronal and oblique sagittal images. Many optic nerve tumors exhibit fusiform homogeneous enhancement (Figure 23). Unenhanced portions of optic nerve tumors may represent the sites of gliomatosis, rather than a true neoplasm. Biopsy is essential to make the correct diagnosis,23,24

CRANIAL NERVES INVESTIGATED BY MRI

19

Figure 20 Perineural spread of squamous cell carcinoma. (a) This axial enhanced composite image shows bulging of the left side of the cavernous sinus and tumor extension into the preganglionic portion of the left trigeminal nerve (arrows). (b and c) Axial images. There is well de®ned edema and tumor in®ltration into the left cerebral peduncle, pons, and middle cerebellar peduncle (arrows). The wide extent through the brainstem correlates with the long nuclei of the ®fth cranial nerve

Figure 21 Facial neuritis and geniculate ganglionitis in a patient with Bell's palsy. Axial enhanced images. (a) Note the enhancement in the fundus of the right internal auditory canal with extension into the adjacent geniculate ganglion (arrow). (b) There is enhancement along the tympanic segment of the right facial nerve (arrow)

20 CRANIAL NERVES INVESTIGATED BY MRI

Figure 22 Enhanced images of viral trigeminal neuritis in a patient with trigeminal neuropathy. (a) There is enhancement along the preganglionic portion of the right trigeminal nerve (arrow). (b) This coronal view con®rms the enhancement of the cisternal portion of the right trigeminal nerve (arrow). (c) The enhancement extends along the infraorbital division of the right maxillary nerve into the infraorbital groove (arrow)

Meningiomas are the most common optic nerve sheath tumors; they usually arise from the arachnoidal coverings of the optic nerve. Visual loss and optic atrophy are the usual presenting symptoms. Most cases are seen in middle-aged or elderly patients, more often in women. The tumor may be cufflike surrounding the optic nerves or eccentrically located on only one side of the nerve. CT scans will often demonstrate calci®cations and show typical postcontrast enhancement parallel to the length of the optic nerves. MR scans readily depict the irregular thickening along the optic nerves and spread into adjacent meninges on the postcontrast scans (Figure 24). Parallel optic cysts may be identi®ed surrounding the optic nerve immediately distal to the meningioma. The process causes trapping of the cerebrospinal ¯uid in the subarachnoid space and can add to the mass effect and proptosis.23,25 Optic neuritis is seen on MRI as focal or diffuse enlargement of the optic nerve associated with abnormal signal

intensities and enhancement (Figure 25). This is caused by immunological in¯ammation of the optic nerve, which affects the myelin sheaths with relative preservation of the axons. As optic neuritis is the initial manifestation of multiple sclerosis in about 20% of patients and may occur at some point in the disease in approximately 50%, the role of MRI in the evaluation of acute optic neuritis is changing. A multicenter trial was developed to assess the ef®cacy of corticosteroid treatment for acute optic neuritis. This study showed that the utility of MRI in establishing the diagnosis of optic neuritis was limited; however, with additional views of the cranial cavity, abnormal MRI scans could differentiate those patients who would later develop multiple sclerosis from those that would not. Thus, MRI is a predictor of multiple sclerosis as it can help to identify a subgroup of patients whose risk of developing the disease appears to be low and it can provide prognostic data about the development of multiple sclerosis after optic neuritis.23,26

CRANIAL NERVES INVESTIGATED BY MRI

21

studies. There is some correlation between the severity of the visual loss and the detection of enlarged nerves with reversed nerve heads. Patients with more severe visual loss demonstrate more frequent and more severe reversal of the optic nerve head.27 5.1

Figure 23 A 4-year-old boy with large left optic nerve glioma. There is a pear-shaped densely enhancing tumor mass in the left orbit causing mild proptosis. Note extension of the tumor into the orbital opening of the left optic nerve canal

The papilledema associated with the pseudotumor cerebri may be detected on CT or MRI scans as enlargement of the optic nerve sheaths. If severe, there is a reversal of the optic nerve head with bulging forward into the posterior wall of the globe. This phenomenon is more readily detected on CT than MRI because of the chemical shift artifact inherent to the MR

Optic Nerve Neuropathies

Isolated visual disturbances may result from a variety of optic nerve neuropathies, caused by radiation, chemotherapy, compressive phenomena, or ischemia.28±30 Radiation-induced optic neuropathy (RON) is a rare catastrophic complication of radiation therapy regimens used to treat a variety of neoplasms of the skull base, sella, and parasellar regions (Figure 26). There is typically a latency period of 6 to 36 months following treatment. Clinically, the nerve head may appear normal, but gadolinium-enhanced MRI will show patchy, linear, or con¯uent enhancement along the portions of the optic nerve, chiasm, or optic tract.25,28 These ®ndings precisely correlate as the cause of delayed visual loss following radiation therapy. Treatment of RON with corticosteroids may be helpful, although optic atrophy may develop with permanent visual loss.28 Compressive optic nerve neuropathies can be seen with a variety of lesions in the orbital apex [including Idiopathic Orbital Pseudotumor (IOP), thyroid ophthalmopathy with muscular hypertrophy, and edema causing compression of the intraocular portion of the optic nerve] or other systemic diseases with common orbital manifestations, such as sarcoidosis or Wegener's granulomatosis.29±36 Compression of the optic nerve may occur as a result of cavernous carotid ®stulae, arteriovenous malformations, or orbital varices. Such vascular anomalies may produce retrograde ¯ow through the ophthalmic vessels, with subsequent dilatation of the orbital veins and passive congestion of the orbital tissues. This leads to progressive increases in the intraocular pressure and subsequent decreases in visual acuity

Figure 24 An 80-year-old female with left optic sheath meningioma. (a) Coronal enhanced view shows an enlarged left optic nerve with near complete enhancement of the lesion. (b) Axial enhanced MR scan shows the extent of the tumor in a `railroad track' manner about the left optic nerve, with extension into the intracanalicular portion of the left optic nerve. Incidental note is made of a retinal banding procedure around the right globe

22 CRANIAL NERVES INVESTIGATED BY MRI

Figure 25 Right optic nerve neuritis in a 40-year-old female. (a) Axial enhanced MR scan showing no enlargement of the right optic nerve. There is enhancement along a major portion of the intraorbital part of the optic nerve. (b) Coronal enhanced scan showing the normal size but enhancing nerve in the right eye

leading to blindness. Spontaneous thrombosis of the ophthalmic veins occasionally occurs and may aggravate the mass effect within the apex of the orbit and the compressive neuropathy.23,33 MRI will demonstrate the dilated ophthalmic veins, facial veins, and other regional venous structures along with enlargement of the cavernous sinus. Large edematous extraocular muscles and other periorbital structures may be identi®ed. These ®ndings are optimally seen with MRI, particularly as the addition of MR angiography allows for ¯ow assessments along with the static morphologic changes. In some cases, conventional angiography may be required to make the de®nitive diagnosis, although most commonly this is used in conjunction with therapeutic interventional procedures. 6 VERTIGO AND HEARING LOSS 6.1 Dizziness and Vertigo Dizziness is a common clinical complaint. It accounts for 1% of visits to US of®ce-based physicians. Vertigo is a form of dizziness in which there is an illusion of movement (rotation, tilt, or linear translation). Vertigo occurs through an imbalance of tonic vestibular signals; consequently, it is a hallucination of movement and represents a symptom of a disturbed vestibular system.37,38 The complete vestibular system comprises the end organs in the temporal bone, the vestibular components of the VIIIth cranial nerve, and the central connections in the brainstem. The end organs in the temporal bones are the cristae of the three semicircular canals, which respond to movement of the head, and the macula of the utricle, which records the position of the head. The semicircular canals record dynamic actions while the utricle records static function. Vertigo is subdivided into peripheral vertigo (caused by failure of the end organs) or central vertigo (caused by failure of the vestibu-

lar nerves or central connections to the brainstem and the cerebellum.38,39 6.2

Peripheral Vestibular Disorders

Patients with benign positional vertigo describe episodic vertigo lasting less than a minute, brought on by movements of the head and without other associated symptoms. There are no radiological ®ndings in patients with benign positional vertigo.37,39 In Meniere's disease, paroxysmal attacks of whirling vertigo are usually accompanied by nausea; the attacks are transient in nature, lasting a few hours but not days. The severe episodic vertigo is accompanied by tinnitus, ¯uctuating hearing loss, and a feeling of fullness in the affected ear or ears. Typically, hearing decreases and tinnitus increases during the attack. Hearing may improve between attacks in early stages of the disease. Generally, the hearing loss begins unilaterally and affects primarily the lower frequencies, with mid and high frequencies being affected in later stages of the disease.37±39 Meniere's disease is most common in middle age and may become bilateral in up to 50% of the affected patients. The etiology of Meniere's disease is a failure of the mechanism regulating the production and disposal of endolymph, resulting in recurrent attacks of endolymphatic hydrops. Since the endolymphatic duct and sac are the sites of resorption of endolymph, these structures play an important role in the pathogenesis of endolymphatic hydrops. The success of various surgical procedures in relieving the symptoms of Meniere's disease has led to great interest in using CT and/or MRI to evaluate the vestibular aqueduct and the endolymphatic duct and sac.39±41 Unfortunately, there is no unanimity on the value of imaging in cases of Meniere's disease. Some investigators have used CT or MRI to evaluate the potential success of shunt surgery, based on showing patency of the vestibular aqueduct.39,41

CRANIAL NERVES INVESTIGATED BY MRI

23

Figure 26 Parasellar and suprasellar Rathke's pouch cyst. (a) Fragile unenhanced scan showing the mass causing elevation of the optic chiasm. (b) Coronal enhanced scan at the level of the chiasm. Note that the non-enhanced cyst causes elevation and deformity of the optic chiasm. (c) Coronal enhanced scan at the level of the pituitary stock shows displacement of the stock to the right by the left parasellar mass

Other investigators, however, report that the size, shape, and patency of the vestibular aqueduct are of no value in predicting surgical results in shunt procedures or in predicting occurrence of bilateral disease.40 MRI with its ability to detect the endolymphatic duct and sac, separate from the bony vestibular aqueduct, may offer more useful information than can CT.41 The value of CT and MRI may be in their ability to exclude associated infectious or neoplastic disease processes.38,39 Vestibular neuronitis is a clinical diagnosis based on an aggregate of speci®c symptoms. The disease is characterized

by an acute onset of severe vertigo, lasting several days, followed by gradual improvement of several weeks. Hearing is typically unaffected. The history includes onset of vertigo following an illness such as an upper respiratory infection. Most patients become completely symptom free following central compensation.39 Vestibular labyrinthitis is similar in that the disease presents with the acute symptoms of vertigo, but this disease is always associated with hearing loss. Labyrinthitis is usually viral in origin but may result from acute or chronic bacterial middle ear infections. Unlike viral labyrinthitis,

24 CRANIAL NERVES INVESTIGATED BY MRI be detected by MRI through loss of the signal intensity of the ¯uid contents. Later, more complete obliteration of all the labyrinthine structures occurs, leading to labyrinthitis obliterans, which is readily diagnosed on CT or conventional tomography.42 With MRI, there may be postgadolinium enhancement of the labyrinthine structures or vestibular nerves during the acute or subacute stages of vestibular neuronitis and/or labyrinthitis.43,44 Such results must be interpreted with care, since sudden labyrinthine dysfunction may be caused by spontaneous hemorrhage or injury, which result in abnormal signal intensities within the labyrinthine structures secondary to the presence of blood products.45 Diseases of the internal auditory canal and cerebellopontine angle are generally not characterized by severe attacks of vertigo but rather with intermittent dizziness and/or exacerbated periods of dizziness.37,39 A variety of benign or malignant tumors of the petrous temporal bone, such as paragangliomas, carcinomas, or metastatic tumors, may directly involve the labyrinthine structures, causing vertigo. Such processes are readily evaluated with MRI (Figures 27±29). Figure 27 Tumor of the left internal auditory canal. Enhanced coronal image documents a linear enhancing tumor that ®lls the left auditory canal

labyrinthitis associated with suppurative ear disease may progress to develop partial or complete occlusion of the lumen of the affected labyrinth.38,39 Early on, the obstructed lumen may

6.3

Central Vestibular Disorders

Lesions of the brainstem or cerebellum that result in central vertigo can readily be diagnosed by MRI. Vascular insuf®ciency in the vertebrobasilar circulation is a common cause of vertigo in patients over 50 years of age. Thrombosis of the

Figure 28 Meningioma of the right cerebellopontine angle. (a) Axial MR scan without contrast. The nearly isointense tumor can be seen to indent the left cerebellopontine angle. (b) Enhanced axial CT scan showing the full extent of the tumor mass. Note that the mass bends up to the level of the left ®fth nerve (arrowhead). (c) Enhanced coronal image shows that mass to extend into the internal auditory canal, simulating an acoustic nerve tumor

CRANIAL NERVES INVESTIGATED BY MRI

25

Figure 29 Lipoma of the right cerebellopontine angle and vestibule. (a) Saggital unenhanced image shows the fat intensity mass in the cerebellopontine angle. (b) Coronal unenhanced images document the lesion in the right cerebellopontine angle with a small focus of a high signal in the region of the right vestibule. (c) Axial enhanced CT scan again shows the tumor masses both in the cerebellopontine angle and the vestibule. The patient complained of severe dizziness that was more likely related to the small lesion in the vestibule

labyrinthine artery or infarction of the lateral medulla from vertebral or posterior inferior cerebellar artery insuf®ciency may cause severe vertigo. The subclavian steal syndrome can cause a variety of symptoms, including vertigo.38,46,47 Such conditions can be carefully evaluated with MR angiography or conventional angiography of the posterior fossa vasculature. A variety of other central nervous diseases may produce vertigo or dizziness. These include seizure disorders, multiple sclerosis, ataxic diseases, head injuries, or any cause for increased intracranial pressure. Vertigo may result from stroke and transient ischemic attacks may present as episodic dizziness.39 Various metabolic disorders may result in dizziness. These include thyroid disorders, hyperlipidemia, diabetes, and hypoglycemia. Autoimmune diseases or diseases that affect the proprioceptive system may be the cause of vertigo. In many cases, the possibility of function neurotic symptoms must be considered in patients in whom no diseases can be found. Finally cervical spondylosis is thought to cause vertigo by disc

degeneration and narrowing of the disc space, which affects the nerves in close proximity, or by osteophyte formation, which compresses the blood vessels. In such cases, conventional radiographs and/or cross-sectional imaging procedures may be helpful.38,39,42 6.4

Sensorineural Hearing Loss

Sensorineural hearing loss may be sudden, ¯uctuating, or progressive. Sudden sensorineural hearing loss is a manifestation of viral infections, vascular occlusive diseases, or inner ear membrane ruptures.48±52 As discussed above, there may be vertigo associated with these conditions, which would help to de®ne whether the lesion is peripheral or central. In order to discriminate idiopathic or viral infections from other causes of sensorineural hearing loss, auditory brainstem responses and gadolinium enhanced MRI may be utilized.48±50 Patients with cochleitis or cochlear nerve neuritis typically have abnormal

26 CRANIAL NERVES INVESTIGATED BY MRI auditory brainstem responses and may be helped by a tapering course of oral corticosteroids.49,50 Whether gadolinium enhanced MRI shows enhancement of the cochlear nerve or cochlea is not a helpful indicator for or against using corticosteroid therapy. However, one study suggests that sudden deafness associated with MRI changes is more dif®cult to cure with steroid therapy than that without associated MRI changes.49 Fluctuating neurosensory hearing loss is a dif®cult disease to assess properly. The audiometric examination does, of course, indicate the level of dysfunction but not the likely cause. In patients where MRI indicates large vestibular aqueducts (apertures greater than 4 mm), this may indicate a ¯uctuating frequency loss more often than low frequency loss. Fluctuating sensorineural hearing loss resulting from an enlarged vestibular aqueduct appears to be more common in children and young adults, which is an important point in differentiating this disease from Meniere's disease in which the majority of patients are middle aged or older. The vestibular aqueduct of patients with Meniere's disease may be small rather than large.38,39 There is speculation on the causes of a sudden drop in hearing in patients with large vestibular aqueducts.53 Two possible causes are re¯ux of hyperosmolar ¯uid from the endolymphatic sac to the inner ear and rupture of the membranous labyrinth or a perilymphatic ®stula caused by transmission of intracranial pressure to the inner ear through the enlarged vestibular aqueduct. It is well recognized that patients sustaining relative minor head trauma, or patients who are subjected to extreme barotrauma (scuba diving), may have aggravated episodes of hearing loss. Consequently, it may be worthwhile to image the temporal bones in order to detect enlarged vestibular aqueducts and thus advise the patients or their parents of the dangers of contact sports or activities that entail extreme barometric pressure changes. The imaging ®ndings must be correlated with audiometry, since the ¯uctuating sensorineural hearing loss of large vestibular aqueduct patients does not resemble the low frequency changes characteristic of Meniere's disease which may also be associated with ¯uctuating hearing loss.54,55 The pathophysiologic basis of disease in patients with isolated large vestibular aqueducts may differ from that in patients whose large aqueducts are associated with other inner ear malformations. Patients with complex inner ear malformations may be subject to recurrent episodes of meningitis and/or the `gusher' syndrome, resulting in a dead ear at the time or surgical intervention such as a stapedectomy.52,54 Asymmetric sensorineural hearing loss or gradually declining unilateral sensorineural hearing loss is a common symptom that may be ascribed to a variety of pathologic processes. Initial evaluation attempts to ®nd the site of the lesion (i.e. cochlear or retrocochlear). All retrocochlear lesions are associated with an abnormal auditory brainstem response, which is often obtained prior to an imaging study. Whether auditory brainstem response testing should be eliminated as a cost-saving measure is a subject of considerable debate. It seems unlikely that clinicians will refer patients directly to MRI without at least preliminary audiometric and/or auditory brain response testing.48,50,56 A complete MR study of the head should be performed in addition to the studies of the internal auditory canal and temporal bones. The MR examination should include complete evaluation of the central nuclei in the brainstem as well as the

auditory pathways extending upwards into the cerebral hemispheres.7 Whether or not gadolinium contrast enhancement is routinely utilized depends on a variety of factors including coil size, ®eld of view, ®eld strength, and pulse sequences. CT is diagnostic in lesions 1±1.5 cm or greater in diameter but does not readily detect small brainstem lesions such as infarctions or demyelination.7,56

7

RELATED ARTICLES

Eye, Orbit, Ear, Nose, and Throat Studies Using MRI; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Head and Neck Investigations by MRI.

8

REFERENCES

1. L. A. Hayman, L. A. Rolak, J. M. Gundzik, and R. B. Lufkin, in `Clinical Brain Imaging', ed. L. Anne Hayman, Mosby, St. Louis, 1992, Chap. 8. 2. W. G. Bradley, Jr, Radiology, 1991, 179, 319. 3. W. M. Kelly, Neuroimaging Clin. North Am., 1993, 3, 1. 4. L. R. Gentry, R. C. Mehta, R. E. Appen, and J. E. Weinstein, Am. J. Neuroradiol., 1991, 12, 707. 5. L. G. Hutchins, H. R. Harnsberger, C. W. Hardin, W. P. Dillon, W. R. Smoker, and A. G. Osborn, Am. J. Neuroradiol., 1989, 10, 1031. 6. S. S. Gebarski, S. A. Telian, and J. K. Niparko, Radiology, 1992, 183, 391. 7. S. S. Gebarski, D. L. Tucci, and S. A. Telian, Am. J. Neuroradiol., 1993, 14, 1311. 8. C. J. Jacobs, H. R. Harnsberger, R. B. Lufkin, A. G. Osborn, W. R. Smoker, and J. L. Parkin, Radiology, 1987, 164, 97. 9. A. N. Hasso and D. S. Smith, Semin. Ultrasound, CT, MR, 1989, 10, 280. 10. L. G. Hutchins, H. R. Harnsberger, J. M. Jacobs, and R. I. Apfelbaum, Radiology, 1990, 175, 837. 11. N. Martin, O. Sterkers, D. Mompoint, and N. Nahum, Neuroradiology, 1992, 34, 62. 12. A. N. Hasso, in `MRI Atlas of the Head and Neck', Martin Dunitz, London, 1993, Chaps 1 and 6. 13. M. F. Mafee, C. S. Lachenauer, A. Kumar, P. M. Arnold, R. A. Buckingham, and G. E. Valvassori, Radiology, 1990, 174, 395. 14. E. T. Tali, W. T. C. Yuh, H. D. Nguyen, G. Feng, T. M. Koci, J. R. Jinkins, R. A. Robinson, and, A. N. Hasso, Am. J. Neuroradiol., 1993, 14, 1241. 15. A. N. Hasso and K. D. Brown J. Med. Radiol. Imag., 1993, 3, 247. 16. D. S. Russell and L. J. Rubinstein, in `Pathology of Tumours of the Nervous System', Williams & Wilkins, Baltimore, OH, 1989, Chap. 5. 17. Y. Y. Lee, R. D. Tien, J. M. Bruner, C. A. DePena, and P. VanTassel, Am. J. Roentgenol., 1990, 154, 351. 18. F. J. Laine, I. F. Braun, M. E. Jensen, L. Nadel, and P. M. Som, Radiology, 1990, 17, 65. 19. G. D. Parker and H. R. Harnsberger, RadioGraphics, 1991, 11, 383. 20. R. D. Tien, G. J. Felsberg, and A. K. Osumi, Am. J. Roentgenol., 1993, 161, 167. 21. A. Vahlne, S. Edstrom, P. Arstila, M. Beran, H. Ejnell, O. Nylen, and E. Lycke, Arch. Otolaryngol., 1981, 107, 79.

CRANIAL NERVES INVESTIGATED BY MRI 22. R. Tien, W. P. Dillon, and R. K. Jackler, Am. J. Neuroradiol., 1990, 11, 735. 23. L. T. Bilaniuk, R. A. Zimmerman, and T. H. Newton, in `Modern Neuroradiology, Vol. 4. Radiology of the Eye and Orbit', ed. T. H. Newton, L. T. Bilaniuk. Clavadell Press, New York, pp. 5.1±5.84. 24. S. R. Kodsi, D. J. Shetlar, R. J. Campbell, J. A. Garrity, and G. B. Bartley, Am. J. Ophthalmol., 1994, 117, 177. 25. B. J. Goldsmith, S. A. Rosenthal, W. M. Wara, and D. A. Larson, Radiology, 1992, 185, 71. 26. M. C. Brodsky and R. W. Beck, Radiology, 1994, 192, 22. 27. W. A. Gibby, M. S. Cohen, H. I. Goldberg, and R. C. Sergott, Am. J. Roentgenol., 1993, 160, 143. 28. J. Guy, A. Mancuso, R. Beck, M. Moster, L. A. Sedwick, R. G. Quisling A. L. Rhoton Jr, E. E. Protzko and J. Schiffman, J. Neursurg., 1991, 74, 426. 29. N. Hosten, B. Sander, and M. Cordes, Radiology, 1989, 172, 759. 30. H. Tonami, H. Tamamura, K. Kimizu, A. Takarada, T. Okimura, I. Yamamoto, and K. Sasaki, Am. J. Roentgenol., 1990, 154, 385. 31. R. F. Carmody, M. F. Mafee, J. A. Goodwin, K. Small, and C. Haery, Am. J. Neuroradiol., 1994, 15, 775. 32. N. A. Courcoutsakis, C. A. Langford, M. C. Sneller, T. R. Cupps, K. Gorman, and N. Patronas, J. Comput. Assist. Tomogr., 1997, 21, 452. 33. A. S. Cytryn, A. M. Putterman, G. L. Schneck, E. Beckman, and G. E. Valvassori, Ophthal. Plastic Recons. Surg., 1997, 13, 129. 34. P. de Potter, J. A. Shields, and C. L. Shields, MRI of the Eye and Orbit, J. B. Lippincott Company, Philadelphia, PA, USA, 1995 35. R. A. Nugent, R. I. Belkin, J. M. Neigel, J, Rootman, W. D. Robertson, J. Spinelli, and D. A. Graeb, Radiology, 1990, 177, 675. 36. T. Ohnishi, S. Noguchi, N. Murakami, J. Tajiri, M. Harao, H. Kawamoto, H. Hoshi, S. Jinnouchi, S. Futami, S. Nagamachi, and K. Watanabe, Radiology, 1994, 190, 857. 37. S. R. McGee, West. J. Med., 1995, 162, 37. 38. P. D. Phelps and G. A. S. Lloyd in `Radiology of the Ear', Blackwell Scienti®c, Boston, MA, 1983, pp. 137±141. 39. J. R. E. Dickins and S. S. Graham, Ear Hearing, 1986, 7, 133. 40. E. M. Kraus and P. J. DuBois, Arch. Otolaryngol., 1979, 105, 91. 41. F. W. J., Albers, R. V. Van Weissenbruch, and J. W. Casselman, Acta Otolaryngol., 1994, 114, 595. 42. A. N. Hasso and J. A. Ledington, Otolaryngol. Clin. North Am., 1988, 21, 219. 43. S. Seltzer and A. S. Mark, Am. J. Neuroradiol., 1991, 12, 13. 44. A. S. Mark, S. Seltzer, J. Nelson-Drake, J. C. Chapman, D. C. Fitzgerald, and A. J. Gulya, Ann. Otol. Rhinol. Laryngol., 1992, 101, 459. 45. J. L. Weissman, H. D. Curtin, B. E. Hirsch, and W. L. Hirsch Jr, Am. J. Neuroradiol., 1992, 13, 1183.

27

46. S. Kikuchi, K. Kaga, T. Yamasoba, R. Higo, T. O'uchi, and A. Tokumaru, Acta Otolaryngol., 1993, 113, 257. 47. B. Norving, N. Magnusson, and S. HoltaÊs, Acta Neurol. Scand., 1995, 91, 43. 48. R. A. Hendrix, R. M. DeDio, and A. P. Sclafani, Otolaryngol. Head Neck Surg., 1990, 103, 593. 49. K. Kano, T. Tono, Y. Ushisako, T. Morimitsu, Y. Suzuki, and T. Kodama, Acta Otolaryngol., 1994, 514, 32. 50. N. Y. Busaba and S. D. Rauch, Otolaryngol. Head Neck Surg., 1995, 113, 271. 51. M.-H. Huang, C.-C. Huang, S. J. Ryu, and N.-S. Chu, Stroke, 1993, 24, 132. 52. J. S. Reilly, Laryngoscope, 1989, 99, 393. 53. G. E. Valvassori and J. D. Clemis, Laryngoscope, 1978, 88, 723. 54. M. F. Mafee, D. Charletta, A. Kumar, and H. Belmont, Am. J. Neuroradiol., 1992, 13, 805. 55. T. Okumura, H. Takahashi, I. Honjo, A. Takagi, and K. Mitamura, Laryngoscope, 1995, 105, 289. 56. S. H. Selesnick, R. K. Jackler, and L. W. Pitts, Laryngoscope, 1993, 103, 431.

Acknowledgements The authors sincerely thank W. G. Bradley, Jr, M.D., Ph.D., for reviewing the manuscript and for granting permission to reproduce Figure 2.

Biographical Sketches Anton N. Hasso. b 1940. B.S., 1962, M.D., 1967, Loma Linda University. Faculty, UCLA, 1972±74. Faculty, Loma Linda University, 1974±96, Professor and Chairman, Department of Radiological Sciences, University of California, Irvine, 1996±present. Approx. 130 publications including 15 in NMR; senior author of MRI Atlas of the Head and Neck (1993) and MRI of Brain: Neoplastic Disease (1991); co-author of MRI of the Brain, Head and Neck, A Text Atlas (1985). Research interests include applications of NMR in neuroradiology/head and neck radiology, particularly magnetic resonance angiography and the use of gadolinium contrast agents. Peggy J. Fritzsche. b 1941. B.S., Andrews University, 1962; M.D., Loma Linda University, 1966. Faculty, UCLA, 1973±74. Faculty, Loma Linda University, 1974±91, Medical Director, Riverside MRI, 1991±present. Approx. 85 publications including eight in NMR; coauthor of MRI of the Body (1992). Research interests include applications of NMR in genitourinary and abdominal radiology, breast imaging and pelvic imaging.

DEGENERATIVE DISK DISEASE STUDIED BY MRI

Degenerative Disk Disease Studied by MRI Michael T. Modic Cleveland Clinic Foundation, OH, USA

1 INTRODUCTION Degeneration of the intervertebral disk complex is a process that begins early in life and is a consequence of a variety of environmental factors as well as normal aging. The pathophysiology of this disorder is complicated and poorly understood. It has been stated1 that degeneration as commonly applied to the intervertebral disk covers such a wide variety of clinical, radiologic, and pathologic manifestations as to be really `only a symbol of our ignorance'. It is estimated that low back pain affects 5% of the adult population each year and that the lifetime incidence of LBP in adults is 60±80%.2 Fortunately, the natural history of back pain is such that the vast majority of back pain sufferers improve with little or no medical intervention; only 14% of patients have episodes of pain lasting more than 2 weeks.3 Despite the self-limited nature of most LBP, the total societal costs of LBP have been estimated at US$16±100 billion annually,2 and, as back pain is the second leading cause of physician visits among American adults, it is estimated to account for more than US$10 billion in direct medical care costs each year.4 Much of the direct cost of treating patients with LBP is related to diagnostic tests;5 it is estimated that MRI alone contributes US$2 billion dollars in direct medical care costs (Modic MT, 1993, projected estimate). The sequelae of disk degeneration remain among the leading causes of functional incapacity in both sexes and are an all too common source of chronic disability in the working years. However, by the age of 50 years, 85±95% of adults show evidence of degenerative disk disease at autopsy;6 thus the jump from identifying an anatomic derangement to proposing a symptom complex must be made with caution, since to date only a moderate correlation has been found between imaging evidence of disk degeneration and symptomatology.7 There is no unifying theory regarding the cause of disk degeneration. It is likely that degeneration and aging are multifactorial processes that encompass a wide spectrum of changes and sequelae. Nevertheless, in patients in whom surgical therapy is planned, or where the diagnosis remains elusive and patient treatment and outcome is dependent upon a more de®nitive diagnosis, imaging studies are an important component in the evaluation of patients with symptoms of disk disease.

2 MORPHOLOGY The contrast sensitivity and multiplanar imaging capability of proton magnetic resonance (MR) place this modality in a position to provide a unique noninvasive means of imaging

1

intervertebral disks. The implementation of surface coil technology,8 cardiac gating,9 gradient refocusing,10 paramagnetic contrast agents,11 saturation pulses,12 gradient echo volume imaging, turbo (fast) T2-weighted spin echo sequences, and magnetization transfer techniques continue to re®ne the capability of MRI for degenerative disk disease. When a combination of imaging planes and pulse sequence parameters is used, the anatomy of the intervertebral disk (Figure 1), spinal nerves, dural sac, and adjacent structures can be clearly depicted. From a morphological aspect, MRI may be the most accurate means of evaluating the intervertebral disk. Accordingly, most research to date has been directed clinically towards optimizing anatomic image display in a fashion similar to that of computerized tomography (CT) for assessment of disk contour.13±19 Unlike CT and conventional radiography, however, which are dependent on information related to electron density, proton MR signals are in¯uenced by the T1 and T2 relaxation times, and by proton density, providing greater tissue contrast. Thus the role of this technique may go beyond gross anatomic appraisal to actual tissue characterization of pathology and biochemical change.20 The relationship among the vertebral body, endplate, and disk has been studied21±24 using both degenerated and chymopapain-treated disks as models. Signal intensity changes in vertebral body marrow adjacent to the endplates of degenerated disks are a common observation in MRI. Work using T2weighted spin echo sequences20 further suggests that MRI is capable of detecting changes in the nucleus pulposus and anulus ®brosus relative to degeneration and aging based on a loss of signal presumed to be secondary to known changes in hydration that occur within the intervertebral disk (Figure 2). However, the correlation is not straightforward, since differences in signal intensity appear to be somewhat exaggerated for the degree of water loss noted with degeneration (about 15%).25 At present the role that speci®c biochemical changes (proteoglycan ratios, aggregating of complexes) play in the altered signal intensity is not well understood. In fact, the important factor may not be the total quantity of water but the state that the water is in. Sodium images suggest that the T2 signal intensity in the disk tracks the concentration and regions of high glycosamino glycans (GAG) percentages. Thus it seems likely that the health and status of the proteoglycans determine the signal intensity. The vacuum phenomenon within a degenerative disk is represented on spin echo images as areas of signal void.26 Gradient echo sequences demonstrate this better than do conventional spin echo sequences, plain radiographs and CT. This is due to the magnetic susceptibility effects caused by the intradiskal gas collection. Whereas the presence of gas within the disk is usually suggestive of degenerative disease rather than an infected process, spinal infection may (rarely) be accompanied by intradiskal or intraosseous gas.27 A gas density cleft within a transverse separation of the vertebral body appearing in extension and disappearing in ¯exion is characteristic of the vacuum phenomenon within a region of ischemic vertebral collapse. Rarely, the phenomenon has been identi®ed with vertebral body neoplasms such as multiple myeloma.28±32 Hyperintense intervertebral disks are not an infrequent ®nding on T1-weighted MR images of the spine, and it has been suggested33 that the relatively `bright' intervertebral disk may re¯ect diffuse abnormality and loss of normal signal in the

2 DEGENERATIVE DISK DISEASE STUDIED BY MRI

Figure 1 Axial L3±L4 disk: (a) and (b) are normal. Axial T1-weighted spin echo (500/17), and T2-weighted spin echo (2000/90) images through the axial midplane of the intervertebral disk. On the T1-weighted axial image (a) the disk has a homogeneous soft tissue signal intensity. On the T2weighted image (b) three distinct components of the disk can be identi®ed: ®rst, a decreased signal intensity is noted in the outer annulus and ligamentous region; second, the inner annulus has a slightly increased signal intensity, and third, the region of the nucleus pulposus has the highest signal intensity consistent with the highest concentration of proteoglycans

marrow of the adjacent vertebral bodies. Calci®cation has usually been described on MR images as a region of decreased or absent signal. The loss of signal is attributed to a low mobile proton density as well as, in the case of gradient echo imaging, to its sensitivity to the heterogeneous magnetic susceptibility found in calci®ed tissue. There is, however, variability of signal intensity of calcium on various sequences; the type and concentration of calci®cation are probably important factors. Multiple examples of a hyperintense signal on T1-weighted spin echo images in areas that contain calci®cation on CT have been reported in the literature. These hyperintensities have been attributed to the paramagnetic effects of methemoglobin,34±36 melanin,37 and trace elements,38,39 as well as to the T1-shortening effects of lipids and cholesterol,40 proteins,41,42 and laminar necrosis associated with infarction and calci®cation.43±47 Focal or diffuse areas of hyperintensity on T1-weighted spin echo sequences may also be encountered in the intervertebral disk. Hyperintensities that are affected by fat-suppression techniques have also been noted within intervertebral disks.48 These are presumably related to areas of ossi®cation with formation of a lipid marrow in the ossi®ed disk space; they also appear calci®ed on conventional studies. 3 SEQUELAE OF DISK DEGENERATION Disk herniation, especially in the lumbar and thoracic regions, is probably better depicted by MRI than by other mod-

alities.14±19,49,50 It is, again, important to bear in mind that not all morphological changes are a cause of symptoms. In a prospective study of individuals who never had low back pain, sciatica or neurogenic claudication,51 one-third of the subjects were found to have substantial abnormalities. In patients who were less than 60 years of age, 20% had a herniated nucleus pulposus. In the group that was 60 years or older, ®ndings were abnormal on 57% of the scans. Thirty-six per cent of the subjects had a herniated nucleus pulposus, and 21% had spinal stenosis. More recent studies52,53 have con®rmed the observation that disk bulges and protrusions are common ®ndings on MRI in the asymptomatic population. These latter studies, however, suggest that frank disk extrusion is rare; a disk bulge was observed to be age related, but disk protrusion was not. Multidimensional imaging allows the direct acquisition of orthogonal views covering long segments of the spine, without requiring secondary reconstructions. Alternatively, threedimensional datasets can be acquired with or without contrast, which allows multiplanar reformation with variable partition thicknesses for improved overall spatial resolution and reduced examination times. The outer anulus±posterior longitudinal ligament (PLL) complex can usually be seen as an area of decreased signal relative to the inner anulus± nucleus pulposus, which helps in characterizing the type of herniation (protrusion, extrusion and/or sequestration).54 This ability to characterize and differentiate the various subgroups of disk herniation has been proposed to have certain diagnostic and therapeutic rami®cations, particularly in the lumbar region.

DEGENERATIVE DISK DISEASE STUDIED BY MRI

3

Figure 2 Spin echo and gradient sagittal images. (a) Intermediate spin echo (2000/20). (b) T2 weighted spin echo (2000/90). (c) FLASH (50/13/10 degrees). (d) FLASH (50/13/60 degrees). (b) There is decreased signal intensity in the L4±L5 and L5±S1 disk, consistent with degeneration. The L2±L3 and L3±L4 disks have a normal signal intensity. The signal intensity changes consistent with degeneration are not as well appreciated from sequences (a), (c) and (d). (Reproduced with permission from Magnetic Resonance Imaging of the Spine, 2nd edn, Chap. 2, p. (52, Fig. 2-14)

Abnormal disks can be classi®ed as an anular bulge or herniated (protruded, extruded, or free fragment). Concurrently, herniated disk disease should be described by contour, size, location, and presence or absence of enhancement. 4 BULGE An anular bulge is a result of disk degeneration with a grossly intact, albeit lax, anulus, usually recognized as a generalized extension of the disk margin beyond the margins of the adjacent vertebral endplates, regardless of the signal of the interspace. An index of disk bulging using sagittal anatomic

sections from 149 lumbar disks55 has been studied. The largest disk bulgings were always associated with radial tears of the anulus, which contradicts the previously held concept that the anulus ®brosus remains intact with a bulging disk. Some investigators would recognize the anular tear as an intermediate category between bulge and herniation. The signi®cance of the anular tear, however, remains extremely unclear; it is perhaps best considered as a sequela of disk degeneration rather than as a discrete category that may imply some type of clinical signi®cance. On nonenhanced T2weighted MR images anular tears are best appreciated as an area of high signal extending into the area of decreased signal of the anulus±ligamentous complex (Figure 3).

4 DEGENERATIVE DISK DISEASE STUDIED BY MRI

Figure 2 continued

5 LUMBAR DISK HERNIATIONS Obviously some degree of anular disruption is an intrinsic component of any disk herniation. A protrusion represents herniation of nuclear material through a defect in the anulus, producing a focal or broadbased extension of the disk margin. The extension is less than that which occurs with an extrusion. At least some ®bers of the overlying anulus and PLL remain intact, and the disk is described as `contained'. The signal intensity of the parent nucleus is usually decreased. The next two categories, extrusion and sequestration, represent herniations that are no longer contained by the overlying anulus and ligament (Figure 4). In an extrusion, nuclear material becomes an anterior extradural mass that remains attached to the nucleus of origin, often via a high-signal pedi-

cle on the T2-weighted image. The signal intensity of the extruded portion may be increased or decreased. The disk usually appears contained by the PLL and remaining contiguous portions of the anulus, which show up as curvilinear areas of decreased signal. The term `free fragment' or `sequestrated disk' refers to disk material external to the anulus ®brosus and no longer contiguous with the parent nucleus. A frequent ®nding with both extruded and free fragments is the presence of a high signal intensity extradural defect often surrounded by a curvilinear area of decreased signal that is distinct from the interspace of origin. Whereas all of the foregoing has become a part of our clinical lore, it is not at all clear whether these ®ndings are clinically relevant. It has been proposed, for example, that differentiating between various degrees of herniation is critically

DEGENERATIVE DISK DISEASE STUDIED BY MRI

Figure 3 Sagittal T2-weighted spin echo image (2000/90) through a cadaveric spine. Note the loss of signal intensity at the L4±L5 disk and the region of high signal extending into the anulus±ligament complex (arrow). This represents an annular tear. (Reproduced with permission from Magnetic Resonance Imaging of the Spine, 2nd edn, Chap. 3, p. 96, Fig. 3-16)

important; yet the reality of the situation is that any disk herniation most likely represents a spectrum or continuum rather than a discrete entity with speci®c clinical relationships. One may view the continuum of herniated disk disease as starting with anular disruption, proceeding to a small focal herniation that does not break through the anulus±ligamentous complex, and winding up as a frank herniation (extrusion) that does indeed dissect through the anulus±posterior ligamentous complex. Traversing nerve roots within the thecal sac above a particular level of impingement from a herniated disk or stenosis have been noted to be enlarged compared to the contralateral side. This may be a re¯ection of nerve root edema. Enhancement of traversing nerve roots within the thecal sac has also been reported56 on MRI, the majority of cases being associated with focally protruding disk pathology. A signi®cant number of these patients, however, had isolated enhancement of multiple nerve roots without signi®cant associated

5

anatomic pathology. The mechanism of such enhancement may be related to the blood±nerve barrier of this spinal nerve root, which is altered by compression.57 It has also been suggested that the enhancement may be related to the vascular anatomy as a variation of normal, rather than as a pathologic reaction.58 In a prospective evaluation of surface coil MRI, CT, and myelography for lumbar herniated disk disease and canal stenosis16 there was 82.6% agreement between MRI and surgical ®ndings regarding type and location of disease, 83% agreement between CT and surgical ®ndings, and 71.8% agreement between myelography and surgical ®ndings. In another study,59 the accuracy of CT, myelography, CTmyelography and MRI for lumbar herniated nucleus pulposus was compared prospectively in 59 patients, all of whom underwent surgical exploration. MRI was the most accurate (76.5%), CT-myelography next (76%), and then CT (73%) and myelography (71%). The false-positive rate was lowest for MRI, at 13%, followed by myelography and CT.59 Similar data from a carefully controlled prospective study in Wisconsin60 suggest the same equivalence for plain CT, CT-myelography, and MRI. The natural history of lumbar disk herniation has engendered recent interest and MRI has been an excellent tool for these investigations. Multiple studies have demonstrated that the size of the disk herniation can change with time, and studies of patients treated conservatively have demonstrated that the majority will show a reduction in disk herniation size of 30±100%. The larger herniations show the most signi®cant decrease in size. A resulting impression has also been that patients who do well conservatively will show this reduction in herniation size, yet the correlation of disk changes with time and symptoms resolution has not yet been carefully worked out. As part of a long term natural history study, patients with acute radiculopathy were evaluated clinically and with contrast enhanced MRI to determine if there was a correlation between presenting symptoms and the type, size, location and enhancement of the disk herniations at presentation. In this group, 72% had a herniated disk at one or more levels at presentation. There was excellent agreement between MRI and clinical ®ndings for level and size, but no correlation of pain and disability with disk size, type or enhancement. At the 6-week follow-up, 36% of large herniated disks demonstrated a signi®cant reduction in size (Figure 5), but there was no difference in clinical outcome based on disk size, change, or type. At 6 months, 60% of herniated disks had reduced signi®cantly in size, but again there was no correlation of pain and disability with disk size, change, or type. Enhancement of herniated disk was almost a constant feature. The degree of enhancement was variable and probably related to granulation tissue which develops around and through herniated disk. The granulation tissue response itself may comprise a fair percentage of the herniated disk mass. One possible explanation for the change of a disk with time then is that the granulation tissue undergoes cicatrization and retraction, a normal phenomenon for reparative tissue. Although less common than lumbar or cervical herniations, thoracic herniations have been noted more frequently with the advancement of imaging techniques.19 A higher prevalence of

6 DEGENERATIVE DISK DISEASE STUDIED BY MRI

Figure 4 L4±L5 disk herniation. (a) Pre- and (b) post-contrast sagittal T1-weighted spin echo images demonstrate an extradural mass at the L4±L5 disk level. Axial T1-weighted spin echo (500/17 images) before (c) and after (d) contrast show the peripheral enhancement of disk material, which is also appreciated on the sagittal images. The enhancing areas of granulation tissue contribute to the overall size of the mass. (Reproduced with permission from Magnetic Resonance Imaging of the Spine, 2nd edn, Chap. 3, p. 102, Fig. 3-26)

asymptomatic morphologic change is also present in the thoracic spine. Symptomatic cervical disk herniations are most common in young patients (third and fourth decades) and frequently occur without recognized trauma. In a prospective study of patients with no symptoms of cervical disease,61 10% of subjects less than 40 years of age had a herniated nucleus pulposus and 4% had foraminal stenosis. Of the subjects who were older than 40 years of age, 5% had a herniated nucleus pulposus, 3% a bulging disk, and 20% foraminal stenosis. Narrowing of the disk space, degeneration of a disk, spurs, or compression of the cord were noted at one or more levels in 25% of the subjects less than 40 years of age and in almost 60% of those who were older than 40. Cervical disk herniation, especially when central or large, is well appreciated on routine sagittal and axial MR images.

Again, thin slices (3 mm or less) are critical for accurate diagnosis. This may necessitate the use of three-dimensional volume sequences with partitions of 2 mm or less and/or reconstructions in other planes. The T2 signal intensity of intervertebral disks in the cervical region is not as helpful as that in the lumbar region for identifying the presence or absence of degeneration. Cervical disk herniation is usually identi®ed on sagittal images as an anterior or anterior±lateral extradural defect that may indent or compress the cervical cord. In a prospective study to compare the accuracy of surface coil MRI with metrizamide myelography (MM) and CT with metrizamide (CTM),17 there was surgical agreement in 74% of patients with surface coil MRI, 85% with CTM, and 67% with MM. When surface coil MRI and CTM were used jointly, 90% agreement with surgical ®ndings was seen, and when CTM and MM were used jointly there was 92% agreement. In gen-

DEGENERATIVE DISK DISEASE STUDIED BY MRI

7

Figure 4 continued

eral, surface coil MRI was as sensitive as CTM for identifying disease level, but not as speci®c for type of disease. In another study comparing MRI and CT, plain ®lm myelography, and CT-myelography in 35 patients operated on for cervical radiculopathy and myelopathy,62 MRI correctly predicted 88% of the surgically proved lesions. The corresponding rates were 81% for CT myelography, 58% for plain myelography, and 50% for CT. Although we would maintain that most evaluations of the cervical spine for extradural disease can be done in an adequate fashion using spin echo T1-weighted sagittal and axial images, the introduction of fast gradient echo images with small ¯ip angles (less than 15 ) has improved the accuracy of the examination by increasing the conspicuousness of extradural defects. Recent work63±65 has con®rmed the value of this technique. This capability, coupled with the potential of utilizing very short TRs (50 ms or less), has provided the stimulus for using volume imaging in the evaluation of extradural disease, in the hope of shortening the examination time and decreasing slice thickness. The disadvantages of gradient echo imaging relate to problems with ®eld inhomogeneity and contrast detectability of pathologic processes within the cord itself. Furthermore, depending upon the echo time, the ability accurately to characterize morphologic defects, foraminal narrowing, and bony ridges is inferior to that for viewing bone by CT. An adjunct to conventional MRI in evaluating degenerative disk disease is the utilization of gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA). Studies with virgin disks and

postoperative disk herniation11 indicate that it is the most accurate means of separating epidural ®brosis and disk tissue. There is consistent enhancement of peridiskal ®brosis early (less than 15 min after injection), and variable enhancement occurs in herniated or degenerated parent disks late (30 min after injection). Enhancement of intervertebral disks does not appear to occur in the normal state and is probably a sequela of the degenerative process.11,66 Thus it seems likely that GdDTPA may play a role in identifying the sequelae of degenerative disk disease by enhancing the reactive granulation tissue that forms secondary to disruption of the disk and associated structures. In addition to changes within the disk, including herniation, secondary changes are noted both in animal models and in humans following disk degeneration. The stability of a motion segment depends on the integrity of all its components. Diseases occurring in this area are circumferential, with that in one joint affecting another, and degeneration of one disk leads to a loss of disk height and forces the facet joints into malalignment, so-called `rostrocaudal subluxation'. This leads to increased biomechanical forces at the facet joint with increasing joint relaxation and instability, secondary facet and arthritic changes, and potential fractures. Similarly, abnormal movements allowed by disk degeneration and facet changes add stress to the posterior ligaments and can result in hypertrophy. A vicious degenerative cycle is established which includes degenerative disk disease, facet arthrosis, ligamentous and capsular hypertrophy, spinal instability, and lumbar stenosis.67

8 DEGENERATIVE DISK DISEASE STUDIED BY MRI

Figure 5 L3±L4 disk herniation. (a) Pre- and (b) post-contrast axial T1-weighted spin echo images through the L3±L4 disk in a patient who presented with the onset of acute right-sided radiculopathy. A moderate sized extradural mass is identi®ed on the right which shows peripheral enhancement following the administration of contrast. (c) (d) Axial T1-weighted spin echo (500/17) images at the same level as (a) and (b) following 6 weeks of conservative management. There is a marked reduction in the size of the extradural mass, although there is still some enhancement in this region

6 SPINAL STENOSIS Spinal stenosis results from an overall diminution of the spinal canal, lateral recesses, or neural foramina. It occurs more commonly in the lumbar and cervical regions. The symptoms are usually a re¯ection of the compressive pathology. In the lumbar spine, central stenosis tends to occur at multiple levels but is most frequent and usually most severe at the L4±L5 level, where it may occur alone. The transverse and sagittal dimensions of the central neural canal are best depicted by integrating orthogonal (e.g. axial and

sagittal) planes. Gradient echo images with low ¯ip angles or more T2-weighted spin echo sequences provide the best views of thecal sac dimensions by furnishing a gray-scale inversion of the cerebrospinal ¯uid and extradural elements.63,64 Peripheral stenosis is best appreciated on T1-weighted images, which maintain the separation between neural structures and epidural fat quite well. The signal intensity between neural elements and fat is less well seen on more T2-weighted and gradient echo images, although bony overgrowth can be identi®ed. In the cervical spine degenerative changes affect all major spinal articulations, including the intervertebral disks, apophy-

DEGENERATIVE DISK DISEASE STUDIED BY MRI

seal joints, ligamentous connections between vertebral bodies, and the vertebral bodies themselves. The suggestion has been made68 that myelopathy or stenosis is associated with an average canal diameter of less than 12 mm. Care must be taken, however, when using gradient echo images since the degree of spinal narrowing can be overestimated because of susceptibility effects. Another method suggested for determining cervical spinal stenosis69 is the vertebral body/canal ratio, which has been suggested to correct for body size. A spinal canal/vertebral body ratio of less than 0.82 would indicate signi®cant cervical spinal stenosis. MRI and myelographic CT studies appear to be equivalent in measuring the degree of cord compression. There may be a correlation between compression and the amount of neurologic dysfunction.70 Although there are no good studies to indicate that MRI has led to improved therapeutic choices, there is at present suf®cient evidence to suggest that MRI is adequate for most therapeutic planning to allow physicians to stop further testing, obviating the need for a myelogram or CT scan.71,72

7 FACET JOINTS Although facet arthrosis constitutes an important cause of acquired lumbar stenosis, degenerative change can occur independently of it and may be a cause of low back pain and radiculopathy.67,73 On MRI the facet joints are best evaluated by visually integrating sagittal and axial images. Again, osteophytes are easier to identify on gradient echo or spin echo images with long relaxation times. Those that contain marrow are also better demarcated than are those that are sclerotic, which can be confused with the adjacent capsular ligamentous structures. Altered signal intensity in bone marrow consistent with fatty replacement has also been reported to be commonly associated.74 The articular cartilage can be directly visualized on both T1 and T2-weighted spin echo images, but thinning is dif®cult to measure accurately because of variable axial obliquity and chemical shift artifact from the adjacent facet. Gradient echo examinations have demonstrated that they provide better conspicuousness of the articular cartilage itself as distinct from adjacent osseous structures. In summary, degenerative disk changes appear to be a normal consequence of the aging process. MRI is an excellent modality for depicting them. However, the clinical relevance of these morphologic changes remains to be established and the value of MRI as a prognostic indicator for patient outcome needs to be studied more fully. At the present time, one can clearly recommend MRI as the single best presurgical decisionmaking tool, but its role in the evaluation of patients who are going to be treated in a conservative fashion appears to be much more limited.

8 RELATED ARTICLES Contrast Agents in Magnetic Resonance: Operating Mechanisms; Contrast Agents in Whole Body Magnetic Resonance: An Overview; Gadolinium Chelate Contrast Agents in MRI:

9

Clinical Applications; Head and Neck Investigations by MRI; Lung and Mediastinum MRI.

9

REFERENCES

1. K. P. H. Pritzker, Orthop. Clin. North Am. 1977, 8, 66. 2. J. W. Frymoyer and W. L. Cats-Baril, Orthop. Clin. North Am., 1991, 22, 263. 3. R. A. Deyo and Y. J. Tsui-Wu, Spine, 1987, 12, 264. 4. B. K. Cypress, Am. J. Public Health, 1983, 73, 389. 5. R. A. Deyo, J. Gen. Intern. Med. 1986, 1, 328. 6. R. J. Quinet and N. M. Hadler, Semin. Arthritis Rheum. 1979, 8, 261. 7. A. A. White and S. L. Gordon, Spine, 1982, 7, 141. 8. L. Axel, J. Comput. Assist. Tomogr., 1984, 8, 381. 9. D. R. Enzmann, J. B. Rubin, and A. Wright, Radiology, 1987, 162, 763. 10. E. M. Haacke and G. Lenz, Am. J. Roentgenol. 1987, 148, 1251. 11. M. G. Hueftle, M. T. Modic, J. S. Ross, T. J. Masaryk, J. R. Carter, R. G. Wilber, H. H. Bohlman, P. M. Steinberg, and R. B. Delamartor, Radiology 1988, 167, 817. 12. R. R. Edelman, D. J. Atkinson, M. S. Silver, F. L. Hoaiza, and W. S. Warren, Radiology 1988, 166, 231. 13. N. I. Chafetz, H. K. Genant, K. L. Moon, C. A. Helms, and J. M. Morris, Am. J. Roentgenol., 1983, 141, 1153. 14. R. R. Edelman, G. M. Shoukimas, D. D. Stark, R. R. Davis, P. F. New, S. Saini, D. I. Rosenthal, G. L. Wismer, and T. J. Brady, Am. J. Roentgenol., 1985, 144, 1123. 15. K. R. Maravilla, H. P. Lesh, J. C. Weinreb, D. K. Selby, and V. Mooney, Am. J. Neuroradiol., 1985, 6, 237. 16. M. T. Modic, T. J. Masaryk, F. Boumphrey, M. Goormastic, and G. Bell, Am. J. Roentgenol., 1986, 147, 757. 17. M. T. Modic, T. J. Masaryk, G. P. Mulopolos, C. V. Bundschuh, J. J. Han, and H. Bohlman, Radiology, 1986, 161, 753. 18. M. T. Modic, T. J. Masaryk, J. S. Ross, G. P. Mulopolos, C. V. Bundschuh, and H. Bohlman, Radiology, 1987, 163, 227. 19. J. S. Ross, N. Perez-Reyes, T. J. Masaryk, M. Bohlman, and M. T. Modic, Radiology, 1987, 165 511. 20. M. T. Modic, W. Pavlicek, M. A. Weinstein, F. Boumphrey, F. Ngo, R. Hardy, and P. M. Duchesneau, Radiology, 1984, 152, 103. 21. A. deRoos, H. Kressel, C. Spritzer, and M. Dalinka, Am. J. Roentgenol., 1987, 149, 531. 22. M. T. Modic, P. M. Steinberg, J. S. Ross, T. J. Masaryk, and J. R. Carter, Radiology, 1988, 166, 193. 23. J. Aoki, I. Yamamoto, N. Kitamura, T. Sone, H. Iroh, K. Torizuka, and K. Takasu, Radiology, 1987, 164, 411. 24. T. J. Masaryk, F. Boumphrey, M. T. Modic, C. Tamborrello, J. S. Ross, and M. D. Brown, J. Comput. Assist. Tomogr. 1986, 10, 917. 25. V. M. Haughton, Radiology, 1988, 166, 297. 26. N. Grenier, R. I. Grossman, M. L. Schiebler, B. A. Yeager, H. I. Goldberg and H. Y. Kressel, Radiology, 1987, 164, 861. 27. D. K. Bielecki, D. Sartoris, D. Resnick, K. Van Lom, J. Fierer, and P. Haghighi, Am. J. Roentgenol., 1986, 147, 83. 28. W. Kumpan, E. Salomonowitz, G. Seidl, and G. R. Wittich, Skeletal Radiol. 1986, 15, 444. 29. D. Resnick, G. Niwayama, J. Guerra, V. Vint, and J. Usselman, Radiology, 1981, 139, 341. 30. L. G. Naul, G. J. Peet, and W. B. Maupin, Radiology, 1989, 172, 219. 31. B. E. Maldague, H. M. Noel, and J. J. Malghem, Radiology, 1978, 129, 23. 32. F. Gagnerie, B. Taillan, L. Euller-Ziegler, and G. Ziegler, Clin. Rheumatol., 1987, 6, 597.

10 DEGENERATIVE DISK DISEASE STUDIED BY MRI 33. M. Castillo, J. A. Malko, and J. C. Hoffman, Am. J. Neuroradiol., 1990, 11, 23. 34. J. M. Gomori, R. I. Grossman, H. I. Goldberg, R. A. Zimmerman, and L. T. Bilaniuk, Radiology, 1985, 157, 87. 35. J. M. Gomori, R. I. Grossman, D. B. Hackney, H. I. Goldberg, R. A. Zimmerman, and L. T., Bilaniuk Am. J. Neuroradiol., 1987, 8, 1019; Am. J. Roentgenol., 1988. 36. H. Nabatame, N. Fujimoto, K. Nakamura, Y. Imura, Y. Dodo, H. Fukuyama, and J. Kimura, J. Comput. Assist. Tomogr., 1990, 14, 521. 37. S. A. Mirowitz, K. Sartor, and M. Gado, Am. J. Neuroradiol., 1989, 10, 1159; Am. J. Roentgenol., 1990, 154, 369. 38. S. A. Mirowitz, T. J. Westrich, and J. D. Hirsch, Radiology, 1991, 181, 117. 39. S. A. Mirowitz and T. J. Westrich, Radiology, 1992, 185, 535. 40. P. P. Maeder, S. L. Holtas, L. N. Basibuyuk, L. A. Salford, V. A. Tapper, and A. Bruh, Am. J. Neuroradiol., 1990, 11, 575. 41. P. M. Som, W. P. Dillon G. D. Fullerton, R. A. Zimmerman, B. Rajagopalan, and Z. Marom, Radiology, 1989, 172, 515. 42. K. Abe, H. Hasegawa, Y. Kobayashi, H. Fujimura, S. Yorifuji, and S. Biton, Neuroradiology, 1990, 32, 166. 43. O. B. Boyko, P. C. Burger, J. D. Shelburne, and P. Ingram, Am. J. Neuroradiol., 1992, 13, 1439. 44. R. M. Henkelman, J. F. Watts, and W. Kucharczyk, Radiology, 1991, 179, 199. 45. L. A. Dell, M. S. Brown, W. W. Orrison, C. G. Eckel, and N. A. Matwiyoff. Am. J. Neuroradiol., 1988, 9, 1145. 46. R. D. Tien, J. R. Hesselink, and A. Duberg, Am. J. Neuroradiol., 1990, 11, 1251. 47. Y. Araki, T. Furukawa, K. Tsuda, T. Yamamoto and I. Tsukaguchi, Neuroradiology, 1990, 32, 325. 48. B. A. Bangert, M. T. Modic, J. S. Ross, N. A. Obuchowski, J. Perl, P. M. Ruggieri, and T. J. Masaryk, Radiology, 1995. 49. R. P. Jackson, J. E. Cain, R. R. Jacobs, B. R. Cooper, and G. E. McManus, Spine, 1989, 14, 1362. 50. J. S. Ross, M. T. Modic, T. J. Masaryk, J. Carter, R. E. Marcus, and H. Bohlman, Am. J. Neuroradiol., 1989, 10, 1243. 51. S. D. Boden, D. O. Davis, T. S. Dina, N. J. Patronas, and S. W. Wiesel, J. Bone Joint Surg (Am.), 1990, 72, 403. 52. M. C. Jensen et al., N. Engl. J. Med. 2, 1994, 331, 69. 53. D. Fardon, S. Pinkerton, R. Balderston, S. Gar®n, R. Nasca, and R. Salib, Spine, 1993, 18, 274. 54. T. J. Masaryk, J. S. Ross, M. T. Modic, F. Boumphrey, H. Bohlman, and G. Wilberg, Am. J. Neuroradiol., 1988, 150, 1155. 55. S. W. Yu, V. M. Haughton, L. A. Sether, and M. Wagner, Radiology, 1988, 169, 761. 56. J. R. Jinkins, Am. J. Neuroradiol., 1993, 14, 193. 57. S, Kobayashi, H. Yoshizawa, and Y. Hachiya in `18th Annual Meeting of the International Society for the Study of the Lumbar Spine, 1991'. 58. R. Quencer Am. J. Neuroradiol., 1994, 59. R. P. Jackson, J. E. Cain, R. R. Jacobs, B. R. Cooper, and C. E. McManus, Spine 1989, 14, 1362. 60. J. R. Thornbury, D. G. Fryback, P. A. Turski, M. J. Jarid, J. V. McDonald, B. R. Bernlich, L. R. Gentry, J. F. Saekott, E. J. Dosbach, and P. A. Martin, Radiology, 1993, 186, 731. 61. S. D. Boden, P. R. McCowin, D. O. Davis, T. S. Dina, A. S. Mark, and S. Wiesel, J. Bone Joint Surg. (Am.), 1990, 72, 1178.

62. B. M. Brown, R. H. Schwartz, E. Frank, and N. K. Blank, Am. J. Neuroradiol., 1988, 9, 859. 63. M. C. Hedberg, B. P. Drayer, R. A. Flom, J. A. Modak and C. R. Bird, Am. J. Roentgenol., 1988, 150, 683. 64. D. R. Enzmann, and J. B. Rubin, Radiology, 1988, 166, 467. 65. D. M. Yousem, S. W. Atlas, H. I. Goldberg, and R. I. Grossman, Am. J. Neuroradiol., 1991, 12, 229. 66. E. E. Awwad, D. S. Martin, K. R. Smith, and R. D. Bucholz, J. Comput. Assist. Tomogr., 1990, 14, 415. 67. D. Schellinger, L. Wener, B. D. Ragsdale, and N. J. Patronas, RadioGraphics, 1987, 7, 944. 68. T. B. Freeman and C. R. Martinez, Perspect. Neurol. Surg., 1992, 3, 34. 69. H. Pavlov, J. S. Torg, B. Robie, and C. Jahre, Radiology, 1987, 164, 771. 70. T. Fukushima, T. Ikata, Y. Taoka, and S. Takata, Spine, 1991, 16, (Suppl. 10), 5534. 71. P. F. Statham, D. M. Hadley, P. MacPherson, R. A. Johnston, I. Bone, and G. M. Teasdale, J. Neurol. Neurosurg. Psychiatry., 1991, 54, 484. 72. B. M. Brown, R. H. Schwartz, E. Frank, and N. K. Blank, Am. J. Roentgenol., 1988, 151, 205. 73. R. I. Harris and I. McNab, J. Bone Joint Surg. (Br.), 1954, 36, 304. 74. N. Grenier, H. Y. Kressel, M. L. Schiebler, R. I. Grossman, and M. K. Dalinka, Radiology, 1987, 165, 517.

Biographical Sketch Michael T. Modic received his M.D. degree from Case Western Reserve School of Medicine in 1975. He completed his residency in radiology and fellowship in neuroradiology at the Cleveland Clinic Foundation. He spent a year as assistant professor of Radiology and as a staff neuroradiologist at University Hospitals in Cleveland from 1979±1980 and in 1980 returned to the Cleveland clinic as a staff neuroradiologist. In 1982 he was appointed head of the section of Magnetic Resonance. In 1985 he returned to Case Western Reserve School of Medicine/University Hospitals as Director of Magnetic Resonance and Neuroradiology, positions he held through 1989. During that time he also held the rank of Professor of Radiology, Neurology, and General Medical Sciences. In 1988 he was given a joint appointment as Professor of Neurosurgery as well. In 1989, Dr. Modic returned to the Cleveland Clinic Foundation as Chairman of Radiology and in 1993 was appointed as Professor of Radiology, Ohio State University. Dr. Modic has served on the Editorial Boards of the journals Radiology, American Journal of Neuroradiology, Neurology, Magnetic Resonance in Medicine, and Magnetic Resonance Imaging. He has served as a member of the Board of Trustees of the Society of Magnetic Resonance in Medicine and Board of Directors of the Society of Magnetic Resonance Imaging. He was President of the Society of Magnetic Resonance in Medicine for the 1992±1993 year and was the recipient of the Society Gold Medal in Clinical Science for his research activities related to the spine in 1991. He is co-author of the text `Magnetic Resonance Imaging of the Spine' which is in its second edition and is the author or co-author of over 125 peer reviewed articles related to neuroradiology.

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

Eye, Orbit, Ear, Nose, and Throat Studies Using MRI Mahmood F. Mafee University of Illinois at Chicago, Chicago, IL, USA

1 MRI OF THE EYE 1.1 Introduction The eye consists of three primary layers: (1) the sclera, or outer layer, composed primarily of collagen-elastic tissue; (2) the uvea or middle layer, a richly vascular and pigmented tissue consisting of the iris, ciliary body, and choroid; and (3) the retina or inner layer, the neurosensory stratum of the eye.1 The Tenon's capsule (bulbar fascia) is a ®broelastic membrane that covers the sclera and envelops the eyeball. Tenon's capsule encloses the posterior four-®fths of the eyeball separating it from the orbital fat. Tenon's capsule is separated from the sclera by the potential Tenon's space.2,3 1.2 Anatomy of the Globe 1.2.1

Anterior and Posterior Segments of the Globe

The ciliary body and the iris divide the globe into an anterior and a posterior segment. The anterior segment which includes the anterior and posterior chambers is the space lying between the cornea and the crystalline lens. The posterior (vitreous) segment contains the vitreous chamber (body), which acts as a biologic shock absorber. The vitreous chamber is a gel-like, transparent, extracellular matrix, composed of a meshwork of 0.2% collagen ®brils interspersed with 0.2% hyaluronic acid polymers, 98±99% water, and small amounts of soluble protein.4,5 1.2.2

Lens

The lens is one of the least hydrated organs of the body containing only 66% water.5 The remainder is composed of long, thin cells (lens ®bers) whose main cytoplasmic content is protein. 1.2.3

Sclera

The sclera, the outer, white, leathery coat of the globe is primarily composed of cellular bundles of collagen. Its external surface lies adjacent to Tenon's capsule. Its internal surface blends with the uveal tissues. 1.2.4

Uvea (Iris, Ciliary Body and Choroid)

The uvea lies between the sclera and retina and provides vascular supply to the eye and regulates ocular temperature. The choroid is a portion of the uvea that lies between the sclera and the retinal pigment epithelium (RPE). The ciliary body is a direct anterior continuation of the choroid, and the iris is a further extension of the ciliary body itself.

1.2.5

1

Retina

The external surface of the retina is in contact with the choroid and the internal surface is adjacent to the vitreous body; posteriorly, the retina is continuous with the optic nerve. Grossly the retina has two layers: (1) the inner layer is the sensory retina, composed of photoreceptors, ®rst and second order neurons (ganglion cells) and neuroglial elements and (2) the outer layer is the RPE, which consists of a single lamina of cells whose nuclei are adjacent to the basal membrane of the choroid, the so-called Bruch's membrane. 1.3

MRI Anatomy of the Globe

The globe is unique in that it contains the most (vitreous) and the least (lens) water concentration of all the tissues in the body.4,5 The eye is an ideal organ to be evaluated by MRI because the wide variations in water content produce different water proton relaxation times of the various tissues. The eye can be imaged using head coil or surface coil. Images obtained with surface coil provide better spatial resolution. A routine MRI of the eye usually includes single echo short repetition time (TR), short echo time (TE) (TR/TE, 400±600/20±25 ms), sagittal and axial views and an axial view SE double echo (TR/ TE, 2000/20±25; 70±100 ms) pulse sequences. Following the intravenous (i.v.) administration of gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA), single echo, short TR, short TE images in axial and other views are obtained with or without fat suppression techniques. The section thickness for MRI of the eye should be 3 mm. The normal lens has long T1 and short T2 relaxation properties on MRI (Figure 1). The nucleus of the lens has both lower water content and shorter T2 than the cortex of the lens. The sclera has a low signal intensity on various pulse sequences which is probably related to the short T2 caused by a greater proportion of bound water.5 Differentiation of individual layers of the retina, choroid, and posterior hyaloid membrane is impossible by MRI in the normal eye; however, in pathological conditions, detachment of different layers of the eye can be visualized on MR images. 1.4

MRI of Ocular Pathology

MRI has served as an important diagnostic method in ophthalmology for the last decade.2,6±14 Excellent contrast between various normal structures and high sensitivity in detection of pathologic states is due to the intrinsic differences in proton density and in proton relaxation times.2,4 1.5

Posterior Hyaloid, Retinal, and Choroidal Detachment

The posterior hyaloid space is a potential space between the posterior hyaloid (vitreous) membrane and the sensory retina. The posterior hyaloid membrane is thin and is invisible on MRI. It becomes visible when blood or other ¯uid ®lls the potential posterior hyaloid space. On MRI, a detached posterior hyaloid membrane is seen as a thin membrane within the globe, in front of the optic disk. Retinal detachment results from separation of the sensory retina from the RPE with accumulation of ¯uid in the potential subretinal space.

2 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI of the vortex veins and short posterior ciliary arteries. This restriction results in a characteristic appearance in which the leaves of the detached choroid, unlike the detached retina, frequently do not extend to the region of the optic disc. The MRI appearance of hemorrhagic posterior hyaloid, retinal and choroidal detachments varies with the age and the degree of organization of the hemorrhage. A hematoma less than 48 h old usually appears iso- to slightly hypointense relative to the normal vitreous body on T1-weighted (T1W) and proton-weighted (PW) MR images, but is markedly hypointense on T2-weighted (T2W) images. When the hematoma is approximately 5 days old, it appears relatively hyperintense on PW and T1W images but is hypointense on T2W images. At this stage, hematoma may be confused with an ocular melanoma.6,8,9 The hematoma usually continues to increase in signal intensity on T1W, PW, and T2W images and usually becomes markedly hyperintense by day 14 on all MR pulse sequences. 1.6

Figure 1 Normal MRI anatomy of the eye and orbit. Sagittal PW (top) and T2W (bottom) MR images. Ac = anterior chamber, EL = eyelids, FB = frontal bone, IO = inferior oblique muscle, IR = inferior rectus muscle, L = lens, Lc = lens capsule, LPS = levator palbebra superioris, MA = maxillary antrum, OA = ophthalmic artery, ON = optic nerve, OO = ®bers of the orbicularis oculi muscle, SOV = superior ophthalmic vein, SR = superior rectus muscle, STP = superior tarsal plate, Tc = Tenon's capsule

Since the retina is very thin, it cannot be directly visualized on MRI scans. However, it may be shown when it is outlined by signi®cant contrast differences between the subretinal effusion and the vitreous chamber. On axial MRI, the detached retina is seen as a V-shaped structure, with its apex at the optic disk and its extremities toward the ora serrata just behind the ciliary body, where the sensory retina ends. On coronal MR scans, retinal detachment appears as a characteristic folded membrane. The subretinal exudate is rich in protein, therefore has higher MR signals than the vitreous cavity on T1- and often on T2-weighted MR images. Choroidal detachment results from the accumulation of ¯uid (serous) or blood (hemorrhagic choroidal detachment) within the potential space between the choroid and the sclera. MRI is an excellent technique in the evaluation of choroidal detachment, particularly when ultrasonography or computerized tomography (CT) in conjunction with the clinical examination have not provided suf®cient information. Serous choroidal detachment appears as a semilunar or ring-shaped area of higher MR signal than vitreous on T1- and T2-weighted MR images.9 The appearance of choroidal detachment on MR images may be confused with retinal detachment. However, choroidal detachment is often restricted by the anchoring effect

Retinoblastoma

Retinoblastoma, the most common intraocular tumor of childhood, is a highly malignant retinal tumor. The tumor arises from neuroectodermal cells (nuclear layer of the retina) destined to become retinal photoreceptors.2,10 Leukokoria, a pink±white or yellow±white pupillary re¯ex (cat's eye), is the most common presenting sign of retinoblastoma. Computerized tomography is most sensitive in demonstrating the presence of the typical deoxyribonucleic acid (DNA)±calcium complexes within the tumor.2,10 Magnetic resonance imaging is not as speci®c as CT scanning in the diagnosis of retinoblastoma due to its lack of sensitivity in detecting calci®cation. However, it is more speci®c than CT in differentiating retinoblastoma from other lesions that simulate retinoblastoma clinically, such as persistent hyperplastic primary vitreous (PHPV), Coats' disease, retinopathy of prematurity (ROP), toxocariasis, retinal detachment, and other so-called pseudogliomas and leukokorias.2,10,12,13 Lesions less than 2±3 mm in thickness are not recognized as discrete areas by MR technology to date. At times lesions less than 3 mm thick may be better visualized on post Gd-DTPA contrast MR scans due to increased contrast resolution. On MRI, retinoblastomas appear slight to moderately hyperintense to vitreous on T1W and slightly or moderately hypointense to vitreous on T2W MR scans. Even extensive calci®cation noted on CT scans may be missed in all MR pulse sequences. Retinoblastomas demonstrate moderate to marked contrast enhancement following i.v. administration of Gd-DTPA contrast material. Involvement of the Tenon's capsule, optic nerve and retrobulbar space is best evaluated on post Gd-DTPA fat-suppression T1W MR images. MRI including Gd-DTPA contrast study appears to be the study of choice for the evaluation of intracranial spread of retinoblastoma and the association of bilateral retinoblastoma, suprasellar and pineal primitive neuroectodermal tumors (tetra lateral retinoblastoma).14 1.7

Persistent Hyperplastic Primary Vitreous (PHPV)

Persistent hyperplastic primary vitreous is caused by failure of the embryonic hyaloid vascular system to regress normally. The ocular malformation can re¯ect either an isolated congeni-

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

tal defect or may be a manifestation of more extensive ocular or systemic involvement.2,15,16 The MRI ®ndings of PHPV include: (1) microphthalmos, (2) lack of calci®cation, (3) posterior hyaloid or retinal detachment, (4) tubular, cylindrical or other discrete intravitreal images suggestive of persistence of fetal tissue, (5) ¯uid level within the globe, re¯ecting the presence of serosanguinous ¯uid in either the subhyaloid or subretinal space, (6) enhancement of abnormal intravitreal tissue following i.v. administration of Gd-DTPA contrast material, (7) altered signal of lens of the involved eye compared with the other normal eye.

1.8 Coats' Disease Coats' disease is a primary vascular anomaly of the retina, characterized by idiopathic retinal telangiectasia and exudative retinal detachment (exudative retinopathy). It is almost always unilateral and generally affects boys of 4 to 6 years of age, although it may be seen in younger children. In the early stage of the disease, MRI may yield little information. In the later stages, the exudative retinal detachment is seen as hyperintense images on all MRI pulse sequences. At times, due to lipoproteinaceous exudate, typically seen in Coats' disease, the subretinal exudate may be less hyperintense than the vitreous body on T2W MR scans. The leaves of the detached retina in Coats' disease may be very thickened and may show moderate to signi®cant enhancement following i.v. administration of GdDTPA contrast material. 1.9 Retinopathy of Prematurity (ROP) Retinopathy of prematurity, formerly termed retrolental ®broplasia, is believed to be related to a combination of developmental (prematurity) and environmental (supplemental oxygen therapy) factors that prevent normal retinal vasculogenesis. The MRI ®ndings of the early stage of ROP are nonspeci®c. The eyes may be microphthalmic. In the more advanced stages the differential diagnoses include PHPV, retinoblastoma, endophthalmitis as well as a number of pathologic conditions in which retinal detachment is a common feature.2,17 Retinoblastoma has rarely been reported in microphthalmic eyes with or without ROP or PHPV.

3

mm in thickness are usually well visualized on MR scans.6,17 Smaller lesions are better studied with ultrasonography. The MR characteristics of melanotic lesions are thought to be related to the paramagnetic proton relaxation by stable radicals in melanin.18 The stable radicals cause a proton relaxation enhancement that shortens both T1 and T2 relaxation times.11 Hence, uveal melanomas appear as areas of moderately to markedly high signal (greater signal intensity than vitreous) on T1W and PW MR images. On T2W images, melanomas appear as areas of moderately to markedly low signal (lesser signal intensity than vitreous). These MR characteristics are similar to those of retinoblastoma. Some of the melanomas may not be hypointense on T2W images. Melanomas demonstrate moderate enhancement following i.v. administration of Gd-DTPA contrast media (Figure 2). Exudative retinal detachment, associated with choroidal melanoma is seen on MRI as dependent areas of moderate to high intensity on T1W, PW and T2W images. Organized exudative chronic retinal detachment and hemorrhagic subretinal ¯uids may simulate melanoma on MRI scans. Invasion of the sclera, extension of tumor into the Tenon's capsule or optic disk and extraocular invasion can be best detected by MR scans, particularly on post Gd-DTPA fat-suppression T1W MR images. 1.12 Choroidal Hemangioma Choroidal hemangiomas are congenital vascular hamartomas. Two different forms have been described: a circumscribed or solitary type and a diffuse angiomatosis.19 Diffuse choroidal hemangiomas are usually seen in association with Sturge± Weber disease. Choroidal hemangiomas can be diagnosed by ophthalmoscopy, ¯uorescein angiography, ultrasonography or CT. However, misdiagnoses are not uncommon, particularly when the angiomas are concealed by a detachment of the retina.2 On MRI, choroidal hemangiomas appear hypointense to slightly hyperintense to vitreous on T1W images and hyperintense on T2W images. They show intense enhancement on post Gd-DTPA T1W MR scans. MRI scans are, therefore, extremely

1.10 Ocular Toxocariasis Toxocariasis results from ingestion of eggs of the nematode, Toxocara canis. The death of the larva results in a wide spectrum of intraocular in¯ammatory reactions. MRI is superior to CT for the evaluation of ocular toxocariasis. The MR ®ndings of ocular toxocariasis consist of localized or diffuse intravitreal mass with moderate to marked contrast enhancement. The subretinal exudate has variable hyperintense signal on T1W, PW, and T2W MR scans.17 1.11 Malignant Uveal Melanoma Malignant uveal melanoma is the most common intraocular tumor in adults. They are uncommon in blacks, the white:black ratio being about 15 : 16 Metastasis primarily involves the liver. Uveal melanomas and other intraocular lesions more than 3

Figure 2 Malignant choroidal melanoma. Axial T1W postcontrast MR image. A hyperintense enhancing mass (solid arrows) is demonstrated. The melanoma has caused retinal detachment (hollow arrow). The tumor can be easily differentiated from subretinal exudate

4 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI useful in differentiating choroidal hemangiomas from choroidal melanomas.

1.13 Uveal Metastases A most common source of secondary carcinoma within the eye is from breast or lung. Metastatic lesions of the uvea extend chie¯y in the plane of the choroid, causing relatively little increase in its thickness. Unlike uveal melanomas, which tend to form a protuberant mass, uveal metastases usually have a mottled appearance and a diffuse outline on MRI scans. Most uveal metastases appear as areas of low signal on T1W MR scans. On T2W images, metastases appear as areas of high signal intensity. A mucin-producing metastatic lesion (adenocarcinoma) may simulate a uveal melanoma because the proteinaceous ¯uid tends to decrease T1 and T2 relaxation times.2 Uveal metastatic carcinoid lesions may simulate uveal melanomas on MRI scans.20 Uveal metastases demonstrate moderate to marked enhancement following i.v. administration of Gd-DTPA contrast material. Exudative retinal detachment associated with uveal metastases can be best differentiated from metastasis on postcontrast T1W images. 1.13.1 Choroidal Hemorrhage and Detachment Choroidal hemorrhage and detachment may be easily mistaken for choroidal tumors.2,8,9 MRI is invaluable in differentiating choroidal detachments from uveal melanomas.8 1.13.2 Other Uveal Tumors Neuro®broma and schwannoma of the choroid and ciliary body, adenoma, leiomyoma, and lymphoreticular proliferative disorders of the uveal tract are rare lesions involving the choroid and ciliary body which can cause diagnostic confusion with uveal melanoma. If these lesions are large enough (3 mm or more) they can be detected by MRI.10

2.2 2.2.1

2.1 Introduction The orbits are two recesses that contain the globes, the extrinsic ocular muscles, the blood vessels, lymphatic, nerves (II, III, IV, and V), adipose and connective tissues, and most of the lacrimal apparatus.1 The orbit is bounded by the periosteum (periorbita) and separated from the globe by Tenon's capsule. Anteriorly are the orbital septum and the lids. The orbital cavity is pyramidal; its apex is directed posteriorly and medially and its base is directed anteriorly and laterally as the orbital opening. Its bony walls separate it from the anterior cranial fossa superiorly, the ethmoid and sphenoid sinuses and nasal cavity medially, the maxillary sinuses inferiorly, and the lateral surface of the face and temporal fossa laterally and posteriorly. At the apex of the orbit, the optic canal and the superior and inferior orbital ®ssures allow various structures to enter and leave the orbit.1,21

Orbital Septum and Eyelids

The orbital septum is a weak, membranous sheet that forms the ®brous layer of the eyelids and is attached to the margins of the bony orbit where it is continuous with the periorbita. Each eyelid or palpebra from without inwards consists of skin, subcutaneous areolar tissue, ®bers of the orbicularis oculi, tarsus, and orbital septum. 2.2.2

Orbital Fatty Reticulum, Extraocular Muscles, and Optic Nerve

Within the orbit all structures are embedded in a fatty reticulum. The ®broelastic tissue that makes up the reticulum divides the fat into lobes and lobules.21 The six striated extraocular muscles, including the four recti and the two oblique muscles, control eye movement. The rectus muscles arise from the annulus of Zinn, which is a tunnel-shaped tendinous ring that encloses the optic foramen and the medial end of the superior orbital ®ssure.22 The optic nerve is a nerve ®ber tract that conveys visual sensation. It traverses the optic canal with the ophthalmic artery. The optic nerve is covered by three layers, the pia mater, the arachnoid, and the dura, all of which sheathe the nerve. 2.2.3

Lacrimal Apparatus

The lacrimal gland lies in the superolateral angle of the orbit in the lacrimal fossa. Tears produced by the gland are drained into the lacrimal sac. The lacrimal sac lies in the lacrimal groove, located in the anterior-inferomedial angle of the orbit. From the sac, the tears drain through the nasolacrimal duct into the inferior meatus of the corresponding nasal cavity.

2.3 2 MRI OF THE ORBIT

Anatomy of the Orbit

MRI Techniques

Excellent natural contrast between various orbital tissues (water, fat, muscles, and nerves) makes the orbit an ideal organ to study by MRI. The three-dimensional capability of MRI along with its superior contrast resolution is of great assistance in localizing orbital pathology. Using a head coil, a routine MRI of the orbit usually includes a single echo short TR, short TE (TR/TE, 500±800/20±25 ms), SE sagittal view, an axial SE double echo (TR/TE, 2000±2500/20±25, 70±100 ms) and a single echo, short TR, short TE, SE axial or coronal view. Following the intravenous i.v. administration of Gd-DTPA, single echo, short TR, short TE images are obtained with or without fat-suppression techniques. Images obtained with a surface coil provide better spatial resolution; however, the apical lesions and intracranial extension of the orbital lesions may not be fully evaluated because of signal dropout, proportional to the distance from the surface coil. We use the surface coil for those lesions situated in the anterior orbit. In general, the orbital MRI study and the use of paramagnetic contrast material should be tailored according to the suspected problem of individual patients.22

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

5

its clear demonstration of the relationship between nasal, sinus, orbital and cranial disease.22 2.5.3

Figure 3 Normal orbital anatomy. Coronal 3 mm thick MR scan, demonstrating various anatomical structures of the orbits, nasal and paranasal sinuses. E = ethmoidal sinus, IR = inferior rectus muscle, IT = inferior turbinate, LPS = levator palbebra superioris, LR = lateral rectus muscle, MA = maxillary antrum, MR = medial rectus muscle, MT = middle turbinate, OA = ophthalmic artery, ON = optic nerve, so = superior oblique muscle, Sov = superior ophthalmic vein, SR = superior rectus muscle

Idiopathic Orbital In¯ammations (Pseudotumors)

Pseudotumors are de®ned as nonspeci®c, idiopathic, orbital in¯ammatory conditions for which no local identi®able cause or systemic disease can be found.22±24 Classically the rapid development of unilateral, painful ophthalmoplegia, proptosis and chemosis, with a rapid and lasting response to steroid therapy in an otherwise healthy patient, is highly suggestive of the diagnosis of pseudotumor. On MRI, pseudotumors usually appear as an in®ltrative process which is isointense or hypointense to muscles on T1W and hyperintense on T2W images. On MRI, the enhancement of pseudotumors with Gd-DTPA is best demonstrated on fat-suppressed T1W images. There are no speci®c imaging characteristics or clinical signs to establish the diagnosis of pseudotumor with absolute certainty. Idiopathic orbital myositis may simulate thyroid myopathy. However, typically, enlargement of the muscles in thyroid myopathy involves the muscle belly, sparing its tendinous portion. 2.6

Orbital Tumors

2.4 MRI Anatomy of the Orbit

2.6.1

MRI provides the best cross sectional imaging anatomy of the orbit. The SE, short TR and short TE pulse sequences provide the most spatial resolution with exquisite anatomical details.22 The extraocular muscles, the intermuscular septae, the optic nerve, superior ophthalmic vein, ophthalmic artery, the lacrimal gland, lacrimal sac, and nasolacrimal duct are very well visualized on routine MR images (Figure 3). Smaller structures such as divisions of the ophthalmic and third cranial nerves, lacrimal vein and artery, and even ciliary vessels can be visualized on high resolution MR images, obtained with a surface coil. The orbital septum, structures of the eyelids including tarsal plates, the annulus of Zinn, and most of the tendinous portion of the extraocular muscles can be best visualized on images obtained with a surface coil.22

Lymphomas, the most common malignant orbital tumors in adults, are solid tumors of the immune system. Most are composed of monoclonal B cells. The most common cytologic forms of malignant lymphoma involving the orbit are histiocytic and lymphocytic in various degrees of differentiation. True lymphoid tissue in the eye is found in the subconjunctival and lacrimal gland. This explains why most orbital lymphomas are commonly seen in the anterior portion of the orbit. The MRI ®ndings of orbital lymphomas are usually nonspeci®c and at times are impossible to differentiate from those of orbital pseudotumors. Both pseudotumors and lymphomas may have an intermediate or hypointense signal in T1W images (Figure 4) and appear isointense or hyperintense and even hypointense (although less common) to fat in T2W images (Figure 4). Both pseudotumors and lymphomas demonstrate moderate contrast enhancement on postcontrast MRI scans.

2.5 MRI of Orbital Pathology 2.5.1

Developmental Orbital Cysts

MRI is particularly helpful in the evaluation of congenital anomalies of the orbit and optic system. The most frequent developmental cysts involving the orbit and periorbital structures are epidermoid and dermoid cysts and teratomas. Epidermoid and dermoid cysts appear hypointense to brain on T1W and hyperintense on T2W MR images. Portions of the dermoids which contain fatty tissue appear hyperintense on T1W and hypointense on T2W images. Epidermoid and dermoid cysts do not show enhancement with Gd-DTPA. 2.5.2

In¯ammatory Conditions

Preseptal and postseptal orbital in¯ammation, subperiosteal phlegmon and abscess, ophthalmic vein and cavernous sinus thrombosis can be demonstrated by MRI. The main contribution of MRI to the diagnosis of sinonaso-orbital infections is

2.6.2

Lymphoma

Rhabdomyosarcoma

Rhabdomyosarcoma is the most common malignant mesenchymal tumor of childhood, as well as the most common primary malignant tumor of the orbit in children.22,25 The tumor is notorious for producing a rapidly developing unilateral proptosis. Any rapidly developing proptosis in childhood must presumptively be diagnosed as rhabdomyosarcoma. The differential diagnosis also includes leukemia and metastatic deposits (neuroblastoma), Langerhans histiocytosis, ruptured dermoid cyst, hemorrhage in a pre-existing lymphangioma, subperiosteal infection and hematoma after trauma. Rhabdomyosarcomas are both aggressive bone-destroying lesions and bone-pushing lesions. Rhabdomyosarcoma appears hypointense to brain on T1W and isointense to slightly hyperintense to brain on T2W images. The use of paramagnetic contrast media results in moderate to marked enhancement.25 Magnetic resonance imaging is the most sensitive imaging study to differentiate rhabdomyosarcoma from most simulating lesions.22,25

6 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI middle-aged and elderly women. ONSM presents as very slowly progressive axial proptosis and loss of vision. On MRI scans, ONSM can be seen as localized or fusiform enlargement of the optic nerve. ONSM frequently appears hypointense on T1W and hyperintense on T2W images as compared with the brain. ONSM demonstrates moderate to marked enhancement on post Gd-DTPA MR scans.22 Postcontrast fat-suppressed T1W MR scans are the most sensitive images to detect ONSM. Involvement of the optic nerve by sarcoidosis may result in focal or diffuse enlargement of the optic nerve, mimicking an optic nerve meningioma on all MR pulse sequences. 2.6.5

Figure 4 Small cell lymphoma of the orbit. Coronal T1W (top) and axial T2W (bottom) MR scans demonstrating normal right lacrimal gland (LG) and abnormal left lacrimal gland (hollow arrows) due to involvement by lymphoma. Notice lymphomatous involvement of left superior rectus muscle (SR) as well as extraconal fat (curved arrow)

2.6.3

Hemangiomas and Lymphangiomas

Capillary hemangiomas are the most common orbital vascular tumors that occur in infants during the ®rst year of life. The tumor often increases in size for 6 to 10 months and then gradually involutes. On MRI, capillary hemangiomas have long T1 and long T2 characteristics. At times, prominent vessels can be identi®ed within the lesion. They demonstrate marked contrast enhancement. On MR angiography, the vascular blush may not be appreciated. Cavernous hemangiomas, the most common orbital vascular tumor in adults, appear as well-de®ned masses, frequently occurring within the intraconal space. On MRI, they appear as well-de®ned, sharply marginated, homogeneous, rounded, ovoid, or lobulated masses, which appear hypointense to brain on T1W and hyperintense on T2W images. They demonstrate moderate to marked contrast enhancement on MRI. The MRI characteristics of orbital schwannomas and hemangiopericytomas may be similar to cavernous hemangiomas.22,26 Orbital lymphangiomas occur in children and young adults. Lymphangiomas may have distinct borders but are typically diffuse and not well capsulated. Spontaneous hemorrhage within the lesion is common.22,26 On MRI scans lymphangiomas are seen as homogeneous or heterogeneous, and hypointense to relatively hyperintense on T1W and, usually, heterogeneously very hyperintense on T2W scans. Their MRI characteristics usually help to differentiate them from hemangiomas, pseudotumors, rhabdomyosarcoma and many other lesions.22,26 2.6.4

Optic Nerve Sheath Meningioma

Optic nerve sheath meningioma (ONSM) arises from the meningoendothelial cells of the arachnoid that are situated along the optic nerve sheath. Meningiomas are usually seen in

Optic Nerve Glioma

Optic nerve glioma is a tumor arising from the neuroglia (glial cells) of the optic nerve. It is usually a tumor of childhood (2 to 8 years of age). In general, optic nerve glioma in children is considered a benign, well-differentiated, slowly growing, and noninvasive astrocytoma (pilocystic astrocytoma). However, optic nerve glioma in adults is a rare but invasive malignant astrocytoma (usually glioblastoma multiformis). Bilateral optic nerve gliomas are characteristic of neuro®bromatosis type 1 (NF1). Optic nerve gliomas are seen on MRI as a well-de®ned, fusiform enlargement of the optic nerve. Diffuse, tortuous enlargement of the optic nerve with a characteristic kinking and buckling appearance is a characteristic feature of the childhood form of optic glioma. On T1W MR scans, gliomas appear isointense or slightly hyperintense compared with the white matter. On T2W images they appear hyperintense compared with the white matter.22 The use of paramagnetic contrast media results in moderate to marked enhancement. MRI remains the study of choice for the evaluation of optic gliomas. Intracranial extension of optic nerve gliomas and associated intracranial pathological changes in patients with neuro®bromatosis can be best appreciated on MRI scans.22

2.7

Optic Neuritis

Optic neuritis is an acute in¯ammatory process involving the optic nerve. The process may be idiopathic, immunemediated, metabolic, and infective in nature.27,28 Multiple sclerosis is the most common cause of optic neuritis;28,29 visual loss is typically unilateral in patients with demyelinative optic neuritis. A recent study by Beck et al.29 has shown that intravenous methylprednisolone followed by oral prednisone will speed the recovery of visual loss due to optic neuritis. On MR, the optic nerve in patients with optic neuritis may infrequently appear diffusely enlarged. On T2W MR scans, the involved optic nerve may be slightly or moderately more hyperintense than the one on the normal side. Most patients with acute idiopathic optic neuritis and with no other neurologic disorder have the same areas of high-intensity signal in the cerebral white matter on T2W MR scans that are found in patients with multiple sclerosis. In patients with optic neuritis, contrast enhancement on MRI is often subtle or present in a short segment of the optic nerve. It is best demonstrated by comparing pre- and post-contrast T1W MR images or on postcontrast fatsuppresed T1W MR scans.22

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

7

3 MRI OF THE EAR

tion is believed to be the degradation and absorption of the endolymph, produced in the sensory labyrinth, i.e. the cochlea and vestibule. All of the structures of the membranous labyrinth are enclosed within hollowed-out bony cavities that are considerably larger than their membranous contents. These bony cavities assume the same shape as the membranous chambers and are referred to as the osseous labyrinth. The bony cavities of the osseous labyrinth contain ¯uid, known as perilymph, that bathes the external surface of the membranous labyrinth. The sound vibrations travel through the middle ear and then through the ossicular chain to the oval window. This in turn sets the perilymph in the cochlear labyrinth in motion. Stimulated by the ¯uid motion, sensory receptors (hair cells) in the cochlea generate nerve impulses that are transmitted by the cochlear nerve and auditory pathway to the auditory center in the temporal lobe of the brain. Impulses generated in the sensory cells of the semicircular canals and vestibule are carried to the brain by the vestibular nerve. The essential function of the vestibular labyrinth is to provide the central nervous system (CNS) with a constant ¯ow of information concerning the static position of the head in space, or its state of linear or angular acceleration or deceleration.32,33

3.1 Introduction: Anatomy of the Ear

3.2 Acoustic and Facial Nerves

The ear is essentially a distance receptor concerned with the collection, conduction, modi®cation, ampli®cation and parametric analysis of the complex soundwaves which impinge on the head. The ear is subdivided into three major areas: the outer ear, the middle ear, and the inner ear. Each region has a speci®c function. The outer ear channels sound to the tympanic membrane (ear drum). The middle ear is an air-®lled space (tympanic cavity), connected via the eustachian tube to the nasopharynx. Lying across the tympanic cavity are three small bones; the malleus, the incus and the stapes. The stapes footplate touches a membrane called the oval window, located between the middle ear and the inner ear.32 A second membrane-covered opening called the round window is located in the cochlea below the oval window. This connects the inner ear with the middle ear and maintains a constant (perilymphatic ¯uid) pressure within the cochlea. The essential function of the tympanic cavity and its ossicles is the ef®cient transfer of energy from the air to the perilymphatic ¯uids which surround the cochlea.32 The inner ear contains sensory receptors for hearing and balance. It consists of three main parts: the cochlea, the vestibule, and the semicircular canals. The inner ear is composed of the membranous labyrinth and the osseous labyrinth. The membranous labyrinth has two major subdivisions, a sensory portion called the sensory labyrinth and a nonsensory portion designated as the nonsensory labyrinth. The sensory labyrinth lies within the petrous portion of the temporal bone. It contains two intercommunicating portions: (1) the cochlear labyrinth that consists of the cochlea and is concerned with hearing, and (2) the vestibular labyrinth that contains the three semicircular canals, the saccule and utricle, (two small sacs), occupying the vestibule, all of which are concerned with equilibrium. These hollow chambers are ®lled with ¯uid, known as endolymph, that resembles intracellular ¯uid. The nonsensory element of the membranous labyrinth is formed by the endolymphatic duct and sac, whose main func-

Acoustic nerve consists of the cochlear nerve and the vestibular nerves (superior and inferior). These are joined into a common trunk that enters the internal auditory canal with the facial nerve. The facial nerve then travels within the temporal bone for some 33 mm through a tortuous bony canal, known as the fallopian canal, before it enters the parotid gland.

2.8 Lacrimal Gland Tumors Epithelial tumors represent approximately 50% of masses involving the lacrimal gland. Half of these are pleomorphic (benign mixed) adenomas; the other half are malignant.30 Of the malignant tumors, adenoid cystic (adenocystic) are the most common.31 The nonepithelial masses of lacrimal gland fossa include lymphoid-in¯ammatory lesions, that is, benign dacryoadenitis, pseudotumor, and malignant lymphoma.31 MRI is the study of choice for the evaluation of lacrimal gland tumors. Benign, mixed tumors appear as well-de®ned encapsulated masses and have long T1 and long T2 MR signal characteristics. The tumor may be homogeneous or heterogeneous. The use of paramagnetic contrast material results in moderate to marked enhancement. The heterogeneity within the tumor may be better appreciated on T2W and postcontrast T1W MR images. Malignant lacrimal gland tumors may show poor margins and an in®ltrative character. Invasion of the surrounding tissues and intracranial extension can be best evaluated on MRI scans.22,31

3.3 MRI Anatomy of the Ear MRI remains the study of choice for imaging evaluation of the membranous labyrinth, and the vestibulo-cochlear and facial nerves. The osseous labyrinth, cortical bones, and the air ®lled mastoid cells appear hypointense in all pulse sequences. The endolymphatic and perilymphatic ¯uid within and surrounding the membranous labyrinth provides MR signal, resulting in visualization of the membranous labyrinth on various MR pulse sequences (Figure 5). MRI remains the study of choice for the evaluation of the membranous labyrinth, and the vestibulo-cochlear and facial nerves. Cochlear and vestibular divisions of the acoustic nerve and the entire course of facial nerve including its parotid segment can be visualized with exquisite details on MR scans. 3.4 Pathology using MRI of the Ear The main contribution of MRI to the diagnosis of ear disorders is its clear demonstration of the membranous labyrinth, cochlear, vestibular, and facial nerves. The use of paramagnetic contrast material enhances the sensitivity and at times speci®city of MRI in the evaluation of certain pathological conditions. Gd-DTPA contrast has been shown to play an important role in the detection of in¯ammatory (Figure 6), autoimmune, and neoplastic diseases of the membranous labyrinth as well as soft tissue tumors and in¯ammatory and in®ltrative processes of the ear.34,35

8 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

Figure 5 Normal anatomy of the ear: 3 mm axial proton-weighted (top) and T2W (bottom) MR scans. The cochlea (c), vestibule (V), lateral semicircular canal (L), posterior semicircular canal (P), endolympathic sac (ES), cochlear nerve (CN) and inferior vestibular nerve (vn) are shown

3.5 Developmental Anomalies of the Ear Computerized tomography scanning remains the study of choice for the evaluation of ear anomalies. MRI provides little information in the evaluation of patients with anomalies of the external and middle ears. In these patients MRI may be used to evaluate the facial nerve. Inner ear anomalies such as Mondini anomaly, large vestibular aqueduct, and Michel anomaly are best evaluated on CT scans. In Mondini anomaly, MRI reveals ¯uid-®lled cystic dilatation of the membranous labyrinth. In patients with large vestibular aqueduct, MRI reveals ¯uid-®lled cystic dilatation of the endolymphatic duct and sac.33 3.6 In¯ammatory Conditions of the Ear CT remains the study of choice for the evaluation of acute and chronic otomastoiditis and their complications, including acquired cholesteatomas.34 The main contribution of MRI to the evaluation of the in¯ammatory conditions of the ear is its clear demonstration of the involvement of membranous labyrinth and associated intracranial complications (Figure 6). T1weighted MRI sequences obtained following the use of paramagnetic contrast material are most informative, as they readily demonstrate abnormal intralabyrinthine enhancement (Figure 6). In patients with vestibulocochlear or facial neuronitis, MRI may demonstrate moderate to marked contrast enhancement of the involved nerve (Figure 7).

Figure 6 Acute otomastoiditis and labyrinthitis: 3 mm axial post-GdDTPA T1W MR scans. There is marked enhancement of granulation tissues within the mastoid air cells (M). Notice marked enhancement of left membranous cochlea (white, short and long arrows). There is slight enhancement of the right cochlea (black arrow); patient had bilateral acute otomastoiditis, left more so than right

lesterol crystals and other blood byproducts. Cholesteatomas and cholesterol granulomas of the petrous apex are the most common expansile, slowly destructive, lesions of the petrous bone.34 On CT scans they are very dif®cult to differentiate from each other. However, on MRI, cholesteatomas have a long T1 and long T2 characteristics, whereas, cholesterol granulomas have a short T1 and long T2 characteristic. Both lesions if not infected demonstrate no contrast enhancement. CT scanning is superior to MRI for the detection of middle ear cholesteatomas.34 MRI however, is superior to CT scanning for the evaluation of infected cholesteatomas, petrous apex and intradural cerebellopontine (CPA) cholesteatomas (epidermoids), as well as for the evaluation of cholesteatomatous involvement of the facial nerve, membranous labyrinth, and intracranial structures.

3.7 Cholesteatomas and Cholesterol Granulomas Cholesteatomas of the ear may be congenital or acquired. Histologically the pathology of congenital cholesteatomas is similar to that of the acquired varieties, namely, a cavity, lined with keratinizing strati®ed squamous epithelium which encapsulates desquamated epithelial cells, keratomatous materials, and minimal cholestrin. Cholesterol granulomas of the ear are believed to be a tissue response to hemorrhage and an irritant foreign body, i.e. cho-

Figure 7 Ramsey hunt syndrome: 3 mm axial post-Gd-DTPA T1W MR scan. There is marked enhancement of the meatal segment (arrow head) and tympanic segment (solid arrows) of the facial nerve. Note enhancement of the superior vestibular nerve (hollow arrow)

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

3.8 Tumors of the Ear The diagnostic approach to evaluating a patient suspected of having a lesion of acoustic and facial nerves or a CPA tumor has changed as the resolution of MRI has improved and as the sensitivity and speci®city of MRI has been established. MRI remains the initial imaging study of choice for the evaluation of acoustic and facial neuromas, and most CPA tumors. The most common tumors of the temporal bone are Schwann-cell tumors that arise from the vestibular divisions of the acoustic nerve; as such, they are actually vestibular schwannomas. Bilateral vestibular schwannomas are characteristic of neuro®bromatosis type 2 (NF2). Vestibular schwannomas are the most common CPA masses.36 Meningiomas are the second and epidermoid cysts are the third most common CPA masses. Magnetic resonance imaging is the most useful diagnostic test to diagnose as well as to differentiate these CPA masses.37 The use of paramagnetic contrast material signi®cantly increases the sensitivity of MRI. Tumors as small as 2 mm in size can be detected along the course of facial and acoustic nerves. The use of paramagnetic contrast has allowed for the ®rst time the detection of tiny intralabyrinthine schwannomas,38 and in¯ammatory processes of the facial nerve which were extremely dif®cult to diagnose even with high-resolution CT scan.39±41 Paragangliomas or chemodectomas (glomus tumors) are the second most common tumors of the temporal bone.42 Paragangliomas appear as regions of hypointensity on T1W and hyperintensity on T2W MR scans.42 There are frequently areas of signal void related to rapid arterial ¯ow present in the matrix of these hypervascular tumors. Paragangliomas demonstrate signi®cant enhancement following intravenous (i.v.) administration of paramagnetic contrast material.42 Other benign and malignant tumors that involve the ear, including adenoma, carcinomas, adenocarcinomas, papillary adenocarcinomas, adenoid cystic carcinomas, rhabdomyosarcomas, chondrosarcomas, osteosarcomas, osteoclastomas, chordomas, and other mesenchymal tumors can be adequately evaluated by MRI.

the superior turbinate from the anterior surface of the body of the sphenoid, through which each sphenoidal sinus opens into the nasal cavity.43 The nasal mucous membrane lines the nasal cavity. It is continuous with the mucous membrane of the nasopharynx and paranasal sinuses. The paranasal sinuses are air-®lled cavities and include the frontal, ethmoidal, sphenoidal and maxillary sinuses. They are lined with upper respiratory mucosa, continuous with that of the nasal cavity.43 The frontal sinuses are rudimentary or absent at birth, they are usually fairly well developed by the seventh or eighth years, and they reach their full size after puberty. The ethmoidal sinuses are small, but of clinical importance, at birth; they grow rapidly between the sixth and eighth years and after puberty.43 The sphenoidal sinuses are present at birth, as minute cavities within the body of the sphenoid bone, but their main development takes place between the third and tenth years and after puberty. The maxillary sinuses are present as minute cavities at birth, but do not, however, reach full size until after the eruption of all the permanent teeth. The natural opening of the maxillary sinus is high above its ¯oor and communicates with the lower part of the hiatus semilunaris to be drained into the middle meatus. 4.2

4.1 Introduction The nasal cavity, the ®rst of the respiratory passages, extends from the roof of the mouth upwards to the base of the skull. It is divided by the nasal septum into two halves which open on the face through the nostrils, and communicates behind with the nasopharynx.43 The lateral wall of the nasal cavity is irregular, owing to the presence of three bony projections, the inferior, middle, and superior nasal conchae (turbinates). The conchae project downward and slightly medially and divide the passageway into meatuses, or channels of air. The inferior meatus receives the ori®ce of the nasolacrimal duct. The middle meatus receives the opening of the ipsilateral anterior ethmoidal air cells, frontonasal duct of the frontal sinus and the ostium of the maxillary sinus. The superior meatus receives the opening of the ipsilateral posterior ethmoidal air cells. A narrow interval, the spheno-ethmoidal recess, separates

MRI Anatomy of Nasal and Paranasal Sinuses

The nasal cavity and paranasal sinuses can be visualized by MRI, using head coil or surface coil. Images obtained with head coil are preferred because of the size and deep location of posterior sinonasal cavities. The cortical bone and air of the sinonasal structures do not provide MR signal, and therefore they appear hypointense in all pulse sequences. The mucosa of nasal structures, however, appear as intermediate signal on T1W and increased signal (hyperintense) on T2W images. The normal mucosa of the paranasal sinuses are very thin and cannot be visualized on MR scans. 4.3 4.3.1

4 MRI OF THE NASAL CAVITY AND PARANASAL SINUSES

9

MRI of Sinonasal Pathology Congenital Anomalies

Magnetic resonance imaging remains the study of choice for the evaluation of congenital sinonasal anomalies such as sphenoidal, sphenoethmoidal, ethmoidal, frontonasal, and intraorbital encephaloceles. Magnetic resonance imaging is very useful for the evaluation of intranasal±extranasal dermoids and nasal gliomas. 4.3.2

In¯ammatory Conditions

Standard sinus roentgenograms, CT and, more recently, MRI are currently the most used imaging methods to assess paranasal pathology.44 Computerized tomography remains the imaging study of choice for appropriate evaluation of patients with acute sinusitis associated with complications such as subperiosteal or orbital abscesses and for patients with chronic sinusitis.44,45 Magnetic resonance imaging is also very sensitive in the evaluation of acute sinonasal pathology. The main contribution of MRI to the diagnosis of sinonasal-orbital infections is its clear demonstration of the relationship between nasal, sinus, orbital and cranial diseases. Orbital cellulitis, subperiosteal phlegmon and abscess can be readily visualized on MR scans. Magnetic resonance imaging is superior to CT in

10 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI demonstrating sinogenic complications such as thrombosis of the superior ophthalmic vein, thrombosis of the cavernous sinus, and intracranial complications such as epidural abscess, focal or diffuse encephalitis and brain abscesses. 4.3.3

Tumors

Magnetic resonance imaging provides excellent delineation of tumor from surrounding soft tissue, in¯ammatory tissue, and retained secretions within the sinuses.46±52 It might be anticipated that the typical appearance of edema or retained secretions within the sinuses on MR images would be of low intensity on T1W and high intensity on T2W images, re¯ecting the high water content associated with the excessive interstitial ¯uid or retained secretions. However, because of the frequent chronic nature of these benign processes in patients who have a nasal cavity or paranasal sinus tumor, especially in those patients in whom an advanced tumor has been diagnosed, suf®cient time has elapsed to allow for the concentration of high water af®nity mucoproteins and the absorption of free water. This leads to various degrees of shortening of both T1 and T2 relaxation times.45 The areas of complete desiccation appear as low signal on T1W, PW, and T2W MR images. Sinonasal carcinomas are highly cellular tumors with little free water; as such, they appear as low to intermediate signal intensity on T1- and T2W images (Figure 8). In¯ammatory diseases have intermediate signal intensity on T1W and high signal intensity on T2W images [Figure 8(b)]. Benign pathological processes within the nasal cavity and paranasal sinuses, such as polyps, papillomas, low grade adenocarcinomas, (minor salivary gland tumors), and schwannomas have higher signal intensity than carcinomas on T2W images (Figure 9). The use of paramagnetic contrast material often offers additional information.52 In most instances, Gd-DTPA contrast enhancement of the tumor is less than the enhancement of the normal mucosa and associated in¯ammatory changes. Gd-DTPA is extremely helpful in detecting intracranial extension of a sinonasal tumor. Magnetic resonance imaging has proved valuable in the diagnosis of most other tumors of the sinonasal cavity such as rhabdomyosarcomas, lymphomas, angiomas, angio®bromas, hemangiopericytomas, chondrosarcomas, osteogenic sarcomas malignant melanomas, and metastases.45

Figure 8 (a) Sinonasal squamous cell carcinoma. Sagittal T1W MR image. A hypointense mass (M) is demonstrated, involving the nasal cavity and ethmoidal sinus. The tumor has caused obstruction of the ostium of the sphenoid sinus, resulting in hyperintense retained secretions within the adjacent sphenoid sinus (S). (b) Coronal T2W MR images. A relatively hypointense mass (M) is demonstrated, compared with hyperintense retained secretions in the ethmoidal sinuses (E)

5.2 5 MRI OF THE NECK 5.1 Introduction: Anatomy of the Neck The neck is the junctional region between the head and thorax and upper limbs. Its rostral limits are the base of the skull posteriorly and the inferior border of the mandible anteriorly. Caudally it is bounded by the thoracic inlet and pectoral girdle (clavicle, manubrium of sternum, acromion and spine of the scapula), a plane which parallels the ®rst rib. The neck contains the cervical vertebrae and muscles, the pharynx, the larynx, cervical esophagus, cervical trachea, thyroid gland, salivary glands lymph nodes, and vessels and nerves that either supply the various organs of the neck or are in transit between the head, thorax, and upper limbs.53

MRI Techniques

Our protocol for the upper neck (above the angle of the mandible) uses a head coil and a neck coil for the lower neck. The nasopharynx, parapharyngeal space, infratemporal fossa, masticatory space, oropharynx, tongue, and salivary glands are ideally suited for evaluation by head coil. The larynx is best evaluated by using a surface coil. Thyroid gland and parathyroid glands and lower neck may be evaluated by both head and surface coils. A routine MRI of the neck should include at least a sagittal T1W localizer and axial SE double echo pulse sequences. At times, additional coronal T1W or T2W images may be obtained. The use of Gd-DTPA contrast material, gradient echo images, and fat-suppression pulse sequences should be tailored according to the clinical information and initial evaluation of the sagittal T1W and axial SE double echo pulse sequences.

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI

5.3

11

MRI Anatomy of the Neck

Magnetic resonance imaging offers the best soft tissue contrast resolution and therefore remains the study of choice for imaging the anatomy of the soft tissue of the neck.54 The larynx, pharynx, tongue, ¯oor of the mouth, salivary glands, thyroid gland, parapharyngeal space, infratemporal fossa, masticatory space, and vascular structures of the neck can be visualized with exquisite anatomical detail.54±56 5.4 5.4.1

MRI Pathology of the Neck Developmental Neck Cysts

Magnetic resonance imaging is extremely helpful in the evaluation of congenital anomalies of the neck.55 The most frequent developmental conditions involving the neck are brachial cleft anomalies, cystic hygromas (lymphangiomas), dermoids and epidermoids, and teratomas. Most of these lesions have characteristic MRI ®ndings that should help the radiologists make the diagnosis.55 Figure 9 Frontal hyperintense polyp aerated frontal sinus. is greater than that Figure 8(b)

sinus polyp. Coronal T2W MR image. A (arrows) is demonstrated within an otherwise The signal intensity of the polyp (benign process) of the carcinoma (malignant process), seen in

5.4.2

In¯ammatory Conditions

The main contribution of MRI to the diagnosis of neck infections is its clear demonstration of the relationship between various potential spaces within the neck and adjacent structures

Figure 10 (a) Chronic laryngeal and perilaryngeal infection: 5 mm axial post-Gd-DTPA T1W MR scans. A large enhancing in¯ammatory induration (arrows) is demonstrated, associated with marked thickening of the left aryepiglottic fold (AE), and an area of ¯uid loculation, proved to be pus (P), in the paralaryngeal space. Notice the normal right thyroid cartilage (arrowhead) and partial destruction on the left side. c = common carotid, E = external jugular vein, IJ = internal jugular vein, P = pus, SM = sternocleidomastoid muscle. (b) Coronal post-Gd-DTPA MR scan (4 mm). This coronal MR image demonstrates the caudal and cephalad extension of laryngeal and paralaryngeal in¯ammation (arrows). Notice marked thickening of the left aryepiglottic fold (AE), lateral displacement of the left internal jugular vein (IJ), and bilateral laryngoceles (L). LV = laryngeal vestibule, MA = masseter muscle, MP = medial pterygoid muscle, SG = submandibular gland, SM = sternocleidomastoid muscle, T = palatine tonsil, U = uvula

12 EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI including base of the skull and intracranial structures. Soft tissue induration, edema, and ¯uid collections (pus) can be readily demonstrated on MRI (Figure 10). Magnetic resonance imaging is highly sensitive for the detection of early and late in¯ammatory changes of the bone marrow of the bony structures of the head and neck as well as the temporo-mandibular joint. 5.5 Neck Tumors Magnetic resonance imaging is capable of depicting most benign and malignant tumors of the neck. On MRI, most tumors have long T1 and T2 characteristics.54±57 The T2W images are most valuable for detecting differences in signal intensity of the pathological process.54±60 Magnetic resonance imaging remains the study of choice for imaging evaluation of nasopharyngeal, and parapharyngeal, salivary gland tumors. In most patients with laryngeal tumor, MRI appears to be more appropriate than a computerized tomography (CT) scan for evaluating the extent of laryngeal tumor.57 Involvement of the pre-epiglottic space, paralaryngeal space, postcricoid region, as well as cartilage invasion by laryngeal tumor can be readily visualized on MR scans.57 Magnetic resonance imaging is as sensitive as CT for demonstrating nodal (lymph) diseases of the neck.59 Gd-DTPA contrast studies have been shown to play an important part in the detection and diagnosis of certain lesions of the neck.61

6 RELATED ARTICLES Cranial Nerves Investigated by MRI; Head and Neck Investigations by MRI; Hemorrhage in the Brain and Neck Observed by MRI; Temperomandibular Joint MRI.

7 REFERENCES 1. R. Warwick and P. L. Williams, eds. `Gray's Anatomy', 35th edn. (British), Saunders, Philadelphia, 1973. 2. M. F. Mafee, in `Radiology of the Eye and Orbit', eds. T. H. Newton and L. T. Bilaniuk, Raven Press, New York, 1990 p. 2. 1±23.1. 3. M. F. Reeh, J. L. Wobij and J. D. Wirtschafter, `Ophthalmic Anatomy: A Manual With Some Clinical Applications', American Academy of Ophthlmology, San Francisco, 1981, p. 11. 4. J. B. Aguayo, B. Glaser, A. Mildvan, H. M. Cheng, R. G. Gonzalez, and T. Brady, Invest. Ophthalmol. Visual Sci., 1985, 26, 692. 5. B. J. terPenning, H. M. Cheng, P. Barnett, J. Seddon, D. Sang, M. Latina, J. B. Aguayo, R. G. Gonzalez and T. Brady, J. Comput. Assist. Tomogr., 1986, 10, 551. 6. M. F. Mafee, G. A. Peyman, J. E. Grisolano, M. E. Fletcher, D. G. Spigos, F. W. Wehrli, F. Rasouli, and V. Capek, Radiology, 1986, 160, 773. 7. M. F. Mafee, G. A. Peyman, J. H. Peace, S. B. Cohen, and M. W. Mitchell, Ophthalmology, 1987, 94, 341. 8. M. F. Mafee, B. Linder, G. A. Peyman, B. G. Langer, K. H. Choi, and V. Capek, Radiology, 1988, 168, 781. 9. M. F. Mafee and G. A. Peyman, Radiol. Clin. North Am., 1987, 25, 487. 10. M. F. Mafee, M. F. Goldberg, S. B. Cohen, E. D. Gotsis, M. Safran, L. Chekuri, and B. Rao®, Ophthalmology, 1989, 96, 965.

11. J. M. Gomori, R. I. Grossman, J. A. Shields, J. J. Augsburger, P. M. Joseph, and D. DeSimeone, Radiology, 1986, 158, 443. 12. R. G. Peyster, J. J. Augsburger, J. A. Shields, B. L. Hershey, R. Eagle, and M. E. Haskin, Radiology, 1988, 168, 773. 13. B. G. Haik, L. Saint Louis, M. E. Smith, R. M. Ellsworth, D. H. Abramson, P. Cahill, M. Deck, and J. Coleman, Ophthalmology, 1985, 92, 1143. 14. M. F. Mafee, in, `Head and Neck Disorders (Fourth Series) Test and Syllabus', ed. P. M. Som, American College of Radiology, Reston, VA, 1992, p. 71. 15. M. F. Mafee, M. F. Goldberg, G. E. Valvassori, and V. Capek. Radiology, 1982, 145, 713. 16. M. F. Mafee and M. F. Goldberg, Radiol. Clin. North Am., 1987, 25, 683. 17. M. F. Mafee, Radiol. Clin. North Am. 1998, 36, 1083. 18. R. Damadian, K. Zaner, D. Hor, and T. Dimaio, Physiol. Chem. Phys. 1973, 5, 381. 19. M. F. Mafee, D. J. Ainbinder, A. A. Hidayat, and S. M. Friedman, Int. J. Neuroradiol. 1995, 1, 67. 20. R. G. Peyster, M. D. Shapiro, and B. G. Haik, Radiol. Clin. North Am. 1987, 25, 647. 20. R. Warwick and P. L. Williams, eds. `Gray's Anatomy', 35th edn. (British), Saunders, Philadelphia, 1973, p. 262±264. 21. M. F. Reeh, J. L. Wobig, and J. D. Wirtschafter, `Ophthalmic Anatomy: A Manual with Some Clinical Applications', American Academy of Ophthalmology, San Francisco, 1981, p. 11. 22. M. F. Mafee, in `Imaging of the head and neck', eds. G. E. Valvassori, M. F. Mafee, and B. L. Carter, Thieme, Stuttgart, 1995, p. 158. 23. F. C. Blodi and J. D. Gass, Trans. Am. Acad. Opthalmol. Otolaryngol., 1967, 71, 303. 24. A. E. Flanders, M. F. Mafee, and V. M. Rao, J. Comput. Assist. Tomogr, 1989, 13, 40. 25. M. F. Mafee, E. Pai, and B. Philip, Radiol. Clin. North Am., 1998, 36, 215. 26. M. F. Mafee, A. Putterman, G. E. Valvassori, M. Campos, and V. Capek, Radiol. Clin. North Am., 1987, 25, 529. 27. S. Lessell, N. Engl. J. Med., 1992, 329, 634. 28. G. C. Ebers, Arch. Neurol., 1985, 42, 702. 29. R. W. Beck, P. A. Cleary, M. M. Anderson, J. L. Keltner, W. T. Shults, D. I. Kaufman, E. G. Buckley, J. J. Corbett, M. J. Kupersmith, and N. R. Miller, N. Engl. J. Med., 1992, 326, 581. 30. M. F. Mafee and B. G. Haik, Radiol. Clin. North Am., 1987, 25, 767. 31. M. F. Mafee, D. P. Edward, K. Keller, and S. Darodi, Radiol. Clin. North Am., 1998, 37, 219. 32. R. Warwick and P. L. Williams, eds. `Gray's Anatomy', 35th edn. (British), Saunders, Philadelphia, 1973, p. 1134. 33. M. F. Mafee, D. Charletta, A. Kumar, and H. Belmont, Am. J. Neuroradiol., 1992, 13, 805. 34. M. F. Mafee, J. Otolaryngol., 1993, 22, 240. 35. A. S. Mark and D. Fitzgerald, Am. J. Neuroradiol., 1993, 14, 991. 36. M. F. Mafee, Otolaryngol. Clin. North Am., 1995, 28, 407. 37. G. E. Valvassori, Otolaryngol. Clin. North Am., 1988, 21, 337. 38. M. Brogan and D. W. Chakeres, Am. J. Neuroradiol., 1990, 11, 407. 39. M. F. Mafee, G. E. Valvassori, A. Kumar, et al. Otolaryngol. Clin. North Am., 1988, 21, 349. 40. M. F. Mafee, C. S. Lachenauer, A. Kumar, P. M. Arnold, R. A. Buckingham, and G. E. Valvassori, Radiology, 1990, 174, 395. 41. D. L. Daniels, L. F. Czervionke, S. J. Millen, T. J. Haberkamp, G. A. Meyer, L. E. Hendrix, L. P. Mark, A. L. Williams, and V. M. Haughton. Radiology, 1989, 171, 807. 42. P. D. Phelps, Clin. Radiol., 1990, 41, 301. 43. eds. R. Warwick and P. L. Williams, `Gray's Anatomy', 35th edn. (British), Saunders, Philadelphia, 1973, p. 1087. 44. M. F. Mafee, J. Am. Med. Assoc., 1993, 269, 2608.

EYE, ORBIT, EAR, NOSE, AND THROAT STUDIES USING MRI 45. P. M. Som and H. D. Curtin, Radiol. Clin. North Am., 1993, 31, 33. 46. J. M. Chow, J. P. Leonetti, and M. F. Mafee, Radiol. Clin. North Am., 1993, 31, 61. 47. M. F. Mafee, Radiol. Clin. North Am., 1993, 31, 75. 48. P. M. Som, M. D. Shapiro, H. F. Biller, C. Sasaki, and W. Lawson, Radiology, 1988, 167, 803. 49. G. A. S. Lloyd, V. J. Lund, P. O. Phelps, and D. J. Howard, Br. J. Radiol., 1987, 60, 957. 50. M. G. M. Hunink, R. G. M. de Slegte, G. J. Gerritsen, and H. Speelman, Neuroradiology, 1990, 32, 220. 51. P. van Tassell and Y. Y. Lee, J. Comput. Assist. Tomogr., 1991, 15, 387. 52. C. F. Lanzieri, M. Shah, D. Krauss, and P. Lavertu, Radiology, 1991, 178, 425. 53. J. A. Gosling, P. F. Harris, and J. R. Humpherson, eds. `Atlas of Human Anatomy with Integrated Text', Lippincott, Philadelphia, 1985. 54. M. F. Mafee, F. Rasouli, D. G. Spigos, G. E. Valvassori, H. Friedman, and M. Capek, Otolaryngol. Clin. North Am., 1986, 19, 523. 55. M. F. Mafee, in `Head and Neck Imaging', eds. G. E. Valvassori, R. A. Buckingham, B. L. Carter, W. N. Hanafee, and M. F. Mafee, Thieme Medical, New York, 1988, p. 253.

13

56. R. B. Lufkin and W. N. Hanafee, Invest. Radiol., 1988, 23, 162. 57. J. A. Castelijns, R. P. Golding, C. Van Schaik, J. Valk, and G. B. Snow, Radiology, 1990, 174, 669. 58. H. D. Curtin, Radiology, 1989, 173, 1. 59. P. M. Som, Am. J. Radiol., 1992, 158, 961. 60. L. M. Teresi, R. B. Lufkin, D. G. Worthman, E. Abemayor, and W. N. Hanafee, Radiology, 1989, 163, 405. 61. T. Vogl, S. Dresel, L. T. Bilaniuk, G. Grever, K. Kang, and J. Lissner, Am. J. Neuroradiol., 1990, 154, 585.

Biographical Sketch Mahmood F. Mafee. b 1941. M.D., 1969 Tehran University, Iran; Resident in Radiology, 1973, Albert Einstein University Hospital, New York; Resident in Radiology, University of Illinois Hospital at Chicago, 1974±76, Fellowship in Neuroradiology-Head and Neck, University of Illinois Hospital and Eye and Ear In®rmary, 1976±1977. Faculty in Radiology, Illinois State University 1976±present. Approx. 170 publications. Research interests: the clinical applications of MRI in the diagnosis of head and neck disorders and neuro-otological and neuro-ophthalmological disorders.

HEAD AND NECK INVESTIGATIONS BY MRI

Head and Neck Investigations by MRI Yoshimi Anzai University of Michigan, USA

and Robert Lufkin University of California, Los Angeles, CA, USA

1 INTRODUCTION MRI has revolutionized head and neck imaging and has replaced computerized tomography (CT) as the study of choice for many lesions of the extracranial head and neck. MRI easily surpasses CT in its ability to differentiate subtle differences in soft tissue boundaries and extensions of tumors of the head and neck. The beam hardening artifacts on CT images from dental amalgam and dense cortical bone of the mandible, skull base, shoulders, and other areas are also not a problem with MRI. Multiplanar imaging capabilities and lack of ionizing radiation make MRI the preferred imaging study for many head and neck processes. The exceptions where CT is still indicated are many in¯ammatory diseases, congenital malformations of the temporal bone, ®bro-osseous lesions, and fractures. In patients with a contraindication to MRI, dif®cult or confusing cases where the MRI ®ndings are inconclusive, or in situations where MRI is not available, CT is often very valuable. In the paranasal sinuses and middle ear disease where lesions are often de®ned by their relationship to thin cortical bone and air, CT is the study of choice. However, MRI is the preferred imaging modality for the majority of neoplastic conditions in most areas of the head and neck. Cancers in the head and neck are often more accessible by visual inspection or palpation than similar lesions in brain, lung, abdomen, and pelvis. Therefore the role of any imaging modality in the evaluation of head and neck cancer is usually to de®ne the deep extension of tumors, rather than to detect their presence. Mucosal changes are usually best evaluated by direct inspection during physical examination. Some submucosal tumors do not show ®ndings other than fullness or bulging on physical examination. MRI is a powerful tool for the determination of deep disease extension and the relationship of the tumor to signi®cant structures such as major vessels and nerves, which helps de®ne therapeutic options. 2 TECHNIQUE Surface coils greatly increase the MRI signal-to-noise ratio and allow higher spatial resolution of most examinations of the head and neck. While acceptable studies of the sinuses and naso- and oropharynx may be obtained with the standard head coil, the use of surface coils for examination of the larynx,

1

hypopharynx, neck, and temporal bone is practically essential.1,2 The new bilateral phased array surface coil provides high resolution with a higher signal-to-noise ratio compared with standard surface coils. Each phased array coil element has its own preampli®er, receiver channel, and digitizer, allowing the collection of data and generation of images individually and subsequent combination into a composite image. This approach is now routinely used in many centers for MRI of the temporal bone and temporomandibular joint (Figure 1). While specialized surface rf coils are important factors that make quality magnetic resonance images of the head and neck possible, perhaps the most signi®cant factor in the superiority of MRI over CT scanning in this area is the high soft tissue contrast resolution of MRI. Unlike the central nervous system where virtually no MRI signal from fat is present, the abundance of fat±water interfaces in the extracranial head and neck, which literally de®nes the anatomy of this region, greatly in¯uences the selection of pulse sequences for optimum image contrast. Although there is considerable variation in imaging approaches to the head and neck, a standard MRI sequence would include T1- and T2-weighted axial images, and T1weighted coronal images for most lesions. T1-weighted sagittal images are necessary for the larynx, mastoid portion of the facial nerve, and midline lesions. Currently, T2-weighted fast spin echo (FSE) imaging has replaced the regular T2-weighted spin echo image for most studies of the head and neck at many centers.3 In FSE acquisition, multiple refocused echoes are generated during a single TR (usually 8±16 echoes train), resulting in a short scan time and a better signal-to-noise ratio. The reduced scan time allows higher image matrix acquisition, such as 512 instead of 256, for improved spatial resolution. The drawbacks of FSE include fewer slices obtained for the same TR and less sensitivity to susceptibility effects. Other parameters commonly used in head and neck MRI are an FOV (®eld of view) of 18±20 cm and slice thickness of 3±4 mm. A 192  256 matrix is routinely used in most of head and neck MRI, except for high resolution temporal bone MRI where a 384  512 matrix is used. The role of gadolinium contrast agents in MRI examinations of the head and neck has been investigated.4 Although gadolinium does enhance most head and neck tumors, it may also obscure the tumor margin in some cases because of the presence of enhancement within normal tissue resulting in reduced fat/tumor contrast. Fat suppression gadolinium enhancement MRI, therefore, was introduced to improve the lesion conspicuity following contrast administration. This technique is most valuable in regions with large amounts of fat, such as the orbit. In other regions of the head and neck the advantages of this technique are less well de®ned. Although gadolinium administration adds information about whether the suspected tumor does or does not enhance, the clinical relevance of this information has been questioned in some cases. When intracranial extension is present, however, gadolinium can clearly improve the detection of the blood± brain barrier and leptomeningeal pathology. For these reasons many centers now limit their use of gadolinium to studies of the orbit and other regions close to the skull base where leptomeningeal and perineural pathology is a possibility.

2 HEAD AND NECK INVESTIGATIONS BY MRI

Figure 1 Intracanalicular acoustic neuroma. (a) Axial T1-weighted image without contrast barely resolves the abnormality. (b) Following contrast administration, there is dense enhancement by the tumor. (c) High resolution (512 matrix) surface coil T2-weighted axial image through the contralateral side shows clearcut depiction of the normal internal auditory canal contents. (d) View through the abnormal side shows the tumor without the use of a contrast agent. Some investigators are now questioning the added value of the contrast study

3 TEMPORAL BONE MRI is the modality of choice to evaluate sensorineural hearing loss and to a large extent to evaluate the facial nerve in its course through the temporal bone. High-resolution technique is critically important in the imaging of the temporal bone.5 In our institution, a bilateral phased array coil is used with 3 mm interleaved slices. T1-weighted spin echo and T2weighted FSE images as well as gadolinium enhanced T1weighted images are obtained. Gadolinium may also help in the detection of small intracanalicular acoustic tumors, although the possibility of in¯ammatory enhancement simulating a neoplasm must always be considered.6 The most common CP angle tumor is the acoustic neuroma that appears as an enhancing mass often extending into the internal acoustic canal (IAC) (Figure 2). While gadolinium enhanced MRI clearly delineates the tumor extension and improves detection of the intracanalicular acoustic neuroma, recent studies using high-resolution T2-weighted images now question its value.7 Meningioma is the second most common CP angle tumor, and characteristically shows broad dural attachment. They are isointense to the cerebral cortex on T2-weighted images in the majority of cases. Hyperostosis and dural thickening are often associated with meningioma. Following gadolinium administration, meningioma typically shows enhancing dural extension of the tumor.

Epidermoid (primary cholesteatoma) is a congenital lesion resulting from inclusion of ectoderm during early fetal life. The epidermoid contains keratin debris and solid cholesterin. The signal intensity of the tumor depends on the proportion of these two elements. When keratin debris predominates, the tumor appears of low signal on T1- and of high signal on T2-weighted images. The cholesterin portion gives a high signal on T1weighted images. The absence of gadolinium enhancement is characteristic. CT remains the imaging study of choice at the present time for in¯ammatory disease and most other conditions resulting in conductive hearing loss. Important information about the status of the ossicles and bony covering of the facial nerve canal is still best obtained with CT. CT can also best visualize tumors and metabolic diseases affecting the middle ear and mastoid air cells. For the remainder of lesions of the temporal bone, CT and MRI will continue to play complementary roles in the imaging workup. Depending on the location and particular pathology, each of the imaging modalities will contribute information. 4

SALIVARY GLANDS

The preferred method of imaging the salivary glands has evolved with new developments in imaging technology as well as clinical needs. While sialography is still used to evaluate in-

HEAD AND NECK INVESTIGATIONS BY MRI

3

¯ammatory or autoimmune diseases of the salivary glands, MRI has replaced sialo CT scanning for the imaging evaluation of the majority of masses in the major salivary glands.8±10 While 80% of major salivary gland tumors occur in the parotid gland, in¯ammatory lesions are the most common in the submandibular gland. The sublingual gland is the smallest major salivary gland. Tumors originating from the sublingual gland are relatively rare. Therefore, MRI is most frequently indicated for evaluation of parotid masses. T1-weighted images in the axial plane provide excellent contrast among most tumors, parotid gland, and parapharyngeal fat, allowing differentiation of an intraparotid from extraparotid gland masses (Figure 3). These views are supplemented with coronal or sagittal views when there is a suggestion of temporal bone involvement. Although poor tumor margin and low signal intensity on T2-weighted images has been reported to predict the high grade of malignancies in parotid gland tumors to some extent this is not universally accepted, and determinations of histology by imaging studies is generally not possible. This is in part due to the unusual, and variety of, histology found in parotid tumors. Fortunately, ®ne needle aspiration (FNA) is a simple, lower cost and relatively noninvasive method to determine the nature of the tumor, and has been widely applied in large clinical series. The real advantage of MRI in evaluating parotid masses is its ability to determine more accurately whether the intraparotid tumor is within the super®cial or deep lobe. The super®cial and deep lobes of the parotid gland are not based on an actual anatomical structure but, rather, are de®ned by their location relative to the main trunk of the facial nerve. Coronal images are sometimes useful to determine the distance of the parotid mass from the stylomastoid foramen, which is crucial information for a surgeon. Malignant tumors in close relation to the facial nerve may require excision of part of that structure and subsequent microsurgical repair or transposition of nerve VII. Access to larger, deeper lesions may require division and re¯ection of the ramus of the mandible. This valuable information is easily obtained with MRI.11±13

5

Figure 2 Acoustic neuroma. This middle-aged man presented with left sided sensorineural hearing loss. (a) Axial T1-weighted image shows a rounded soft tissue mass (arrowed) centered over the right internal auditory canal extending to the cerebellopontine angle. (b) Axial T2-weighted image shows similar ®ndings. (c) Axial T1-weighted image following contrast administration shows uniform enhancement

PARANASAL SINUSES

As stated before, MRI is primarily indicated for sinus tumors rather than in¯ammatory or ®bro-osseous diseases. Although cortical bone and air do not return a signal on MRI, cortical bone can often be visualized as a lower signal between the layers of high signal soft tissue and mucosa covering the sinus wall. MRI allows excellent delineation of solid tissue masses surrounded by secretions within the paranasal sinuses.14,15 Erosions of the bony walls can be demonstrated on MRI as the absence of signal void normally present in cortical bone (Figure 4). For extensions beyond the bony sinuses, MRI is the clear study of choice because it differentiates skeletal muscle from tumor extension. In cases where there is a question of extension to the anterior or middle cranial fossa, MRI with gadolinium enhancement remains the study of choice. MRI is also valuable for repeated follow-up studies for diseases of children such as juvenile angio®broma where multiple studies are required and radiation dose becomes a factor. The high soft tissue contrast capabilities of MRI in distinguishing low protein ¯uids, high protein ¯uids, soft tissues, and normal

4 HEAD AND NECK INVESTIGATIONS BY MRI musculature make MRI ideal to differentiate sinus tumor from secondary obstructed sinuses.14 One possible pitfall encountered in MRI of in¯ammatory sinus disease is signal void seen in patients with desiccated lesions such as fungus infection. Unlike nonfungus sinusitis, mycetoma is of low signal intensity on both T1- and T2weighted images, simulating well pneumatized normal sinuses. This is due to the presence of paramagnetic substances such as manganese, iron, and calcium as well as desiccated debris within mycetoma which do not contain any free water. The fungus infection can be clearly demonstrated on CT, which is another reason that the mainstay for imaging in¯ammatory sinus disease will continue to be CT.

6

Figure 3 MRI guided aspiration cytology of a parotid mass. Several attempts at needle biopsy in the clinic without imaging guidance were uninformative. (a) Axial T1-weighted image shows the super®cial parotid mass (arrowed). (b) Axial T2-weighted image shows heterogeneous areas of high signal which are nonspeci®c. This may be seen with a variety of benign and malignant conditions. (c) Axial T1weighted image following MRI compatible needle placement shows the linear signal defect of the needle entering the mass (arrowed). The cytology revealed adenocarcinoma

NASOPHARYNX

The lack of motion and abundant facial planes of the nasopharynx result in high quality MRI.16 Retropharyngeal adenopathy, tumor in®ltration beyond the pharyngobasilar fascia, and hypertrophic lymphoid tissue are all better identi®ed on MRI than on CT. The main goal of MRI examination is to evaluate the extension of primary tumor as well as nodal metastases. T1-weighted images are usually adequate for examining the nasopharynx because of the abundance of loose areolar tissue between various muscle groups and bundles. Tumors can be identi®ed as low signal regions on T1-weighted images and high signal on T2-weighted images in these loose areolar planes (Figure 5). The con®guration of the nasopharynx is dominated by a very tough fascial membrane called the pharyngobasilar fascia. This tough fascia represents a continuation of the pharyngeal constrictor muscles and extends from the base of the skull down to the level of the hard palate. Its function is to maintain the airway as an open channel for breathing during normal activities and during chewing. Only malignant tumors and very aggressive in¯ammatory processes such as mucormycosis will pass from the mucosa of the nasopharynx through the pharyngobasilar fascia to involve the structures within the parapharyngeal space. MRI can routinely show the pharyngobasilar fascia and destruction of this tough fascia which indicates the aggressive nature of mass. The nasopharynx remains an area that is obscure to casual clinical examination due to gag re¯ex or poor patient cooperation. Its proximity to the skull base makes cancers in this region particularly devastating. Most malignancies of the nasopharynx are squamous cell carcinoma. The presenting symptoms of nasopharyngeal carcinoma vary widely. The most common complaints of patients presenting with nasopharyngeal tumors are nasal obstruction, local invasion of cranial nerves, serous otitis media, and cervical lymph node metastases. MRI is particularly well suited to complement the clinical examination in this situation. Other nasopharyngeal tumors, such as lymphoma, plasmacytoma, and occasionally rhabdomyosarcomas are also encountered. These tumors tend to be bulkier and in®ltrate more widely than squamous carcinoma. If a patient has symptoms of cranial nerve involvement, the skull base, jugular foramen, and the cavernous sinus should be thoroughly studied with MRI with administration of gadolinium. In particular, the direct coronal and sagittal MR scans are valuable to assess craniocaudal extension of tumor in skull

HEAD AND NECK INVESTIGATIONS BY MRI

5

Figure 4 Maxillary sinus carcinoma. (a) Axial T1-weighted image shows opaci®cation of the maxillary sinus with associated bone destruction. The amount of extension posteriorly is unclear. (b) Axial T2-weighted image at the same level shows improved muscle/tumor contrast. (c) Axial T1weighted image at a higher level shows invasion of the pterygopalatine fossa. (d) Coronal T1-weighted image better de®nes the craniocaudal extension of the mass which involves the ethmoid sinus

base involvement, which can help to de®ne radiation ports. Abnormalities of the skull base are detected by replacement of the normal low signal cortical bone and normal high signal bone marrow with the invading neoplasm. The tensor veli palatini muscle is enveloped in a fascial plane of its own which divides the parapharyngeal space into two compartments.17 The tensor veli palatini muscle fascia passes posterior to the styloid process and divides the space into a lateral compartment which is called the pre-styloid space and a medial compartment which is called the post-styloid space. This helps narrow down the list of differential diagnosis, since depending on the anatomical location the possible pathologies are different. For example, the majority of pre-styloid space masses are salivary gland tumors, and most post-styloid masses are of neurovascular origin, such as neuroma and paraganglioma. Metastatic tumor can be seen in both compartments. The capability of multiplanar imaging and the far superior soft tissue resolution make MRI the clear imaging study of choice to evaluate the nasopharynx and the adjacent spaces.

7 TONGUE AND OROPHARYNX In general, MRI produces superior soft tissue detail in evaluating the tongue and oropharynx compared with CT.18,19

Lack of artifacts from dental amalgam and beam hardening artifacts from the mandible on MRI also eliminates two major shortcomings of CT in the examination of this area. Moreover, the ability of MRI to obtain direct coronal and sagittal scan planes is a distinct advantage in recognizing intrinsic tongue musculature and assessing tumor volume and spread for treatment planning. MRI is therefore considered the study of choice in this area. Squamous cell carcinomas account for well over 90% of the malignancies of the tongue. The remaining lesions are lymphomas, rhabdomyosarcomas, adenocarcinoma, adenoid cystic carcinoma originating from the minor salivary gland, and an assortment of benign tumors and in¯ammatory processes. Squamous cell carcinomas of the tongue, tonsillar bed and posterior pharyngeal wall are the lesions most likely to require radiological imaging. The main purpose of the MRI study is to evaluate the extent of the primary tumor and nodal spread. Occasionally an in®ltrating tumor of these regions may have spread to the nasopharynx or the supraglottic larynx so that additional imaging is needed prior to surgical management. The tongue is one area in the head and neck where T2weighted pulse sequences are extremely valuable (Figure 6). The musculature of the tongue is of low signal on both T1- and T2-weighted images. Tumors produce a very high signal on T2weighted images, resulting in excellent delineation of tumor margins. Anatomical details of the midline are best obtained by coronal T1-weighted images. The midline lingual septum

6 HEAD AND NECK INVESTIGATIONS BY MRI

Figure 5 Nasopharynx carcinoma. This man presented with a neck mass. (a) T1-weighted axial image through the nasopharynx shows a right sided fullness. (b) T2-weighted image at the same level shows that the mass crosses the normal pharyngobasilar fascia and invades the parapharygeal space. (c) Higher axial T1-weighted image shows the mass extending cephalad and involving the pterygopalatine fossa. (d) Axial T1weighted image shows the large [presenting] posterior triangle lymph node

appears as high signal ®brofatty tissue separating the two lateral halves of the tongue. If the tumor crosses this midline, the relationship of the tumor margin to the lingual artery and hypoglossal nerve should be accessed. At least one hypoglossal nerve and one lingual artery must be retained in order to perform a hemiglossectomy. A total glossectomy is a much more involved procedure for selected patients, and ideally requires specialized preoperative planning. The lingual artery is easily visible between the genioglossus muscle and the interdigitation of the styloglossus and hyoglossus muscles. Disruption of the diaphragm of the ¯oor of the mouth (the mylohyoid muscle) in a bilateral fashion clearly indicates a far advanced tumor. Adenoid cystic carcinomas almost invariably extend along perineural lymphatics. In this case, gadolinium enhanced MRI is helpful to demonstrate the enhancement of the involved nerve. Carcinomas of the base of tongue (posterior to the foramen cecum) are particularly troublesome. Because of their relative inaccessibility, they are usually not discovered until relatively late. Of these tumors, 76% are reported to have metastases at the time of initial examination. Nodal disease is frequently bilateral because of the bilateral lymphatic drainage of the posterior third of the tongue. Tumors of the base of the tongue have a propensity to spread laterally into the glossopharyngeal sulcus and tonsillar bed regions and anteriorly into the vallecula and preepiglottic space. Since the mucosal surface of the base of tongue varies and asymmetry does not necessary indicate the presence of tumor,

the MRI ®nding which more strongly suggests the presence of an early base of tongue tumor is the disruption of the high signal intrinsic musculature. This is well visualized on axial and sagittal T1-weighted images. Sagittal images are particularly helpful in showing tumor extension to the supraglottic larynx. The remaining anterior lesions of the cheek, retromolar trigone, alveolar ridge, and lips are readily visible by inspection or available to the palpating ®nger, and imaging studies are seldom necessary, except for advanced cases that require the investigation of deep extension.

8

LARYNX AND HYPOPHARYNX

Rarely does any radiological imaging modality play a signi®cant role in reaching a diagnosis of malignancy in the larynx. The larynx is so readily accessible to clinical examination that the combination of biopsy and visual inspection usually strongly indicates the diagnosis of cancer. The very small true cord cancer may not be detected by MRI. While laryngoscopy can show mucosal surfaces and masses involving the lumen, deep extensions are dif®cult to detect from clinical examination alone, yet, in several areas, these extensions have profound implications on the management of disease.1,2,20±23 MRI to an even greater extent can de®ne this important deep anatomy.

HEAD AND NECK INVESTIGATIONS BY MRI

7

Figure 6 Tongue carcinoma. (a) Axial T1-weighted image shows an artifact from ferromagnetic dental work. Conventional dental amalgam does not create an artifact. (b) Axial T2-weighted image shows a small region of increased signal laterally which represents squamous carcinoma. It does not cross the midline nor invade the mandible. (c) Axial T1-weighted image following contrast shows enhancement of the lesion. The information is similar to that in the T2-weighted image. (d) Coronal T1-weighted image following contrast administration again shows the lesion

T1-weighted images with three scanning planes, axial, coronal, and sagittal, are ideal for the study of the larynx (Figure 7).2,20 It is valuable to start with sagittal images which show the level of true vocal cords. Axial images are obtained parallel to the true cord which is readily visualized in the paramedian sagittal image. Then coronal images are acquired perpendicular to the true vocal cord. This will provide improved delineation of three important anatomical divisionsÐthe glottic, subglottic and supraglottic regions. T1-weighted sequences maximize contrast between the loose areolar tissue of the parapharyngeal and preepiglottic spaces, and most tumors. In addition, the laryngeal skeleton consists of several cartilages, which contain high signal marrow on T1-weighted images. The level of the true vocal cord can be identi®ed by the intrinsic vocalis muscle attached to the vocal process of arytenoid cartilage, which rests on top of the cricoid lamina.24 The supraglottic larynx consists of the laryngeal ventricle, the false cord, aryepiglottic folds, the epiglottis, and arytenoid cartilages. The supraglottic larynx is embryologically part of the buccopharyngeal anlage, and its lymphatic drainage is shared with the tongue, extending superiorly to the internal jugular chain. The preepiglottic space, anterior to the epiglottis, has a normal high signal which is displaced by low signal when tumors in®ltrate this space. The ®nding of the vertical spread of supraglottic tumors to the glottic region is even more important. Only advanced lesions spread down the inferior margin of the epiglottis to the

anterior commissure or from superior to inferior in the paralaryngeal space. For the treatment of supraglottic tumors involving the tongue base, a partial glossectomy may have to be performed in addition to the primary surgery.1 Coronal and sagittal MRI scans with T1-weighted pulse sequences readily demonstrate these spreads. The true vocal cords and subglottic region act more like the trachea with the majority of the lymphatic drainage directed posteriorly and inferiorly. Planning for any voice conservation laryngeal surgery depends on an accurate pre-operative knowledge of the precise extent of the disease within the larynx. Speci®cally, all techniques require an intact cricoid cartilage and at least one mobile arytenoid cartilage. MRI can provide this essential information. To plan this type of operation, the direct coronal and direct sagittal imaging capabilities of magnetic resonance far surpass axial CT images in the ability to de®ne critical information regarding the cranial±caudal extent of the tumors. Compared to CT, MRI consistently shows superior soft tissue de®nition in cooperative patients. The use of direct coronal and sagittal scan planes allows the visualization of cranial±caudal tumor extension. 9

IMAGING GUIDED ASPIRATION CYTOLOGY

Imaging guided aspiration cytology has been applied to the nonpalpable deep sited head and neck lesion and has shown

8 HEAD AND NECK INVESTIGATIONS BY MRI

Figure 7 Supraglottic larynx carcinoma. (a) Axial T1-weighted image shows a large right sided supraglottic mass. (b) The T2-weighted image at the same level shows similar ®ndings but improved muscle/tumor contrast. (c) More cephalad axial T1-weighted image shows extralaryngeal spread to the adjacent neck. (d) Coronal T1-weighted image shows that the mass does not involve the cricoid cartilage of subglottic region

promising results. This was initially performed under CT or ultrasound guidance. Imaging guided aspiration cytology is now a standard procedure in the evaluation of head and neck tumors that cannot be sampled by more blind approaches. Aspiration cytology is now possible with MRI guidance using specially developed MRI compatible needles (E-Z-EM Corporation, Westbury, New York).25±28 The advantages of using MRI as an imaging guidance are lack of radiation, high soft tissue contrast allowing better delineation and detection of pathology, and multiplanar capability of showing the needle track in a single plane (see Figure 3). 10

RELATED ARTICLES

Eye, Orbit, Ear, Nose, and Throat Studies Using MRI; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Multi Echo Acquisition Techniques Using Inverting Radiofrequency Pulses in MRI; Surface and Other Local Coils for In Vivo Studies; Temperomandibular Joint MRI; Whole Body Machines: NMR Phased Array Coil Systems. 11

REFERENCES

1. R. B. Lufkin, W. N. Hanafee, D. Wortham, and L. Hoover, Radiology, 1986, 158, 747.

2. C. B. McArdle, B. J. Bailey, and E. G. Amparo, Arch. Otolaryngol. Head Neck Surg., 1986, 112, 616. 3. G. H. Zoarshi, J. R. Maskey, Y. Anzai, W. N. Hanafee, P. S. Melki, R. V. Mulkern, F. A. Jolesz, and R. B. Lughin, Radiology, 1993, 188, 323. 4. J. D. Robinson, S. C. Crawford, M. Teresi, V. L. Schiller, R. B. Lufkin, H. R. Harnsberger, R. B. Dietrich, J. R. Grim, G. R. Duckwiler, and E. M. Spiekler, Radiology, 1989, 172, 165. 5. J. Schwartz and H. Harnsberger, `Imaging of the Temporal Bone', 2nd edn., Thieme, New York, 1992. pp. 192±246. 6. M. H. Han, B. A. Jabour, J. C. Andrews, R. F. Canelis, F. Chen, Y. Arzai, D. P. Becker, R. B. Lughin, and W. N. Hanafee, Radiology, 1991, 179, 795. 7. R. W. Allen, H. R. Harnsberger, W. L. Davis, B. D. King, J. L. Parkin, and R. I. Affelbaum, Radiology, 1993, 189(P), 141. 8. S. M. Mandelblatt, I. F. Brain, P. C. Davis, S. M. Fry, L. H. Jacobs, and J. Hoffman, Jr., Radiology, 1987, 163, 411. 9. L. M. Teresi, E. Kolin, R. B. Lythin, and W. Hanafee, Am. J. Neuroradiol., 1987, 148, 995. 10. L. M. Teresi, D. Wortham, E. Abemayor, and W. Hanafee, Radiology, 1987, 163, 405. 11. J. W. Casselman and A. A. Mancuso, Radiology, 1987, 165, 183. 12. D. R. Mirich, C. B. McArelle, and M. V. Kulkarni, J. Comput. Assist. Tomogr., 1987, 11, 620. 13. D. H. Rice, Arch. Otolaryngol. Head Neck Surg., 1987, 113, 78. 14. M. P. Som, W. P. Dillon, G. D. Fullerton, R. A. Zimmeroner, B. Ragagopalan, and Z. Maron, Radiology, 1989, 172, 515. 15. M. Shapiro and P. Som, Radiol. Clin. N. Am., 1988, 27, 447. 16. W. P. Dillon, C. M. Mills, B. Kjos, J. De Groot, and M. BrantZawalzki, Radiology, 1984, 152, 731.

HEAD AND NECK INVESTIGATIONS BY MRI 17. H. D. Curtin, Radiology, 1987, 163, 195. 18. R. B. Lufkin, G. D. Wortham, R. B. Dietrich, A. L. Hoover, S. G. Larsson, H. Kangarloo, and W. N. Hanafee, Radiology, 1986, 161, 69. 19. J. M. Unger, Radiology, 1985, 155, 151. 20. R. B. Lufkin, S. G. Larsson, and W. N. Hanafee, Radiology, 1983, 148, 173. 21. R. B. Lufkin and W. N. Hanafee, Am. J. Neuroradiol., 1985, 145, 483. 22. J. A. Castelijns, J. Doomber, B. Verbeeten, Jr., G. J. Vielroye, and J. L. Bloem, J. Comput. Assist. Tomogr., 1985, 9, 919. 23. H. D. Curtin, Radiology, 1989, 173, 1.

9

24. C. R. Archer, S. S. Sagel, V. L. Yeager, S. Montin, and W. H. Friedman, Am. J. Roentgenol., 1981, 136, 571. 25. P. R. Mueller, D. D. Stark, J. F. Simeone, S. Saini, R. J. Butch, R. R. Edelmann, J. Wittenberg, and J. T. Ferrucci, Jr. Radiology, 1986, 161, 605. 26. R. Lufkin and L. Lay®eld, J. Comput. Assist. Tomogr., 1989, 13, 1105. 27. R. Lufkin, L. Teresi, and W. Hanafee, Am. J. Roentgenol., 1987, 149, 380. 28. G. Duckwiler, R. B. Luphin, L. Jeresi, E. Spickler, J. Dion, F. Vinuela, J. Bentson, and W. N. Hanafee, Radiology, 1989, 170, 519.

HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI

Hemorrhage in the Brain and Neck Observed by MRI Robert I. Grossman University of Pennsylvania Medical Center, Philadelphia, PA, USA

1

improving the detectability of hemorrhage at all ®eld strengths.12,13 This review begins with the biophysical principles relevant to hemorrhage. These include a brief discussion of paramagnetism and magnetic susceptibility followed by structure of hemoglobin, the effect of protein on MRI signal intensity, and the importance of heterogeneity of magnetic susceptibility in the appearance of hemorrhage. The bene®ts and limitations of common imaging pulse sequences is then described. Lastly, we shall apply those principles to understand and interpret the diverse, yet characteristic, ®ndings produced by blood on MRI.

1 HISTORY Any new imaging modality that purported to compete with computerized tomography, in the late 1970s and early 1980s, had to be capable of detecting hemorrhage reliably. The success of MRI was intimately related to its ability for characterizing all stages of hemorrhage. Initially, at low imaging ®eld strengths (up to 0.3 T), demonstration of acute hemorrhage (within the ®rst 24±48 hours) was problematic at best.1 With the introduction of higher ®eld (0.5±1.5 T) MRI scanners in approximately 1984, it was clear that MRI could reliably image all stages of hemorrhage from the acute stage (within the ®rst 3 days or so after the acute event) to chronic hemorrhages (those that occurred months to years following the event). Both the ability to recognize older hemorrhages and localize them precisely were major factors enabling MRI to become the primary imaging modality for the vast majority of hemorrhagic conditions affecting the central nervous system. It is interesting to note the magnetic properties of dried blood had been ®rst studied by Faraday almost 150 years ago.2 Pauling and Coryell in 1936 recognized that deoxyhemoglobin was paramagnetic and oxyhemoglobin was diamagnetic.3 The T1 relaxation mechanisms of solutions of methemoglobin were investigated by Davidson and Gold in 1957 and further elucidated by Koenig et al. in 1981.4,5 These studies were based on the Bloembergen et al. theory (1948) of `outer sphere' relaxation of protons by paramagnetic centers.6 Thulborn et al. in 1982 observed that, at Larmor frequencies from 80 to 469 MHz, deoxyhemoglobin inside intact red blood cells (RBCs) would cause signi®cant heterogeneity of magnetic susceptibility resulting in selective T2 shortening.7 They also demonstrated that this phenomenon was proportional to the square of the magnetic ®eld. De La Paz et al. attempted to characterize the MRI features of acute intracranial hemorrhage.8 They appreciated that the relaxation times of hemorrhagic tissue could be related to several interacting factors including red cell integrity, oxygen saturation, and hemoglobin concentration. Bradley and Schmidt in 1984 modeled subarachnoid hemorrhage on a spectrometer operating at 20 MHz and concluded that methemoglobin may be at least partially responsible for the observed high intensity on some magnetic resonance images.9 This was con®rmed experimentally by Di Chiro et al. in an animal model of intraparenchymal hemorrhage.10 Gomori et al. unequivocally showed on a 1.5 T MRI scanner that a variety of different stages in the evolution of hemorrhage could be detected.11 The gradient echo technique was initially utilized clinically by Bydder and Young in 1985 and then applied by Edelman et al. to increase the sensitivity to susceptibility differences, thus

2 2.1

BIOPHYSICAL PRINCIPLES Paramagnetism and Magnetic Susceptibility

The magnetic moment of an electron is the result of its momentum and electric charge. Electrons have very large magnetic moments because of their light mass (~1/2000th of the proton's mass). The term `magnetic dipole' is applied to the electrons because of their north and south magnetic poles which are separated by a distance. Magnetic ®elds are induced by moving electric charges. Diamagnetic substances have paired electrons in their atomic and molecular orbitals. Paramagnetic molecules or atoms, on the other hand, contain unpaired orbital electrons. This produces a situation in which the magnetic moment of the unpaired electron is unopposed. Paramagnetic and diamagnetic molecules do not have magnetic ®elds of their own. In the absence of an external magnetic ®eld, the electron dipoles of these substances randomly align, resulting in zero net magnetization. However, when placed in the magnetic ®eld of an MRI scanner the electron dipoles of the paramagnetic substances line up in such a manner that the majority of the electrons point in the same direction as the external ®eld. This results in augmentation of the external magnetic ®eld. The ratio of the additional magnetic ®eld strength to the strength of the applied magnetic ®eld is termed the magnetic susceptibility of the paramagnetic substance. Magnetic susceptibility is thus a measure of how easily the substance may be magnetized. Diamagnetic substances, which possess paired electrons, may still produce a magnetic ®eld (when placed in an external magnetic ®eld) by virtue of the fact that they are orbiting around the nucleus. This magnetic ®eld is oriented to oppose the main magnetic ®eld, and is called a Lenz ®eld. It is far weaker than the ®eld generated by the unpaired electron spins of paramagnetic substances. Paramagnetic substances, by virtue of having unpaired electrons, have the effect of magnetic susceptibility which augments the main magnetic ®eld and also the much smaller effect of the Lenz ®eld opposing the main magnetic ®eld. In the absence of unpaired electrons, protons relax (realign with the main magnetic ®eld) by ¯uctuations in their magnetic ®elds caused by the motion of adjacent protons. Because of their large magnetic moments, unpaired electrons create ¯uctuations in local magnetic ®elds. These enable protons to realign faster with the external magnetic ®eld producing a shortened T1 relaxation time. This phenomenon is termed proton±electron dipole±dipole proton relaxation enhancement (PEDD PRE). In

2 HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI order for this interaction to occur the water proton must come within 0.3 nm of the unpaired electron. If a paramagnetic substance is nonuniformly distributed and the unpaired electrons are con®gured in a position such that water protons cannot come close (within 0.3 nm) to the unpaired electrons then the only interaction the water protons will appreciate is the locally increased magnetic ®eld produced by the unpaired electrons (paramagnetic susceptibility). Protons will precess at a rate proportional to the local ®eld strength which varies with the local susceptibility. The precession rate (Larmor frequency) is proportional to the local ®eld strength. If a paramagnetic substance is heterogeneously distributed then the local ®eld strength will vary throughout the region in which it is distributed. Protons, diffusing through these different areas of local ®eld variation produced by susceptibility differences, will precess at different rates. Thus, protons sensing a slightly higher local ®eld will precess more rapidly than those in a slightly lower magnetic ®eld environment. The net effect is for the protons to develop phase differences with shortening of the apparent T2 relaxation time (T2 proton relaxation enhancement or T2 PRE). The T2 PRE occurs when there is a heterogeneous distribution of paramagnetic substances with diffusion of water protons. In summary, paramagnetic substances enhance relaxation by two mechanisms, either PEDD PRE or T2 PRE. The former requires the water protons to approach the unpaired electrons and produces shortening of both T1 and T2; however, the effect is predominantly noted as a T1 effect. This relates to the fact that the T1 relaxation rate (1/T1) is much smaller than the T2 relaxation rate (1/T2). A paramagnetic relaxation effect added equally to either relaxation rate would contribute proportionally more to the much smaller term (1/T1). Shortening of the T1 relaxation time produces high intensity on the short TR/short TE (so-called T1 weighted) images. The T2 PRE produces low intensity on long TR images (so-called proton density and T2 weighted images). Hypointensity from T2 PRE can, at times, also be observed on T1 weighted images. This is because there is a T2 component to intensity even on a T1-weighted image, and substances with a very short T2 (heterogeneously paramagnetic) can produce hypointensity on a T1-weighted image. The presence of susceptibility effects can be visualized by observing signi®cant increases in hypointensity as the images become more T2-weighted (i.e. as the TE is increased). 2.2 Structure of Hemoglobin Hemoglobin, the primary oxygen carrier in the bloodstream, is composed of four protein subunits weighing approximately 16 000 Da apiece. Each subunit contains one heme molecule consisting of a porphyrin ring and an iron atom which provides the binding site for oxygen.14 Binding of oxygen to the heme molecule of an individual subunit produces a conformational change in that subunit and adjacent subunits. The iron atom sits near the center of the porphyrin ring and is bound to one of the imidazole nitrogens of histidine-92. Molecular oxygen binds to the iron atom on the opposite side of the porphyrin ring. In oxyhemoglobin, the oxidation state of the Fe is formally in the ferrous state which has six d orbital electrons. Molecular oxygen has two unpaired electrons. Oxyhemoglobin is diamagnetic, indicating that there must be pairing of the unpaired

electrons of molecular oxygen and the d electrons of the heme iron atom. Suf®ce to say that theories have been proposed to conceptualize the electronic state of oxyhemoglobin which are beyond the scope of this discussion.15,16 There are four unpaired electrons in deoxyhemoglobin making it a paramagnetic molecule. When a hemoglobin subunit loses its oxygen molecule to form deoxyhemoglobin, the heme protein undergoes a small but signi®cant change in its tertiary structure. The histidine ligand of the heme pulls the Fe2+ atom of deoxyhemoglobin out of the plane of the porphyrin ring and causes the porphyrin itself to dome. As a consequence, water molecules are unable to bind to the heme iron as they do in methemoglobin (see below), effectively preventing water from approaching close enough to the paramagnetic iron (<0.3 nm) to undergo PEDD PRE. However, deoxyhemoglobin can still produce local magnetic susceptibility changes (T2 PRE). Deoxyhemoglobin can be oxidized by one electron to methemoglobin by a number of different mechanisms. The Fe3+ is closer to the plane of the porphyrin ring in methemoglobin (in contrast to the Fe2+ atom of deoxyhemoglobin) so that water molecules can rapidly bind to the heme iron. Water must virtually bind to the heme iron in order to experience signi®cant PEDD PRE. As the number of heme molecules is relatively small compared with the number of water molecules, this effect would be small except that the exchange rate of the water molecules is rapid relative to TR, allowing many water molecules to bind to the heme during the course of imaging. Normally an enzyme system within the RBCs rapidly reduces methemoglobin back to deoxyhemoglobin, but this process requires glucose and NADPH, which most likely are in short supply in sequestered hematomas.

2.3

The Effect of Protein

Hemorrhagic lesions generally contain signi®cant concentrations of proteins. The effect of such protein is to slow the rotation of the water molecules. The much larger protein molecule has an intrinsically slower rate of rotation arising from Brownian motion due to its much larger mass. A strongly bound hydration shell of water molecules which surrounds the protein molecule with a zone of `structured' water possesses slower rotational rates than the `bulk' water molecules.17,18 In hemorrhagic situations the concomitant increased protein results in a greater number of water molecules rotating more slowly, and associated T1 shortening. A physical correlate of the change in rotational rate that occurs with increasing the protein concentration is the increased viscosity of solutions with high protein concentrations (i.e. acute and subacute intraparenchymal hematomas). The rotational correlation time is a measurement of the time it takes for this rotation to occur. The T1 relaxation time is noted to decrease with increasing protein concentration. At very high concentrations, the T1 shortening produced by the addition of protein molecules plateaus. Slowing of the rotational rate of the water molecules by the addition of protein also produces a decrease in T2. There are thus two major effects produced by the hemoglobin (deoxyhemoglobin or methemoglobin) molecule: (1) the paramagnetic effects secondary to the iron within the heme molecule and (2) the diamagnetic effects of the protein portion (apoprotein) of the hemoglobin molecule.

HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI

3

2.4 Heterogeneity of Hemorrhage

3.3

Heterogeneity of an RBC solution is proportional to Hct  (100 ÿ Hct), where Hct is the hematocrit. Thus heterogeneity would be maximal at Hct = 50 and nil at Hct = 0 or Hct = 100. Since the Hct of clot is ~90, heterogeneity is present with a considerable susceptibility effect (36% of the maximal effect). Furthermore, retracted in vivo clot packs less ef®ciently and regularly than under experimental conditions. This further increases this predicted heterogeneity of magnetic susceptibility.

In the case of this pulse sequence there are up to 16 pulses of 180 per TR interval. This decreases the diffusion time of water protons and hence decreases the T2 PRE. Susceptibility differences are decreased, and those elements of hemorrhage whose detection depends upon such differences (i.e. acute and chronic hemorrhage) become slightly more dif®cult to detect.19

4 3 IMAGING PULSE SEQUENCES IN HEMORRHAGE 3.1 Spin Echo Imaging The 90 rf pulse in a spin echo sequence rotates the protons into the transverse plane, and subsequently they precess at rates proportional to the ®eld strength they sense. If there is heterogeneous distribution of paramagnetic molecules the protons will rapidly develop phase incoherence. This is because protons diffusing to different regions are experiencing varying magnetic strengths and precessing at slightly different rates. The 180 rf pulse in the spin echo sequence was designed to obviate local static magnetic inhomogeneities. The analogy is made to slow and fast runners that must reverse course half-way through a race so that all runners wind up at the ®nish line at precisely the same time. If, however, the protons diffuse randomly to different regions containing varying magnetic ®elds, then, the 180 rf pulse cannot rephase these diffusing protons to precisely the same location they had previously been at just prior to the pulse. This loss of coherence of the spins in the transverse plane results in the T2 PRE and hypointensity on T2weighted images. The longer the interecho interval (the time between 180 pulses) the shorter the T2 in regions of heterogeneous distribution of paramagnetic molecules. This is because the protons have more time to diffuse through different regions of varying ®eld strengths when the interecho times are increased. Protons will thus lose phase coherence faster and have a shorter T2 under conditions of heterogeneity of magnetic susceptibility and long interecho times. The selective shortening of T2 varies as the square of the magnitude of the applied magnetic ®eld, and the square of the variation in the magnetic susceptibility. 3.2 Gradient Echo Imaging Gradient refocused echo images are more sensitive to susceptibility changes because the resultant local ®eld gradients superimpose upon the applied phasing and rephasing gradients. The presence of local ®eld gradients from some blood products adds linearly to the phasing and rephasing gradients. This leads to rapid dephasing of the protons and signal loss on these T2*weighted images. This imaging technique is most sensitive for detection of heterogeneity of magnetic susceptibility but lacks some speci®city with respect to particular hemorrhagic lesions. It is also problematic for routine imaging because of susceptibility differences at the skull base imposed by the air±tissue interfaces.

4.1

Fast Spin Echo Imaging

CLINICAL SITUATIONS Acute Hemorrhage

The biophysical effects just outlined sum to produce the appearance of acute hemorrhage as hypointensity on T2weighted images. In a simple intraparenchymal hemorrhage, blood is extravasated into the brain parenchyma. The RBCs are isolated from the circulation and cannot be reoxygenated. The RBCs which initially contain oxyhemoglobin become desaturated. Oxyhemoglobin has no unpaired electrons and is not paramagnetic. Imaging a hematoma containing only oxyhemoglobin will be iso- or slightly hypointense on short TR images and hyperintense on long TR images. This is related to the water content in the serum of the extravasated blood. Occasional images of hyperacute hematomas (within a few hours after an ictus) have noted a rim of hypointensity on long TR/long TE images. This rim has been postulated to be the result of early accumulation of iron or deoxyhemoglobin.20 Over hours the RBCs become desaturated. In addition, the inability of the isolated hematoma to rid itself easily of metabolic wastes with the possibility of increased lactic acid and CO2 will to tend to shift the oxyhemoglobin dissociation curve to the right (Bohr effect) which at any pO2 will result in less hemoglobin saturation and more deoxyhemoglobin. Such desaturation can usually be appreciated very early following an ictus (within a few hours). The high protein content of the clot, as evidenced by the hyperdensity noted on CT, shortens T1 relative to cerebrospinal ¯uid causing the hematoma to be hyperintense relative to CSF and isointense to the brain parenchyma on the short TR/short TE images (T1-weighted images). Since solvent water molecules are unable to approach close enough to the Fe2+ heme of deoxyhemoglobin, there is no additional T1 shortening due to PEDD PRE interactions. On long TR/long TE images (T2-weighted images), the protein also shortens T2, rendering the clot hypointense relative to cerebrospinal ¯uid. Superimposed is a marked hypointensity secondary to susceptibility effects arising from the local ®eld inhomogeneity produced by encapsulation of the paramagnetic deoxyhemoglobin within the RBCs. This effect is magni®ed on gradient refocused images, which as stated above are more sensitive to susceptibility effects. The end-result is the profound hypointensity of the acute hematoma on long TR/long TE and gradient refocused images (Figure 1). There is usually a penumbra of high intensity on long TR images surrounding the acute hematoma. This is a consequence of clot retraction with extrusion of serum into the surrounding normal brain and, later, the development of vasogenic edema (over days from the initial ictus). The high intensity of the surrounding edema and

4 HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI 4.2

Early Subacute Hemorrhage

Within 3±7 days there is oxidation of the deoxyhemoglobin to methemoglobin inside of the RBCs, occurring initially at the periphery of the clot. Unlike deoxyhemoglobin, water molecules are able to approach within 0.3 nm of the paramagnetic heme of methemoglobin, permitting PEDD PRE interactions which shorten T1. It is this effect, in concert with the T1 shortening from the high protein concentration, that gives methemoglobin its characteristic hyperintensity on T1-weighted images. Because the paramagnetic methemoglobin remains encapsulated within the RBCs, marked hypointensity is also seen on the long TR/long TE and gradient refocused images due to the same susceptibility mechanism described above for deoxyhemoglobin. The PEDD PRE interaction also shortens T2, but this effect is insigni®cant compared with the much larger susceptibility effects.

4.3

Mid Subacute Hemorrhage

The oxidation of deoxyhemoglobin proceeds centrally with time as evidenced by the development of hyperintensity within the center of the hematoma on the short TR/short TE images (Figure 2). Again the hematoma remains hypointense on the long TR, long TE and gradient refocused images due to susceptibility effects.21

4.4

Figure 1 Acute hematoma. (a) Short TR/short TE image of right cerebellar hemorrhage. This image was performed approximately 72 hours after the ictus. Note the hypointensity of the hematoma compared with normal cerebellum. There is a slight rim of hyperintensity representing methemoglobin which is just beginning to form at the periphery of the hematoma. (b) Long TR/long TE image of (a). There is striking hypointensity of the hematoma which is the result of heterogeneity of magnetic susceptibility due to deoxyhemoglobin in intact RBCs. Surrounding the hematoma is a penumbra of high intensity from serum and/or vasogenic edema

the mass effect of the hemorrhage should disappear within 4±6 weeks. If persistent high intensity remains surrounding the hemorrhage, then one should be suspicious of an intratumoral hemorrhage.

Late Subacute Hemorrhage

Methemoglobin is less stable than deoxyhemoglobin, and the heme group can be lost spontaneously from the protein molecule. This free heme and/or other exogenous compounds (including peroxide and superoxide) can produce RBC lysis. Concomitantly there is protein breakdown and dilution of the remaining extracellular methemoglobin. Hyperintensity persists on the short TR/short TE images despite the decrease in the protein concentration due to the PEDD PRE of methemoglobin even at relatively low concentrations. The hematoma signal intensity increases on long TR/long TE images, approaching that of cerebrospinal ¯uid as there is loss of local ®eld inhomogeneity upon RBC lysis as well as a decrease in the protein concentration. Also, the T2 shortening effects of the paramagnetic heme due to PEDD PRE are rather small and unimportant except at the highest concentrations. Paralleling the breakdown of the methemoglobin is an accumulation of the iron molecules as hemosiderin and ferritin within macrophages at the periphery of the lesion which can be identi®ed in the mid subacute stage of hemorrhage as well [Figure 2(b)].22 The iron cores of hemosiderin and ferritin contain ~2000 iron molecules ferromagnetically coupled to produce a `super paramagnetic' substance which exhibits a very large susceptibility effect. The result is a black ring surrounding the lesion visible on short TR/short TE images but increasingly prominent on long TR, long TE spin echo and gradient refocused images. 4.5

Chronic Hemorrhage

After months, there is near complete breakdown and resorption of the ¯uid and protein within the clot. The iron atoms

HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI

5

are most prominent on the long TR/long TE and gradient refocused images. There is no edema or mass effect associated with the hemorrhage. The methemoglobin (central high intensity) is gradually resorbed over months to years while the hemosiderin remains permanently in the brain. Very old hemorrhages often appear as slit-like cavities lined with hemosiderin and ferritin (Figure 3).

5

OTHER CLINICAL CONCEPTS

MRI is the technique of choice in virtually all hemorrhagic conditions of the head and neck. With its ability to detect hemorrhages of various ages, diagnosis of recurrent hemorrhagic events is facilitated. Conditions in which these situations are important include child abuse, siderosis of the central nervous system, and amyloid angiopathy. It should be emphasized that the pO2 of the hemorrhagic environs is important. Tumors are generally more hypoxic than normal brain so that an acute hemorrhage into a tumor may be more hypointense on long TR images because of increased amounts of deoxyhemoglobin. Conversely, in subarachnoid hemorrhage the red blood cells are bathed in the cerebrospinal ¯uid which has a pO2 of approximately 43 mmHg. At this pO2, 72% of the blood is in the oxyhemoglobin form, and 28% in the deoxyhemoglobin state. As the amount of T2 shortening varies as the square of the concentration of the paramagnetic (0.28)2  100% = 7.8% of the T2 shortening would be expected in this case when compared with that from 100% deoxyhemoglobin.23 Thus, acute subarachnoid hemor-

Figure 2 Subacute hematoma. Same patient as in Figure 1 except approximately 7 days after ictus. (a) Short TR/short TE image revealing ring of high intensity. This high intensity from the PEDD PRE of methemoglobin is the hallmark of the subacute stage of hemorrhage. (b) Long TR/long TE image demonstrates a rim of low intensity representing peripheral hemosiderin. Free methemoglobin is observed inside the hemosiderin rim. In the centermost portion of the hematoma is deoxyhemoglobin. There is edema surrounding the hemosiderin rim

from the metabolized hemoglobin molecules are deposited in hemosiderin and ferritin molecules unable to exit the brain parenchyma due to an intact blood±brain barrier. The susceptibility effects of the `super paramagnetic' iron cores of hemosiderin produce hypointensity on all spin sequences but

Figure 3 Chronic hemorrhage. Coronal long TR/long TE image of an old right occipital hemorrhagic cavity delineates the hypointensity of hemosiderin lining the walls of the collapsed hematoma cavity. Note that there is no mass effect or edema associated with this chronic hemorrhage

6 HEMORRHAGE IN THE BRAIN AND NECK OBSERVED BY MRI rhage is the lone condition in which MRI is not the imaging technique of choice. In general the vast majority of hemorrhagic lesions follow the continuum of MRI changes that have just been described. This makes MRI an elegant probe for most hemorrhagic conditions including intraparenchymal hemorrhage, intratumoral hemorrhage, traumatic hemorrhage, hemorrhagic cortical infarction, vascular dissection, venous thrombosis, and cavernous hemangioma.

6 RELATED ARTICLES Brain Neoplasms Studied by MRI; Contrast Agents in Magnetic Resonance: Operating Mechanisms; Contrast Agents in Whole Body Magnetic Resonance: An Overview; MRI in Clinical Medicine.

7 REFERENCES 1. J. T. Sipponen, R.E. Sepponen, and A. Sivula, J. Comput. Assist. Tomogr., 1983, 7, 954. 2. L Pauling and C. Coryell, Proc. Natl. Acad. Sci. USA, 1936, 22, 210. 3. L Pauling and C. Coryell, Proc. Natl. Acad. Sci. USA, 1936, 22, 159. 4. S. H. Koenig, R. D. Brown III, and T. R. Lindstrom, Biophys. J; 1981, 34, 397. 5. N. Davidson and R. Gold, Biochim. Biophys. Acta, 1957, 26, 370. 6. N. Bloembergen, E. M. Purcell, and R. V. Pound, Phys. Rev., 1948, 73, 679. 7. K. R. Thulborn, J. C. Waterton, P. M. Matthews, and G. K. Radda, Biochim. Biophys. Acta, 1982, 714, 265. 8. R. L. De La Paz, P. F. J. New, F. S. Buonanno, J. P. Kistler, R. F. Oot, B. R. Rosen, J. M. Taveras, and T. J. Brady, J Comput. Assist. Tomogr., 1984, 8, 599. 9. W. G. Bradley Jr and P. G. Schmidt, Radiology, 1985, 156, 99. 10. G. Di Chiro, R. A. Brooks, M. E. Girton, T. Caporale, D. C. Wright, A. J. Dwyer, and M. K. Harne, Am. J. Neuroradiol., 1986, 7, 193.

11. J. M. Gomori, R. I. Grossman, H. I. Goldberg, R. A. Zimmerman, and L. T. Bilaniuk, Radiology, 1985, 157, 87. 12. G. M. Bydder and I. R. Young, J. Comput. Assist. Tomogr., 1985, 9, 1020. 13. R. R. Edelman, K. Johnson, R. Buxton, G. Shoukimas, B. R. Rosen, K. R. Davis, and T. J. Brady, Am. J. Neuroradiol., 1986, 7, 751. 14. R. Dickerson and I. Geis, `Hemoglobin: Structure, Function, Evolution and Pathology', Benjamin/Cummings, Menlo Park, CA, 1983. 15. J. Weiss, Nature, 1964, 202, 83. 16. W. A. Goddard and B. D. Olafson, Proc. Natl. Acad. Sci. USA, 1975, 72, 1335. 17. S. Koenig, `The Dynamics of Water-Protein Interactions', in ACS Symposium Series, No. 127, ed. S. P. Rowland, American Chemical Society, 1980, p. 157. 18. G. Fullerton, M. Finnie, K. Hunter, V. A. Ord, and I. L. Cameron, Magn. Reson. Imaging, 1987, 5, 353. 19. K. M. Jones, R. V. Mulkern, M. T. Mantello, P. S. Melki, S. S. Ahn, P. D. Barnes, and F. A. Jolesz, Radiology, 1992, 182, 53. 20. K. R. Thulborn and S. W. Atlas, in `Magnetic Resonance Imaging of the Brain and Spine', ed. S. W. Atlas, Raven Press, New York, 1991, p. 175. 21. J. M. Gomori, R. I. Grossman, and D. B. Hackney, Am. J. Neuroradiol., 1987, 8, 1019. 22. K. R. Thulborn, A. G. Sorensen, N. W. Kowall, A. McKee, A. Lai, R. C. McKinsky, J. Moore, B. R. Rosen, and T. J. Brady, Am. J. Neuroradiol., 1990, 11, 291. 23. R. I. Grossman, S. S. Kemp, I. C. Yu, J. E. Fishman, J. M. Gomori, P. M. Joseph, and T. Asakura, Acta Radiol., 1986; 369 (suppl), 56.

Biographical Sketch Robert I. Grossman. b 1947. B.S., 1969, M.D., 1973 University of Pennsylvania School of Medicine, USA. Internship Beth Israel Hospital, Boston, USA, 1973±74. Neurosurgical Residency, University of Pennsylvania, USA, 1974±76. Radiology Residency, University of Pennsylvania, USA, 1976±79. Neuroradiology Fellowship, Massachusetts General Hospital, USA, 1979±81. Professor of Radiology, Neurosurgery, and Neurology, Associate Chairman of Radiology, Section Chief Neuroradiology, University of Pennsylvania Medical Center, 1981±present. Approx. 200 publications. Current research interests: white matter diseases, stroke, head trauma.

INTRACRANIAL INFECTIONS

Intracranial Infections Spyros K. Karampekios University Hospital of Crete, Heraklion, Greece

and John R. Hesselink UCSD Medical Center, San Diego, CA, USA

1 INTRODUCTION Despite the development of many effective antibiotic therapies and the evolution of new neurosurgical techniques, central nervous system (CNS) infections persist. Also the increasing, devastating effect of acquired immune de®ciency syndrome (AIDS) has created a whole group of life-threatening opportunistic infectious diseases. CNS involvement usually occurs as a result of an infection from another organ system, or as a manifestation of a systemic disease. The brain, particularly, is well protected from invading agents by the calvarium, the meninges and the blood±brain barrier. However, different types of pathogens including bacteria, viruses, fungi, and parasites, can reach the brain by hematogenous spread (related to septicemia or endocarditis) and less likely by direct extension (bony erosion from an adjacent infected paranasal sinus, mastoid or middle ear). Other possible pathways for intracranial infections are via anastomotic veins from the face and scalp, along peripheral and cranial nerves (viruses) and after a penetrating head trauma. Once in the intracranial cavity, pathogens can involve the parenchyma (cerebritis±abscess, encephalitis), the meninges (meningitis±ependymitis), and other extraaxial spaces (subdural and epidural empyemas). The brain's unique response to infection is due in part to the absence of draining lymphatics, the differentiation of vascular supply in gray and white matter, the presence of the blood±brain barrier, and the existence of perivascular cerebrospinal ¯uid (CSF) containing spaces (Virchow± Robin spaces). CSF assists in the dissemination of an infectious disease, acting as a perfect culture medium for microbial growth. Magnetic resonance imaging (MRI) provides the most sensitive imaging modality to detect cerebral infection because of its optimal contrast resolution, the multiplanar imaging capability and the absence of signal from the surrounding bone. Intravenous contrast administration of gadolinium±diethylenetriaminepentaacetic acid (Gd-DTPA) increases the sensitivity of MRI and has allowed earlier detection of many infections compared to computed tomography (CT).1 The only disadvantage of MRI is its poorer ability to detect calci®cations, an important ®nding in some cerebral infections. 2 CEREBRITIS AND BRAIN ABSCESS Cerebritis and abscess formation constitute a spectrum of the same process. Regardless of the pathogen, the brain tissue

1

reacts in a predictable way to focal parenchymal infection, developing initially an area of focal cerebritis, which consists of vascular congestion, petechial hemorrhage, and brain edema. Progression from the cerebritis stage to an encapsulated abscess with a central area of lique®ed, necrotic material requires 10± 14 days, although the rapidity of this process depends on patient's immunocompetence and the organism.2 In the late mature abscess stage, when the collagenous capsule is fully formed, the surrounding edema decreases, and surgical drainage is facilitated. A brain abscess is usually caused by direct extension from an adjacent infected sinus or mastoid, or by hematogenous spread from an extracerebral source. Rarely is an abscess secondary to meningitis. In up to 20% of cases the source of infection is not discovered. In children the majority of cerebral abscesses are associated with cyanotic congenital heart disease. Anaerobic bacteria are isolated most frequently in the brain abscesses, although a mixture of pathogens are often found. Overall, in otherwise immunocompetent individuals, the commonest cultured organisms are Staphylococcus and Streptococcus species.3 Patients present usually in the late cerebritis±early abscess phase with nonspeci®c symptoms of headache, confusion, seizures or focal neurologic de®cits. Fever and leukocytosis are common during the invasive phase of an abscess, but may resolve as it matures. The MRI features of cerebritis±brain abscess depend on the stage of the infectious process at the time of imaging. In the early cerebritis stage MRI depicts the changes earlier than CT because of its superior sensitivity to alterations in water content. The area of cerebritis appears mildly hypointense on T1-weighted images. On the same sequence the early, subtle mass effect is best demonstrated. The infectious focus depicts high signal on T2-weighted images, both centrally from in¯ammation and peripherally from edema. As the infection matures, it increases in size due to an increase in edema and necrotic debris accumulates centrally, while the body attempts to isolate the infection by forming a capsule. The abscess capsule is thicker along the cortical surface because the vascularity of gray matter is much greater than that of white matter, resulting in increased local cortical reaction. The thinner portion of the capsule toward the white matter accounts for the predilection of abscesses to rupture into the ventricles producing ependymitis. At this stage of abscess formation, MRI demonstrates lengthening of T1 and T2 relaxation times in the core of the lesion. Peripherally there is moderate amount of vasogenic edema, which is mildly hypointense on T1- and hyperintense on T2-weighted images. On T1-weighted images, against the hypointense areas of the necrotic center and the surrounding edema, the abscess capsule stands out as an iso- or slightly hyperintense ring. On T2-weighted images the ring is markedly hypointense (Figure 1(a)). There is still discussion about the causes of capsular intensity. The hyperintensity on T1-weighted images is attributed to capsular hemorrhage by some authors. More recently the signal properties of the abscess capsule have been attributed to paramagnetic hemoglobin degradation products, or free radicals within macrophages. Macrophages are abundant in the capsule, and their activity is highest at the late cerebritis±early abscess phase, exactly when the signal intensity of the capsule on the T2-weighted images is particularly low.4 Contrast administration is very helpful in the evaluation of brain abscesses. Gd-DTPA produces mottled, heterogenous

2 INTRACRANIAL INFECTIONS

Figure 1 Streptococcal abscesses. (a) Axial, T2-weighted (SE 2500/70) image shows the abscess capsules as hypointense rings surrounding the necrotic central core. There is also signi®cant amount of peripheral edema. (b) Axial, postcontrast T1-weighted (SE 800/20, Gd-DTPA) image one week later demonstrates intense ependymal enhancement of the right lateral ventricle from the resulting ependymitis secondary to rupture of an abscess into the ventricular system. (Reproduced by permission of W. B. Saunders from R. R. Edelman and J. R. Hesselink (eds), Clinical MRI, 1990, Chap. 19, p. 578)

areas of enhancement in the cerebritis stage, with an enhancing rim developing as the abscess matures. The ring-like enhancement re¯ects the damaged blood±brain barrier and is typically smooth, well de®ned and thin walled. The ring is thinnest at its medial margin and often points to the adjacent ventricle. If an abscess ruptures into a ventricle and secondary ependymitis develops, there is enhancement of the ventricular wall, suggesting a very poor prognosis [Figure 1(b)]. Brain abscesses produced by nonpyogenic organisms typically occur in immunocompromised patients; these will be discussed in the section of AIDS-related infections.

3 MENINGITIS Meningitis is an acute or chronic in¯ammation of the piaarachnoid (leptomeninges) and the adjacent CSF. Patients with meningitis present with fever, headache, neck stiffness, photophobia, and altered consciousness. Acute meningitis (purulent) is usually caused by bacteria. The most common pathogens are Haemophilus in¯uenza, Neisseria meningitidis, and Streptococ-

cus pneumonia. Signi®cant morbidity and mortality occurs with meningitis. The overall mortality rate ranges from 5% to 15% and severe, persistent neurologic de®cits may be seen in 10± 25%.5 In neonates and immunosuppressed patients Gram-negative meningitis is common, caused by Escherichia coli, Pseudomonas aeruginosa and Klebsiella. Viruses can also cause an acute meningitis (lymphocytic), particularly enteroviruses and the mumps virus. Viral meningitis is usually selflimited, with less signi®cant symptoms and complications than those of bacterial origin. Chronic meningitis is usually of tuberculous origin and presents as a long-standing, indolent process, in which vasculitis and cerebral infarctions from basal meningeal in¯ammation are more prevalent. In patients with an immunologic dysfunction, meningitis can be the result of fungal infection, with the main representatives being cryptococcosis, coccidiomycosis, and blastomycosis. Cryptococcal meningitis will be discussed in the section of AIDS-related infections. Finally, sarcoidosis is a noninfectious granulomatous disease that involves the CNS in 5% of patients, producing in¯ammation of the leptomeninges in the basal cisterns, the optic chiasma and the infundibulum. Other noninfectious processes which cause meningeal disease

INTRACRANIAL INFECTIONS

and simulate infections are meningeal carcinomatosis, postoperative meningeal irritation, chemical meningitis, meningeal reaction adjacent to cerebral infarction, and subarachnoid hemorrhage. Neuroimaging plays a limited role in the diagnosis of meningitis, which is made by history, physical examination and laboratory CSF ®ndings. CT and MRI are mainly focused on the detection of associated complications, which include vascular thrombosis, infarctions, cerebritis±brain abscess, ventriculitis, hydrocephalus, empyemas of epidural and subdural space and subdural effusions. In case of uncomplicated acute meningitis, unenhanced MRI scans are usually unremarkable. More severe chronic cases may disclose hyperintense CSF on T1-weighted and proton density (PD) images in the basal cisterns secondary to obliteration from the in¯ammatory exudate and meningeal hyperemia. Contrast administration is very helpful in the evaluation of suspected meningeal infection because the involved meninges enhance diffusely and intensely. In the majority of cases of acute bacterial or viral meningitis, the meningeal enhancement occurs predominantly over the cerebrum, especially involving the frontal and parietal lobes, and the interhemispheric and sylvian ®ssures. However, in cases of tuberculous, fungal and sarcoid meningitis, the meningeal enhancement is more prominent in the basal cisterns (Figure 2). Gd-DTPA enhanced MRI appears to be much more sensitive than postcontrast CT in the detection of the meningeal enhancement, particularly when it occurs near the convexity.6 However, according to some authors, postcontrast MRI does not completely correlate with the extent of the in¯ammatory cell in®ltration. Using animal models they have proved that above a threshold of in¯ammation, meningeal enhancement was visualized, but areas that exhibited mild meningitis histologically did not enhance.7 While more sensitive, postcontrast MRI is no more speci®c, in that any process that causes meningeal irritation can also cause meningeal enhancement. In untreated or more severe cases of meningitis signi®cant complications may occur in the following days or weeks. The exposure of the blood vessels to the in¯ammatory exudate may result in vasculitis and thrombosis, a common complication in cases of tuberculous meningitis. Occlusion of small perforating arteries results in focal infarcts in the basal ganglia, while involvement of larger vessels can produce massive infarcts. Multiple hemorrhagic infarcts in the white matter are the result of cortical venous or dural sinus thrombosis.8 Another severe complication of meningitis is cerebritis and brain abscess formation. Sometimes, rupture of a preexisting brain abscess into the CSF spaces can produce secondary meningitis. Ventriculitis and ependymitis can result from direct extension of an abscess, progression of meningitis, or from an infected intraventricular shunt. In cases of ependymitis, MRI exhibits increased periventricular signal on T2- and PD-weighted images, that often has a nodular and irregular character to distinguish it from transependymal CSF ¯ow associated with obstructive hydrocephalus. Also, on postcontrast scans ependymal enhancement outlines the ventricles. If purulent debris ®lls the ventricles, the CSF may give a higher signal than normal. It has been proved more accurate to compare the intraventricular CSF with the vitreous of the globe, since associated meningitis may prevent comparison with CSF of the basal cisterns.9 Another frequent complication of meningitis is either obstructive, or communicating hydrocephalus, which occurs secondary

3

Figure 2 Coccidiomycosis meningitis. Contrast-enhanced T1weighted (SE 500/15, Gd-DTPA) axial (a) and coronal (b) scans demonstrate abnormal, extensive meningeal enhancement in the basal and ambient cisterns (arrows). (Reproduced by permission of the American Society of Neuroradiology from C. J. Wrobel, S. Meyer, R. H. Johnson, and J. R. Hesselink, Am. J. Neuroradiol., 1992, 13, 1243

4 INTRACRANIAL INFECTIONS to cellular debris obstructing the CSF pathways, or to arachnoid adhesions impairing extraventricular CSF ¯ow and absorption. Ventricular dilatation may be the only abnormal ®nding in patients with meningitis and is adequately evaluated with both CT and MRI. Subdural collections can also complicate meningitis; these are discussed in the section on extraaxial empyemas.

4 ENCEPHALITIS Encephalitis refers to a diffuse parenchymal in¯ammation of the brain caused primarily by viruses. Clinically, acute encephalitis should be suspected if the patient presents with convulsions, altered consciousness, delirium, aphasia, or ataxia. Particularly in herpetic encephalitis the symptoms re¯ect the propensity to involve the subfrontal and temporal lobes, with hallucinations, seizures, and personality changes. Viral encephalitis is usually acute, although it can occur from reactivation of a latent virus. The most common invading viruses are herpes simplex type 1 and type 2, herpes zoster, arbo- and enteroviruses, which produce almost the same reaction in brain tissue and appear similar on CT and MRI.10 In patients with AIDS, viral encephalitis may be caused by the human immunode®ciency virus (HIV), the cytomegalovirus (CMV) and the papovavirus (PML); these are discussed in the section on AIDS-related infections. Acute disseminated encephalomyelitis (ADEM) represents an immune-mediated complication following an antecedent viral infection, especially measles, mumps, rubella, varicella, or a preceding vaccination. Various forms of viral encephalitis (slow virus) are described in association with Creutzfeldt±Jacob disease and subacute sclerosing panencephalitis (SSPE). Finally, neonatal encephalitis is caused frequently by the group of TORCH pathogens (from the initials of toxoplasma, rubella, cytomegalovirus, and herpes simplex type 2) and share some common features like microcephaly, brain atrophy, hydrocephalus, and cerebral calci®cations. Herpes simplex virus type 1 (HSV-1) accounts for 95% of herpetic encephalitis in adults. The mortality rate approaches 70% and most of the survivors exhibit severe, persistent neurologic impairment. The virus usually invades the brain after reactivation of a latent form, which is frequently located in the trigeminal (gasserian) ganglion. The marked predilection for temporal lobe involvement supports the proposed theory that the infection spreads intracranially from the trigeminal ganglion along the meningeal branches of the trigeminal nerve. The resulting necrotizing encephalitis rapidly disseminates in the brain, sparing the basal ganglia and producing mass effect and edema as well as small petechial hemorrhages. Early diagnosis is paramount in order to institute effective medical therapy (vidarabine or acyclovir) and avoid devastating and irreversible brain damage. De®nitive diagnosis of HSV-1 encephalitis is done after isolation of the virus from brain biopsy. However, given the appropriate clinical presentation, with or without MRI con®rmation, treatment should be instituted immediately. In the early course of the infection the characteristic distribution is almost pathognomonic for HSV-1 encephalitis. By MRI, the early edematous changes appear as ill-de®ned areas of low signal on T1- and high signal on T2-weighted images, usually beginning unilaterally but rapidly progressing to both hemispheres (Figure 3). Variable mass effect and gyral

Figure 3 Herpes simplex encephalitis. (a) Sagittal, T1-weighted (SE 800/20) image shows area of low signal intensity in the right temporal lobe and frontal operculum. (b) Axial, T2-weighted (SE 2800/80) image demonstrates extensive hyperintense areas in the subfrontal and temporal lobes bilaterally, though more severe on the right hemisphere. (Reproduced by permission of W. B. Saunders from R. R. Edelman and J. R. Hesselink, (eds), Clinical MRI, 1990, Chap. 19, p. 584)

INTRACRANIAL INFECTIONS

enhancement may also be present. Occasionally, foci of hemorrhage are visualized as areas of high intensity on both T1- and T2-weighted images.11 Note that, unlike HSV-1, HSV-2 encephalitis, the most common neonatal viral encephalitis, is a panencephalitis without any predilection in the temporal lobes.

5 EXTRA-AXIAL EMPYEMAS Accumulation of in¯ammatory debris may occur in the subdural or, less frequently, in the epidural space. Most of the extraaxial empyemas present acutely secondary to sinusitis or mastoiditis. Empyemas that occur secondary to an infected posttraumatic extra-axial hematoma or postcraniotomy cavity have a more prolonged and indolent course. Empyemas of the extra-axial compartments can also occur (particularly in infants) as complications of meningitis. 5.1 Subdural Empyemas Pathogens may enter the subdural space via retrograde thrombophlebitis, via local dural erosion, or after contamination of a meningitis-induced subdural effusion. Most frequently, subdural empyemas are secondary to frontal sinusitis (40%).12 Patients usually present with fever, seizures, focal neurologic de®cit, or coma, and require urgent surgical and medical intervention. If completely or partially untreated, the empyema may be complicated with venous thrombosis, infarcts and parenchymal abscesses. The long TR, long TE sequence is the most sensitive in detecting small, crescentic ¯uid collections, especially when they are located near the inner table or along the falx. MRI can also distinguish subdural empyemas from subdural effusions or hematomas, depending on signal differences. T1-weighted images demonstrate the purulent collections as areas with more signal than pure CSF, due to the increased content of proteins and in¯ammatory debris. Subdural hematomas (subacute) can be easily distinguished because the extracellular methemoglobin exhibits markedly hyperintense signal on both T1- and T2-weighted images. MRI can also evaluate mass effect on the adjacent brain or CSF spaces, as well as the underlying parenchymal abnormalities. 5.2 Epidural Empyemas In the case of epidural empyema, the purulent collection tends to localize outside the inelastic and ®rm dura, which protects the underlying brain from undesirable concomitant abnormalities. Thus, patients usually have a more silent clinical course. As with subdurals, MRI is the most sensitive method for the evaluation of epidural disease. The signal characteristics of the lentiform extraaxial collections are similar to subdural empyemas on both T1- and T2-weighted images. Mass effect on the adjacent brain can also be seen, usually without any evidence of parenchymal changes. On postcontrast scans there is profound enhancement of the in¯amed dura, often thicker than that observed in subdurals.

6

5

CYSTIC LESIONS

Parasitic infections of the brain can commonly produce focal, cystic areas. We review the imaging features of the most frequent parasitic diseases with emphasis on neurocysticercosis. Toxoplasmosis is included in the section on AIDS-related infections.

6.1

Cysticercosis

This is the most common parasitic infection of the CNS in immunocompetent individuals and is caused by the pork tapework Taenia solium. Humans ingest the eggs with food contaminated by human or pig feces. In the stomach the eggs release embryos (oncospheres) which enter the intestinal wall and spread hematogenously, invading any human tissue and developing the larvae (cysticerci). Skeletal muscles and CNS are more frequently affected by cysticercosis. CNS infection is reported in 70±90% of cases and constitutes the commonest cause of seizures in young patients in developing countries with poor hygiene. Neurocysticercosis may involve the brain parenchyma, the ventricles, or the subarachnoid space. In parenchymal cysticercosis as the larvae initially invade the brain, they produce a mild in¯ammatory reaction, which on MRI can be demonstrated as focal, nonenhancing area of edema. Typically, at this very early, active stage of the disease, CT ®ndings are unremarkable. In the next stage (3±12 months after infection) cystic lesions with thin walls and clear ¯uid are formed. They are usually located at the gray-white matter junction and appear as 1±2 cm spherical cysts with signal intensity identical to CSF on all MRI sequences (Figure 4). There is no evidence of associated edema or contrast enhancement. At this stage, a mural nodule (scolex), which is pathognomonic for cysticercosis, can be seen as a small (1±2 mm) focal, mural projection, isointense to the brain parenchyma. Visualization of the scolex is much easier with magnetic resonance than CT, due to the superior contrast resolution of MRI. Later on, as the parasites degenerate and die, they induce an in¯ammatory response with associated thickened cystic walls, increased signal intensity from the cyst's content owing to protein accumulation, marked surrounding edema and ring-like enhancement around the cyst. Finally, at the end-stage of the disease process, the cysts are completely mineralized and only punctate foci of calci®cations may be seen. Obviously CT is more effective than MRI in the detection of calci®cations, although speci®c MR sequences (gradient echo) are relatively sensitive to calci®ed lesions.13 Intraventricular cysticercosis occur in approximately 20% of cases. Cysts located near the aqueduct of sylvius or other CSF pathways can cause acute, obstructive hydrocephalus. The signal characteristics of the intraventricular cysts also follow CSF. Furthermore, on T1- and PD-weighted images, the cystic wall and the scolex can be more apparent against the adjacent ventricular CSF, even though the cystic component is indistinguishable from CSF.14 The unique feature in the ventricular variety of cysticercosis is that the degenerated cysticerci rarely calcify. In about 10%, cysticercosis may occur in the subarachnoid space, where the cysts have a racemose form and are larger, sterile, and usually without scoleces. Sometimes the only imaging ®nding is asymmetry of the CSF spaces. In the case of a profound in¯ammatory reaction, adja-

6 INTRACRANIAL INFECTIONS cent brain abnormalities such as edema, gliosis, or communicating hydrocephalus can be seen.

6.2

Hydatid Disease

This is a parasitic infection which results from the ingestion of contaminated dog feces, containing eggs of Echinococcus granulosus. After ingestion, the eggs release embryos (oncospheres) in the gastrointestinal tract that spread hematogenously via the portal circulation to other human tissues. The majority of the lesions affect the liver and the lungs, although some can reach the brain (2±5%), where the larvae develop into large, unilocular cysts. Either CT or MRI can demonstrate adequately the large, commonly solitary cerebral hydatid cyst as a spherical thin-walled structure containing clear, sterile ¯uid with CSF imaging characteristics. Calci®cation of the cystic wall may occasionally occur and is better shown by CT. Hydatid (echinococcal) cysts are not associated with edema or contrast enhancement, except in case of cystic rupture and leakage, when an extensive in¯ammatory reaction is created.15

7

AIDS-RELATED INFECTIONS

Patients with AIDS often develop neurologic complications. At least 10% of all AIDS patients present with problems related to the nervous system, and more than one-third of them will manifest a clinically apparent neurologic disorder during the course of the disease.16 At autopsy it was revealed that 80± 90% of patients with AIDS had neuropathologic abnormalities, and usually more than one disease process was present.17 AIDS is due to the HIV, which on the one hand damages the nervous system by direct infection (HIV encephalitis), and on the other produces immune system dysfunction, especially of cell-mediated immunity. The resultant immunode®ciency leaves the person with AIDS vulnerable to many opportunistic CNS infections. The most important and the commonest of these will be further analyzed below.

7.1

Figure 4 Parenchymal cysticercosis. (a) Noncontrast CT scan demonstrates multiple cysts in the brain parenchyma, several of which are calci®ed. (b) Axial, T2-weighted (SE 2500/70) image at approximately the same level, reveals high signal intensity within the cysts. Small calci®ed foci are dif®cult to identify, although the larger calci®cations exhibit markedly low signal intensity. (Reproduced by permission of W. B. Saunders from R. R. Edelman and J. R. Hesselink, (eds), Clinical MRI, 1990, Chap. 19, p. 580)

Toxoplasmosis

Toxoplasmosis is the most frequent opportunistic brain infection in AIDS patients. It is caused by the parasite Toxoplasma gondii which is an obligate intracellular protozoan. Patients may present with clinical manifestations of focal mass effect such as seizures, focal neurologic de®cits, or cranial nerve palsies, as well as more generalized symptoms like headache, confusion, lethargy, and declining mental status. MRI is the most sensitive imaging modality, demonstrating lesions in patients with normal CT scans, particularly lesions at an earlier stage or located in the posterior fossa. Postcontrast MRI scans show multiple ring or nodular enhancing lesions surrounded by vasogenic edema and an associated mass effect. Toxoplasma lesions are typically located at the corticomedullary junction and the basal ganglia. Hemorrhage is uncommon. On T2weighted images the lesions are either hyperintense or iso- to

INTRACRANIAL INFECTIONS

7

hypointense to brain parenchyma.18 The neuroimaging ®ndings with both CT and MRI are not pathognomonic for toxoplasmosis and can also be seen in various infectious or noninfectious processes, such as brain metastases, intracerebral lymphoma, Kaposi's sarcoma, cryptococcoma, and tuberculoma. If the imaging studies are suggestive of toxoplasmosis and con®rmatory toxoplasma titers are present, empiric antitoxoplasma medication is begun and the response of the treatment can be monitored with follow-up CT or MRI scans. If any or all of the lesions fail to respond to therapy by imaging criteria, toxoplasma may not be the causative infective agent, or there may be another, concurrent disease process present, and brain biopsy should be considered.19 7.2 Cryptococcosis Cryptococcus neoformans causes the most common CNS fungal infection in AIDS patients, which in terms of relative frequency ranks third after HIV encephalitis and toxoplasmosis. It involves the CNS most commonly as a meningitis with minimal in¯ammatory response and extends mainly in the distribution of the perforating brain arteries and the perivascular (Virchow±Robin) spaces. On MRI there are four patterns: focal cryptococcomas; dilated Virchow±Robin spaces in the basal ganglia and midbrain; miliary enhancing nodules in the brain parenchyma and leptomeningeal-cisternal spaces; and a mixed pattern.20 7.3 Other Brain Abscesses Brain abscesses caused by nonpyogenic organisms typically occur in immunocompromised patients and share almost the same characteristics as the bacterial ones. Some of their unique features have been attributed to the diminished host response. Thus, multiplicity and extraparenchymal spread are frequently encountered. Tuberculous brain abscesses usually spread hematogenously from the lungs. They contain encapsulated pus with viable tubercle bacilli and differ from the more common tuberculomas (granulomas) which are smaller and contain caseous debris. Other associated lesions such as basal meningitis, multiple granulomas, and deep cerebral infarctions are very helpful clues to the diagnosis of tuberculosis.21 Another cause of nonbacterial brain abscess is fungal infections, which have recently increased substantially because of the large number of AIDS patients. The fungi may involve intracranial blood vessels, meninges, or the brain parenchyma (granulomas or abscesses). Fungal abscesses are caused primarily by Aspergillus, Mucor and Candida species, owing to their large hyphal forms which permit only limited access to the meningeal microcirculation. Aspergillus spreads to the brain hematogenously from the lung, gastrointestinal tract, or directly from the nasal cavity (paranasal sinuses). It readily invades vascular structures and causes hemorrhagic infarcts. In that case the resulting brain abscesses may mimic other hemorrhagic lesions such as metastases. CNS mucormycosis occurs in patients with compromised natural defense and uncontrolled diabetes and is spread by direct extension from adjacent facial compartments. Like Aspergillus, Mucor invades blood vessels and causes abscesses, more often at the inferior temporal and frontal lobes. Candidiasis may involve the brain parenchyma, producing multiple, scattered microabscesses.22

Figure 5 HIV encephalitis. The axial, T2-weighted (SE 2500/70) image demonstrates multiple, ill-de®ned, hyperintense lesions in the periventricular white matter, becoming con¯uent in the right hemisphere. (Reproduced by permission of W. B. Saunders from R. R. Edelman and J. R. Hesselink (eds), Clinical MRI, 1990, Chap, 19, p. 590)

7.4

HIV Encephalitis

This is the most common CNS complication in AIDS patients, resulting in a progressive subcortical dementia with associated motor and behavioral abnormalities (AIDS dementia complex).23 The hallmark of HIV encephalitis is the isolation of multinucleated giant cells and microglial nodules in the brain. Neither CT nor MRI are sensitive enough to detect those characteristic pathologic changes, showing only a nonspeci®c diffuse brain atrophy with a central predominance (ventricular enlargement). HIV leukoencephalopathy, which occurs in more severe cases of the encephalitis, is frequently located in the periventricular white matter and centrum semiovale. MRI depicts the white matter disease as extensive, nearly symmetric, ill-de®ned areas of high intensity on T2-weighted images, without any evidence of mass effect or contrast enhancement24 (Figure 5).

8 INTRACRANIAL INFECTIONS 7.5 CMV Encephalitis CMV is a herpes virus which can reactivate in the immunosuppressed host and produce a necrotizing encephalitis and ependymitis. CNS infection by CMV is found in more than one-third of autopsies of AIDS patients. Pathologically, there is necrosis of the periventricular parenchymal tissue, associated with accumulations of enlarged cells containing the typical inclusion bodies of CMV. However, the correlation between clinical diagnosis and postmortem results is very poor in cases of CMV encephalitis, for three reasons: ®rstly, the signs and symptoms of the encephalitis are subtle and nonspeci®c; secondly, there is no characteristic CSF pro®le; and, ®nally, the neuroimaging ®ndings are usually absent or nonspeci®c. PD- and T2-weighted magnetic resonance images may demonstrate a periventricular, thick, or nodular hyperintensity, which often involves the corpus callosum. CMV infection usually has a centrifugal spread from the ventricular ependyma involving diffusely the gray and white matter. On postcontrast scans there is diffuse subependymal enhancement around the lateral ventricles, representing the changes of ependymitis and making this an important differentiating point from HIV encephalitis or PML. The common simultaneous occurrence of both HIV and CMV viruses within the brain of AIDS patients supports

the proposed theory that an interaction between them may play a role in the pathogenesis of encephalitis in patients with AIDS. There is also some evidence from recent studies that CMV by itself has immunosuppressive properties which may compromise a patient's natural defense against HIV or other opportunistic pathogens.25,26 7.6

Progressive Multifocal Leukoencephalopathy (PML)

PML is a progressive demyelinating disease of the brain, caused by reactivation of a latent papovavirus, and affecting 3± 5% of all AIDS patients. The main target of PML is the oligodendrocyte which is the myelin-producing cell. The resulting demyelination develops a rapidly deteriorating clinical syndrome, with mental status changes, limb weakness, visual loss, or ataxia, and death usually ensues within 6 months. Initially, T2-weighted images demonstrate focal, round, or oval areas of high intensity which become larger and con¯uent with time. These areas are predominantly located in the white matter of the parieto-occipital region. However, involvement of the cortical gray matter, as well as the basal ganglia, the cerebellum, or the brainstem, is not uncommon.27 Generally the lesions in PML do not enhance and there is no evidence of signi®cant edema or mass effect (Figure 6).

Figure 6 Progressive multifocal leukoencephalopathy. (a) Axial, proton density (Fast SE 3600/17) image demonstrates area of high signal intensity in the subcortical white matter of the left, posterior temporal lobe. (b) Postcontrast, T1-weighted (SE 550/20, Gd-DTPA) coronal image discloses low signal intensity in the same region without any evidence of enhancement or mass effect

INTRACRANIAL INFECTIONS

8 RELATED ARTICLES Brain Neoplasms Gadolinium Chelate cations; Hemorrhage Magnetic Resonance Clinical Medicine.

Studied by MRI; CSF Velocity Imaging; Contrast Agents in MRI: Clinical Appliin the Brain and Neck Observed by MRI; Imaging of White Matter Disease; MRI in

9 REFERENCES 1. G. Schroth, K. Kretzschmar, J. Gawehn, and K. Voigt, Neuroradiology, 1987, 29, 120. 2. D. R. Enzmann, R. H. Britt, and R. Placone, Radiology, 1983, 146, 703. 3. G. Sze and S. H. Lee, in `Cranial MRI and CT', 3rd edn, eds. S. H. Lee, K. C. V. G. Rao, and R. A. Zimmerman, McGraw-Hill, New York, 1992, Chap. 13. 4. A. B. Haimes, R. D. Zimmerman, S. Morgello, K. Weingarten, R. D. Becker, R. Jennis, and M. D. F. Deck, Am. J. Roentgenol., 1989, 152, 1073. 5. Z. A. McGee and J. R. Baringer, in `Principles and Practice of Infectious Diseases', 3rd edn., eds. G. L. Mandell, R. G. Douglas, and J. E. Bennett, Churchill Livingstone, New York, 1990, Chap. 66. 6. K. H. Chang, M. H. Han, J. K. Roh, I. O. Kim, M. C. Han, and C. W. Kim, Am. J. Neuroradiol., 1990, 11, 69. 7. V. P. Mathews, M. A. Kuharik, M. K. Edwards, P. G. D'Amour, B. Azzarelli, and R. G. Dreesen, Am. J. Neuroradiol., 1988, 9, 1045. 8. G. Sze and R. D. Zimmerman, Radiol. Clin. N. Am., 1988, 26, 839. 9. T. J. Barloon, W. T. C. Yuh, L. E. Knepper, J. Biller, T. J. Ryals, and Y. Sato, J. Comput. Assist. Tomogr., 1990, 14, 272. 10. G. Sze, in `Magnetic Resonance Imaging', 2nd edn., eds. W. G. Bradley and D. D. Stark, Mosby, 1992, Vol. 1, Chap. 22. 11. R. D. Tien, G. J. Felsberg, and A. K. Osumi, Am. J. Roentgenol., 1993, 161, 167. 12. K. Weingarten, R. D. Zimmerman, R. D. Becker, L. A. Heier, A. B. Haimes, and M. D. F. Deck, Am. J. Roentgenol., 1989, 152, 615. 13. H. R. Martinez, R. Rangel-Guerra, G. Elizondo, J. Gonzalez, L. E. Todd, J. Ancer, and S. S. Prakash, Am. J. Neuroradiol., 1989, 10, 1011. 14. G. P. Teitelbaum, R. J. Otto, M. Lin, A. T. Watanabe, M. A. Stull, H. J. Manz, and W. G. Bradley, Am. J. Roentgenol., 1989, 153, 857.

9

15. K. H. Chang, S. Y. Cho, J. R. Hesselink, M. H. Han, and M. C. Han, Neuroimag. Clin. N. Am., 1991, 1, 159. 16. R. M. Levy, D. E. Bredesen, and M. L. Rosenblum, J. Neurosurg., 1985, 62, 475. 17. P. L. Lantos, J. E. McLaughlin, C. L. Schoitz, C. L. Berry, and J. R. Tighe, Lancet, 1989, i, 309. 18. B. C. Bowen and M.J.D. Post, in `MRI of the Brain and Spine', ed. S. W. Atlas, Raven, New York, 1991, Chap. 16. 19. J. A. Cohn, A. McMeeking, W. Cohen, J. Jacobs, and R. S. Holzman, Am. J. Med., 1989, 86, 521. 20. R. D. Tien, P. K. Chu, J. R. Hesselink, A. Duberg, and C. Wiley, Am. J. Neuroradiol., 1991, 12, 283. 21. C. Campi de Castro and J. R. Hesselink, Neuroimag. Clin. N. Am., 1991, 1, 119. 22. C. Bazan III, M. G. Rinaldi, R. R. Rauch, and J. R. Jinkins, Neuroimag. Clin. N. Am., 1991, 1, 57. 23. B. A. Navia, B. D. Jordan, and R. W. Price, Ann. Neurol., 1986, 19, 517. 24. H. S. Chrysikopoulos, G. A. Press, M. R. Grafe, J. R. Hesselink, and C. A. Wiley, Radiology, 1990, 175, 185. 25. C. A. Wiley and J. A. Nelson, Am. J. Pathol., 1988, 133, 73. 26. R. L. Yarrish, in `AIDS and other Manifestations of HIV Infection', 2nd edn, ed. G. P. Wormser, Raven, New York, 1992, Chap. 17. 27. A. S. Mark and S. W. Atlas, Radiology, 1989, 173, 517.

Biographical Sketches Spyros K. Karampekios. b 1959. M.D., 1984, M.D. Thesis, 1989, University of Athens, Greece, 1989. Faculty in the Department of Radiology, University Hospital of Crete, Greece, 1990±present. Currently lecturer in radiology, University of Crete, School of Medicine. Postdoctoral work as a research fellow in neuroradiology, University of California, San Diego. Approx. 15 publications. Research interests include MRI of central nervous system infections and magnetic resonance evaluation of spinal cord cavities. John R. Hesselink. b 1945. M.D., University of Wisconsin, 1971. Neuroradiology fellowship, Massachusetts General Hospital, 1979. Assistant Professor of Radiology, Harvard Medical School, 1979±84. Professor of Radiology and Neurosciences, University of California, San Diego, 1984±present. Approx. 150 publications and coeditor of textbook with Robert Edelman entitled Clinical MRI. Research interests include MRI of infections in immune compromised patients, fat suppression imaging techniques, magnetic resonance angiography and functional magnetic resonance.

ISCHEMIC STROKE

Ischemic Stroke

1

syncope, and psychiatric disorders), neuroimaging, particularly MRI with contrast injection, becomes essential for early diagnosis and prompt treatment.

William T. C. Yuh and Toshihiro Ueda The University of Iowa College of Medicine, Iowa City, IA, USA

and J. Randy Jinkins and Ronald A. Rauch University of Texas Health Science Center, San Antonio, TX, USA

1 INTRODUCTION Stroke affects 750 000 people annually in the USA and is the third leading cause of death.1 However, only a small percentage of patients die as a result of stroke, making this the leading cause of disability and the third most costly disease affecting adults in this country.2 Because many diseases can mimic ischemic stroke symptoms (epilepsy, migraine, tumor,

2

ETIOLOGY OF STROKE

Ischemic stroke is the result of failure to perfuse and/or oxygenate the brain. This is usually caused by arterial occlusion. However, it may also be seen after systemic hypotension and/or hypoxia. The underlying pathophysiology during acute brain ischemia can be explained by the hypothetical ischemic models proposed by Virapongse et al.: complete ischemia (no perfusion) and incomplete ischemia (some perfusion).3 The MRI appearance and clinical outcome are usually distinctly different in these two ischemic models (Figures 1±4).4±6 Because of the paucity of collateral circulation, complete ischemia, caused by severe compromise of the arterial supply, generally involves most of the cerebral tissue supplied by the vessel and leads to complete infarction (Figures 1 and 2). Limited or no contrast agent can reach the ischemic tissue, thus early parenchymal enhancement is not expected. Incomplete ischemia, on the other hand, caused by a transient vascular

Figure 1 Complete ischemia in a 62-year-old man who presented with acute left hemispheric stroke symptoms. (a) Axial T1-weighted contrastenhanced image (SE700/20) obtained 3 hours after onset showing abnormal arterial enhancement (AE) in the distribution of the left middle cerebral artery. No abnormal parenchymal enhancement (PE) is noted (AE ! 1/PE). (b) The axial T2-weighted image (SE2200/90) at the corresponding level shows no parenchymal T2 signal changes (T2
values) at this time. (c) Axial T2-weighted image (SE2000/100) obtained 24 hours after (b) showing extensive T2 signal changes consistent with ischemia of the left middle cerebral artery (AE ! T2
! 1/PE). (Reproduced by permission of the American Society of Neuroradiology from W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J. Neuroradiol., 1991, 12, 565)

2 ISCHEMIC STROKE

Figure 2 Complete ischemia with an inverse relationship between arterial enhancement (AE) and parenchymal enhancement (PE). (a) Axial T1-weighted contrast-enhanced image (SE583/20) and (b) the corresponding T2-weighted image (SE2000/100) obtained within the ®rst 24 hours of ischemic symptoms. Arterial enhancement (arrowed) is demonstrated in the distribution of the right middle cerebral artery (a). An intense signal on T2-weighted images (T2
values) without parenchymal enhancement is present at this time (b). These ®ndings are typical of complete ischemia expressed as T2
! AE ! 1/PE. (c, d) Follow-up axial T1-weighted contrast-enhanced images (SE583/20) obtained 7 days after the onset of acute ischemic symptoms when arterial enhancement (AE) has completely resolved and signi®cant parenchymal gyriform enhancement (PE) has developed (PE ! 1/AE). (Reproduced by permission of the American Society of Neuroradiology from Crain et al.6)

occlusion (complete or partial), tends to produce a less severe ischemic insult and may be associated with a potentially reversible or minimal neurological de®cit (Figures 3 and 4). Contrast material may be able to reach the ischemic tissue, and therefore early parenchymal enhancement is possible. Watershed infarction has a somewhat different pathophysiological basis from that of complete or incomplete ischemia.4±6 This region is located at the edges of the vascular territory between major cerebral arteries (Figure 5). A decrease in perfusion pressure from either systemic hypotension or partial occlusion may therefore affect only the regions of the brain that have the most marginal blood supply. Watershed infarction is a hybrid of complete and incomplete ischemia in that collateral circulation and/or antegrade arterial supply are intact but inadequate. Despite the severe ischemic insult, contrast material can still be delivered to the ischemic tissue as in incomplete ischemia. The clinical outcome is less favorable, however, and is more characteristic of complete ischemic infarction.

3 ACUTE CEREBRAL ISCHEMIA The MRI appearance of acute cerebral ischemia is dependent on several physiological factors (Table 1) that are often

Figure 3 Incomplete ischemia in a patient with transient neurological symptoms and early parenchymal enhancement. This patient developed transient left hemispheric symptoms after 2 minutes of test balloon occlusion of the left internal carotid artery. (a) Coronal T1-weighted contrast-enhanced image (SE538/20) obtained at 2 hours showing diffuse parenchymal enhancement (PE) (arrows) in the distribution of the left middle cerebral artery without arterial enhancement (AE) (PE ! 1/AE). Also noted is enhancement of the caudate nucleus. (b) The axial T2-weighted images (SE2000/100) show no apparent signal change at this time. (c) Axial T1-weighted contrast-enhanced images (SE583/20) at 24 hours show no evidence of parenchymal or arterial enhancement. The resolution of parenchymal enhancement by 24 hours parallels the rapid neurological improvement in this patient. (Reproduced by permission of the American Society of Neuroradiology from Crain et al.6)

ISCHEMIC STROKE

3

Figure 4 Incomplete ischemia with early, intense parenchymal enhancement (PE) in a patient who had a transient ischemic episode with right-sided weakness and aphasia that resolved almost completely in 2 hours. (a) Axial T1-weighted contrast-enhanced image (SE700/20) showing diffuse, intense cortical enhancement in the distribution of the left middle cerebral artery without evidence of arterial enhancement (AE) (PE ! 1/AE). (b) The corresponding axial T2-weighted image (SE2000/100) shows a much less extensive area (T2
values) of signal abnormality (PE ! 1/AE ! 1/T2
). (Reproduced by permission of the American Society of Neuroradiology from Crain et al.6)

Figure 5 Watershed infarction in a patient with severe neurological sequelae and early, intense parenchymal enhancement. (a) Axial T1weighted contrast-enhanced image (SE516/20) showing early and intense parenchymal enhancement in the right posterior watershed zone without arterial enhancement. (b) Axial T2-weighted image (SE2000/ 100) showing an area of signal abnormality approximating the area of enhancement in (a). (Reproduced by permission of the American Society of Neuroradiology from Crain et al.6)

coexistent.5 Four types of MRI abnormality occur soon after an acute cerebral insult: (1) alteration in arterial blood ¯ow, (2) parenchymal swelling, (3) parenchymal signal change, and (4) abnormal parenchymal enhancement.2,4,7±19 Vascular ¯ow abnormalities can be detected by MRI as an absence of ¯ow void and/or abnormal arterial enhancement (Figures 1, 2, 6, and 7). Normal rapid ¯ow in large arteries, such as the internal carotid or basilar arteries, typically produces a ¯ow void on standard spin echo MRI, especially when studied with long TE pulse sequences. The loss of a ¯ow void denotes thrombosis or abnormally slow ¯ow (Figures 6 and 7).5±9 Although angiography is necessary to discriminate slow ¯ow from thrombosis, the loss of ¯ow void on MRI suggests cerebral vascular disease. In certain cases, contrast enhanced MRI can further accentuate the underlying vascular ¯ow abnormalities. On T1-weighted spin echo images (TE 5 20 ms) obtained without ¯ow compensation, arterial enhancement usually does not occur after the intravenous injection of paramagnetic contrast agents. In vessels where there is abnormally slow ¯ow, caused either by

a high degree of stenosis or by occlusion without signi®cant collateral supply, enhancement of arterial structures may be evident (see Figures 1 and 2).2,5,6,10,11 Although the absence of a ¯ow void may represent either slow ¯ow or thrombosis, arterial enhancement has been shown to correlate well with angiographic evidence of slow ¯ow.12 In cortical ischemia, patients with arterial enhancement typically have more severe clinical symptoms and a poorer outcome, whereas absent or minimal symptoms are usually seen in patients with evidence of cerebral ischemia without arterial enhancement. Although arterial enhancement occurs early, it generally lasts only about 1 week and seldom persists beyond 11 days (Figure 8).6 This contrasts with absence of ¯ow void, which can be permanent.5,7±9 The disappearance of arterial enhancement probably indicates the reestablishment of rapid ¯ow in the involved vessels and coincides with the `luxury perfusion' demonstrated on computerized tomography or scintigraphy (occurring at 7± 14 days). The early disappearance of arterial enhancement within 1 or 2 days has been seen in patients with transient ischemic attacks and coincides with the rapid resolution of the

Table 1

Pathophysiological Mechanisms for MRI Findings in Ischemia

Mechanism Flow kinetics Biophysiological

MRI Findings Absent ¯ow void Arterial enhancement T1 morphological change (swelling) T2 signal change T1 signal change

Combination

Progressive parenchymal enhancementb Early exaggerated enhancementd

Possible causes Slow ¯ow; occlusion Slow ¯ow Cytotoxic edema (free water) BBB breakdown; vasogenic edema; macromolecular binding BBB breakdown; vasogenic edema; macromolecular binding Impaired delivery of signi®cant contrast agent Intact delivery of contrast agent; focal hyperemia

Estimated time (h)a Immediately Immediately 2±4 8 16±24 >120c 2±4

BBB, blood±brain barrier. Time at which ®ndings generally could ®rst be detected by available MRI examinations; this does not necessarily imply the exact time of onset. b Typical ®ndings in completed cortical infarctions. c Usually not detected before 5±7 days. d Found in cases with transient or partial occlusions and in watershed infarctions. (Reproduced by permission of the American Society of Neuroradiology from W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J. Neuroradiol., 1991, 12, 621) a

4 ISCHEMIC STROKE

Figure 6 Arterial thrombosis in a 50-year-old man with symptoms of brain stem ischemia. (a) Parasagittal T1-weighted image (SE350/26) obtained 4 hours after the onset showing a linear signal isointense to brain in the prepontine region (arrows) along the course of the basilar artery, suggesting an intraluminal clot. (b) Axial T2-weighted image (SE2000/100) showing the absence of a ¯ow void in the basilar artery (arrow). No apparent T2 signal abnormality is detected within the pons at this time. (c) Repeat axial T2-weighted image (SE2000/100) obtained 48 hours after (a) showing the interval development of ischemic changes in the pons. (Reproduced by permission of the American Society of Neuroradiology from W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J. Neuroradiol., 1991, 12, 565)

underlying proximal vascular disorder.6 Because watershed infarction usually has intact although insuf®cient blood ¯ow, arterial enhancement is seen infrequently. 4 BRAIN SWELLING Another early ®nding in ischemic stroke is brain swelling. Mass effect or brain swelling is probably caused by an abnormal accumulation of tissue water related to a complex combination of intra- and extracellular edema.5 This swelling can be recognized by the distortion of normal brain anatomy, with sulcal obliteration being the ®rst observable sign of cortical ischemia (see Figure 7). T1-Weighted images offer the best de®nition of anatomy with minimal interference from cerebrospinal ¯uid and are generally superior for recognition of early swelling. Brain swelling often occurs as early as 2 hours before the onset of T2 signal changes, presumably caused by cytotoxic edema (see Figures 3 and 7). This early (cytotoxic) edema primarily represents a shift of free water from the extracellular space into the intracellular space without an associated protein shift. The swelling may then progress over several days, mostly caused by the development of vasogenic (interstitial) edema, and is associated with an abnormal signal on long TE sequences.5,13±18 In the early phase of acute ischemia, the signal change due to vasogenic edema is generally not observed until 8 hours5 (see Figures 1, 3, and 6), and is not fully developed until 24 hours after the infarction.5,19 The fact that the

signal change usually occurs after tissue swelling has begun supports the hypothesis of two phases of edema (cytotoxic and vasogenic). In transient ischemic attacks, swelling can be seen transiently on T1-weighted images without the development of T2 hyperintensity.20 In fact, reversible ischemia seldom is associated with major T2-weighted parenchymal signal changes. The absence of a T2 signal change in cytotoxic edema is probably related not only to the small amount of free water shift (estimated at 3%) but also to the absence of a major change in the interactions between the water protons and the macromolecular proteins.5,15 Vasogenic edema, by comparison, is readily detected on T2-weighted images and usually becomes visible 8 hours after the onset of symptoms (see Figures 1, 2, and 6). It is associated with a signi®cant amount of water and protein shift (exudate) from the intravascular space to the extracellular space. Because signal changes usually do not occur until 2 hours after the onset of the blood±brain barrier breakdown, appreciable signal changes detected by MRI may require a gradual accumulation of a suf®cient amount of water in the extracellular space (the amount of water protons) as well as an alteration in the relaxation time of water molecules (the binding state of proton molecules). The maximal signal changes noted on T2-weighted images usually occur in 24±48 hours (see Figures 1, 5, and 6).5,21 Signal changes are usually best seen on standard T2weighted images. Abnormalities in the cortex or near the ventricle may be dif®cult to separate from the normal hyperintensity of adjacent cerebrospinal ¯uid. For this reason, the abnormal tissue signal may be easier to recognize on the

ISCHEMIC STROKE

Figure 7

(a,b)

5

6 ISCHEMIC STROKE

Figure 7 (c) Early development of brain swelling associated with complete ischemia in a 35-year-old woman. (a) Parasagittal T1weighted nonenhanced images (SE450/20) obtained 2.5 hours after the onset showing signals isointense to brain in the prepontine cistern (arrow). (b) Corresponding parasagittal T1-weighted contrast enhanced images (SE450/20) obtained 40 minutes after (a) showing development of massive brain swelling in the occipital lobes (asterisk) and cerebellum and progression of a blood clot in the basilar artery (arrows). (Reproduced by permission of the American Society of Neuroradiology from W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J. Neuroradiol., 1991, 12, 565)

®rst echo of a double-echo T2-weighted study, also referred to as a proton density weighted image.

5 ABNORMAL ENHANCEMENT FOLLOWING ADMINISTRATION OF A CONTRAST AGENT Abnormal enhancement of cerebral tissue with intravenously administered gadolinium after an ischemic insult is associated with a breakdown of the blood±brain barrier and/or a loss of arterial autoregulation.4±6 Parenchymal enhancement usually occurs late in complete ischemia. During the acute phase of complete ischemia, severe arterial obstruction coupled with de®cient collateral supply prevents both blood and contrast 20

Number of lesions

15

Positive (37) Negative (45)

10 5 0 –5 –10 <24 h

2

3

4–5

6–9 10–14 Time (days)

1 month

2 months

material from reaching the ischemic tissue (see Figures 1, 2, and 7). Therefore, no parenchymal enhancement is expected during the early phase of complete ischemia. By 5±7 days, the parenchymal enhancement will appear due to the reestablishment of the blood supply by recanalization of the occluded vessel and/or improved collateral circulation (see Figure 2). This reestablished blood supply will not only permit contrast material to reach the ischemic tissue, resulting in parenchymal enhancement, but also allow suf®ciently rapid ¯ow to cause the disappearance of arterial enhancement. This abnormal parenchymal enhancement will continue until the blood±brain barrier is reinstated with the maturation of the neovasculature, a process that normally takes several months after the infarction.20 Early enhancement can be seen in either incomplete ischemia or watershed infarction (see Figures 3±5). In incomplete ischemia, the intact delivery of contrast material may allow for the possibility of early parenchymal enhancement (see Figure 2). We have seen parenchymal enhancement within a few hours after the onset of symptoms. Such enhancement is most likely caused by a loss of autoregulation or hyperemia and typically resolves on subsequent scans performed 24 hours later. Minimal or no T2 signal abnormality develops. Similarly, in a watershed infarction (hypoperfusion with intact but inadequate blood supply), it is uncommon to see early parenchymal enhancement (see Figure 5). Unlike incomplete ischemia, however, the early enhancement observed in a watershed infarction does not indicate a favorable clinical outcome. The complex relationship between the abnormal MRI ®ndings, including extent of brain swelling (EBS), T2 signal changes (T2
values), arterial enhancement (AE), and parenchymal enhancement (PE), and clinical severity (CS) in acute ischemia generally can be expressed by the following formula: CS / T2
/ AE / EBS / 1=PE

…1†

This formula expresses the direct relationship between clinical severity, signal changes, extent of brain swelling, and arterial enhancement, which in turn have an inverse relationship to parenchymal enhancement. The two exceptions that do not apply to equation (1) are the relationship between clinical severity in a watershed infarction and arterial enhancement in a small noncortical infarction. In a watershed infarction there is intact delivery of the contrast agent as in incomplete ischemia; however, it is insuf®cient. Therefore, despite the fact that the MRI ®ndings suggest incomplete ischemia by equation (1), the prognosis and clinical severity are more like those of complete ischemia. The other exception is the lack of arterial enhancement in noncortical ischemia. Because the blood vessels involved in these types of infarction tend to be small, arterial enhancement usually is not seen. Nevertheless, the clinical severity of such cases may be quite grave.

6 weeks

Figure 8 Graph depicting the number of lesions that demonstrated arterial enhancement at the time of the initial MRI examination. Arterial enhancement frequently was detected within the ®rst 24 hours, rarely after 7 days, and not at all after 11 days, suggesting the reestablishment of the circulation or development of collateral circulation. (Reproduced by permission of the American Society of Neuroradiology from Crain et al.6)

6 6.1

NEW MAGNETIC RESONANCE MODALITIES Diffusion MRI

Recent advances in MRI techniques now provide new information directly related to the underlying pathophysiology at

ISCHEMIC STROKE

the cellular and microscopic levels and may allow the diagnosis of acute ischemia and estimation of infarction size. Diffusion imaging re¯ects abnormal water movement in acute ischemic tissues caused by the failure of the high-energy Na+± ATP pump and is sensitive to hyperacute ischemia within a few minutes after onset.22 In both animal models and clinical cases, early ischemic changes have been reported on diffusion imaging when standard T2-weighted imaging has shown no abnormalities.22,23 The abnormality in diffusion imaging is shown as a high signal intensity area. This increase in signal demonstrates a region of diminished net diffusion, which is dark on the apparent diffusion coef®cient (ADC) images. The ADC is obtained from an exponential ®t of the signal intensities of a series of diffusion imaging as a function of b factor. Diffusion abnormality can be identi®ed at high b values. ADC changes are a sensitive marker of ionic equilibrium because they show a function of intra±extracellular water homeostasis. In animal studies, the signal abnormality in diffusion imaging appears a few minutes after occlusion of the middle cerebral artery and enlarges over the next 24 hours.22,24 The abnormality in diffusion imaging is equivalent to that in T2weighted imaging by 24 hours after the onset of ischemia;25 however, cytotoxic and vasogenic edema may cause an overestimation of lesion volume. Kohno et al. reported regional correspondence of hyperintensity in diffusion imaging with perfusion de®cits at cerebral blood ¯ow (CBF) thresholds of 34 ml/min per 100 g tissue after 30 minutes and 41 ml/min per 100 g tissue after 2 hours of a major coronary artery occlusion.26 Their reports suggest that the threshold for ADC changes at a given CBF depending on the duration of ischemia. In clinical studies, diffusion imaging has also been reported to have a high sensitivity and speci®city for acute ischemia. LoÈvblad et al. reported that the sensitivity and speci®city of diffusion imaging in acute ischemic patients within 24 hours of onset were 88% and 95%, respectively.27 However, the pathophysiological changes in acute ischemia in humans are likely to be more heterogeneous than those found in animal models. Human subjects reportedly have two phases in the time course of ADC changes: a signi®cant reduction for 96 hours from stroke onset and an increasing trend from reduction to pseudonormalization to elevation at later subacute to chronic time points (> 7 days).28 Previous reports do not conclude whether the diffusion imaging signal changes are indicative of reversibly or irreversibly injured tissue. The diffusion abnormality in global ischemia is reversible by early reperfusion within 12 minutes.29 The regression of the diffusion abnormality was also reported in reperfusion models. Miyabe et al. indicated that if reperfusion occurred before ADC value decreased to approximately 70% or less of control values for 10±20 minutes, the ADC changes were usually reversible.30 Ueda et al. suggested that diffusion imaging had the highest sensitivity but was not as speci®c as the regional cerebral blood volume (rCBV) map in predicting acute ischemic injury and tended to overestimate infarction size in patients studied within 72 hours of stroke onset.31 In addition, 25% (4 out of 16 lesions) of ADC abnormalities were false positives or reversible ischemia. These results may support the suggestion that the diffusion abnormality indicates early changes of both reversible and irreversible ischemia.

6.2

7

Perfusion MRI

Perfusion imaging provides direct information related to a reduction in blood ¯ow that re¯ects the primary underlying pathophysiology of acute ischemia. Compared with the conventional spin-echo pulse sequence, echo-planar imaging is more susceptible to ®eld inhomogeneity and is more advantageous in the evaluation of the T2* effect. The combination of dynamic echo-planar T2*-weighted imaging and intravenous bolus injection of contrast material produces hemodynamic information such as mean transit time (MTT) and CBV. Although perfusion imaging cannot produce absolute values but only semiquantitative data in estimating MTT and CBV maps, it can be performed quickly and has proven essential in the emergent management of patients with acute ischemic stroke. Furthermore, higher temporal resolution and multisection images can be achieved by the echo-planar imaging techniques. In an animal model, perfusion imaging demonstrated diminished perfusion within minutes of arterial occlusion.32 Perfusion imaging shows a signal or a delay in peak signal loss in the vascular distribution when an artery is occluded. In clinical studies, perfusion imaging was reported to be superior to diffusion imaging in the assessment of hemodynamic changes of chronic cerebral hypoperfusion. Maeda et al. indicated that ischemic tissue with prolonged regional MTT (rMTT) and a marked decrease in rCBV tended to suffer irreversible damage.33 A mild decrease in rCBV with prolonged rMTT may suggest an area of reversible ischemia. A marked increase in rCBV may show the state of luxury perfusion in subacute ischemia. Combined diffusion and perfusion imaging can provide more important information in the management of patients with acute ischemic stroke. The volume of ischemic tissue demonstrated by both diffusion and perfusion MRI has been reported to have a high correlation with neurologic outcome as measured by the National Institutes of Health (NIH) stroke scale, and these imaging techniques, therefore, can play a useful prognostic role in acute ischemic stroke patients (Figure 9).34 Currently, the mismatch between diffusion and perfusion imaging in patients with acute ischemic stroke has been reported in several studies. Sorenson et al. studied patients within 10 hours of onset and showed that the abnormality in rMTT maps was larger than those in rCBV maps and diffusion imaging.35 Rordorf et al. demonstrated that diffusion lesion volumes were smaller than the volumes of rCBV map abnormality in patients studied within 12 hours of onset, and the early CBV abnormality was slightly better than the diffusion abnormality as a predictor of ®nal infarction size.36 RoÈther et al. suggested that it was possible to differentiate between severely ischemic tissue and peri-infarct parenchyma by rCBV maps in hyperacute ischemia.37 A recent report by Ueda et al. is similar to those reports in which the rMTT map overestimated the ®nal infarction volume and the rCBV map provided the best estimation but differed in that their diffusion imaging underestimated ®nal infarction volume.31 The size of the abnormality in diffusion and perfusion imaging depends on the imaging time from the onset of symptoms. Quantitative assessment of ischemic tissue viability and/or reversibility requires further study.

8 ISCHEMIC STROKE

Figure 9 MRI within 12 hours in a 50-year-old male with acute onset of left hemiparesis. (a) Diffusion-weighted imaging; (b) apparent diffusion coef®cient (ADC) map; (c) relative cerebral blood volume (rCBV); (d) relative mean transit time (rMTT) map. Both (a) and (b) demonstrate right hemisphere lesions in the middle cerebral artery distribution. The rMTT map (d) shows an area with hypoperfusion (arrow heads) that is much larger than that demonstrated by (a) and (b). The infarction core (arrow) may show the highest signal intensity (most severe hypoperfusion). In (c), the infarction core has a depletion of blood volume consistent with an inadequate collateral circulation (arrow)

6.3 MR Spectroscopy Clinical application of MR spectroscopy (MRS) in the diagnosis and management of acute stroke has not been well established. There are several possible causes for the underutilization of spectroscopy in patients with strokes. (1) Most MR centers do not have suf®cient scienti®c background and clinical expertise to facilitate such an application, and frequently avoid using such a technique. (2) The acquisition time for clinical MRS, which in the past has been long and not very reproducible, is not adequate for the management of acute stroke. (3) The diagnosis of acute stroke is usually quite

straightforward by the conventional clinical examination and radiological means. Consequently, clinical MRS is usually only applied for diagnosis of problematic cases. The capability of MRS in the assessment of tissue viability/reversibility has not been well established, although it may have great potential if the acquisition times can be shortened with improved techniques. MRS provides a means to assess the biochemical characteristics of brain disease through direct and noninvasive assay of cerebral metabolites.38,39 Practically, phosphorus and protons are being measured in clinical applications for central nervous system disease. In ischemic brain tissue, MRS provides infor-

ISCHEMIC STROKE

9

Figure 10 MRI examination of a 45-year-old patient with acute right side weakness: (a) T1-weighted image; (b) T2-weighted image. A rounded lesion at the left-periventricular region has hypointensity on the T1-weighted image (a) and hyperintensity on the T2-weighted image (b). This lesion does not demonstrate any contrast enhancement. Although the referring physician has high suspicion of acute stroke, the MR ®ndings are not characteristic of acute stroke and are more consistent with a mass lesion. Proton spectra were obtained from normal parenchyma on the contralateral side (c) and from the lesion (d). This lesion shows a large lactate (La) peak and a diminished N-acetyl aspartate (NAA) peak at 1.3 and 2.0 ppm, respectively (d), compared with the normal tissue (c). (Ch, choline; Cr, creatine.) Follow-up MR examination obtained several months later demonstrated atrophy and gliosis at the same location, consistent with chronic infarction. In this particular patient, spectroscopy was valuable not only because it enabled surgical biopsy to be avoided, but also because it allowed a correct diagnosis for acute stroke

mation about energy status and oxidative metabolites related to the underlying pathophysiology (ischemia). Proton MRS is more effective because hydrogen atoms are more abundant than phosphorus in the brain parenchyma and proton MRS is easier to perform with the clinical unit.

There are two types of approach: localized and spectroscopic imaging (MRSI). Localized proton MRS methods can be separated into those with long and short echo times. Long TE (135 or 270 ms) acquisitions generally have proved easiest to use in clinical practice. Localization methods commonly

10 ISCHEMIC STROKE

Figure 11 MRI examination 24 hours after a patient presented with acute stroke symptoms. (a) T1-Weighted and T2-weighted images were unremarkable. (b), Spectroscopic imaging of N-acetyl aspartate (NAA) did not demonstrate abnormality (area with decreased NAA). (c) Spectroscopy imaging of lactate showed an area of increased lactate in the right basal ganglia. (d) Follow-up spectroscopic imaging of NAA at 8 days demonstrated a focal area with decreased NAA at the right basal ganglia that was much smaller than the initial abnormalities demonstrated on the spectroscopy imaging of lactate (c) and was consistent with a focal infarction demonstrated on the T2-weighted imaging obtained at the same day (not shown). (e) Spectroscopy imaging of lactate at the same time again showed persistent elevations of lactate in the infarcted tissue. (Courtesy of Dr Nick Bryan, Diagnostic Radiology Department, NIH)

used in clinical proton MRS include depth-resolved surface coil spectroscopy (DRESS), point-resolved surface coil spectroscopy (SPARS), and the stimulated-echo method (STEAM).40±42 For practical purposes, STEAM allows for shorter echo times, thereby improving resolution of metabolites (e.g. myoinositol, glutamate, glutamine, and glycine).

Magnet designs have until recently favored the use of long TE sequences that are insensitive to eddy currents and can be easily implemented on the commercial scanner. Using long TE, the signal from most metabolites in the brain is lost except for that from four: choline, creatine, N-acetyl aspartate (NAA), and lactate. However, improvements in MRS techniques have

ISCHEMIC STROKE

allowed short TE sequences taking 10±15 minutes; consequently these may realistically be incorporated into a routine imaging study without a signi®cant time penalty. The characteristic ®ndings of a NAA and lactate peak can re¯ect the underlying pathophysiology of acute ischemia through a biochemical parameter. The NAA peak is attributable to its N-acetylmethyl group, which resonates at 2.0 ppm (Figure 10(c,d)). This peak also contains contributions from less-important N-acetyl groups. NAA is considered as a neuronal marker and is not present in tumors outside the central nervous system. Its concentration decreases with many diseases of the brain.43 Similar to NAA, glutamate and N-acetylaspartyl glutamate are also localized in neurons. Glutamate is an excitatory neurotransmitter that plays a role in mitochondrial metabolism.44 Glutamine plays a role in detoxi®cation and regulation of neurotransmitter activities. These two metabolites resonate closely together and they are commonly represented by their sum as peaks located between 2.1 and 2.5 ppm. Breakdown of N-acetylaspartyl glutamate releases both NAA and glutamate, and subsequent breakdown of NAA leads to aspartate. The compounds are excitatory amino acids that increase with ischemia and cause `toxic' effects, resulting in expanded tissue damage. Therefore, the concentrations of N-acetylaspartyl glutamate and glutamate may serve to monitor treatments designed to protect brain tissues by blocking excitatory amino acids. The lactate peak has a special con®guration and occurs at 1.32 ppm. It consists of two distinct, resonant peaks called a `doublet' and is caused by the magnetic ®eld interactions between adjacent protons (J coupling). Lactate levels of the normal brain parenchyma are low. The presence of lactate generally indicates that the normal cellular oxidative respiration mechanism is no longer in effect, and that carbohydrate catabolism is taking place.45 Con®rmation that a peak at 1.32 ppm corresponds to lactate may be obtained by altering the TE. At a TE of 272 ms, lactate projects above the baseline, whereas at a TE of 136 ms the lactate doublet is inverted below the baseline. In humans, proton MRS performed within the ®rst 24 hours after a stroke shows elevation of lactate, suggesting that anaerobic glycolysis is occurring as a result of ischemia (Figures 2 and 3).46 Decreased NAA can be seen as early as 4 days after acute ischemia, suggesting neuronal loss (infarction) (Figures 10 and 11).47 In chronic infarctions, there is a decrease in NAA, creatine, and choline, but no evidence of lactate.48 Experimentally, an increase in lactate may be detected after only 2 to 3 minutes of cerebral ischemia.49 In these animals, the lactate returned to normal when the underlying ischemia was reversed. In the evaluation of human hyperacute cerebral infarction, shortening of examination time as well as the better tolerance of the motion artifact will be needed in order to make spectroscopy a realistic tool in the management of the stroke patient. The age-related white matter changes contain normal levels of NAA and creatine50 but do not contain lactate. The increased choline levels are suggestive of an alteration of the white matter phospholipids.

7 CONCLUSIONS An understanding of the spectrum of MRI ®ndings in acute ischemia may facilitate a correct diagnosis of stroke, particularly in the hyperacute stage, and allow differentiation from

11

other etiologies. Recent advances in diffusion and perfusion MRI indicate a potential for providing important information concerning factors that determine tissue viability and/or reversibility; this will assist clinical decisions in selecting the appropriate patients for thrombolytic therapy.

8

RELATED ARTICLES

Anisotropically Restricted Diffusion in MRI; Brain MRS of Human Subjects; Brain MRS of Infants and Children; Diffusion: Clinical Utility of MRI Studies; Hemorrhage in the Brain and Neck Observed by MRI; ; Magnetic Resonance Imaging of White Matter Disease.

9

REFERENCES

1. Stroke update, US Department of Public Health, HHS, National Institutes of Health, 1989, p. 3. 2. S. Imakita, T. Nishimura, H. Naito, N. Yamada, K. Yamamoto, M. Takamiya, Y. Yamada, Y. Sakashita, J. Minamikama, and H. Kikuohi, Neuroradiology, 1987, 29, 422. 3. C. Virapongse, A. Mancuso, and R. Quisling, Radiology, 1986, 161, 785. 4. W. T. C. Yuh and M. R. Crain, Neuroimaging Clin. North Am., 1992, 2, 421. 5. W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J. Neuroradiol., 1991, 12, 621. 6. M. R. Crain, W. T. C. Yuh, G. M. Greene, D. J. Loes, T. J. Ryals, Y. Sato, and M. N. Hart, Am. J. Neuroradiol., 1991, 12, 631. 7. J. Biller, W. T. C. Yuh, G. W. Mitchell, A. Bruno, and H. P. Adams Jr., Stroke, 1988, 19, 297. 8. A. Uchino, T. Mori, and M. Ohno, Clin. Imaging, 1991, 15, 176. 9. J. I. Lane, A. E. Flanders, H. T. Doan, and R. D. Bell, Am. J. Neuroradiol., 1991, 12, 819. 10. A. D. Elster and D. M. Moody, Radiology, 1990, 177, 627. 11. S. Imakita, T. Nishimura, N. Yamada, H. Naito, M. Takamiya, Y. Yamada, J. Minamikama, H. Kikuchi, M. Nakamura, and T. Swada, Neuroradiology, 1988, 30, 372. 12. D. P. Mueller, W. T. C. Yuh, D. J. Fisher, K. B. Chandran, M. R. Crain, and Y.-H. Kim, Am. J. Neuroradiol., 1993, 14, 661. 13. M. Brant-Zawadzki, B. Pereira, P. Weinstein, S. Moore, W. Kucharczyk, T. Berry, M. McNamara, and N. Derugin, Am. J. Neuroradiol., 1986, 7, 7. 14. K. A. Hossmann and F. J. Schuier, Stroke, 1980, 11, 583. 15. F. J. Schuier and K. A. Hossmann, Stroke, 1980, 11, 593. 16. B. A. Bell, L. Symon, and N. M. Branston, J. Neurosurg., 1985, 62, 31. 17. O. Gotoh, T. Asano, T. Koide, and K. Takakura, Stroke, 1985, 16, 101. 18. M. Brant-Zawadzki, P. Weinstein, H. Bartkowski, and M. Moseley, Am. J. Neuroradiol., 1987, 8, 39. 19. R. N. Bryan, Radiology, 1990, 177, 615. 20. T. M. Simonson, T. J. Ryals, W. T. C. Yuh, G. P. Farrar, K. Rezai, and H. T. Hoffman, Am. J. Roentgenol., 1992, 159, 1063. 21. T. J. Masaryk, Curr. Opin. Radiol., 1992, 4, 79. 22. M. E. Moseley, Y. Cohen, J. Mintorovitch, L. Chileuitt, H. Shimizu, J. Kucharczyk, M. F. Wendland, and P. R. Weinstein, Magn. Reson. Med., 1990, 14, 330. 23. M. P. Marks, A. de Crespigny, D. Lentz, D. R. Enzmann, G. W. Albers, and M. E. Moseley, Radiology, 1996, 199, 403. 24. R. A. Knight, M. O. Dereski, J. A. Helpern, R. J. Ordidge, and M. Chopp, Stroke, 1994, 25, 1252. 25. I. Loubinoux, A. Volk, and J. Borredon, Stroke, 1997, 28, 419.

12 ISCHEMIC STROKE 26. K. Kohno, M. Hoehn-Berlage, G. Mies, T. Back, and K. A. Hossmann, Magn. Reson. Imag., 1995, 13, 73. 27. K. O. LoÈvblad, H. J. Laubach, A. E. Baird, F. Curtin, G. Schlaug, R. R. Edelman, and S. Warach, Am. J. Neuroradiol., 1998, 19, 1061. 28. G. Schlaug, B. Siewert, A. Ben®eld, R. R. Edelman, and S. Warach, Neurology, 1997, 49, 113. 29. D. Davis, J. Ulatowski, S. Eleff, M. Izuta, S. Mori, D. Shungu, and P. C. M. van Zijl, Magn. Reson. Med., 1994, 31, 454. 30. M. Miyabe, S. Mori, P. C. M. van Zijl, J. R. Kirsch, S. M. Eleff, R. C. Koehler, and R. J. Traystman, J. Cereb. Blood Flow Metab., 1996, 16, 881. 31. T. Ueda, W. T. C. Yuh, J. E. Maley, J. P. Quets, P. Y. Hahn, and V. A. Magnotta, Am. J. Neuroradiol., 1999, 20, 983. 32. D. A. Finelli, A. L. Hopkins, W. R. Selman, R. C. Crumrine, S. U. Bhatti, and W. D. Lust, Magn. Reson. Med., 1992, 27, 189. 33. M. Maeda, W. T. C. Yuh, T. Ueda, J. E. Maley, D. L. Crosby, M. W. Zhu, and V. A. Magnotta, Am. J. Neuroradiol., 1999, 20, 43. 34. D. C. Tong, N. A. Yenari, G. W. Albers, M. O'Brien, M. P. Marks, and M. E. Moseley, Neurology, 1998, 50, 864. 35. A. G. Sorensen, F. S. Buonanno, R. G. Gonzalez, L. H. Schwamm, M. H. Lev, F. E. Huang-Hellinger, T. G. Reese, R. M. Weisskoff, T. L. Davis, N. Suwanwela, U. Can, J. A. Moreira, W. A. Copen, R. B. Look, S. P. Finklestein, B. R. Rosen, and W. J. Koroshetz, Radiology, 1996, 199, 391. 36. G. Rordorf, W. J. Koroshetz, W. A. Copen, S. C. Cramer, P. W. Schaefer, R. F. Budzik, L. H. Schwamm, F. Buonanno, A. G. Sorenson, and G. Gonzalez, Stroke, 1998, 29, 939. 37. J. RoÈther, F. GuÈckel, W. Neff, A. Schwartz, and M. Hennerici, Stroke, 1996, 27, 1088. 38. B. Ross and T. Michaelis, Magn. Reson. Q., 1994, 10, 191. 39. M. Castillo, L. Kwock, S. K. Mukherji, Am. J. Neuroradiol., 1996, 17, 1. 40. P. A. Bottomley, T. B. Foster, and R. B. Darrow, J. Magn. Reson., 1984, 59, 338. 41. P. R. Luyten, A. J. H. Marien, B. Systma, et al., J. Magn. Reson., 1989, 9, 126. 42. J. Frahm, H. Bruhn, M. L. Gyngell, K. D. Merboldt, W. Hanicke, and R. Sauter, Magn. Reson. Med., 1989, 9, 79. 43. B. L. Miller, NMR Biomed., 1991, 4, 46. 44. M. S. van der Knapp, B. Ross, and J. Valk, in `Magnetic Resonance Neuroimaging', ed. J. Kucherarczyk, M. Mosely, A. J. Barkovich, CRC Press, Boca Raton, FL, 1994, pp. 245±318. 45. J. A. Sanders, in `Functional Brain Imaging', ed. W. W. Orrison, J. D. Lewine, J. A. Sanders, M. F. Harthshorne, Mosby, St Louis, MO, 1995, pp. 419±467. 46. P. B. Barker, J. H. Gillard, P. C. M. van Zijl, B. J. Seher, D. F. Hanley, A. M. Agildere, S. M. Oppenheimer, and R. N. Bryan, Radiology, 1994, 192, 723. 47. H. Bruhn, J. Frahm, M. L. Gyngell, K. D. Merboldt, W. Hanick, and R. Sauter, Magn. Reson. Med., 1989, 9, 126.

48. J. H. Duijn, G. B. Matson, A. A. Maudsley, J. W. Hugg, M. W. Weiner, Radiology, 1992, 183, 711. 49. K. L. Behar, J. A. den Hollander, M. E. Stromski, T. Ogino, R. G. Shulman, O. A. Petroff and J. W. Prichard, Proc. Natl. Acad. Sci. USA, 1983, 80, 4945. 50. Sappey-Marinier, G. Calabrese, H. P. Hetherington, S. N. Fisher, R. Deicken, C. Van Dyke, G. Fein, and M. W. Weiner, Magn. Reson. Med., 1992, 26, 313.

Biographical Sketches William T. C. Yuh. b 1947. B.S., 1971, Chiao-tung University, Taiwan. M.S.E.E., 1974, Auburn University. M.D., 1980, University of Alabama-Birmingham, USA. Internship, Lloyd Noland Hospital. Radiology residency, UCLA Medical Center. Nuclear medicine fellowship, V. A. Wadsworth-UCLA. Magnetic resonance fellowship, UCLA Medical Center. Instructor and assistant professor, neuroradiology fellowship, associate professor, professor, Department of Radiology, The University of Iowa, 1994±present. Approx. 150 publications. Research specialties: magnetic resonance contrast agents, central nervous system ischemia. Toshihiro Ueda. b 1960. M.D., 1987, Ehime University School of Medicine, Japan, Neurosurgery residency and fellowship at Ehime University School of Medicine, Ph.D., 1995, Postgraduate School of Ehime University School of Medicine. Associate (Staff Physician), Department of Neurosurgery, Ehime University School of Medicine, 1995±96. Research fellow, Department of Radiology, the University of Iowa, 1996±98. Visiting Assistant Professor, Department of Radiology, the University of Iowa, 1998±present. Approx. 40 publications. Research interests: cerebral ischemia, neurointervention, diffusion perfusion MRI, thrombolytic therapy. J. Randy Jinkins. b 1949. B.A., 1971, Biology, University of Texas, Austin, USA. M.D., 1975, University of Texas, Galveston, USA. Radiology Residency, Emory University, Atlanta, USA. Neuroradiology Fellowship, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA. Currently Associate Professor of Radiology, University of Texas, San Antonio, USA. Approx. 125 publications. Current research specialty: pathophysiologic aspects of disease as they pertain to medical neuroradiologic imaging. Ronald A. Rauch. b 1953. B.A. (Biochemistry), 1975, University of Kansas M.D., 1979, Baylor College of Medicine. Neurology Resident, Stanford, 1981±84. Radiology Resident, University of California, Irvine, 1985±88. Neuroradiology fellow, Long Beach Memorial, 1988±89, and UCLA, 1989±90. Currently assistant professor of radiology, University of Texas Health Science Center at San Antonio. Approx 20 publications. Current research interests: use of MRI to quantify corpus callosum morphology, MRI of white matter changes, especially those associated with dementia, and MRI of spondylolisthesis associated with spondylolysis.

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

Magnetic Resonance Imaging of White Matter Disease Donald M. Hadley Institute of Neurological Sciences, Glasgow, UK

1 INTRODUCTION 1.1 Characteristics The white matter of the brain constitutes the core of the hemispheres, brainstem, and cerebellum. It is composed of axons, which transmit chemically mediated electrical signals, and glial supporting cells set in a mucopolysaccharide ground substance. The glial cellsÐoligodendrocytes, astrocytes, ependymal cells, and microgliaÐaccount for about half the brain's volume and 80±90% of its cells. The oligodendrocytes provide an insulating sheath of myelin by invagination, wrapping concentric layers of their cell membrane around the axons. Astrocytes are now known to in¯uence and communicate through their long foot processes, which are in intimate contact with capillaries, neurones, synapses, and other astrocytes. The ependymal cells form the lining of the brain's internal cavities, while the microglia, normally relatively inconspicuous, are capable of enlarging and becoming active macrophages. The white matter ®bers are grouped into location-speci®c tracts, which can be divided into three main types: (a) projection ®bers that allow efferent and afferent communication between the cortex and target organ; (b) long and short association ®bers, which connect cortical regions in the same hemisphere; and (c) commissural ®bers, which connect similar cortical regions between hemispheres. The formation and maturation of axons has been reviewed by Barkovich et al.1 After development of the axons and their synapses, the ®nal process of myelination occurs. This is crucial to the appearance on MRI. 1.2 Evolution and Imaging The contrast obtained between gray and white matter is largely due to the myelination of the white matter tracts. Myelin is composed of a bilayer of lipids (phospholipids and glycolipids), cholesterol, and large proteins. In 1974, Parrish et al.2 showed differences in the relaxation times of gray and white matter by spectroscopy before imaging was possible. These differences were later con®rmed by MRI. In white matter, the myelin lipids themselves contain few mobile protons visible to routine MRI, but they are hydrophobic, and therefore, as myelination progresses, there is loss of brain water and a decrease in T2 signal. Cholesterol tends to have a short T1, and the increased protein also decreases the T1 of water. This results in

1

white matter having a reduced intensity on T2-weighted images and increased intensity on T1-weighted images compared with unmyelinated ®bers or gray matter. Myelination of the white matter is ®rst noted in the cranial nerves during the ®fth fetal month, and continues throughout life.3 By birth, myelination is present in the medulla, dorsal midbrain, cerebellar peduncles, posterior limb of the internal capsule, and the ventrolateral thalamus. In general, myelination is completed from caudal to cephalad, from central to peripheral, and from dorsal to ventral. Key landmarks include the pre- and post-central gyri, which are myelinated at one month, with the motor tracts completed by three months. At this time, myelination is completed in the cerebellum, and progresses through the pons in the corticospinal tracts, cerebral peduncles, the posterior limb of the internal capsule, and up to the central portion of the centrum semiovale, to be completed by six months. The optic radiations and the anterior limb of the internal capsule are myelinated by three months. Myelination in the subcortical white matter is ®rst noted in the occipital region at three months, and proceeds rostrally to the frontal lobes. This posterior-to-anterior maturation is noticeable in the corpus callosum, with the splenium ®rst showing myelin at four months, progressing to the genu, where it is complete at six months. This normal development is best visualized by MRI with age-related heavily T1-weighted (e.g., inversion±recovery) sequences for the ®rst six months, by which time the appearance is close to adult; after this, T2-weighted sequences are most helpful, with all the major tracts assuming an adult appearance by 18 months. The cause of these differences is not fully understood, but is thought to be related to the initial hydrophilicity, with its associated increase in hydrogen bonding with water. Next, the T2 shortening may be caused by the subsequent tightening of the myelin sheath, further redistributing the free water components. It has been shown that the very earliest changes of myelination are shown better by diffusion- or ¯uid-attenuated T2-weighted sequences than by conventional T1- or T2-weighted sequences.4,5 It must be noted, however, that some areas around the trigones of the lateral ventricles may not fully myelinate in normal children until they are 10 years old. 1.3

Classi®cation of Abnormalities

There are a bewildering number of white matter diseases with multiple etiologies and pathological mechanisms. Although MRI is very sensitive to any white matter abnormality, it is rarely possible for the radiologist to make a speci®c diagnosis.6 It is, however, useful to divide them into three main groups: (a) a dysmyelinating group in which there is a biochemical defect in the production or maintenance of normal myelin; some of the individual enzyme de®ciencies have been identi®ed and will be discussed below; (b) a demyelinating group in which myelin is formed normally but is later destroyed; (c) a vascular group in which normally myelinated white matter is destroyed by a critical reduction in blood ¯ow to a particular region; this may also involve the adjacent gray matter or a large segmental part of the brain, depending on

2 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

These are diseases where dysmyelination occurs as a result of the production and maintenance of abnormal myelin. Becker7 and Kendall8 have produced excellent reviews of this subject.

myelin remaining is normal. Changes become evident between one and six months old, although it is occasionally noted earlier, and leads to death within one to three years. The clinical diagnosis is based on an assay of -galactosidase from leukocytes or skin ®broblasts. Early in the disease process, MRI can be normal12 and a spectrum of lesions then develops over several months. These are nonspeci®c symmetrical patchy changes in the periventricular white matter similar to many other demyelinating diseases such as multiple sclerosis, with increased T1 and T2 signals.8,13 The thalami, central white matter, and cerebellar white matter may show decreased T1 and normal or slightly decreased T2 signals.14 These changes are re¯ections of the increased attenuation sometimes seen on computerized tomography (CT), and are probably the result of paramagnetics such as crystalline calci®cation. In advanced disease, there is diffuse cerebral atrophy.12

2.1.1

2.1.4

the extent and severity of the reduction in blood ¯ow and the susceptibility of the cells involved. In many of these conditions, the diagnosis is biochemical, but the radiologist has an important role in suggesting the diagnosis, and documenting progression, response to therapy, or complications.

2 DYSMYELINATION DISEASE 2.1 Leukodystrophies

Alexander's Disease (Fibrinoid Leukodystrophy)

This usually presents in the ®rst few weeks of life with macrocephaly and failure to attain developmental milestones. There is progressive spastic quadriparesis and intellectual failure. Death ensues in infancy or early childhood, although cases have been reported in adolescents and adults. An enzyme defect has not yet been identi®ed. MRI shows increased T1 and T2 relaxation times, starting in the frontal lobes, and progressing to the parietal and capsular regions.9,10 With the accumulation of Rosenthal ®bers around blood vessels, there may be disruption of the blood±brain barrier, producing frontal periventricular enhancement. Frank cystic changes develop in the frontal lobes in the later stages, with atrophy of the corpus callosum. 2.1.2

Canavan's Disease (Spongiform Leukodystrophy)

This is a lethal autosomal recessive neurodegenerative disorder of Jewish infants caused by a de®ciency of aspartoacylase. The disease progresses with marked hypotonia, macrocephaly, seizures, and failure to attain motor milestones in the ®rst few months of life, although sometimes it starts as early as a few days, progressing to spasticity, intellectual failure, and optic atrophy. Death usually occurs in the second year of life. The radiological features may be seen before the full clinical picture has developed, but the diagnosis depends on the biochemical testing. The demyelinated white matter shows increased T1 and T2 relaxation times, preferentially in the arcuate U ®bers of the cerebral hemispheres. The occipital lobes are more involved than the frontal, parietal, and temporal lobes. Initially, it may spare the corpus callosum, deep white matter, and internal and external capsules, but, as it progresses, diffuse white matter involvement occurs, which leads to eventual cortical atrophy.11 2.1.3

Krabbe's Disease (Globoid Cell Leukodystrophy)

This is a rare, lethal, autosomal recessive leukodystrophy (locus now mapped to chromosome 14) due to a de®ciency of the ®rst of the two galactocerebroside -galactosidases. This arrests the normal breakdown of cerebroside, disrupts the turnover of myelin, and results in the accumulation of galactosylsphingosine. This is toxic to oligodendrocytes, and causes a marked loss of myelin, although the minute amount of

Pelizaeus±Merzbacher Disease

This term has been used to cover the ®ve subtypes of sudanophilic leukodystrophy,15 but here it will be taken to mean the slowly progressive X-linked recessive leukodystrophy. The dysmyelination is now thought to be due to a point mutation in the PLP gene coding for the myelin±protein proteolipid protein. It presents in infancy, and runs a very chronic course leading to death in adolescence or early adulthood. On MRI, there is a general lack of myelination without white matter destruction. The brain has the appearance of the newborn, with high signal intensity only appearing in the internal capsule, optic radiations, and proximal corona radiata on T1-weighted images, and practically no low signal in the supratentorial region on T2-weighted sequences.16 A `tigroid' pattern consisting of normal myelinated white matter within diffuse dysmyelination can be seen later on T2-weighted sequences. When severe, there may be a complete absence of myelin. Cortical sulcal enlargement may be seen. 2.1.5

Metachromatic Leukodystrophy

The commonest of the sphingolipidoses is due to a de®ciency in the activity of arylsulfatase A. This enzyme is responsible for normal metabolism of sulfatides, which are important constituents of the myelin sheath. The disease is subdivided into: (a) neonatal, with a rapid downhill course leading to early death; (b) infantile, presenting between one and four years with polyneuropathy, ataxia, progressive retardation, and spastic tetraparesis; (c) juvenile, with dementia17 and behavioral disorders progressing to spastic tetraparesis; (d) the rare adult type, presenting at any age with dementia and spastic paraparesis. The imaging ®ndings are nonspeci®c, with symmetrical areas of increased T1 and T2 relaxation times in the centrum semiovale, representing progressive dysmyelination and gliosis within areas of normal myelination. The peripheral white matter, including the arcuate U ®bers, is spared until late in the disease. As there is no in¯ammation, enhancement is not a feature. These appearances allow differentiation from the gross lack of myelination seen in Pelizaeus±Merzbacher disease. As

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

3

the disease progresses, brain atrophy becomes more prominent than the white matter signal changes. Proton MRS may have a clinical role in the diagnosis.18 2.1.6

Adrenoleukodystrophy (Childhood Type: X-Linked)

This is seen exclusively in males, and was thought to be due to a de®ciency of acyl-CoA synthetase. Long-chain fatty acids are incorporated in cholesterol esters, replacing the normal nonesteri®ed cholesterol. It usually presents between the ages of ®ve and ten years, with a disturbance of gait and intellectual impairment, with fairly rapid progression and the development of hypotonia, seizures, visual impairment, and bulbar symptoms. Neurological complaints are classically preceded by adrenal insuf®ciency and skin pigmentation, which may be precipitated by an intercurrent infection; however, they sometimes may never appear. Symmetrical long T1 and T2 signals are usually ®rst seen in the peritrigonal regions extending into the splenium of the corpus callosum. Although typical, these may rarely be seen in other white matter diseases such as Krabbe's. These signal changes gradually extend to involve the occipital lobes and more anterior regions such as the medial and lateral geniculate bodies, thalami, and the inferior brachia. The pyramidal tracts and the occipito-temporo-parieto-pontine ®bers show progressive alteration, with sparing of the fronto-pontine ®bers in the medial part of the crus cerebri. The lateral lemnisci and the cerebellar white matter may also become involved. Although this is the usual pattern, atypical symmetrical or asymmetrical involvement of other lobes sometimes occurs19 (Figure 1). Three zones of abnormal long T1 and T2 signals can be recognized: (a) a central region of gliosis with necrosis and cavitation, next to (b) an intermediate region of active in¯ammatory demyelination that shows enhancement due to blood±brain barrier breakdown, surrounded by (c) a peripheral less marked zone of demyelination without in¯ammatory reaction. As the disease progresses, atrophy becomes the more dominant feature on MRI.8 2.1.7

Adrenoleukodystrophy (AdrenomyeloneuropathyÐAdult Type)

This often presents in the same family as the childhood type, but occurs in adult life. It is caused by a similar enzyme defect. The abnormal myelination is most marked in the corticospinal and spinocerebellar tracts, but can extend into the brainstem, involving the pyramidal tracts running into the posterior limb of the internal capsules, and the frontopontine and occipito-temporo-parieto-pontine ®bers. The cerebellar white matter is usually affected, with sparing of the cerebral white matter. MR changes usually appear late in the course of the disease. 2.1.8

Adrenoleukodystrophy (Neonatal)

Several disorders have been grouped under this heading, but with active research at present underway, the classi®cation of the enzyme defects may change. Severe progressive neurological impairment occurs, with psychomotor retardation,

Figure 1 Adrenoleukodystrophy: (a) T2-weighted and (b) gadoliniumenhanced T1-weighted sections of a ®ve-year-old boy showing bilateral focal areas of white matter abnormality with marginal enhancement (biopsy-proven)

dysmorphic facial features, hypotonia, seizures, and defective liver function. In contradistinction to the childhood type, these abnormalities are present from birth. The enzyme defect may be con®ned to fatty acyl-CoA oxidase resulting in defective very long chain fatty acid oxidation. There is diffuse degeneration of cerebral white matter, causing atrophy at a very early age. Progressive MRI changes have

4 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE been described20 in a single case followed for three years, with delayed myelination followed by symmetrical demyelination of the corona radiata, optic radiations, and pyramidal tracts. 2.1.9

Phenylketonuria

This is an autosomal recessive metabolic encephalopathy due to a defect in phenylalanine hydroxylase conversion of phenylalanine to tyrosine, resulting in hyperphenylalaninemia. Strict dietary restrictions must be maintained from early infancy to prevent profound mental retardation. Other cofactor defect variants may result in lesser or greater degrees of encephalopathy.21 When there is a defect of dihydropteridine reductase, there are severe neurological and cognitive abnormalities in spite of adequate dietary restrictions. Severe white matter changes have been noted,22 with cystic degeneration and loss of parenchyma. Subtle abnormalities of white matter have been shown by MRI in older children and adults who have classical phenylketonuria despite having maintained a degree of dietary restriction.23,24 This possibly provides some evidence for continuing the restrictive diet and phenylalanine-free protein supplements and into adulthood.25 Varying degrees of periventricular white matter abnormality have been shown, with focal and diffuse lengthening of T1 and T2 relaxation times most easily seen on the T2-weighted images (Figure 2). In some studies, these changes were found to correlate loosely with the adequacy of the reduction and maintenance of serum phenylalanine levels,25,26 while other workers showed no clear relationship.24

2.2

Miscellaneous

Over 600 individual dysmyelination disorders that may affect MRI appearances have been identi®ed in childhood alone. Continuing research is progressively isolating the individual enzyme or gene defects, which will in time allow more speci®c classi®cation. Meanwhile, the following less well de®ned groups of disorders will be considered. 2.2.1

Neurodegenerative Disorders

These occur in a number of devastating developmental disorders of childhood, which are either congenital or acquired. Clinical ®ndings are usually nonspeci®c, and laboratory tests have to be selected carefully. Imaging demonstrates the results of abnormal cellular function on parenchymal morphology. MRI is sensitive, but speci®city is limited, and it must be integrated with the other clinical ®ndings. Proton MRS may be able to detect abnormal metabolite levels and allow an earlier and more speci®c determination of neurodegeneration.27 2.2.2

Lysosomal Disorders

Several of these have been mentioned under speci®c enzyme defects above, such as metachromatic leukodystrophy and Krabbe's disease. All lack activity of a speci®c lysosome enzyme, which is inherited in an autosomal recessive manner except Hunter II, which is X-linked. Abnormal materials build up in the lysosomes. The CNS is affected directly or secondary to the metabolic abnormality in adjacent structures. Dysmyelination is shown as an increase in T1 and T2 relaxation times in the white matter, with a variable degree of involvement of the arcuate U ®bers. In Fabry's disease, involvement of the small arteries may cause multifocal small infarcts visible on MRI. The gangliosidosis in addition may show focal decreases of T2 relaxation time in the thalami, possibly re¯ecting the calci®cation seen on CT.7,8 2.2.3

Peroxisomal Disorders

These relate to de®ciencies in the activity of respiratory enzymes and organelles of most cells. In most of these diseases, the CNS is involved. Adrenoleukodystrophy has already been discussed above, but there are multiple other rarer diseases that belong to this classi®cation. In general, they produce dysmyelination. Several also show disturbances of neuronal migration. In some, there is an additional in¯ammatory response. 2.2.4

Figure 2 Phenylketonuria: T2-weighted section showing subtle white matter hyperintensities in the optic radiations in spite of apparently adequate dietary control in a 13-year-old

Mitochondrial Encephalopathies

These group of disorders are characterized by functionally or structurally abnormal mitochondria in the CNS or muscle. They are transmitted by non-Mendelian maternal inheritance, resulting in slowly progressive multisystem diseases with a wide range of clinical presentations, usually appearing in childhood but showing considerable variability depending on their severity. Imaging is nonspeci®c.28 There is diffuse but variable white matter atrophy and lengthened T1 and T2 signals in the basal ganglia. Focal infarcts may also be seen (Figure 3). Abnormal metabolites, including lactate, have been shown by MRS in the brain lesions. This is thought to be due to impaired aerobic metabolism of pyruvate.29,30 In Leigh's disease, there may be spongy degeneration with astroglial and microglial reaction, with vascular proliferation

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

5

affecting the basal ganglia, brainstem, and spinal cord. Cerebellar and cerebral white matter undergoes demyelination, with preservation of the nerve cells and axons resulting in hyperintensity and hypointensity on T2- and T1-weighted images, respectively. 3

DEMYELINATION DISEASE

3.1 3.1.1

Figure 3 Mitochondrial encephalopathy: (a) T2-weighted and (b) T1weighted sections in a nine-year-old boy showing increased T1 and T2 relaxation times representing focal infarction in the posterior limb of the left internal capsule and thalamus

Idiopathic Multiple Sclerosis

Multiple sclerosis (MS) is an idiopathic in¯ammatory and demyelinating disorder of the central nervous system (CNS). The de®nitive clinical and pathological features of the disease were established by Charcot31 as long ago as 1868. Since then, the disorder's characteristics have been re®ned, with improvements in imaging giving the most recent insights into its pathophysiology. It is now one of the commonest reasons given for requesting MRI in the northern latitudes of the Western world, and this diagnosis has huge social and economic consequences. It is therefore considered in some depth in the following paragraphs. Clinically, MS usually follows a ¯uctuating course, with symptoms varying from paroxysmal and brief to slowly progressive and chronic. The lesions affect single or multiple sites simultaneously, usually involving long white matter tracts, but clinical±pathological correlation is often poor. The disorder leads to visual loss, numbness, tingling in the limbs, spastic weakness, and ataxia.32 Diagnosis is allowed when a combination of signs and symptoms localize lesions in separate and distinct areas of the CNS disseminated in time and space.33 Supportive laboratory and imaging data can now be de®ned for research studies, dividing the disease into clinically de®nite and probable MS, with or without laboratory support.34,35 Although some CT studies using high-dose iodine-enhanced delayed imaging have reported sensitivities as high as 72% when patients are in an acute relapse,36 generally MRI has proved to have considerably greater sensitivity, and, by using intravenous gadolinium-based contrast agents, can separate acute from subacute and chronic lesions.37 Initially, MS lesions were shown at low ®eld on T1weighted (inversion±recovery) sequences,38 but within four years several rigorously controlled studies demonstrated the effectiveness of spin echo sequences where the T2-dependent contrast can be organized to maximize the sensitivity between normal and abnormal tissue while minimizing partial volume effects between cerebrospinal ¯uid (CSF) and adjacent lesions.39,40 Although no single sequence will detect all lesions, multifocal supratentorial white matter abnormalities have been shown on moderately T2-weighted images in 96.5% of a group of 200 consecutive patients with clinically de®nite MS.41 One or more periventricular lesions were seen in 98%, and lesions discrete from the ventricle in 92.5%, with cerebellar lesions in just over half of the group. Normal scans were found in 1.5% of patients. The majority of these lesions were clinically silent; therefore MRI can produce the extra information that helps to ful®ll the criteria of dissemination in space (Figure 4). It can also exclude other causes of the patients' signs and symptoms, such as Arnold Chiari malformations and spinocerebellar degeneration. Serial studies with careful repositioning may also

6 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

Figure 4 Multiple sclerosis: (a) T2-weighted, (b) proton density, and (c) T1-weighted sections showing the typical signal changes found in the multiple focal and coalescing acute and chronic plaques

ful®ll the criterion of dissemination in time by showing new, often asymptomatic, lesions42,43 This high sensitivity has now been con®rmed by many workers.36,42,44

The T1 of apparently normal white matter in patients with MS may be increased,42,45 and an apparent increase in the iron content has been found in the thalamus and striatum at high

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

7

Table 1 Differential Diagnosis of Multifocal White Matter Lesions Multiple sclerosis Aging, small vessel vascular disease, lacunar infarcts Infarction Acquired immune de®ciency syndrome Encephalitis (ADEM), (SSPE) Progressive multifocal leukoencephalopathy Metastases Trauma Radiation damage Granulomatous disease (e.g., sarcoid) Inherited white matter disease Normal (in healthy elderly, especially hypertensive) Hydrocephalus with CSF interstitial edema

®eld.46 Post-mortem studies have con®rmed that the long T2 lesions found correspond to MS plaques.42 This sensitivity makes MRI the most appropriate modality for examining a patient with suspected MS. Unfortunately, these multifocal white matter lesions may be indistinguishable from other conditions that produce demyelination, gliosis, or periventricular effusions47 (Table 1), and between 5 and 30% of apparently normal controls older than 50 years have been shown to have white matter lesions, probably due to asymptomatic cerebrovascular disease. The patient's age and pattern of lesions can help to improve the speci®city of the MRI examination.48 Fazekas et al.49,50 have shown that if at least three areas of increased T2 signal intensity are present with two of the following featuresÐabutting the body of the lateral ventricles, infratentorial location, and size greater than 5 mmÐthen the sensitivity is decreased to 88%, but the speci®city increases to 96% in patients with clinically de®nite multiple sclerosis. Although the relaxation times of acute plaques have been found to be longer,51 the range was wide, and the age of an individual lesion could not be determined by T1 or T2 measurement alone. It is important to be able to de®ne new and active lesions to differentiate between multiphasic (MS) and monophasic (ADEM) disease and to determine whether there is evidence of continuing disease progression (e.g., in clinical therapeutic trials). The areas of perivenular in¯ammation and edema associated with the acute MS plaque52 cause a transient disruption of the blood±brain barrier53,54 and allow leakage of intravascular contrast agents. This is shown as enhancement on T1-weighted images (Figure 5), and is safer and more effective than highdose delayed contrast-enhanced CT.53 Correlation with the lesions seen on T2-weighted sections is good. Only a small number of cortical or subcortical plaques were seen solely on enhanced T1-weighted images. Enhancement is now considered a consistent feature of recognizably new lesions or new parts of existing plaques,54 although occasionally blood±brain barrier breakdown develops in older previously nonenhancing plaques associated with no increase in their size. The in¯ammatory demyelination has been shown pathologically to begin in a perivenular distribution and spread centrifugally, corresponding with the ring enhancement noted in several studies37,54 (Figure 5). Gadolinium enhancement is particularly useful in the clear delineation of lesions in the spinal cord and optic nerves, especially if T1-weighted fat saturation chemical shift sequences are used to reduce the high signal from surrounding periorbital fat.55,56

Figure 5 Multiple sclerosis: gadolinium-enhanced T1-weighted sections showing (a) multiple enhancing acute lesions including a ring enhancing plaque and (b) a nonenhancing chronic cerebellar peduncular lesion

8 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE Differences in the enhancement pattern of primary and secondary progressive MS, the two major clinical patient groups, have been identi®ed.57 The secondary progressive group had more new lesions (18.2 lesions per patient per year), of which a larger proportion (87%) enhanced. In addition, there was enhancement at the edge of preexisting lesions. This compares with few new lesions (3.3 lesions per patient per year), of which only one enhanced in the primary progressive group at the time when there were no differences over six months in the rates of clinical deterioration between the two groups. This suggests a difference in the underlying dynamics of the in¯ammatory component of the disease. With improved imaging techniques, e.g., fast scanning methods, and particularly with real-time echo planar imaging,58 the complex morphology of the initial phase of gadolinium enhancement after intravenous injection may be further elucidated and related to the lesions on the unenhanced scan. Correlation with lipid imaging may allow study of the relationship between demyelination and in¯ammation. Advances in the use of MRI and MRS have been reviewed by Paty et al.59 Now that new and biologically active lesions can be identi®ed routinely in clinic patients by gadolinium-enhanced T1weighted scans, the association of blood±brain barrier leakage in some but not all of multiple plaques shown on T2-weighted images indicates the presence of dissemination in time60 and re®nes the diagnosis of MS. The method can be used to subdivide clinical groups, and will be useful in monitoring and possibly shortening the time required for therapeutic trials (e.g., with steroids61 and interferon62) in MS. 3.1.2

Schilder's Disease (Myelinoclastic Diffuse Sclerosis)

This is a rare but distinctive acute demyelinating condition, which can be de®ned by biochemical, pathological, and electrophysiological criteria, yet remains faithful to Schilder's original description of 1912.63 There is a severe selective in¯ammatory demyelination with sparing of the subcortical U ®bers, and extensive attempts at remyelination. MR white matter changes have been described,64 with bilateral involvement of the anterior hemispheres, extensive ¯uctuating increased relaxation times, mass effect, and varying partial ring enhancement indicating changes in blood±brain barrier breakdown. Other leukodystrophies with a similar appearance such as adrenoleukodystrophy and Pelizaeus±Merzbacher disease must be excluded by biochemical testing.

increased T1 and T2 relaxation time are found on MRI in both hemispheres, but the effects are usually asymmetrical.65 Although in the acute stage, the demyelinating lesions may enhance, the blood±brain barrier quickly returns to normal. As with MS, MRI is much more sensitive than CT at demonstrating these lesions. 3.2.2

Subacute sclerosing panencephalitis is a rare progressive demyelination resulting from reactivation of the measles virus due to a defect in immunity that allowed the virus to remain latent.66 There is a variable rate of progression, with death between two months and several years after reactivation. There is perivascular in®ltration by in¯ammatory cells, cortical and subcortical gliosis, and white matter demyelination progressing from occipital to the frontal lobes and from the cerebellum to the brainstem and spinal cord. This is re¯ected in an increase in T1 and T2 relaxation times of the multifocal patchy white matter lesions.67,68 3.2.3

3.2.1

ADEM

Acute disseminated encephalomyelitis is a demyelinating disease that is thought to be an immune-mediated disorder secondary to a recent viral infection or more rarely to vaccination. It has an acute onset and a monophasic course, in contradistinction to MS. Most patients make a complete recovery, with no neurological sequelae. This makes it one of the most important differential diagnoses in the acute clinical situation. Pathologically, there is a diffuse perivenous in¯ammatory process resulting in con¯uent areas of demyelination. These frequently occur at the corticomedullary junction, with gray matter much less often involved than white. Large areas of

PML

Progressive multifocal leukoencephalopathy is a demyelinating disease probably caused by the papova viruses (e.g., JC and SV40-PML). These are universal childhood infections that are reactivated in the immunosuppressed patient. It is characterized by demyelination, with abnormalities of oligodendrocytes in the white matter. Initially, the lesions are widely disseminated, but later tend to become con¯uent, producing large lesions. It is now seen with increasing incidence in patients with AIDS69 and those treated with immunosuppressive drugs. It most commonly involves the subcortical white matter of the posterior frontal and parietal lobes extending to the level of the trigones and occipital horns of the lateral ventricles (Figure 6). The lesions give an increased T2 and slightly increased T1 relaxation time, and because there is only occasional perivenular in¯ammation in the acute stage, they do not have mass effect and gadolinium enhancement is not usually a feature. These are useful differentiating features from lymphoma and toxoplasmosis.70

3.3 3.3.1

3.2 Postin¯ammatory: Viral, Allergic, Immune-mediated Responses to Previous Infection

SSPE

Posttherapy Disseminated Necrotizing Leukoencephalopathy

When patients are treated with intrathecal antineoplastic agents such as methotrexate for disseminated lymphoma, leukemia, or carcinomatosis, a necrotizing leukoencephalopathy can occur despite the fact that the agent does not usually cross the blood±brain barrier.71 Radiation therapy potentiates this neurotoxicity. There is endothelial injury, loss of oligodendroglial cells, coalescing foci of demyelination, and axonal swelling. The damaged endothelium responds by attempts at repair, resulting in hyalinization, ®brosis, and mineralization of the vessel walls. This causes a relative tissue ischemia, demyelination, and necrosis. On MRI,72 there is a diffuse increase in T1 and T2 relaxation times, with no mass effect and little or no gadolinium enhancement, re¯ecting only the edema and demyelination present.

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

9

occur. In the weeks or months following radiation, demyelination may occur. MRI may show focal or diffuse increases in T1 and T2 relaxation times in a patchy often periventricular distribution, which may be asymmetrical, affecting the white matter but generally sparing the corpus callosum, internal capsules, and basal ganglia.73,74 The gray matter is only involved in severe cases. Delayed effects can occur months to years after therapeutic irradiation. These are less common, and develop later with hyperfractionated doses. There is endothelial hyperplasia, ®brinoid necrosis of perforating arterioles, and thrombosis. Cerebral necrosis supervenes, with blood±brain barrier disruption, edema, and mass effect. MRI at this stage shows mass effect, increased T1 and T2 relaxation times, and enhancement after intravenous gadolinium. Therefore this cannot be differentiated from recurrent tumor with MRI. There is some evidence that measures of the lesion's metabolism with 18 F-deoxyglucose positron emission tomography, 201Tl or L-3123 [ I]iodo--methyltyrosine single photon emission computerized tomography will selectively differentiate the hypometabolic radiation necrosis from the hypermetabolic malignant tumor.75,76 A reversible acute cerebellar and cerebral syndrome has been reported77 after systemic high-dose cytarabine therapy used for treatment of postremission and refractory leukemia treatment. Diffuse patchy areas of increased intensity on T2weighted images were shown in the deep white matter of the frontal, occipital, and parietal lobes. Punctate enhancement was observed in the occipital lobes. Over a month, the symptoms and white matter abnormalities resolved. At post-mortem later, there was no evidence of white matter disease.

3.4 3.4.1

Toxic and Degenerative Disease Central Pontine Myelinolysis

In this condition, there is loss of myelin and oligodendroglia in the central pons, which may extend to the lateral thalamus and mesencephalon, sparing the ventrolateral longitudinal ®bers.78 It is usually associated with rapidly corrected hyponatremia, often in alcoholics.79,80 When there is only a tiny lesion or the patient is in coma due to the underlying disease process, it may be asymptomatic, but usually there is tetraparesis with a pseudobulbar palsy or a `locked-in' syndrome. Mild cases with full recovery have now been reported.81 The demyelination is depicted as T1 and T2 prolongation with no mass effect, and although there is a single report of ring enhancement, there is usually no blood±brain barrier breakdown.82 Figure 6 PML in AIDS: T2-weighted images showing (a) extensive subcortical demyelination in the trigonal and occipital regions and, in a different patient, (b) gross loss of white matter substance with periventricular coalescing hyperintensities

The patterns of injury to the white matter from radiation therapy on its own are divided into three stages: (a) acute, (b) early, and (c) delayed. In the ®rst days or weeks after therapy, vasogenic edema may be produced because of transient disruption of the blood±brain barrier, and some enhancement will

3.4.2

Marchia¯ava±Bignami Disease

This is characterized by a toxic demyelination of the corpus callosum in alcoholics.83,84 A rapidly fatal form and a more chronic form have been recognized. In the acute form, extensive lesions have been reported in the centrum semiovale and corpus callosum, while at the chronic stage only corpus callosum lesions are seen, and occasionally there is a favorable outcome.85 MRI shows these as small areas of increased T1 and T2 relaxation times, with no mass effect. Enhancement has not been reported.

10 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE 3.4.3

Carbon Monoxide Encephalopathy

As carbon monoxide binds to the hemoglobin molecule and displaces oxygen, it induces hypoxia and vulnerable cells are destroyed. Although the gray matter structures are damaged ®rst, the white matter is also involved, especially when there is episodic or chronic exposure. MRI shows areas of increased T1 and T2 relaxation time in the thalamus, basal ganglia, hippocampus, and centrum semiovale. These areas may show enhancement with gadolinium in the acute stage. The lesions are usually bilateral and symmetrical, but can be patchy. Laminar necrosis has been reported as high signal cortical foci on T2-weighted images.86 3.4.4

Substance Abuse

Inhalation of organic solvents and black market drugs such as heroin vapors (pyrolysate) and cocaine produce a wide variety of acute and chronic neurological signs and symptoms. The effects on the white matter will depend largely on the chemical constituent involved. Xiong et al.87 have shown that in toluene (one of the constituents of paint sprays) abuse, there is generalized cerebral, cerebellar, and corpus callosum atrophy, with a loss of gray± white matter contrast associated with diffuse multifocal hyperintensity of the cerebral white matter on T2-weighted sequences. Additionally, hypointensity of the thalami are also seen. Adulterated and synthetically produced drugs can produce severe leukoencephalopathy. Tan et al.88 reported on four patients who inhaled contaminated heroin vapor and developed extensive, symmetrical lesions of the white matter of the cerebrum, cerebellum, and midbrain. Selective involvement of the corticospinal tract and lemniscus medialis was also found. These have to be differentiated from the effects of cocaine abuse, where there is generally neurovascular damage with vasculitis, vasospasm, and thrombosis. Eventually, cerebral atrophy can be seen.89 3.4.5

Hypoxic±Ischemic Encephalopathy

This generally refers to brain damage in the fetus and infant. It may be focal or diffuse. When focal, it may be the cause of territorial infarcts such as can occur in cyanotic congenital heart disease when emboli can bypass the ®ltering effect of the lungs. This will be discussed in Section 3.4.6. In asphyxia, there is diffuse hypoxia hypercarbia, acidosis, and loss of the brain's normal vascular autoregulation, resulting in pressurepassive ¯ow and reduced perfusion. Capillary permeability is also altered. Sudden reperfusion of these weakened capillaries can result in rupture and intracerebral hemorrhage. The periventricular white matter is particularly susceptible, lying at the distal end of the supply zone of the long narrow centripetal arteries that run from the cerebral surface.90 When 100 high-risk neonates of different gestational ages were followed prospectively with MRI, CT, and ultrasound examinations,91,92 it was found that lesions associated with hypoxic±ischemic encephalopathy such as coagulative necrosis and germinal matrix hemorrhage were best shown on MRI. In the analysis, ultrasound showed 80%, while CT only showed 40% of those lesions depicted on MRI. This has raised interest in medico-legal circles, since the timing of the insult may be more clearly de®ned. The appearances of the brain damage on MRI can now give important

clues as to the time and nature of the asphyxia. Barkovich and Truwit93 found that when the asphyxia occurred before the 26th week of gestation, there was dilatation of the ventricles without any signal changes, whereas in older fetuses there was increasing periventricular gliosis. Both periventricular and more peripheral white matter gliosis with associated general atrophy were found in cases who had been partially asphyxiated, or where asphyxia had occurred near term or in postmature fetuses. Total asphyxia involves the deep gray matter nuclei and the brainstem, and presents a different pattern. These MRI patterns may have prognostic value, with initial studies94 reporting good correlations between imaging ®ndings at 8 months and neurodevelopmental outcome at 18 months. 3.4.6

Trauma (Contusions±Shear Injuries)

In children, the effects of asphyxia and mechanical trauma, be they accidental or nonaccidental, may initially produce the same imaging appearance, with generalized cerebral swelling resulting in blurring of the clear distinction between the gray matter and white matter boundaries, ventricular compression, and loss of CSF from the sulci and cisterns. This is due to a combination of edema and a failure of autoregulation producing an increase in cerebral blood volume. This can result in watershed ischemia and infarction, with eventual loss of white matter producing ventricular and sulcal dilatation. Focal cerebral contusions involve the gyral crests, and can extend into the subcortical and deeper white matter regions, depending on their severity. There is edema and petechial perivascular hemorrhage, but tissue integrity is largely preserved in small lesions. With more severe contusions, the petechial hemorrhages coalesce into focal hematomas, which have some space-occupying effect. These are well shown by MRI at all stages,95,96 although, in practice, CT is easier to perform and gives clinically adequate information in the acute stage. Blunt trauma resulting in sudden acceleration or deceleration of the skull, especially when this is rotational, sets up shear±strain deformation at the moment of impact in response to the inertial differences between tissues of different density and viscosity.97 This can cause immediate and irreversible structural damage to axons, and has been termed diffuse axonal injury.98±100 Diffuse axonal injury is a pathological diagnosis, and imaging may only show a few apparently focal lesions in the lobar white matter. It is, however, very important to recognize these as the `hallmark' of associated widespread microscopic axonal disruption. Although CT is the most commonly conducted examination in the acute situation, MRI is much more sensitive95,101 and essential when the clinical state is not explained by the CT imaging appearances (Figure 7). In the acute situation, foci of edema that may or may not contain macroscopic hematomas can be seen on T2-weighted MRI in the corpus callosum, the parasagittal frontal white matter close to the gray±white matter interface, the basal ganglial regions and the dorso-lateral quadrant of the rostral brainstem.96 At the subacute stage, hemorrhage will be better depicted on T1-weighted images, but both acutely and in the chronic phase T2* gradient echo sequences are most sensitive to deoxyhemoglobin and hemosiderin respectively. MRI can rarely appear entirely normal102 in severe diffuse axonal injury, and it is only on followup that the widespread white matter damage is re¯ected in atrophy with ventricular and cortical sulcal enlargement.100,103,104

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE

4 4.1

Figure 7 Head injury: T2* gradient echo sections in an 25-year-old unconscious patient with a normal CT scan. Multiple focal hypointensities represent hemorrhage at white/gray matter interfacesÐevidence of diffuse axonal injury

11

VASCULAR DISEASE Infarction

Stroke remains one of the commonest causes of hospital admission in the developed world, has a high morbidity and mortality, and consumes more healthcare resources than any other single disease. Ninety percent of cases are due to ischemia (10% to cerebral hemorrhage) from thrombosis of a nutrient artery, with only a small number due to emboli from the heart or other vessels. Despite treatment advances, the mortality remains at around 50%. It is thought that only preventative public health measures and earlier thrombolytic therapy can improve this situation.105 This requires the accurate early identi®cation of patients with acute infarcts and those with transient ischemia at risk of completing their infarcts. Routine unenhanced MRI can detect abnormalities within about 8 h of the onset of symptoms (although changes on MRI with a vessel occlusion stroke model were shown as early as 1±2 h without paramagnetic contrast),106,107 whereas CT is normal for at least 14 h and if perfusion is not re-established remains `bland' for several days. MRI initially shows subtle swelling and an increase in T1 and T2 signals due to failure of the `sodium±potassium' pump and increasing intracellular water±cytotoxic edema. At this stage, function is lost but structure is maintained. It is only with continuing ischemia that blood±brain barrier breakdown occurs, structural integrity is lost, and vasogenic edema supervenes. Although there are anecdotal reports of MRI-de®ned cytotoxic edema being reversed on treatment of ischemia in humans,108,109 and more rigorous demonstrations in cats using diffusion sequences,107 it has not been established in routine clinical practice whether these MRI changes, unlike those on CT, are reversible. Recent gadolinium-enhanced MRI studies110±112 of the ®rst 24±48 h after the ictus in the clinical population has shed light on this crucially important acute stage. Sato and colleagues113 studied six patients within 8 h and a further two between 8 and 26 h. They showed areas of cerebral ischemia/infarction using gadolinium-enhanced T1-weighted spin echo sequences. Abnormal curvilinear areas of enhancement thought to represent cortical arterial vessels with markedly slowed circulation were seen adjacent to affected brain. This tissue was shown to progress to frank infarction on follow-up CT and MRI. These features have been con®rmed and extended by the Iowa group.110,114 They demonstrated the vascular ¯ow-related abnormalities with absence of normal ¯ow voids and the presence of arterial enhancement detected within minutes of the onset of symptoms. Brain swelling on T1-weighted images without signal changes on T2-weighted images was detected within the ®rst few hours. In contrast to the usual absence of parenchymal enhancement typically found in cortical infarctions in the ®rst 24 h, a few lesions showed paradoxical early exaggerated enhancement. These were the transient or partial occlusions and isolated watershed infarcts. Longer term prospective observations through the ®rst fortnight have de®ned the subacute appearances.115 Three stages have been demonstrated: (a) vascular enhancementÐdays 1±3, seen in 77% of cases; (b) leptomeningeal enhancementÐdays 4±7, seen only in larger infarcts; (c) brain parenchymal enhancementÐdays 7±14, seen in 100% of cases studied.

12 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE Enhancement is not noted after two to three months. These changing patterns of enhancement re¯ect the underlying pathophysiology, and may have prognostic signi®cance; if this proves to be so then gadolinium enhancement will be crucial in the evaluation of early ischemia and its response to intervention. Once the parenchymal long T1 and T2 signal changes are established, the differential diagnosis must be made in a similar way to conventional CT, following consideration of (a) the site: vascular territory, watershed region, deep gray or white matter tracts; (b) the shape: wedge, involving gray and white matter with subtle bowing of interfaces; (c) the margins: sulcal effacement, blurring of the gray and white matter borders; (d) the degree of edema; (e) the sequence of resolution of mass effect over three to four weeks.116 Hypointensity on heavily T2-weighted sequences and the use of gradient echoes or susceptibility mapping can often show petechial hemorrhage in the second week that is not seen on CT. While frank hemorrhage correlates with a worsening clinical state, ®ne interstitial bleeding mainly due to diapedesis relates neither to anticoagulation nor to a poorer clinical condition.116 The patency of the extracranial and major intracerebral arteries can be assessed on routine MRI sequences as a `¯ow void' or with slower laminar ¯ow even as echo rephasing. At present, projectional images produced by time-of-¯ight and phase contrast angiographic sequences are being evaluated, and may yet replace preoperative conventional cerebral angiography.117,118 Both diffusion imaging119,120 and spectroscopy121 are being used experimentally in clinical populations to try to gain an understanding of microscopic water shifts as the different types of edema develop and to give an insight into the progressive cycles of bioenergetic exchange as oxidative metabolism breaks down in the ischemic brain cells.

4.2 Ischemic White Matter DiseaseÐNormal Aging Focal and con¯uent white matter abnormalities seen on MRI do not necessarily represent actual necrosis and infarction, but can be due to a spectrum of chronic cerebrovascular insuf®ciency.122 These merge with the changes found in as many as 30% of the normal aging population over 60 years of age who show no clinical cognitive de®cit, but which are seen with increasing frequency in patients with hypertension, diabetes mellitus, and coronary artery disease (Table 1).123 Dilated perivascular spaces give a CSF signal, are usually smaller than lacunar infarcts, and occur in typical locations in the base, deep white matter, and cortex of the brain. Gliosis may become more con¯uent around these vessels as the vascular insuf®ciency progresses and produces an increased signal on proton density images in addition to the increased T2 signal differentiating it from CSF. This has now been con®rmed microscopically. In a small post-mortem study,124 histological examination showed that the larger lesions were characterized centrally by necrosis, axonal loss, and demyelination, and therefore represent true infarcts. Reactive astrocytes oriented along the degenerated axons were identi®ed at distances of up to several

centimeters from the central infarct. This isomorphic gliosis shows hyperintensity on T2-weighted images, and increases the apparent size of the central lesion. Con®rmation has been provided by Munoz et al.,125 who investigated the pathological correlates of increased T2 signal in the centrum ovale in an unselected series of 15 post-mortems. On the basis of size, greater than and less than 10 mm, two types of lesion were described, namely, extensive and punctate. The extensive areas of hyperintensity on T2-weighted images were found to show myelin pallor that spared the subcortical U ®bers. There was diffuse vacuolation and reduction in glial cell density. The punctate abnormalities were less well de®ned, and were found to be due to dilated Virchow±Robin spaces. The white matter changes seen on MRI are therefore nonspeci®c, and although seen with increased frequency in ischemic brains, there is often little or no correlation with the clinical state in the elderly patient. 5

CONCLUSIONS

Over the last 15 years, MRI has become the main diagnostic tool in the investigation of white matter disease. In some conditions, its sensitivity is the key to selecting patients for further attention, while in others it identi®es more speci®c features that in turn lead to further laboratory investigations leading to a ®nal diagnosis. MRI can be used to select patients for treatment and to monitor the effects of this treatment. The implementation of new sequences, faster scanning techniques, and better patient±machine ergonomics will ensure the preemptive position of MRI for the investigation of white matter diseases for the foreseeable future. 6

RELATED ARTICLES

Brain MRS of Infants and Children; Brain Neoplasms Studied by MRI; Diffusion: Clinical Utility of MRI Studies; EchoPlanar Imaging; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Hemorrhage in the Brain and Neck Observed by MRI; Intracranial Infections; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions. 7

REFERENCES 1. A. J. Barkovich, G. Lyon, and P. Evrard, Am. J.N.R., 1992, 13, 447. 2. R. G. Parrish, J. R. Kurland, W. W. Janese, and L. Bakay, Science (Washington, DC), 1974, 483, 349. 3. A. J. Barkovich and T. V. Maroldo, Top. Magn. Reson. Imaging, 1993, 5, 96. 4. Y. Nomura, H. Sakuma, K. Takeda, T. Tagami, Y. Okuda, and T. Nakasasa, Am. J.N.R., 1994, 15, 231. 5. A. Oatridge, J. V. Hajnal, F. M. Cowan, C. J. Baudouin, I. R. Young, and G. R. Bydder, Clin. Radiol., 1993, 47, 82. 6. B. E. Kendall, J. Inherited Metab. Dis., 1993, 16, 771. 7. L. E. Becker, Am. J.N.R., 1992, 13, 609. 8. B. E. Kendall, Am. J.N.R., 1992, 13, 621. 9. G. B. Bobele, A. Garnica, G. B. Schaefer, J. C. Leonard, D. Wilson, W. A. Marks, R. W. Leech, and R. A. Brumback, J. Child. Neurol., 1990, 5, 253.

MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE 10. T. Ichiyama, T. Hayashi, and T. Ukita, Brain Dev., 1993, 15, 153. 11. J. Brismar, G. Brismar, G. Gascon, and P. Ozand, Am. J.N.R., 1990, 11, 805. 12. D. A. Finelli, R. W. Tarr, R. N. Sawyer, and S. J. Horwitz, Am. J.N.R., 1994, 15, 167. 13. T. J. Farley, L. M. Ketonen, J. B. Bodensteiner, and D. D. Wang, Pediatr. Neurol., 1992, 8, 455. 14. S. Choi and D. R. Enzmann, Am. J.N.R., 1993, 14, 1164. 15. J. C. Koetsveld-Baart, I. E. Glaudemans-van-Gelderen, J. Valk, and P. G. Barth, Ned. Tijdschr. Geneeskd., 1993, 137, 2494. 16. M. Ishii, J. Takanashi, K. Sugita, A. Suzuki, M. Goto, Y. Tanabe, K. Tamai, and H. Niimi, No To Hattatsu, 1993, 25, 9. 17. E. G. Shapiro, L. A. Lockman, D. Knopman, and W. Krivit, Neurology, 1994, 44, 662. 18. B. Kruse, F. Hanefeld, H. J. Christen, H. Bruhn, T. Michaels, W. Hanicke, J. Frahm, J. Neurol., 1993, 241, 68. 19. P. J. Close, S. J. Sinnott, and K. T. Nolan, Pediatr. Radiol., 1993, 23, 400. 20. M. S. van der Knaap and J. Valk, Neuroradiology, 1991, 33, 30. 21. J. Brismar, A. Aqeel, G. Gascon, and P. Ozand, Am. J.N.R., 1990, 11, 135. 22. R. Sugita, I. Takahashi, K. Ishii, K. Matsumoto, T. Ishibashi, K. Sakamoto, and K. Narisawa, J. Comput. Assist. Tomogr., 1990, 14, 699. 23. D. W. W. Shaw, K. R. Maravilla, E. Weinberger, J. Garretson, C. M. Trahms, and C. R. Scott, Am. J.N.R., 1991, 12, 403. 24. K. D. Pearsen, A. D. Gean-Marton, H. L. Levy, and K. R. Davis, Radiology, 1990, 177, 437. 25. A. J. Thompson, I. Smith, D. Brenton, B. D. Youl, G. Rylame, D. C. Davidson, B. Kordall, and A. J. Lees, Lancet, 1990, 336, 602. 26. A. J. Thompson, S. Tillotson, I. Smith, B. Kendall, S. G. Moore, and D. P. Brenton, Brain, 1993, 116, 811. 27. A. A. Tzika, W. S. Ball Jr., D. B. Vigneron, R. S. Dunn, and D. R. Kirks, Am. J.N.R., 1993, 14, 1267. 28. A. J. Barkovich, W. V. Good, T. K. Koch, and B. O. Berg, Am. J.N.R., 1993, 14, 1119. 29. P. M. Matthews, F. Andermann, K. Silver, G. Karpati, and D. L. Arnold, Neurology, 1993, 43, 2484. 30. B. Barbiroli, P. Montagna, P. Martinelli, R. Lodi, S. Lotti, P. Cortelli, R. Fanicello, and P. Zaniol, J. Cereb. Blood Flow Metab., 1993, 13, 469. 31. J. M. Charcot, Gaz. des HoÃp. Civ. Mil., Paris, 1868, 41, 554. 32. W. I. McDonald and D. H. Silberberg, `Multiple Sclerosis', Butterworths, London, 1986. 33. G. A. Schumacher, G. Beebe, R. F. Kibler, L. T. Kurland, J. F. Kurtzke, F. McDowell, B. Nagler, W. A. Sibley, W. W. Tourtellotte, and T. L. Willmon, Ann. NY Acad. Sci., 1965, 122, 552. 34. C. M. Poser, D. W. Paty, L. Scheinberg, W. I. MacDonald, F. A. Davis, G. C. Ebers, K. P. Johnson, W. A. Sibley, D. H. Silberberg, and W. H. Tourtellote, Ann. Neurol., 1983, 13, 227. 35. A. K. Asbury, R. M. Herndon, H. F. McFarland, W. I. McDonald, W. J. McIlroy, D. W. Paty, J. W. Prineas, L. C. Scheinberg, and J. S. Wolinsky, Neuroradiology, 1987, 29, 119. 36. D. W. Paty and D. K. B. Li, in `Clinical Neuroimaging', ed. W. H. Theodore, Alan R. Liss, New York, 1988, Vol. 4, Chap. 10. 37. R. I. Grossman, B. H. Braffman, J. R. Brorson, H. I. Goldberg, D. H. Silberberg, and F. Gonzalez-Scarano, Radiology, 1988, 169, 117. 38. I. R. Young, A. S. Hall, C. A. Pallis, N. J. Legg, G. M. Bydder, and R. E. Steiner, Lancet, 1981, ii, 1063.

13

39. S. A. Lukes, L. E. Crooks, M. J. Aminoff, L. Kaufman, H. S. Panitch, C. Mills, and D. Norman, Ann. Neurol., 1983, 13, 592. 40. I. E. C. Ormerod, G. H. du Boulay, and W. I. McDonald, in `Multiple Sclerosis', ed. W. I. McDonald and D. H. Silberberg, Butterworths, London, 1986. 41. D. H. Miller, MRI Decis., 1988, 2, 17. 42. I. E. C. Ormerod, D. H. Miller, W. I. McDonald, E. P. G. H. du Boulay, P. Rudge, B. E. Kendall, I. F. Moseley, G. Johnson, P. S. Tofts, and A. N. Halliday, Brain, 1987, 110, 1579. 43. D. W. Paty, Can. J. Neurol. Sci., 1988, 15, 266. 44. D. W. Paty, J. J. F. Oger, L. F. Kastrukoff, S. A. Hashimoto, J. P. Haage, A. A. Eisen, K. A. Eisen, S. T. Purves, M. D. Low, and V. Brandejs, Neurology, 1988, 38, 180. 45. D. Lacomis, M. D. Osbakken, and G. Gross, Magn. Reson. Med., 1986, 3, 194. 46. B. P. Drayer, P. Burger, B. Hurwitz, D. Dawson, and J. Cain, Am. J.N.R., 1987, 8, 413. 47. D. H. Miller, I. E. C. Ormerod, A. Gibson, E. P. G. H. du Bouley, P. Rudge, and W. I. McDonald, Neuroradiology, 1987, 29, 226. 48. F. Z. Yetkin, V. M. Haughton, R. A. Papke, M. E. Fischer, and S. M. Rao, Radiology, 1991, 178, 447. 49. F. Fazekas, H. Offenbacher, S. Fuchs, R. Schmidt, K. Niederkorn, S. Horners, and H. Lechner, Neurology, 1988, 38, 1822. 50. H. Offenbacher, F. Fazekas, R. Schmidt, W. Freidl, E. Floch, F. Payer, and H. Lechner, Neurology, 1993, 43, 905. 51. I. E. C. Ormerod, A. Bronstein, P. Rudge, G. Johnson, D. G. P. MacManus, A. M. Halliday, H. Barratt, E. P. du Boulay, B. E. Kendall, and I. F. Moseley, J. Neurol. Neurosurg. Psychiatry, 1986, 49, 737. 52. J. Prineas, Hum. Pathol., 1975, 6, 531. 53. R. I. Grossman, F. Gonzalez-Scarano, S. W. Atlas, S. Galetta, and D. H. Silberberg, Radiology, 1986, 161, 721. 54. H. Miller, P. Rudge, B. Johnson, B. E. Kendall, D. G. MacManus, I. F. Moseley, D. Barnes, and W. I. McDonald, Brain, 1988, 111, 927. 55. E.-M. Larsson, S. Holas, and O. Nilsson, Am. J.N.R., 1989, 10, 1071. 56. S. F. Merandi, B. T. Kudryk, F. R. Murtagh, and J. A. Arrington, Am. J.N.R., 1991, 12, 923. 57. A. J. Thompson, A. J. Kermode, D. Wicks, D. G. MacManus, B. E. Kendall, D. P. Kingley, and W. I. McDonald, Ann. Neurol., 1991, 29, 53. 58. M. K. Stehling, P. Bullock, J. L. Firth, A. M. Blamire, R. J. Ordidge, B. Coxon, P. Gibbs, and P. Mans®eld, Proc. VIIIth Ann Mtg. Soc. Magn. Reson. Med., Amsterdam, 1989, p. 358. 59. D. W. Paty, Curr. Opin. Neurol. Neurosurg., 1993, 6, 202. 60. R. Heun, L. Kappos, S. Bittkau, D. Staedt, E. Rohrbach, and B. Schuknecht, Lancet, 1988, ii, 1202. 61. M. J. Kupersmith, D. Kaufman, D. W. Paty, G. Ebers, M. McFarland, K. Johnson, J. Reingold, and J. Whitaker, Neurology, 1994, 44, 1. 62. D. W. Paty and D. K. Li, Neurology, 1993, 43, 662. 63. P. Schilder, Z. Gesamte Neurol. Psychiatr., 1912, 10, 1. 64. M. F. Mehler and L. Rabinowich, Am. J.N.R., 1989, 10, 176. 65. S. W. Atlas, R. I. Grossman, H. I. Goldberg, D. B. Hackney, L. T. Bilanuck, and R. A. Zimmerman, J. Comput. Assist. Tomogr., 1986, 10, 798. 66. A. J. Barkovich, `Pediatric Neuroimaging', Raven Press, New York, 1996, p. 597. 67. R. Murata, H. Hattori, O. Matsuoka, T. Nakajima, and H. Shintaku, Brain. Dev., 1992, 14, 391.

14 MAGNETIC RESONANCE IMAGING OF WHITE MATTER DISEASE 68. S. Yagi, Y. Miura, S. Mizuta, A. Wakunami, N. Kataoka, T. Morita, K. Morita, S. Ono, and M. Fukunaga, Brain. Dev., 1993, 15, 141. 69. A. S. Mark and S. W. Atlas, Radiology, 1989, 173, 517. 70. L. Ketonen and M. Tuite, Semin. Neurol, 1992, 12, 57. 71. F. Ebner, G. Ranner, I. Slavc, C. Urban, R. Kleinert, H. Roulner, R. Ernspieler, and E. Justich, Am. J.N.R., 1989, 10, 959. 72. R. Asato, Y. Akiyama, M. Ito, M. Kubota, R. Okumura, Y. Miki, J. Konishi, H. Mikaua, Cancer, 1992, 70, 1997. 73. J. T. Curnes, D. W. Laster, M. R. Ball, T. D. Koubek, D. M. Moody, and R. L. Witcofski, Am. J.N.R., 1986, 7, 389. 74. W. J. Curran, C. Hecht-Leavitt, L. Schut, R. A. Zimmerman, and D. F. Nelson, Int. J. Radiat. Oncol. Biol. Phys., 1987, 13, 1093. 75. R. B. Schwartz, P. A. Carvalho, E. Alexander III, J. S. Loef¯er, R. Folkerth, and B. L. Holman, Am. J.N.R., 1991, 12, 1187. 76. Karl-J. Langen, H. H. Coenen, N. Roosen, P. Kling, O. Muzik, H. Herzog, T. Kuwort, G. Stocklin, and L. E. Femendegen, J. Nucl. Med., 1990, 31, 281. 77. D. J. Vaughn, J. G. Jarvik, D. Hackney, S. Peters, and E. A. Stadtmauer, Am. J.N.R., 1993, 14, 1014. 78. Y. Korogi, M. Takahashi, J. Shinzato, Y. Sakamoto, K. Mitsuzaki, T. Hirai, and K. Yoshizumi, Am. J.N.R., 1993, 14, 651. 79. M. Mascalchi, M. Cincotta, and M. Piazzini, Clin. Radiol., 1993, 47, 137. 80. R. D. Laitt, M. Thornton, and P. Goddard, Clin. Radiol., 1993, 48, 432. 81. V. B. Ho, C. R. Fitz, C. C. Yoder, and C. A. Geyer, Am. J.N.R., 1993, 14, 163. 82. K. J. Koch and R. R. Smith, Am. J.N.R., 1989, 10, S58. 83. M. E. Charness, Alcohol Clin. Exp. Res., 1993, 17, 2. 84. P. Tomasini, D. Guillot, P. Sabbah, C. Brosset, P. Salamand, and J. F. Briant, Ann. Radiol. (Paris), 1993, 36, 319. 85. S. Canaple, A. Rosa, and J. P. Mizon, Rev. Neurol. (Paris), 1992, 148, 638. 86. A. L. Horowitz, R. Kaplan, and G. Sarpel, Radiology, 1987, 162, 787. 87. L. Xiong, J. D. Matthes, J. Li, and R. Jinkins, Am. J.N.R., 1993, 14, 1195. 88. T. P. Tan, P. R. Algra, J. Valk, and E. C. Wolters, Am. J.N.R., 1994, 15, 175. 89. E. Brown, J. Prager, H. Y. Lee, and R. G. Ramsey, Am. J. Roentgenol., 1992, 159, 137. 90. M. D. Nelson, I. Gonzalez-Gomez, and F. H. Gilles, Am. J.N.R., 1991, 12, 215. 91. S. E. Keeney, E. W. Adcock, and C. B. McArdle, Pediatrics, 1991, 87, 421. 92. S. E. Keeney, E. W. Adcock, and C. B. McArdle, Pediatrics, 1991, 87, 431. 93. A. J. Barkovich and C. L. Truwit, Am. J.N.R., 1990, 11, 1087. 94. P. Byrne, R. Welch, M. A. Johnson, J. Darrah, and M. Piper, J. Pediatr., 1990, 117, 694. 95. E. Teasdale and D. M. Hadley, in `Handbook of Clinical Neurology, 2nd Series: Head Injury', ed. R. Braakman, Elsevier, Amsterdam, 1990, Vol. 13, Chap. 7. 96. D. M. Hadley, Curr. Imaging, 1991, 3, 64. 97. A. H. S. Holbourn, Lancet, 1943, ii, 438. 98. J. H. Adams, D. I. Graham, L. S. Murray, and G. Scott, Ann. Neurol., 1982, 12, 557. 99. T. A. Gennarelli, G. M. Spielman, T. W. Lang®tt, P. L. Gildenberg, T. Harrington, J. A. Jane, L. F. Marshall, J. D. Miller, and L. H. Pitts, J. Neurosurg., 1982, 56, 26.

100. A. D. Gean, `Imaging of Head Trauma', Raven Press, New York, 1994. 101. A. Jenkins, G. M. Teasdale, D. M. Hadley, P. Macpherson, and J. O. Rowan, Lancet, 1986, ii, 445. 102. D. M. Hadley, P. Macpherson, D. A. Lang, and G. M. Teasdale, Neuroradiology, 1991, 33, 86. 103. K. D. Wiedmann, J. T. L. Wilson, D. Wyper, D. M. Hadley, G. M. Teasdale, and D. N. Brooks, Neuropsychology, 1990, 3, 267. 104. J. T. L. Wilson, K. D. Wiedmann, D. M. Hadley, B. Condon, G. M. Teasdale, and J. D. N. Brooks, J. Neurol. Neurosurg. Psychiatry, 1988, 51, 391. 105. C. D. Forbes, Scot. Med. J., 1991, 36, 163. 106. M. Brant-Zawadzki, B. Pereira, P. Weinstein, S. Moore, W. Kusharczyk, I. Berry, M. McNamara, and N. Derugin, Am. J.N.R., 1986, 7, 7. 107. M. E. Moseley, Y. Cohen, J. Mintorovitch, L. Chileuitt, H. Shimizer, W. Kueharczyk, M. F. Wendland, and P. R. Weinstein, Magn. Reson. Med., 1990, 14, 330. 108. A. M. Aisen, T. O. Gabrielsen, and W. J. McCune, Am. J.N.R., 1985, 6, 197. 109. W. G. Bradley, Neurol. Res., 1984, 6, 91. 110. W. T. C. Yuh, M. R. Crain, D. J. Loes, G. M. Greene, T. J. Ryals, and Y. Sato, Am. J.N.R., 1991, 12, 621. 111. S. Warach, W. Li, M. Ronthal, and R. R. Edelman, Radiology, 1992, 182, 41. 112. R. N. Bryan, L. M. Levy, W. D. Whitlow, J. M. Killian, T. J. Preziosi, and J. A. Rosario, Am. J.N.R., 1991, 12, 611. 113. A. Sato, S. Takahashi, Y. Soma, K. Ishii, T. Watanabe, and K. Sakamoto, Radiology, 1991, 178, 433. 114. M. R. Crain, W. T. C. Yuh, G. M. Greene, D. J. Loes, T. J. Ryals, Y. Sato, and M. N. Hart, Am. J.N.R., 1991, 12, 631. 115. A. D. Elster and D. M. Moody, Radiology, 1990, 177, 627. 116. W. G. Bradley, in `MRI Atlas of the Brain', eds. W. G. Bradley and G. Bydder, Martin Dunitz, London, 1990, Chap. 5. 117. T. J. Masaryk, G. A. Laub, M. T. Modic, J. S. Ross, and E. M. Haacke, Magn. Reson. Med., 1990, 14, 308. 118. A. W. Litt, Am. J.N.R., 1991, 12, 1141. 119. M. Doran and G. M. Bydder, Neuroradiology, 1990, 32, 392. 120. R. M. Henkelman, Am. J.N.R., 1990, 11, 932. 121. M. Brant-Zawadzki, P. R. Weinstein, H. Bartkowski, and M. Moseley, Am. J. Roentgenol., 1987, 148, 579. 122. M. L. Bots, J. C. van-Swieten, M. M. Breteler, P. T. de-Jong, J. Van Gijn, A. Hofman, and D. E. Grobbee, Lancet, 1993, 341, 1232. 123. T. Horikoshi, S. Yagi, and A. Fukamachi, Neuroradiology, 1993, 35, 151. 124. V. G. Marshall, W. G. Bradley, C. E. Marshall, T. Bhoopat, and R. H. Rhodes, Radiology, 1988, 167, 517. 125. D. G. Munoz, S. M. Hastak, B. Harper, D. Lee, and V. C. Hachinski, Arch. Neurol., 1993, 50, 492.

Biographical Sketch Donald M. Hadley. b 1950. M.B.Ch.B., 1974, Ph.D., 1980, D.M.R.D., 1981, Aberdeen University, Scotland; F.R.C.R., 1983, London, UK. Introduced to NMR by Professor John Mallard and Dr Francis Smith while carrying out postdoctoral work in the Department of Bio-medical Physics, University of Aberdeen 1981. MRC research fellow, Glasgow University 1984, consultant and director of Neuroradiology 1992, Institute of Neurological Sciences, Glasgow, UK. Approx. 200 publications. Current research interests: MRI and MRS investigation of acute trauma, epilepsy, metabolic white matter diseases and stroke.

MRI AND MRS OF NEUROPSYCHIATRY

MRI and MRS of Neuropsychiatry Basant K. Puri MRC Clinical Sciences Centre, Imperial College School of Medicine, London, UK

1 INTRODUCTION MRI and MRS studies are becoming increasingly important in neuropsychiatry. In this chapter, the contributions will be considered of such studies to our understanding of schizophrenia, mood disorders, anxiety disorders, obsessivecompulsive disorder, eating disorders, attention de®cit hyperactivity disorder (ADHD), psychoactive substance use, Alzheimer's disease, Lewy body disease and Binswanger's disease, Huntington's disease, autism, electroconvulsive therapy, dyslexia, brain changes following incomplete spinal injury in humans, and drug monitoring.

2 SCHIZOPHRENIA Many studies using MRI have been carried out on patients with schizophrenia since 1983. In a well-researched critical review of these by Chua and McKenna,1 it was found that the only well-established structural abnormality in schizophrenia is lateral ventricular enlargement; this is modest and overlaps with ventricular size in the normal population. The authors of the review came to the following conclusions: `there is no consistent evidence from MRI studies for a global reduction in brain size in schizophrenia, and only a minority of studies have pointed to a focal reduction in the size of the frontal lobes. However, the numbers of positive and negative replications are approximately equal for the ®nding of reduced temporal lobe size, and when the hippocampus and amygdala (and perhaps also the parahippocampal gyrus) are speci®cally considered this turns into a slight majority in favour of reduced size. A reasonable conclusion might therefore be that, while not yet established beyond reasonable doubt, it is likely that any brain substance abnormality in schizophrenia will be found to be localised to the temporal lobe, where it will be predominantly subcortical and perhaps also predominantly left-sided.' Recently developed techniques of subvoxel registration of high-resolution three-dimensional (3D) serial MR scans2,3 and quanti®cation of changes thereby discovered4 have just started to be applied to various aspects of this disorder. For example, when ®rst-episode schizophrenic patients were classi®ed according to Gruzelier's syndromal model,5,6 it was found that, compared with normal controls, over an 8-month period patients who were `withdrawn' showed progressive ventricular enlargement, with an increase in ventricle-to-brain volume ratio. In contrast a group of `active' patients showed a reduction in ventricle-to-brain volume ratio, with a change that

1

was opposite in sign and smaller in magnitude.7 These ®ndings suggest that opposite patterns of functional hemispheric activation early in the course of schizophrenia may be associated with strikingly different structural cerebral changes. These techniques have also found application in testing speci®c predictions of Horrobin's neuronal membrane phospholipid model of schizophrenia.8,9 In the ®rst example of this, it has been found that in a patient with long-standing disease not being treated with conventional medication, sustained remission of positive and negative symptoms of schizophrenia associated with treatment with the omega-3 fatty acid eicosapentaenoic acid (EPA; Kirunal) was accompanied by a reversal of cerebral atrophy (Figure 1).10 Using 31P MRS to study the prefrontal cortex in schizophrenia, a number of groups, including Pettegrew and colleagues11 and Stanley and colleagues,12 have reported changes in membrane phospholipid metabolism, irrespective of antipsychotic medication status, with reduced levels of phosphomonoesters (precursors of phospholipid biosynthesis) and increased levels of both phosphodiesters (phospholipid breakdown products) and intracellular magnesium ions. It has been suggested that these ®ndings may be fundamentally related to the pathophysiology of schizophrenia,11±13 with the reduced levels of phosphomonoesters being caused by reduced biosynthesis or altered degradation and the elevated levels of phosphodiesters being associated with increased activity of phospholipase A2 or A1, or perhaps decreased phosphodiesterase activity. An alternative explanation involves a putative disturbance of metabolic compartmentation of phosphatidylcholine biosynthesis.14 Many studies using proton MRS have demonstrated a reduction in the neuronal marker N-acetylaspartate, particularly in the left temporal lobe. In a recent study combining this technique with MRI, the volume of cortical gray matter was found to be reduced in patients with schizophrenia, while the N-acetylaspartate signal intensity from a comparable region was normal; by comparison, the volume of cortical white matter was normal while the N-acetylaspartate signal intensity from a comparable region was reduced.15 The lack of reduction in gray matter N-acetylaspartate signal intensity suggests that the cortical gray matter de®cit involved both neuronal and glial compartments, rather than a neurodegenerative process in which there is a decrease in the neuronal relative to the glial compartment. The reduced white matter N-acetylaspartate signal intensity without a white matter volume de®cit may re¯ect abnormal axonal connections.15

3

MOOD DISORDERS

To date there have been relatively few MR studies of mood disorders and the ®ndings are not consistent. For example, ventricular enlargement is an inconsistent ®nding in depression (using MRI or computed tomography (CT)); when it has been found it has sometimes been shown to be positively correlated with the length of illness. Neither ®rst-episode bipolar disorder nor ®rst-episode major depression appear to be associated with ventricular enlargement, however.16 Another inconsistent ®nding is the possibility of an increased frequency of signal hyperintensities on T2-weighted scans in elderly depressed patients, which may be associated with poor cognitive perform-

2 MRI AND MRS OF NEUROPSYCHIATRY

Figure 1 MRI in a patient with long-term schizophrenia. (a) Transverse image of the brain 12 months prior to commencing treatment with eicosapentaenoic acid (EPA). (b) Transverse image of the brain at baseline (0 months) with respect to EPA treatment. (c) Registered difference image of the baseline scan minus the scan at ÿ12 months ((b) minus (a)). The dark lines around the ventricles are caused by a decrease in brain size. (d) Registered difference image of the scan at 6 months minus the scan at baseline (0 months). The white lines around the ventricles are caused by an increase in brain size. Changes are also seen in the cerebral cortex, with narrowing evident in some sulci and increased volume evident in some gyri

MRI AND MRS OF NEUROPSYCHIATRY

3

ance.17,18 Although such hyperintensities may be a marker of underlying pathology, they are by no means speci®c to depression and indeed may also occur in older normal controls. It has been reported that the presence of such hyperintensities in both the basal ganglia and the pontine reticular formation in patients aged 65 years and over is associated with a poor response to antidepressant monotherapy.19 It has also been suggested that treatment-resistant chronic unipolar depression is associated with reduced gray matter density in the left temporal cortex, including the hippocampus.20 Studies using 31P MRS and 1H MRS have indicated possible abnormalities in membrane phospholipid metabolism, high-energy phosphate metabolism, and intracellular pH in mood disorders.21

re¯ected by endocrinological abnormalities.33 These data suggest that severe malnutrition in patients with anorexia nervosa may result in an abnormality in membrane phospholipid metabolism, which might be related etiologically to the cerebral atrophy of anorexia nervosa. In another study of patients with anorexia nervosa recording proton MR spectra from parietooccipital white matter immediately following an interval of excessive loss of body mass, higher signal intensity ratios of choline-containing compounds relative to total creatine and lower ratios of N-acetylaspartate relative to choline-containing compounds were found compared with controls,34 suggesting that starvation may be associated with an abnormal neuronal membrane turnover in the white matter of the brain.

4 ANXIETY DISORDERS

7

There have been very few MR studies of anxiety disorders, perhaps because anxiety and claustrophobic symptoms constitute a recognized cause of incomplete or cancelled MR examinations.22 In one 31P MRS study of the frontal lobes in panic disorder, no signi®cant differences were found between patients and controls in 31P metabolite levels, although a signi®cant asymmetry (left greater than right) of phosphocreatine concentration was found in the patients; raised intracellular pH in 2 out of 18 of the patients may have resulted from respiratory alkalosis secondary to hyperventilation in the anxiety state.23 5 OBSESSIVE-COMPULSIVE DISORDER Structural neuroimaging studies indicate that at least a subgroup of patients with obsessive-compulsive disorder may have abnormal basal ganglia development.24 Although not all such studies demonstrate reduced volumes of these structures, it is noteworthy that a reduced level of the neuronal marker N-acetylaspartate has been found in either the left25 or right26 corpus striatum in obsessive-compulsive disorder using proton MRS, even when volumetric MRI studies of the same patients do not show reduced volumes.25 Hence the inconsistent volumetric ®ndings may re¯ect the relatively poorer sensitivity of MRI morphometry for detecting neuronal loss compared with proton MRS measurement of N-acetylaspartate. 6 EATING DISORDERS The CT ®nding of cerebral atrophy in patients with eating disorders has been replicated using MRI.27,28 Female patients with anorexia nervosa and bulimia nervosa have been reported to have smaller pituitary glands than matched controls.29,30 In the absence of any other pituitary pathology, this atrophy is likely to be secondary to nutritional or endocrine alterations. Other reported structural abnormalities include enlarged lateral ventricles with dilated cortical and cerebellar sulci,31 and subcortical signal hyperintensities on T2-weighted scans.32 In a small cerebral 31P MRS study of anorexia nervosa before treatment, increased levels of phosphodiesters were found compared with controls, while decreased phosphomonoesters were found that were associated with malnutrition

ATTENTION DEFICIT HYPERACTIVITY DISORDER

Recent MRI studies have shown that some regions of the frontal lobes (anterior superior and inferior) and basal ganglia (caudate nucleus and globus pallidus) are about 10% smaller in ADHD groups than in control groups of children,35 with the right caudate nucleus being larger,36 or left caudate being smaller,37 in children with ADHD. These ®ndings are consistent with theories implicating frontal-striatal circuit abnormalities in this disorder. Also in harmony with this theory is the fact that the corpus callosum has been found relatively consistently to be smaller in children with ADHD, particularly in the region of the genu and splenium.38 Recently, the cerebellum has been systematically studied in this disorder; the vermal volume was found to be signi®cantly smaller in a large sample of boys with ADHD than in matched controls.39 This reduction involved mainly the posterior inferior lobe (lobules VIII to X) but not the posterior superior lobe (lobules VI to VII) and suggests that perhaps cerebello-thalamo-prefrontal circuit dysfunction may subserve the motor control, inhibition, and executive function de®cits seen in this disorder. Advances in genetic studies of ADHD have occurred while these advances in structural neuroimaging have been taking place. An important example of how both of these investigative techniques can complement each other relates to polymorphisms of the D4 dopamine receptor (DRD4). One allele with seven tandem repeats in exon 3 (DRD4*7R) has been associated with ADHD, and when this putative association was investigated by Castellanos and colleagues, it was found that cerebral MRI measures, previously found to discriminate ADHD patients from controls, did not differ signi®cantly between subjects having and those lacking a DRD4*7R allele.40 Hence the MRI results did not support the reported association between DRD4*7R and the behavioral or brain morphometric phenotype associated with ADHD. 8

PSYCHOACTIVE SUBSTANCE ABUSE

Chronic alcoholism is associated with MRI-detectable atrophic changes in many regions of the brain, including the cerebral cortex, cerebellum,41 and corpus callosum.42 Hippocampal volume reduction is proportional to the reduction in volume of the brain as a whole.43 It has been found that over a 5-year period brain volume shrinkage is exaggerated in the pre-

4 MRI AND MRS OF NEUROPSYCHIATRY frontal cortex in normal aging but with additional loss occurring in the anterior superior temporal cortex in alcoholism.44 This association of cortical gray matter volume reduction with alcohol consumption over time suggests that continued alcohol abuse results in progressive cerebral tissue volume shrinkage. MRI, but not CT, has been shown to be useful in con®rming the diagnosis of acute Wernicke's encephalopathy. In one recent study, increased T2 signal of the paraventricular regions of the thalamus and the mesencephalic periaqueductal regions was observed in patients with Wernicke's encephalopathy compared with both controls and asymptomatic chronic alcohol abusers, with the sensitivity of MRI in revealing evidence of this disease being 53% and the speci®city 93%.45 It should be borne in mind, however, that the absence of abnormalities on MRI does not exclude this diagnosis. With MRI, widespread cerebral atrophy is seen in alcoholic Korsakoff patients;46 this is largely subcortical and does not develop independently of the diencephalic pathology. It should be noted that while chronic alcohol abuse is associated with mammillary body and cerebellar tissue volume loss, these markers do not distinguish accurately between amnesic and nonamnesic patients; mammillary body atrophy that is detectable on MRI is not necessary for the development of amnesia in alcoholic patients.47 It has recently been shown that cerebral MRI may be of use clinically in the differential diagnosis of chronic alcohol abuse and schizophrenia.48 In this study, patients with both disorders showed widespread cortical gray matter volume de®cits compared with controls, but only those with chronic alcoholism showed white matter volume de®cits. The patients with schizophrenia had signi®cantly greater volume de®cits in prefrontal and anterior superior temporal gray matter than in more posterior cortical regions. By contrast, the de®cits in the patients with alcoholism were relatively homogeneous across the cortex. For white matter, the de®cits in the patients with alcoholism were greatest in the prefrontal and temporoparietal regions. Although both patient groups had abnormally larger cortical sulci and lateral and third ventricles than the controls, the patients with alcoholism had signi®cantly larger sulcal volumes in the frontal, anterior, and posterior parieto-occipital regions than those with schizophrenia. Reduced levels of N-acetylaspartate and choline have been found using cerebral proton MRS in chronic alcoholism.49 The reduction of N-acetylaspartate is consistent with neuronal loss while the reduction in choline may be related to neuronal membrane lipid changes. In a recent large MRI and proton MRS study comparing asymptomatic abstinent cocaine users with matched controls, it was found that while the ventricle-to-brain ratio and level of white matter lesions did not differ signi®cantly between the two groups, elevated creatine and myo-inositol in the white matter were associated with cocaine use.50 The N-acetylaspartate level was normal in the cocaine users, suggesting that there was no neuronal loss or damage in the brain regions examined. It was, therefore, concluded that the neurochemical abnormalities observed might result from alterations in nonneuronal brain tissue. MRI changes associated with chronic toluene abuse include cerebral atrophy, cerebral and cerebellar white matter T2 hyperintensity, T2 hyperintensity involving the middle cerebellar peduncle and the posterior limb of the internal capsule, and T2

hypointensity involving the basal ganglia and thalamus.51,52 Chronic solvent abusers who have white matter MRI changes have been found have a lower performance intelligence quotient, as measured by the Weschler Adult Intelligence Scale ± Revised, with a particularly low score on the digit symbol subtest.53 In polysubstance abusers, there is MRI evidence of reduced volume of the prefrontal cortex (both left and right) consistent with either atrophy or hypoplasia.54 Ventriculomegaly has not been found to be a feature.55 Abnormal cerebral metabolism has been found using 31P MRS in male polysubstance abusers during early withdrawal: increased phosphomonoesters and decreased -nucleotide trisphosphates were found in the abusers compared with controls, indicating that cerebral highenergy phosphate and phospholipid metabolite changes result from long-term drug abuse and/or withdrawal.56

9

ALZHEIMER'S DISEASE, LEWY BODY DISEASE, AND BINSWANGER'S DISEASE

The ®rst part of this section considers recent studies focusing on the use of MRI in differentiating Alzheimer's disease from both normal aging and other causes of dementia. While the ®nding of cortical or subcortical atrophy on MRI or CT is not pathognomonic of Alzheimer's disease, hippocampal atrophy provides a useful early marker of the disorder, although further longitudinal and neuropathological study is required.57 CT- and MRI-based measurements of hippocampal atrophy may provide useful diagnostic information for differentiating patients with probable Alzheimer's disease from normal elderly individuals. A recent pilot study has indicated that MRI may have a role in assisting with the clinical differentiation between dementia with Lewy bodies and Alzheimer's disease.58 Subjects with known or probable Alzheimer's disease were found to have signi®cantly smaller left temporal lobes and parahippocampal gyri than those with known or probable Lewy body disease. Medial temporal atrophy was present in 9 out of 11 patients with Alzheimer's disease and absent in six out of nine patients with Lewy body disease. While two patients with neuropathologically con®rmed Lewy body disease had severe medial temporal atrophy, in both concurrent Alzheimer's disease-type pathology was present in the temporal lobe. Therefore, this pilot study supports the hypothesis that a greater burden of pathology centers on the temporal lobes in Alzheimer's disease compared with Lewy body disease, unless Lewy body disease occurs with concurrent Alzheimer pathology. Another recent study has suggested that diffusion-weighted MRI may be useful in the differential diagnosis of subcortical arteriosclerotic encephalopathy (vascular dementia of the Binswanger type) and Alzheimer's disease with white matter lesions.59 Apparent diffusion coef®cients in the anterior and posterior white matter and the genu and splenium of the corpus callosum were signi®cantly higher in patients with both these disorders compared with age-matched controls, with apparent diffusion coef®cient values in the groups with Binswanger's disease and those with Alzheimer's disease being almost the same. Apparent diffusion coef®cient ratios, de®ned as diffusionrestricted perpendicular to the direction of nerve ®bers, were also signi®cantly higher in both groups of patients than in the

MRI AND MRS OF NEUROPSYCHIATRY

controls. However, there were regional differences in these ratios in the two disorders, with ratios in Binswanger's disease being higher in the anterior portions of the white matter while ratios in Alzheimer's disease were higher in the posterior portions. In vitro and in vivo 31P MRS studies of the brain in Alzheimer's disease show alterations in membrane phospholipid metabolism and high-energy phosphate metabolism: compared with control subjects, mildly demented patients with Alzheimer's disease have increased levels of phosphomonoesters, decreased levels of phosphocreatine and probably adenosine diphosphate, and an increased oxidative metabolic rate; as the dementia worsens, levels of phosphomonoesters decrease and levels of phosphocreatine and adenosine diphosphate increase.60 The changes in oxidative metabolic rate suggest that the brain in Alzheimer's disease is under energetic stress while the phosphomonoester ®ndings implicate basic defects in membrane metabolism in the brain.60 Thus, in addition to aiding with diagnosis, 31P MRS may provide a noninvasive tool to follow both the progression of this disorder and any response to putative therapeutic interventions. Proton MRS studies of occipital gray matter show that reduced levels of N-acetylaspartate (presumably re¯ecting neuronal loss) and increased levels of myo-inositol characterize Alzheimer's disease.61,62 Studies using proton MRS to measure cerebral amino acids have tended to demonstrate increased glutamate levels and sometimes reduced -aminobutyric acid (GABA); following neuronal loss, the remaining neurons might be exposed to excess glutamate and relatively low levels of GABA, an imbalance that might be neurotoxic.63,64

10

HUNTINGTON'S DISEASE

Initially, structural neuroimaging studies showed atrophy of the caudate and loss of de®nition between the caudate and the adjacent ventricle as Huntington's disease progresses; however more recent studies have also shown cortical atrophy, particularly in the frontal lobes.65 Using proton MRS, Jenkins and colleagues found that lactate concentrations were increased in the occipital cortex of patients with symptomatic Huntington's disease compared with normal controls, with the lactate level correlating with duration of illness.66 Several patients in the same study also showed highly elevated lactate levels in the basal ganglia, while basal ganglia levels of N-acetylaspartate were lowered and choline dramatically elevated, relative to creatine, re¯ecting neuronal loss and gliosis in this brain region. The authors of this study suggested that these ®ndings are consistent with a possible defect in energy metabolism in Huntington's disease, which could contribute to the pathogenesis of the disease, and that the presence of elevated lactate might provide a simple marker that could be followed over time noninvasively and repeatedly to aid in devising and monitoring possible therapies. A more recent proton MRS study by Taylor-Robinson and colleagues found an elevated ratio of glutamine and glutamate relative to creatine in the striatum compared with healthy controls, suggesting disordered striatal glutamate metabolism and possibly supporting the theory of glutamate excitotoxicity in Huntington's disease.67

5

Huntington's disease is now known to result from expanded CAG repeats in a gene on chromosome 4, a possible consequence of which might be progressive impairment of energy metabolism. Jenkins and colleagues have recently extended their previous studies to examine correlations between proton MRS changes and CAG repeat number.68 The spectra in three presymptomatic gene-positive patients were found to be identical to normal control subjects in cortical regions, but three in eight showed elevated lactate in the striatum. Similar to recently reported increases in task-related activation of the striatum in the dominant hemisphere, they found that striatal lactate levels in patients with Huntington's disease were markedly asymmetric (left greater than right). Markers of neuronal degeneration, decreased N-acetylaspartate to creatine and increased choline to creatine ratios, were symmetric. Both decreased N-acetylaspartate and increased lactate in the striatum signi®cantly correlated with duration of symptoms. When divided by the patient's age, an individual's striatal N-acetylaspartate loss and lactate increase were found to correlate with the subject's CAG repeat number, with correlation coef®cients of 0.8 and 0.7, respectively. Similar correlations were noted between postmortem cell loss and age versus CAG repeat length. Together, these data provide further evidence for an interaction between neuronal activation and a defect in energy metabolism in Huntington's disease that may extend to presymptomatic subjects.68

11

AUTISM

MRI studies of individuals with autism have variously and inconsistently shown evidence of hypoplasia of the cerebellum and brainstem with increased size of the fourth ventricle, increased brain volume (though with relative hypofrontality), and smaller size of the body and posterior subregions of the corpus callosum; in addition, previous pneumoencephalographic and CT studies have described lateral ventricular enlargement while MRI studies in general have failed to show abnormality in limbic structures.69 The degree of cerebellar hypoplasia is signi®cantly correlated with the degree of slowed attentional orienting to visual cues in both children and adults with autism.70 It should be noted that even in the absence of abnormal MRI ®ndings, autism may be associated with focal areas of decreased perfusion.71 The ®nding that autism is not necessarily associated with MRI abnormalities is consistent with the results of a recent cerebral proton MRS study comparing 28 patients with autism with both 28 age-matched patients with unclassi®ed mental retardation and 25 age-matched healthy children. The ratio of Nacetylaspartate to choline was lower in the nonautistic patients with mental retardation than in the patients with autism and the controls, and, interestingly, there were no signi®cant differences in this ratio between patients with autism and controls.72 A 31P MRS study of the dorsal prefrontal cortex of 11 highfunctioning autistic adolescent and young adult men and 11 matched normal controls found that the autistic group had decreased levels of phosphocreatine, -ATP, -ADP, dinucleotides, and diphosphosugars compared with the controls.73 When the metabolite levels were compared within each subject group with psychological and language test scores, a common pattern of correlations was observed across measures in the

6 MRI AND MRS OF NEUROPSYCHIATRY

Figure 2 MRI in patients receiving electroconvulsive therapy. (a) Transverse T1-weighted MR baseline scan showing the anatomy. (b) Difference image obtained by subtracting the baseline scan from the registered follow-up scan showing no evidence of acute structural changes in the brain following electroconvulsive therapy

autistic group, but not in the control group. As test performance declined in the autistic subjects, levels of the most labile high-energy phosphate compound and of membrane-building blocks decreased, and levels of membrane breakdown products increased. No signi®cant correlations were present with age in either group or with IQ in the control group, suggesting that these ®ndings were not the consequence of age or IQ effects. This study provides some evidence of alterations in brain energy and phospholipid metabolism in autism that correlate with psychological and language de®cits.

12

ELECTROCONVULSIVE THERAPY

For many years clinicians have been concerned that electroconvulsive therapy may result in acute cerebral structural changes. Indeed, some retrospective imaging studies using MRI and CT have reported an association between a history of electroconvulsive therapy and cerebral change, particularly affecting the lateral ventricles and/or cerebral cortex. However, recently, a prospective MRI study of four electroconvulsive therapy-naõÈve depressed patients in which they underwent scanning 1 week prior to their ®rst treatment with electroconvulsive therapy and then again following this treatment showed that, using accurate subvoxel registration and subtraction of serial MR images,2,3 there was no signi®cant difference in cerebral structure following electroconvulsive therapy, either within 24 h or after 6 weeks (Figure 2).74

A proton and 31P MRS study of three patients found no evidence of changes in lactate or in cerebral energy metabolism following electroconvulsive therapy.75 However, Woods and Chiu have found, using proton MRS, that electroconvulsive therapy reliably induces an elevation in the lipid signal that resonates at approximately 1.2 ppm and observed a similar increase in brain lipids in a patient with temporal lobe epilepsy temporarily off medication, the signal disappearing following restarting medication.76 This is of interest given that elevations of brain concentrations of arachidonic acid and other free fatty acids have been demonstrated to occur after seizures induced in animals. Large shifts of potassium ions from the intra- to the extracellular space occur during seizure activity, and free fatty acids have a direct effect on membrane potassium ion conductance, suggesting that free fatty acids may play a primary role in seizure evolution in brain tissue.76 13

DYSLEXIA

MRI studies have inconsistently shown reversed or diminished asymmetry, compared with normal, in the brain in children with dyslexia, including loss of the usual left greater than right asymmetry of the lateral ventricles and right greater than left asymmetry of the temporal lobes; loss of the normal left greater than right asymmetry of the planum temporale in adolescents, which correlates with the degree of phonological decoding de®cits; reversal of the normal left greater than right asymmetry of the angular gyrus in familial dyslexia; and loss

MRI AND MRS OF NEUROPSYCHIATRY (a) 12250

10

12000

9

Clinically, this ®nding also suggests that MRS might provide a noninvasive method for monitoring such patients.

8

11750

15

Ventricular volume (mm3)

6

11250

5

11000

4 –1

0

1

2

3

Time (h) (b) 13500

10

13000

9

12500

8

12000

7

11500

6

11000

5

10500

4 –1

0

1

2

Blood glucose concentration (mmol /l–1)

7 11500

3

Time (h)

Figure 3 Volume changes in lateral ventricles following oral glucose loading. Blood glucose concentrations (*) and lateral ventricular volumes (*) for two human subjects (a and b) who each ingested 53.7 g glucose at time zero. (After Puri et al.85)

of normal right greater than left asymmetry of the frontal cortices and bilaterally smaller size of the frontal cortices. Inconsistent corpus callosum changes have also been reported.77 In the ®rst 31P MRS study of dyslexia, Richardson and colleagues found elevated phosphomonoesters in the brain in dyslexia compared with that in controls.78 This ®nding is consistent with the hypothesis that neuronal membrane phospholipid metabolism is abnormal in dyslexia, with reduced incorporation of phospholipids into neuronal membranes occurring.79 The ®rst proton MRS study of dyslexia showed lateral differences in the ratios of choline to N-acetylaspartate and of creatine to N-acetylaspartate in the temporo-parietal region and the cerebellum in dyslexic subjects but not in controls.80 14

7

SPINAL INJURY

The ®rst proton MRS study of the human motor cortex following incomplete spinal cord injury showed elevation of Nacetylaspartate in this part of the brain compared with normal controls.81 The authors suggested that this might re¯ect neuronal adaptation to injury, the ®nding being consistent with the hypothesis that dendritic sprouting occurs in the motor cortex following recovery from incomplete spinal injury in humans.

DRUG MONITORING

It is possible to use 7Li MRS directly to measure the cerebral concentration of lithium while 19F MRS can be used to measure the cerebral concentrations of psychotropic drugs containing ¯uorine, for example the selective serotonin reuptake inhibitor ¯uoxetine and the antipsychotics tri¯uoperazine and ¯uphenazine.82

16

FUTURE DIRECTIONS

As mentioned above, the recently developed MRI techniques of subvoxel registration of high-resolution 3D serial MR scans2,3 and quanti®cation of the changes thereby discovered4 are only just starting to be applied in neuroimaging studies. The sensitivity and accuracy of these techniques hold great promise for neuroimaging applications and the discovery of important new facts concerning the central nervous system.83,84 For example, they have been used recently to show that volumetric change takes place in the lateral ventricles in the human brain following oral glucose loading (Figure 3).85 Cerebral MRS is currently used primarily as a research tool in neuropsychiatry; in due course it is likely to become more widely used diagnostically and prognostically. It seems probable that MRI and MRS will interface more often with other disciplines (for instance molecular genetics, as in ADHD) and other investigative tools (such as transcranial magnetic stimulation). In summary, MRI and MRS are proving to be extremely useful in neuroscienti®c and neuropsychiatric research. These powerful noninvasive tools are likely to continue to grow in importance in these ®elds and to gain ever more important clinical applications.

17

RELATED ARTICLES

Brain Infection and Degenerative Disease Studied by Proton MRS; Brain MRS of Human Subjects; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Brain Neoplasms Studied by MRI; Brain Parenchyma Motion Observed by MRI; Hemodynamic Changes owing to Sensory Activation of the Brain Monitored by Echo-Planar Imaging; Central Nervous System Degenerative Disease Observed by MRI; Chemical Shift Imaging; CSF Velocity Imaging; Intracranial Infections; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Magnetic Resonance Imaging of White Matter Disease; Structural and Functional MR in Epilepsy.

18

REFERENCES

1. S. E. Chua and P. J. McKenna, Br. J. Psychiatry, 1995, 166, 563.

8 MRI AND MRS OF NEUROPSYCHIATRY 2. J. V. Hajnal, N. Saeed, E. J. Soar, A. Oatridge, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1995, 19, 289. 3. J. V. Hajnal, N. Saeed, A. Oatridge, E. J. Williams, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1995, 19, 677. 4. N. Saeed, B. K. Puri, A. Oatridge, J. V. Hajnal, and I. R. Young, Magn. Reson. Med., 1998, 16, 1237. 5. J. H. Gruzelier and R. Manchanda, Br. J. Psychiatry, 1982, 141, 488. 6. J. H. Gruzelier `Handbook of Schizophrenia', Vol. 5, ed. S. R. Steinhauer, J. H. Gruzelier, and J. Zubin, Elsevier, London, 1991, p. 599. 7. B. K. Puri, N. Saeed, A. J. Richardson, A. Oatridge, S. B. Hutton, J. V. Hajnal, and G. M. Bydder, Schizophr. Res., 1999. 8. D. F. Horrobin, Lancet, 1977, i, 936. 9. D. F. Horrobin, A. I. M. Glen, and K. Vaddadi, Schizophr. Res., 1994, 13, 195. 10. B. K. Puri, A. J. Richardson, T. Easton, N. Saeed, A. Oatridge, J. V. Hajnal, D. F. Horrobin, and G. M. Bydder, Schizophr. Res., 1999, 36, 312. 11. J. W. Pettegrew, M. S. Keshavan, K. Panchalingam, S. Strychor, D. B. Kaplan, M. G. Tretta, and M. Allen, Arch. Gen. Psychiatry, 1991, 48, 563. 12. J. A. Stanley, P. C. Williamson, D. J. Drost, T. J. Carr, R. J. Rylett, A. Malla, and R. T. Thompson, Arch. Gen. Psychiatry, 1995, 52, 399. 13. D. F. Horrobin, Prostaglandins Leukot. Essent. Fatty Acids, 1996, 55, 3. 14. T. P. George and M. W. Spence, Arch. Gen. Psychiatry, 1996, 53, 1065. 15. K. O. Lim, E. Adalsteinsson, D. Spielman, E. V. Sullivan, M. J. Rosenbloom, and A. Pfefferbaum, Arch. Gen. Psychiatry, 1998, 55, 346. 16. W. G. Iacono, G. N. Smith, M. Moreau, M. Beiser, J. A. Fleming, T. Y. Lin, and B. Flak, Am. J. Psychiatry, 1988, 145, 820. 17. C. E. Coffey, G. S. Figiel, W. T. Djang, W. B. Saunders, and R. D. Weiner, J. Neuropsychiatry Clin. Neurosci., 1989, 1, 135. 18. E. Kramer-Ginsberg, B. S. Greenwald, K. R. Krishnan, B. Christiansen, J. Hu, M. Ashtari, M. Patel, and S. Pollack, Am. J. Psychiatry, 1999, 156, 438. 19. S. Simpson, R. C. Baldwin, A. Jackson, and A. S. Burns, Psychol. Med., 1998, 28, 1015. 20. P. J. Shah, K. P. Ebmeier, M. F. Glabus, and G. M. Goodwin, Br. J. Psychiatry, 1998, 172, 527. 21. T. Kato, T. Inubushi and N. Kato, J. Neuropsychiatry Clin. Neurosci., 1998, 10, 133. 22. S. A. Sarji, B. J. Abdullah, G. Kumar, A. H. Tan, and P. Narayanan, Australas. Radiol., 1998, 42, 293. 23. T. Shioiri, T. Kato, J. Murashita, H. Hamakawa, T. Inubushi, and S. Takahashi, Biol. Psychiatry, 1996, 40, 785. 24. S. Saxena, A. L. Brody, J. M. Schwartz, and L. R. Baxter, Br. J. Psychiatry, 1998, 35 (Suppl.), 26. 25. R. Bartha, M. B. Stein, P. C. Williamson, D. J. Drost, R. W. Neufeld, T. J. Carr, G. Canaran, M. Densmore, G. Anderson, and A. R. Siddiqui, Am. J. Psychiatry, 1998, 155, 1584. 26. D. Ebert, O. Speck, A. Konig, M. Berger, J. Hennig, and F. Hohagen, Psychiatry Res., 1997, 74, 173. 27. G. W. Hoffman Jr, E. H. Ellinwood Jr, W. J. Rockwell, R. J. Herfkens, J. K. Nishita, and L. F. Guthrie, Biol. Psychiatry, 1989, 25, 894. 28. G. W. Hoffman Jr, E. H. Ellinwood Jr, W. J. Rockwell, R. J. Herfkens, J. K. Nishita, and L. F. Guthrie, Biol. Psychiatry, 1989, 26, 321. 29. P. M. Doraiswamy, K. R. Krishnan, G. S. Figiel, M. M. Husain, O. B. Boyko, W. J. Rockwell, and E. H. Ellinwood Jr, Biol. Psychiatry, 1990, 28, 110. 30. P. M. Doraiswamy, K. R. Krishnan, O. B. Boyko, et al., Prog. Neuropsychopharmacol. Biol. Psychiatry, 1991, 15, 351.

31. K. Kingston, G. Szmukler, D. Andrewes, B. Tress, and P. Desmond, Psychol. Med., 1996, 26, 15. 32. K. G. Sieg, M. S. Hidler, M. A. Graham, R. L. Steele, and L. R. Kugler, Int. J. Eat. Disord., 1997, 21, 391. 33. T. Kato, T. Shioiri, J. Murashita, and T. Inubushi, Prog. Neuropsychopharmacol. Biol. Psychiatry, 1997, 21, 719. 34. H. P. Schlemmer, R. Mockel, A. Marcus, F. Hentschel, C. Gopel, G. Becker, J. Kopke, F. Guckel, M. H. Schmidt, and M. Georgi, Psychiatry Res., 1998, 82, 171. 35. J. Swanson, F. X. Castellanos, M. Murias, G. LaHoste, and J. Kennedy, Curr. Opin. Neurobiol., 1998, 8, 263. 36. M. Mataro, C. Garcia-Sanchez, C. Junque, A. Estevez-Gonzalez, and J. Pujol, Arch. Neurol., 1997, 54, 963. 37. P. A. Filipek, M. Semrud-Clikeman, R. J. Steingard, P. F. Renshaw, D. N. Kennedy, and J. Biederman, Neurology, 1997, 48, 589. 38. G. W. Hynd, M. Semrud-Clikeman, A. R. Lorys, E. S. Novey, D. Eliopulos, and H. Lyytinen, J. Learn. Disabil., 1991, 24, 141. 39. P. C. Berquin, J. N. Giedd, L. K. Jacobsen, S. D. Hamburger, A. L. Krain, J. L. Rapoport, and F. X. Castellanos, Neurology, 1998, 50, 1087. 40. F. X. Castellanos, E. Lau, N. Tayebi, P. Lee, R. E. Long, J. N. Giedd, W. Sharp, W. L. Marsh, J. M. Walter, S. D. Hamburger, E. I. Ginns, J. L. Rapoport, and E. Sidransky, Mol. Psychiatry, 1998, 3, 431. 41. K. Hayakawa, H. Kumagai, Y. Suzuki, N. Furusawa, T. Haga, T. Hoshi, Y. Fujiwara, and K. Yamaguchi, Acta Radiol., 1992, 33, 201. 42. R. Estruch, J. M. Nicolas, M. Salamero, C. Aragon, E. Sacanella, J. Fernandez-Sola, and a Urbano-Marquez, J. Neurol. Sci., 1997, 146, 145. 43. I. Agartz, R. Momenan, R. R. Rawlings, M. J. Kerich, and D. W. Hommer, Arch. Gen. Psychiatry, 1999, 56, 356. 44. A Pfefferbaum, E. V. Sullivan, M. J. Rosenbloom, D. H. Mathalon, and K. O. Lim, Arch. Gen. Psychiatry, 1998, 55, 905. 45. E. Antunez, R. Estruch, C. Cardenal, J. M. Nicolas, J. FernandezSola, and A. Urbano-Marquez, Am. J. Roentgenol., 1998, 171, 1131. 46. R. Emsley, R. Smith, M. Roberts, S. Kapnias, H. Pieters, and S. Maritz, Alcohol Alcohol., 1996, 31, 479. 47. P. K. Shear, E. V. Sullivan, B. Lane, and A. Pfefferbaum, Alcohol. Clin. Exp. Res., 1996, 20, 1489. 48. E. V. Sullivan, D. H. Mathalon, K. O. Lim, L. Marsh, and A. Pfefferbaum, Biol. Psychiatry, 1998, 43, 118. 49. N. R. Jagannathan, N. G. Desai, and P. Raghunathan, Magn. Reson. Imag., 1996, 14, 553. 50. L. Chang, C. M. Mehringer, T. Ernst, R. Melchor, H. Myers, D. Forney, and P. Satz, Biol. Psychiatry, 1997, 42, 1105. 51. S. Kojima, K. Hirayama, H. Furumoto, T. Fukutake, and T. Hattori, Rinsho Shinkeigaku, 1993, 33, 477. 52. K. S. Caldemeyer, S. W. Armstrong, K. K. George, C. C. Moran, and R. M. Pascuzzi, J. Neuroimaging, 1996, 6, 167. 53. N. Yamanouchi, S. Okada, K. Kodama, T. Sakamoto, H. Sekine, S. Hirai, A. Murakami, N. Komatsu, and T. Sato, Acta Neurol. Scand., 1997, 96, 34. 54. X. Liu, J. A. Matochik, J. L. Cadet, and E. D. London, Neuropsychopharmacology, 1998, 18, 243. 55. X. Liu, R. L. Phillips, S. M. Resnick, V. L. Villemagne, D. F. Wong, J. M. Stapleton, and E. D. London, Acta Neurol. Scand., 1995, 92, 83. 56. J. D. Christensen, M. J. Kaufman, J. M. Levin, J. H. Mendelson, B. L. Holman, B. M. Cohen, and P. F. Renshaw, Magn. Reson. Med., 1996, 35, 658. 57. P. Scheltens, J. Neurol., 1999, 246, 16. 58. G. T. Harvey, J. Hughes, I. G. McKeith, R. Briel, C. Ballard, A. Gholkar, P. Scheltens, R. H. Perry, P. Ince, and J. T. O'Brien, Psychol. Med., 1999, 29, 181.

MRI AND MRS OF NEUROPSYCHIATRY 59. H. Hanyu, Y. Imon, H. Sakurai, T. Iwamoto, M. Takasaki, H. Shindo, D. Kakizaki, and K. Abe, Eur. J. Neurol., 1999, 6, 195. 60. J. W. Pettegrew, W. E. Klunk, K. Panchalingam, R. J. McClure, and J. A. Stanley, Ann. N. Y. Acad. Sci., 1997, 826, 282. 61. R. A. Moats, T. Ernst, T. K. Shonk, and B. D. Ross, Magn. Reson. Med., 1994, 32, 110. 62. T. K. Shonk, R. A. Moats, P. Gifford, T. Michaelis, J. C. Mandigo, J. Izumi, and B. D. Ross, Radiology, 1995, 195, 65. 63. W. E. Klunk, K. Panchalingam, J. Moossy, R. J. McClure, and J. W. Pettegrew, Neurology, 1992, 42, 1578. 64. R. J. McClure, J. N. Kanfer, K. Panchalingam, W. E. Klunk, and J. W. Pettegrew, Neuroimaging Clin. N. Am., 1995, 5, 69. 65. J. M. Hoffman `Brain Imaging in Clinical Psychiatry', ed. K. R. R. Krishnan and P. M. Doraiswamy, Marcel Dekker, New York, 1997, p. 544. 66. B. G. Jenkins, W. J. Koroshetz, M. F. Beal, and B. R. Rosen, Neurology, 1993, 43, 2689. 67. S. D. Taylor-Robinson, R. A. Weeks, D. J. Bryant, J. Sargentoni, C. D. Marcus, A. E. Harding, and D. J. Brooks, Mov. Disord., 1996, 11, 167. 68. B. G. Jenkins, H. D. Rosas, Y. C. Chen, T. Makabe, R. Myers, M. MacDonald, B. R. Rosen, M. F. Beal, and W. J. Koroshetz, Neurology, 1998, 50, 1357. 69. S. Deb and B. Thompson, Br. J. Psychiatry, 1998, 173, 299. 70. N. S. Harris, E. Courchesne, J. Townsend, R. A. Carper, and C. Lord, Brain Res. Cogn. Brain Res., 1999, 8, 61. 71. Y. H. Ryu, J. D. Lee, P. H. Yoon, D. I. Kim, H. B. Lee, and Y. J. Shin, Eur. J. Nucl. Med., 1999, 26, 253. 72. T. Hashimoto, M. Tayama, M. Miyazaki, Y. Yoneda, T. Yoshimoto, M. Harada, H. Miyoshi, M. Tanouchi, and Y. Kuroda, J. Child Neurol., 1997, 12, 91. 73. N. J. Minshew, G. Goldstein, S. M. Dombrowski, K. Panchalingam, and J. W. Pettegrew, Biol. Psychiatry, 1993, 33, 762. 74. B. K. Puri, A. Oatridge, N. Saeed, J. E. Ging, H. M. McKee, S. K. Lekh, and J. V. Hajnal, Br. J. Psychiatry, 1998, 173, 267. 75. S. R. Felber, R. Pycha, M. Hummer, F. T. Aichner, and W. W. Fleischhacker, Biol. Psychiatry, 1993, 33, 651. 76. B. T. Woods and T. M. Chiu, Adv. Exp. Med. Biol., 1992, 318, 267. 77. J. N. Giedd and F. X. Castellanos, `Brain Imaging in Clinical Psychiatry', ed K. R. R. Krishnan and P. M. Doraiswamy, Marcel Dekker, New York, 1997, p. 126.

9

78. A. J. Richardson, I. J. Cox, J. Sargentoni, and B. K. Puri, NMR Biomed., 1997, 10, 309. 79. D. F. Horrobin, A. I. M. Glen, and C. J. Hudson, Med. Hypotheses, 1995, 45, 605. 80. C. Rae, M. A. Lee, R. M. Dixon, A. M. Balmire, C. H. Thompson, P. Styles, J. Talcott, A. J. Richardson, and J. F. Stein, Lancet, 1998, 351, 1849. 81. B. K. Puri, H. C. Smith, I. J. Cox, J. Sargentoni, G. Savic, D. W. Maskill, H. L. Frankel, P. H. Ellaway, and N. J. Davey, J. Neurol. Neurosurg. Psychiatry, 1998, 65, 748. 82. H. C. Charles, T. B. Snyderman, and E. Ahearn, `Brain Imaging in Clinical Psychiatry', ed. K. R. R. Krishnan and P. M. Doraiswamy, Marcel Dekker, New York, 1997, p. 20. 83. G. M. Bydder and J. V. Hajnal, `Advanced MR Imaging Techniques', ed. W. G. Bradley and G. M. Bydder, Martin Dunitz, London, 1997, p. 239. 84. G. M. Bydder and J. V. Hajnal, `Advanced MR Imaging Techniques', ed. W. G. Bradley and G. M. Bydder, Martin Dunitz, London, 1997, p. 259. 85. B. K. Puri, H. J. Lewis, N. Saeed, and N. J. Davey, Exp. Physiol., 1999, 84, 223.

Biographical Sketch Basant K. Puri. b 1961. B.A. 1982, B.Chir. 1984, M.B. 1985, M.A. 1986, University of Cambridge; M.R.C. Psych. 1989; Dip. Math. 2000. Residency in psychiatry, Addenbrooke's Hospital, Cambridge 1986±88. Research fellow in molecular genetics, MRC Molecular Neurobiology Unit, Cambridge and Peterhouse, Cambridge University 1988±89. Residency in psychiatry, Charing Cross & Westminster Medical School, London 1990±96. Senior Lecturer and Consultant Psychiatrist, MRI Unit, MRC Clinical Sciences Centre, Imperial College School of Medicine, Hammersmith Hospital, London University, and Honorary Consultant, Department of Radiology, Hammersmith Hospital, London 1997±present. Approx. 13 books (psychiatry, neuroscience, statistics) and 60 papers. Research interests: central nervous system MRI and MRS.

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

Pediatric Brain MRI: Applications in Neonates and Infants Jacqueline M. Pennock Royal Postgraduate Medical School, Hammersmith Hospital, London, UK

1 INTRODUCTION The ability of magnetic resonance imaging (MRI) to provide physiological, anatomical and functional information without posing any biological hazard makes it particularly suitable for studying the central nervous system of children where repeated examinations may be critical for diagnosis. Clinical MRI systems are almost always designed for adults and sequences are optimized to maximize contrast in adults. The ®rst task in examining infants is often to modify the equipment to optimize it for small infants and to develop imaging patterns which provide equivalent contrast to that seen in adults. This article covers particular techniques used to examine preterm babies and infants, the normal appearances of the developing brain, and clinical conditions encountered in infants and children.

2 PATIENT PREPARATION For a successful examination, both the child and the parent must be considered. We actively encourage parents to accompany their infants to the imaging unit so that they do not feel excluded and can see that their child is asleep and comfortable. If it is their wish, we invite the parents to watch us putting their baby into the machine, bearing in mind that for many parents seeing their sleeping child enclosed in a head coil and being slid into the machine may be a stressful experience. With older children, parents are asked to stay with them during the examination, although talking is limited to the time between scans. 2.1 Sedation Diagnostic images can only be obtained in a quiet, immobile child, so if possible appointments are made to coincide with the natural sleeping pattern of the child. Children come to a children's day ward where they are seen by a pediatrician who goes through the metal check with the parent and ensures that there are no contraindications to sedation. This is given to children under 2 years of age 20±30 min before the scan time. The child stays on the ward with the supervision of a nurse until she or he is asleep and is then brought to the scanning unit. Once the scan is over, the child returns to the ward until she or he is fully recovered.

1

Preterm infants and neonates are fed prior to the examination and scanned during natural sleep. However, in the irritable stable neonate, oral chloral hydrate via a nasal gastric tube or per rectum (50±80 mg kgÿ1) is used. For infants aged 6 weeks and over, oral chloral hydrate (70±100 mg kgÿ1) is given. This dose is successful in most children up to the age of 3 years or about 15 kg. Once the child is asleep, further immobilization is achieved by partly surrounding the head with a plastic globe ®lled with tiny polystyrene balls which is then evacuated. This minimizes motion, helps to keep the baby warm and provides insulation from sound. Swaddling the infant also prevents movement and provides extra security and warmth. Neonates usually sleep better on their sides, and this position also reduces the risk of inhalation of regurgitated milk or vomit. MRI of the critically ill intubated infant is feasible as long as the imaging unit has the appropriate facilities to offer the same care for the child as does the neonatal intensive care unit (NICU). These include suitably trained staff, gases and suction, and pediatric-sized resuscitation equipment. In our hospital the baby is brought from the NICU in a standard Vickers transport incubator. The incubator is parked and locked in a corner of the room well beyond the 5 G (5  10ÿ4 T) line and, as an extra precaution, attached to a wall by a plastic chain. Infants are ventilated from the Vicker neovent unit via long extension tubes from the incubator to the baby, with appropriate adjustment of the inspiratory and expiratory pressure to take account of the large dead space. 2.2

Monitoring

Monitoring of the sedated and naturally sleeping infant is mandatory. A variety of magnetic resonance (MR) compatible physiological monitoring devices is available for monitoring blood pressure, heart rate, respiration, and blood oxygen saturation. We use electrocardiogram (ECG) monitoring with ®ber optic leads and infant-sized MR-compatible electrodes, as well as pulse oximetry with neonatal probes attached to the infant's foot to display blood oxygen saturation and pulse rate. Further details and information on the sedation and anesthesia of the critically ill infant is available in the radiological literature.1±3 2.3

General Safety

Baby clothing with metal poppers should be removed. Although the fastenings are not ferromagnetic they may conduct eddy currents which can lead to artifacted images. An infant is almost always accompanied by medical staff and parents unaccustomed to an MR unit, and constant care has to be taken to check for loose metal objects on persons in a magnetic ®eld. Pacemakers and aneurysm clips are usually associated with an older population; however, young mothers may also have aneurysm clips. A strict routine and protocol for access is necessary. Scheduling of sedated infants and sleeping children is dif®cult and requires considerable ¯exibility. It is much easier to have exclusively pediatric sessions and not to try to mix sedated and sleeping children with ambulant adults. Prior knowledge of the clinical condition with a presumptive diagnosis is essential so that suitable protocols can be set up before each scan. Protocols which we have found useful in neonatal imaging are given in Table 1.

2 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS Table 1

Suggested Protocol for Imaging Neonates at 1.0 T

Slice orientation Phase-encoding axis Number of slices Echo time (TE) Inversion time (TI) Repeat time (TR) Field-of-view Phase resolution Frequency encode resolution Number of signals (average) Slice thickness (mm) Slice gap Receive coil Flip angle (degrees) Estimated time of scan a

First scan (all ages, short TR/TE SE)

Inversion recovery

T2-weighted spin echo

Transverse Left±right 24 20 ± 860 22±25 192 256 2 4 0 Small diameter 90 4 min

Transverse Left±right 18 30 1500±700a 6000±3500 22±25a 128 256 1 or 2 5±6 0 Small diameter 90 Variable

Transverse Left±right 24 200±120a ± 3500 22±25 128 256 1 or 2 5 0 Small diameter 90 Variable

Choice will depend on age of infant (see Table 2).

3 TECHNICAL CONSIDERATIONS The signal-to-noise ratio is an important factor in producing high quality diagnostic images at all ®eld strengths. Image quality can be improved by using the smallest diameter coil available. Adult knee coils with an internal diameter of approximately 19 cm have proved useful for imaging premature infants (head circumference approximately 30 cm) as well as infants up to 45 weeks gestational age. For children up to 18 months of age we use a receiver coil with an internal diameter of 24 cm. This coil is made in two halves of clear Perspex, making positioning and observation of the baby easier.

4 PULSE SEQUENCES The neonatal brain contains a higher proportion of water (92±95%) than the adult brain (82±85%) and this is associated with a marked increase in T1 and T2.4 The amount of water in

Table 2

the brain falls with increasing age, and by 2 years of age it has almost reached adult levels.5 A brief description of the standard pulse sequences and useful variations for imaging neonates and children are given below. The major pulse sequences and their speci®c variations for neonatal practice are shown in Table 2. 4.1

The Inversion±Recovery (IR) Sequence

The short inversion time (TI) inversion±recovery (STIR) sequence is of value in demonstrating myelination and pathological periventricular changes where the cerebrospinal ¯uid (CSF) signal can be kept less than that of brain.6 This sequence has many features in common with the T2-weighted spin echo sequence, but gives greater gray±white matter contrast (Figure 1). The medium TI version of the IR sequence provides excellent gray±white matter contrast and displays white matter with high signal intensity and gray matter with a lower signal intensity. This sequence is particularly useful for scan-

Inversion Recovery and Spin Echo (SE) sequences at 1.0 T

Age

Type

TR (ms)

TE (ms)

TI (ms)

IR medium TI

30 30 30 30 30 240 160 30

1500 1200 950 800 700 2100 2100 130±150

20 120 20/80

Inversion±recovery sequences 29±34 weeks 35±39 weeks 40 weeks to 3 months >3 months to 2 years >2 years <3 months >3 months All ages

STIR short TI

7000 6500 3800 3500 3200 6000 6000 3800

Spin echo sequences All ages 42 years >2 years

T1-weighted T2-weighted T2-weighted

860 3500 2500

FLAIR long TI

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

3

ning the developing brain, especially for assessment of myelination, but the TI requires adjustment to accommodate the changing water content of the brain with age (see Figure 4 and Section 5). The long TI, long TE version of this sequence is designed to display the CSF with low signal intensity and heavy T2weighting. The acronym for this sequence is FLAIR (¯uid attenuated inversion±recovery sequence) and it provides better lesion conspicuity than conventional T2 spin echo sequences (Figure 2).7 It is not always helpful in the immature brain, but in older children we have found it of great help for lesions with a periventricular distribution. 4.2

Figure 1 Female infant aged 9 months. Short TI inversion recovery (STIR) sequence (IR 3000/30/125). Pathological white matter adjacent to the posterior horns of the lateral ventricles is seen (arrows). The CSF is displayed as moderate signal intensity and the myelin as low signal intensity

The Spin Echo Sequence

The short TR TE T1-weighted spin echo sequence has less gray±white matter contrast than the medium TI inversion recovery sequence. However, the former sequence is quicker and is recommended for use as the ®rst sequence in all studies on sedated infants under 2 years of age. It is extremely useful for the assessment of brain swelling, hemorrhage, cystic change, and contrast enhancement with gadopentetate dimeglumine. With the T2-weighted spin echo sequence it is necessary to increase the TE to 120±200 ms in order to identify long T2 lesions against the background of the long T2 of the immature brain.8 However, the extended TE results in high signal inten-

Figure 2 Male infant aged 1 year. (a) T2-weighted spin echo (SE 2700/120) and (b) FLAIR (IR 7203/240/2100) sequences. A porencephalic cyst is seen in both images in the left hemisphere. Gliosis is shown as high signal intensity in both sequences; however, it appears more extensive and is seen with greater conspicuity in (b) (arrows). Note the low signal intensity of CSF within the porencephalic cyst with the FLAIR sequence

4 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS sity of CSF, which makes it dif®cult to detect lesions in a periventricular distribution. 4.3 Diffusion Weighted Imaging Image contrast in diffusion weighted sequences depends on the molecular motion of water. These sequences are particularly useful for demonstrating early myelination before it can be seen with conventional sequences.9 It also shows regions of focal infarction and diffuse ischemic anoxic brain injury which are not visible on spin echo and inversion±recovery images in the early phase (see Section 9). 4.4 Magnetization Transfer Techniques These sequences have proved useful for improving lesion contrast both for short and long T1 lesions10 and in newborn infants the conspicuity of some cystic lesions is improved. It is also a useful technique for demonstrating myelination. 4.5 MR Angiography MR angiography (MRA) is a noninvasive method of examining cerebral blood ¯ow, although its use in pediatric practice has so far been limited. Cerebral blood ¯ow is slow in infants and the blood vessels are smaller than in adults. A recent innovation in this technique has been the implementation of magnetization transfer MRA which involves the application of a specialized frequency-selective pulse followed by the desired MRA sequence. This results in a darkening of the background static tissue with an increase in contrast with the ¯owing blood. Our initial studies using two- and three-dimensional time-of¯ight angiography have been of value in showing the internal carotid and basilar arteries and the proximal regions of the anterior, middle, and posterior cerebral arteries in the term infant (Figure 3). Large veins are also well shown. 4.6 The Gradient Echo (GE) and Partial Saturation (PS) Sequences The gradient echo and partial saturation sequences have a speci®c role in pediatric practice, especially in the detection of neonatal hemorrhage. Low signal regions in and around hematomas are seen with higher sensitivity with partial saturation sequences than with spin echo sequences, indicating a signi®cant contribution from susceptibility effects.11 Phase maps can be derived from two partial saturation sequences with different values of TE, and are particularly useful for looking at the susceptibility effects of intracerebral hemorrhage.12 The use of these sequences in fast imaging techniques increases the speed of MR examinations, which is of particular interest when scanning infants; however, high signal from CSF remains a problem.13 A GE sequence with a TE of 29 ms and a ¯ip angle that corresponds to the choice of TR is recommended at 1.5 T to optimize the detection of calci®cation in the brain.14 4.7 Volume Imaging This may be of particular value in infants and children because reconstruction and reslicing of images can provide a

Figure 3 Three-dimensional time-of-¯ight (TR 44, TE 8) angiogram: an 18-month-old male infant with known infarction of the left middle cerebral artery. The anterior, posterior, and middle cerebral arteries are seen, but there is a paucity of vessels on the left in the anterior branches from the middle cerebral artery (arrows)

precise correction for differences in registration of follow-up examinations so that subtle differences in growth and development can be recognized.

5

NORMAL DEVELOPMENT AND MRI APPEARANCES

During the ®rst 2 years of life the pediatric brain changes rapidly as physiological myelination takes place. This continues at a slower rate into the second decade. The process of myelination begins in utero and is ®rst seen in the cerebellum and brainstem and then spreads from the posterior limb of the internal capsule to the postcentral gyrus. After birth, white matter develops in a predictable manner with sensory tracts myelinating before motor tracts from dorsal to ventral and from central to the peripheral areas of the brain.15,16 Several descriptions of this process with MRI using a variety of sequences are available in the literature.2,3,17±20 The pattern of development from 29 weeks gestational age to 1 year of age is shown by IR sequences in Figure 4. 5.1

Delays or De®cits in Myelination

Delays or de®cits in myelination are dif®cult to recognize before 3±6 months of age, since relatively little myelin is present. Conversely, after 2 years of age there is time for cases of delayed myelination to `catch up'. As a result, delays or de®-

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

5

Figure 4 Infant born at 29 weeks gestational age and examined at 31 weeks gestational age (IR 7096/30/1500). At the level of the diencephalon (a) no myelin is seen; however, the posterior limb of the internal capsule is seen as low signal intensity (arrows), the remaining white matter is featureless. The cortex is of high signal intensity and the gyral pattern is underdeveloped. At the level of the pons (b) the corticospinal tracts are of low signal intensity (SI); however, myelin is seen in the medial longitudinal fasciculus and the inferior cerebellar peduncle (arrows). Term infant (IR 3500/30/950) (c±d). At the level of the diencephalon (c) the cortex remains of higher SI than unmyelinated white matter, but more structure is seen than in (a). Myelination is present in the ventrolateral nuclei in the thalami and in the posterior limb of the internal capsule (arrows). The lentiform nuclei have relatively high signal compared to the caudate nuclei and the thalami. At the level of the pons (d) myelin is seen in the medial lemnisci (arrows). At the age of 2 months at the level of the pons (e) myelin is present in the corticospinal tracts (arrows) anterior to the medial leminisci and further myelination is seen in the cerebellum. (f, g) Infant aged 12 months (IR 3200/30/800). At the level of the diencephalon (f) myelination is now seen throughout the internal and external capsule, along the occipitothalamic pathways and the corpus callosum. At the level of the mesencephalon (g) myelin is seen in the crus cerebri (arrows)

6 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS cits are most obvious from 6 to 24 months of age. With only limited information about the normal range we have preferred to use age-matched controls and diagnose delays only in the absence of or marked reduction in myelination of named tracts or commissures relative to controls, with both examinations performed using the same technique. Delays or de®cits in myelination have been recognized following probable intrauterine rubella infection, in posthemorrhagic hydrocephalus, after hypoxic ischemic encephalopathy, infarction, periventricular cystic leukomalacia, and metabolic disease. 6 INTRACRANIAL HEMORRHAGE Intraventricular/periventricular hemorrhage (IVH/PVH) is the most important and most common type of hemorrhage in infants of less than 32 weeks gestation, with 90% of the hemorrhage occurring in the germinal matrix adjacent to the heads of the caudate nuclei.21,22 The basal ganglia are the most vulnerable site for hemorrhagic lesions in term infants suffering from severe birth asphyxia.22 Ultrasound is used to detect and monitor the progression of IVH/PVH in the early stages in infants too sick to transport to the MR unit. However, MR is useful in the stable neonate23 and for long-term follow-up. The MR appearance of hemorrhage parallels that seen in adults; however, in neonates hemorrhage occurs against a background of long normal values of brain T1 and T2, so that the T2 of hematoma may appear distinctly shorter than that of the surrounding brain (Figure 5). The signal characteristics of subdural and extradural hematomas are similar to those seen in adults. Subarachnoid blood, especially along the Rolandic ®ssure, is a common ®nding in the newborn term infant (Figure 6).

Figure 6 Normal term infant examined at 24 hours of age (IR 3800/ 30/950). Note the high signal intensity of the subarachnoid blood along the Rolandic ®ssure (arrows)

The ability to demonstrate methemaglobin and hemosiderin by a shortening of T2* and susceptibility effects provides a high degree of sensitivity, and may function as a marker of early hemorrhage years after the event.24 The late sequelae of intracranial hemorrhage may have important consequences such as hydrocephalus.

Figure 5 Female infant with IVH/PVH born at 26 weeks gestational age and examined at 40 weeks gestational age. (a) Inversion±recovery (IR 6000/30/1200) and (b) T2-weighted spin echo (SE 2700/120) sequences. Hemorrhage is seen as high signal intensity on the IR image (a), but is seen more clearly on (b) as low signal intensity (arrows). Myelin in the posterior limb of the internal capsule is better seen in (a) than in (b)

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

7

7 PERIVENTRICULAR LEUKOMALACIA

8

The second most frequent condition to af¯ict premature infants is periventricular leukomalacia (PVL),21,22 which occurs in a characteristic distribution in white matter around the lateral ventricles. Ultrasound scanning is generally used to make the diagnosis of leukomalacia; however, periventricular cysts and subcortical cysts are clearly seen with MRI in the early neonatal period and the changes can be quite extreme. Sometimes the cysts coalesce and become continuous with the adjacent ventricles producing hydrocephalus. The presence of severe or moderately severe cysts in infancy is frequently associated with a delay or de®cit in myelination (Figure 7). The MRI features are closely correlated with clinical outcome.25 For example, infants with mild periventricular change on MRI have a mild spastic diplegia and normal intellectual development, whereas infants with subcortical cysts, diminution of white matter and microcephaly are severely mentally retarded with quadriplegia and seizures. In older children PVL is seen, with a T2-weighted spin echo sequence, as an increased signal intensity usually in white matter in the centrum semiovale and adjacent to the anterior and/or the posterior horns of the lateral ventricles. We have found greater conspicuity of the lesions with the FLAIR and the short TI inversion±recovery sequences. To date, MRI has been most useful for studying the late stages of cystic leukomalacia.26 MRI also permits retrospective diagnosis of the condition in infants who were not scanned either during the neonatal period or were scanned at a stage of evolution when the cysts were not apparent.

Cerebral infarction is most commonly seen in term infants and appears as a region of increased T1 and T2 which may be dif®cult to distinguish from areas of unmyelinated white matter normally present in the brain. There may be a loss of gray± white matter contrast and loss of the normal gyral pattern in a focal area, with or without a slight increase in T1. The use of diffusion weighted imaging, where contrast is mainly determined by differences in the molecular motion of water (and not T1 and T2), may show the extent of the lesion within hours of the onset of injury (Figure 8).27 However, 1±4 days after the onset of symptoms, changes in T1 and T2 occur and the region of infarction may become visible with standard imaging techniques. With increasing age there is some apparent shrinking of the lesion and a porencephalic cyst may develop. Some porencephalic cysts decrease in size, some remain the same size, and others appear smaller due to increasing head size (Figure 9). Associated hydrocephalus with porencephalic cysts communicating with the ventricles may produce an apparent increase in the size of cysts. Wallerian degeneration within the corticospinal tracts can be seen as early as 7 days after the insult in newborn infants, with atrophy of the brainstem occurring by 3 months of age. These ®ndings occur much earlier in children than in adults.28,29

Figure 7 Male infant with severe cystic periventricular leukomalacia born at 33 weeks and examined aged 11 months (IR 1800/44/600). There is a severe delay in myelination compared with that seen in Figure 4(f)

9

INFARCTION

HYPOXIC ISCHEMIC ENCEPHALOPATHY

Hypoxic ischemic encephalopathy (HIE) is most frequently seen in term infants22 and the pattern of damage and its evolution is very variable even in children who appear to be seriously affected immediately after birth. Speci®c early MR features include brain swelling and increased signal intensity on the inversion±recovery sequence in the cortex which is usually the most marked around the Rolandic ®ssure with decreased signal intensity on the T2-weighted spin echo sequence. Loss of the normal signal intensity in the posterior limb of the internal capsules as well as focal hemorrhagic lesions in the basal ganglia, which are most commonly located in the lentiform nuclei, are also seen (Figure 10). Brain swelling is not seen after the ®rst 4 days and the evolution of the early MR ®ndings may include breakdown of white matter into subcortical cysts within the ®rst 2 weeks of life, with diminution of white matter and a severe delay in myelination by 3 months of age. The basal ganglia lesions become less obvious with time and may regress completely. Development of myelination in the posterior limbs of the internal capsules may occur by 3 months of age. Alternatively, the lesions in the basal ganglia and thalami become cystic and this may occur as early as 17 days of age with atrophy of the basal ganglia by 6 weeks of age. The degree of cortical highlighting is also variable and diminishes with time; however, in our experience it may still be present up to 6 months of age. Early diffusion weighted imaging is important in these children (Figure 11) and correctly predicts the sites of injury which become more obvious on standard imaging at a later stage. These striking early MR ®ndings closely correlate with the location of the selective neuronal necrosis21,22 seen at post mortem in asphyxiated infants.

8 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

Figure 8 Cerebral infarction in a male infant born at 40 weeks gestational age. (a) Transverse T1-weighted spin echo (SE 720/20), (b) T2-weighted spin echo (SE 3000/120), (c) diffusion weighted (SE pulse interval/200 ms, A-P sensitization, b = 600 s mmÿ2), and (d) diffusion weighted (SE pulse interval/200 ms through plane sensitization, b = 600 s mmÿ2) images. The infarction is dif®cult to recognize on (a) or (b) but is readily apparent as a high signal region on (c) and (d) (arrows)

The prognosis for infants with global hypoxia can be devastating. However, new cerebroprotective drugs are at present undergoing investigation in stroke models, but these therapies are only effective in the ®rst few hours after birth before the onset of secondary energy failure.30 It is hoped that these MR ®ndings may be used to monitor such treatment in the future. 10

HYDROCEPHALUS

Hydrocephalus can arise in a number of circumstances in children.31 The ventricular size can be readily assessed, and MRI has obvious advantages in the long-term follow-up of children with shunts. It is also possible to recognize periventricular edema both with spin echo and inversion±recovery sequences; this condition may regress following satisfactory ventricular shunting. The periventricular changes are displayed as increased signal intensity in the periventricular regions in some cases of hydrocephalus, probably indicating transependymal spread of ¯uid. We have found the STIR sequence (CSF displayed as moderate signal intensity) and the FLAIR sequence (CSF displayed as low signal intensity) are better than the conventional spin echo sequence for looking at periventricular change in children with hydrocephalus (Figure 12). However, similar changes may be seen in other diseases such as periventricular leukomalacia, and the changes are not

speci®c. It is not always possible to distinguish between acute and chronic hydrocephalus. Hydrocephalus caused by aqueduct stenosis and other obstructive lesions is generally well displayed with MRI and the ability to scan in more than one plane is very useful in these conditions. 11

CONGENITAL MALFORMATIONS

Brain development follows a well de®ned sequence. A disturbance at any particular time may affect one or more stages and result in a developmental anomaly.2,3,32±34 Dorsal induction occurs during week 3±4 of gestation when the nasal plate folds to form the nasal tube. Failure to close caudally results in myelomeningocoele, and failure to close at the cephalad end may result in anencephaly encephalocele, etc. In the next stage the mesencephalon divides to form the telencephalon. Failure at this stage produces prosencephaly. Normal proliferation then follows in which the germinal matrix forms the neurons that form the cortex. Neurons may fail to form the normal cortical layers, or stop along their path resulting in multiple cortical abnormalities including heterotopia. Neuro®bromatosis, tuberous sclerosis, and Sturge±Weber disease are the common neurocutaneous diseases occurring in children.35 However, the MRI features may be dif®cult to de®ne in the neonate, before some degree of myelination has

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

9

Figure 9 Infant aged 9 months with a left middle cerebral infarct. (a, c) Inversion±recovery (IR 3400/30/800) and (b) T2-weighted spin echo (SE 2700/120) images. A large porencephalic cyst is seen in the left hemisphere. Increased signal intensity is seen along the length of the posterior limb of the internal capsule in (b) and low signal intensity is seen in (a) (arrows). The mesencephalon on the left is smaller than on the right (c) (arrows). Less myelin is seen on the left hemisphere than on the right

taken place. MRI is of considerable value in demonstrating tumors associated with the phakomatoses as well as MRI regions of gliosis, hematomas, and cerebral atrophy.36 Calci®cation in tuberous sclerosis is poorly shown, but may be seen.37 Computed tomography (CT) or plain skull X-rays may be more useful in this situation. Obvious anatomical deformations are readily shown, e.g. anencephalopathy, holoprosencephaly, and Dandy±Walker syndrome. The sagittal plane lends itself to demonstration of many of these conditions including, for example, agenesis of the corpus callosum.

12

WHITE MATTER DISEASE

The most common white matter disease to affect infants is periventricular leukomalacia (see Section 7). Recognition of other white matter disease is dif®cult in early life because of the lack of myelin present at birth and in the ®rst 6 months of life. Once myelin has been laid down it can disappear in two principal ways, namely demyelination and dysmyelination.38 In demyelination the breakdown of myelin is caused by extrinsic factors (e.g. infection, trauma, chemotherapy), and in dysmyeli-

10 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

Figure 10 Male infant born at 40 weeks with severe birth asphyxia: T1-weighted spin echo (SE 860/20). Images at age 2 days (a) and 10 days (b); (c) inversion±recovery (IR 3797/30/95) image at age 5 weeks; (d) IR 3405/30/800 at age 6 months. At age 2 days brain swelling is present and there is a loss of sulcal patterns and gray±white matter contrast. The ventricles and the interhemispheric ®ssure are small (a). At 10 days, the brain swelling is no longer present. High signal intensity is seen throughout the cortex and the white matter is featureless. Hemorrhagic lesions are seen with the globus pallidus (arrows). There has been loss of the normal increased signal intensity in the posterior limb of the internal capsule (b). At 5 weeks (c) there has been a diminution in white matter and cystic change is noted in the posterior lobes. Myelin is now present in the posterior limb of the internal capsule. At 6 months (d), there has been further loss of white matter relative to normal. Abnormal signal intensity is seen in the thalami and globus pallidus. Further myelination has taken place

nation there is a genetic disorder of myelin formation which is a feature of metabolic disorders such as the leukodystrophies. Diffuse abnormalities are seen within white matter in leukodystrophy. The changes are usually extensive and not con®ned to the periventricular region. In other forms of white matter disease, such as Alexander's disease, changes may be con®ned to the frontal lobes. A variety of other abnormalities have been described in different forms of leukodystrophy.2,38 We have also seen periventricular abnormalities associated with intrathecal methotrexate therapy in leukemia. Along with delays in myelination, both demyelination and dysmyelination are readily recognized on MRI in children.2,38,39 13

INFECTION

Excellent reviews on in¯ammatory diseases of the brain in childhood are available in the literature.2,40 Cerebral abscess

displays an increase in T1 and T2. Edema is well displayed but the exact margins of the abscess may be dif®cult to de®ne.40 However, ring enhancement after the intravenous injection of gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA) may help in the differential diagnosis of cerebral abscess. Calci®cation associated with abscess is poorly demonstrated in comparison with CT. In two cases of brainstem encephalitis, changes have been seen with very little associated mass effect. This has been the main distinction between tumors at the initial examination and on regression on follow-up examination. It provides strong support for the diagnosis, although a certain amount of caution is necessary as patients are frequently treated with steroids which may result in some regression of edema associated with a tumor. In a case of cytomegalovirus, marked white matter change has been observed in a child of 2 years without any overt clinical signs (Figure 13). In neonatal meningitis, contrast enhancement may be seen in the meninges (Figure 14).

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

11

Figure 11 Hypoxic ischemic encephalopathy on day 1: (a) transverse T1-weighted spin echo (SE 720/20), (b) T2-weighted spin echo (SE 3000/ 120) (c) diffusion weighted (SE pulse interval/200 ms, left±right sensitization, b = 600 s mmÿ2), and (d) diffusion weighted (SE pulse interval/200 ms, through-plane sensitization, b = 600 s mmÿ2) images. High intensity signal is only seen in the medial right occipital lobe in (b) (arrows), suggesting the diagnosis of infarction; much more extensive abnormalities are seen bilaterally in the frontal, temporal, and occipital lobes using diffusion weighted imaging (c and d) (arrows showing some of the high intensity signal areas)

14

TUMORS

In general, the features of tumors in children parallel those in adults; however, there is a higher incidence of tumors in the posterior fossa, and embryological tumors are more common.2 The high incidence of midline tumors lends itself to sagittal imaging, and the clarity with which the posterior fossa is seen is also an advantage. Most tumors display an increased T1 and T2 providing high contrast with long TE long TR spin echo sequences, although distinction between tumor and edema may be dif®cult. Differentiation between brainstem and cerebellar sites is reasonably easy. Craniopharyngiomas and various other lipid-containing tumors may show characteristic features. Hamartomas may not display a signi®cant change in T1 and T2 and may then need to be recognized by their indirect signs. Hypothalamic tumors which may be poorly seen with CT are well recognized with MR. 15

OTHER DISEASE

Certain other conditions are worth reviewing, although they are quite rare. Delays or de®cits in myelination have been

recognized in Hurler's disease, and these may be reversed following successful bone marrow transplantation.41 Hallervorden±Spatz disease is of particular theoretical interest as a condition in which there is abnormal iron deposition in the brain, and in one case abnormalities have been seen in the basal ganglia. In Wilson's disease abnormalities are seen in the lentiform nucleus and within the thalamus;42 however, the ®ndings are not speci®c as similar changes are seen in children with Leigh's disease.2 16

FOLLOW-UP EXAMINATIONS

This is an important aspect of pediatric practice. The normal appearances including the value of T1 and T2, the presence of periventricular long T1 areas, the degree of myelination, as well as the size and the shape of the brain, all change. Pathological changes must be assessed against this changing background. The lack of known hazard is a strong incentive for pediatric MRI. Follow-up examinations in conditions in which long-term survival is expected without accumulating signi®cant X-ray dosage is important. Nevertheless, there are problems in achieving MRI scans at the same level and angulation as in the

12 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

Figure 12 Male infant aged 2 years with posthemorrhagic hydrocephalus: (a) mildly T2-weighted (SE 2500/20) and (b) moderately T2-weighted (SE 2500/80) spin echo sequences; (c) ¯uid attenuated inversion±recovery sequence (FLAIR 6500/160/2100). A shunt artifact is seen. There is an increase in signal intensity around the lateral ventricles in (a), which is less well seen in (b) and best seen in (c) (arrows)

initial studies. There is also a theoretical problem in using ageadjusted sequences since the machine parameters are different. Genuine advances in technique can also make comparison dif®cult. 17

CONCLUSIONS

Developments in pediatric MRI lag behind those of adults, but it is possible to extrapolate ®ndings in adults to children. Clinical correlation has been progressing, but the correlation is not precise and some children may have very large lesions

with relatively small clinical de®cits. Large unsuspected lesions have been found where the clinical signs are quite subtle. The capacity for repeated examination without cumulative radiation dosage problems has been of value in studying the natural history of a variety of neonatal insults. The versatility of MRI with its basic image parameters , T1, T2, chemical shift, ¯ow, susceptibility, and diffusion effects provides a wide variety of options for the various problems encountered in clinical practice. Only a small number of these options has yet been employed in pediatric practice, and a growing role for MRI in this area is certain. The application of MR may be greatly expanded by the installation of suitable

PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS

13

Figure 13 Female infant aged 2 years with cytomegalovirus; ¯uid attenuated inversion±recovery (FLAIR 6500/160/2100) sequences. Areas of long T2 are seen throughout the white matter

Figure 14 Male infant aged 6 weeks with meningitis. T1-weighted spin echo (SE 720/20) sequence before (a) and after intravenous injection of Gd-DTPA (b). Meningeal enhancement is seen in (b) (arrows)

14 PEDIATRIC BRAIN MRI: APPLICATIONS IN NEONATES AND INFANTS systems in neonatal intensive care units. To date only relatively large and stable infants have been examined, but the impact of MRI on the management of the critically ill and very premature infants may be much expanded in the future.

18

RELATED ARTICLES

Brain MRS of Infants and Children; Diffusion: Clinical Utility of MRI Studies; Intracranial Infections; MRI in Clinical Medicine.

19

REFERENCES

1. R. S. Boyer, Am. J. Neuroradiol., 1992, 13, 777. 2. A. J. Barkovich, in `Contemporary Neuroimaging', ed. D. Norman, Raven, New York, 1990, Vol. 1. 3. M. D. Cohen and M. K. Edwards (eds), `Magnetic Resonance Imaging of Children', B. C. Decker, Philadelphia, 1990. 4. M. A. Johnson, J. M. Pennock, G. M. Bydder, R. E. Steiner, D. J. Thomas, R. Hayward, D. J. Bryant, J. A. Payne, M. I. Levene, A. Whitelaw, L. M. S. Dubowitz, and V. Dubowitz, Am. J. Roentgenol., 1983, 141, 1005; Am. J. Neuroradiol., 1983, 4, 1013. 5. J. Dobbing and J. Sands, Arch. Dis. Child., 1973, 48, 757. 6. L. S. de Vries, L. M. S. Dubowitz, V. Dubowitz, and J. M. Pennock, `Colour Atlas of Brain Disorders in the Newborn', Wolfe Medical, Chicago, 1990. 7. B. De Coene, J. V. Hajnal, P. Gatehouse, D. B. Longmore, S. J. White, A. Oatridge, J. M. Pennock, I. R. Young, and G. M. Bydder, Am. J. Neuroradiol., 1992, 13, 1555. 8. M. A. Nowell, D. B. Hackney, R. A. Zimmerman, L. T. Bilaniuk, R. I. Grossman, and H. I. Goldberg, Radiology, 1987, 162, 272. 9. M. A. Rutherford, F. M. Cowan, A. Y. Manzur, L. M. S. Dubowitz, J. M. Pennock, J. V. Hajnal, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1991, 15, 188. 10. J. V. Hajnal, C. J. Baudouin, A. Oatridge, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1992, 16, 7. 11. R. R. Edelman, K. E. Johnson, R. Buxton, G. Shoukimos, B. R. Rosen, K. R. Davis, and T. J. Brady, Am. J. Neuroradiol., 1986, 7, 751. 12. I. R. Young, S. Khenia, D. G. T. Thomas, C. H. Davis, D. G. Gadian, I. J. Cox, B. D. Ross, and G. M. Bydder, J. Comput. Assist. Tomogr., 1987, 11, 2. 13. F. W. Wehrli, `Fast-Scan Magnetic Resonance: Principles and Applications', Raven, New York, 1991. 14. R. M. Henkelman and W. Kucharczyk, Am. J. Neuroradiol., 1994, 15, 465. 15. A. Feess-Higgins and J.-C. Larroche, `Le Developpement du Cerveau Foetal Humain: Atlas Anatomique', Masson, Paris, 1987. 16. P. I. Yakovlev and A. R. Lecours, in `Regional Development of the Brain in Early Life', ed. A. Minkowski, Blackwell Scienti®c, Oxford, 1967, p. 3. 17. C. B. McArdle, C. J. Richardson, D. A. Nicholas, M. Mirfakhraee, C. F. Hayden, and E. G. Amporo, Radiology, 1987, 162, 223. 18. A. J. Barkovitch and C. L. Truwit, `Practical MRI Atlas of Neonatal Brain Development', Raven, New York, 1990. 19. E. C. Prenger, W. W. Beckett., S. S. Koleias, and W. S. Ball, JMRI, 1994, 179.

20. S. M. Wolpert and T. D. Barnes, `MRI in Pediatric Neuroradiology', Mosby, St Louis, 1992. 21. K. E. Pape and J. S. Wigglesworth, `Hemorrhage, Ischemia and the Perinatal Brain', Lippincott, Philadelphia, 1979. 22. J. J. Volpe, `Neurology of the Newborn', Saunders, Philadelphia, 1987. 23. C. B. McArdle, C. J. Richardson, C. K. Hayden, D. A. Nicholas, M. J. Crofford, and E. G. Amparo, Radiology, 1987, 163, 387. 24. J. M. Gomori, R. I. Grossman, H. I. Goldberg, D. B. Hackney, R. A. Zimmerman, and L. T. Bilaniuk, Neuroradiology, 1987, 29, 339. 25. L. S. de Vries, L. M. S. Dubowitz, J. M. Pennock, and G. M. Bydder, Clin. Radiol., 1989, 40, 158. 26. L. L. Baker, D. K. Stevenson, and D. R. Enzmann, Radiology, 1988, 168, 809. 27. F. M. Cowan, J. M. Pennock, D. D. Hanrahan, K. Manji, J. D. Hanrahan, and A. D. Edwards, Neuropediatrics, 1994, 25, 172. 28. J. M. Pennock, M. A. Rutherford, F. M. Cowan, and G. M. Bydder, Clin. Radiol., 1993, 47, 311. 29. M. J. Kuhn, D. J. Mikulis, D. M. Ayoub, B. E. Kosofsky, K. R. Davis, and J. M. Taveras, Radiology, 1989, 172, 179. 30. D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delby, T. Boudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. Reynolds, Pediatr. Res., 1989, 25, 445. 31. C. R. Kitz, Semin. US, CT MR, 1988, 9, 216. 32. S. E. Byrd, R. E. Osborn, M. A. Radkorvoski, C. B. McArdle, E. C. Prengen, and T. P. Naidich, Semin. US, CT MR, 1988, 9, 201. 33. S. R. Pollei, R. S. Boyer, S. Crawford, H. R. Harnsberger, and A. J. Barkovich, Semin. US, CT MR, 1988, 9, 231. 34. M. S. Van der Knaap, and J. Valk, Am. J. Neuroradiol., 1988, 9, 315. 35. S. C. Crawford, R. S. Boyer, H. R. Harnsberger, S. R. Pollei, W. R. T. Smoker, and A. G. Osborn, Semin. US, CT MR, 1988, 9, 247. 36. J. G. Smirniotopoulos and F. M. Murphy, Am. J. Neuroradiol., 1992, 13, 725. 37. B. H. Braffman, L. T. Bilaniuk, and R. A. Zimmerman, Radiol. Clin. North Am., 1988, 26, 773. 38. J. Valk, `MRI of the Brain, Head, Neck and Spine: A Teaching Atlas of Clinical Application', Martinus Nijhoff, Dordrecht, 1987. 39. M. A. Nowell, R. I. Grossman, D. B. Hackney, R. A. Zimmerman, H. I. Goldberg, and L. T. Bilaniuk, Am. J. Roentgenol., 1988, 151, 359. 40. C. R. Fitz, Am. J. Neuroradiol., 1992, 13, 551. 41. M. A. Johnson, S. Desai, K. Hugh-Jones, and F. Starer, Am. J. Neuroradiol., 1984, 5, 816. 42. G. A. Lawler, J. M. Pennock, R. E. Steiner, W. J. Jenkins, S. Sherlock, and I. R. Young, J. Comput. Assist. Tomogr., 1983, 7, 1.

Biographical Sketch Jacqueline M. Pennock. b 1940. M.Phil., 1976. Department of Diagnostic Radiology, Hammersmith Hospital, London 1963±present. Currently, senior scienti®c of®cer. Approx. 80 publications. Metabolic and endocrine disease (with Professor Frank Doyle). Current research interests: pediatric magnetic resonance, speci®cally related to normal development and critically ill preterm and term infants (with Frances Cowan, Mary Rutherford, Lilly Dubowitz and Graeme Bydder).

PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

Pituitary Gland and Parasellar Region Studied by MRI Richard Farb and Walter Kucharczyk University of Toronto, Toronto, ON, Canada

1 ANATOMY AND THE MRI TECHNIQUE The pituitary gland is a small but important organ situated in a small bony depression in the skull base called the sella turcica. Although the average weight of the pituitary gland in the adult is only 0.5 g, it is responsible for the regulation of many of the body's most critical endocrine functions. The gland enlarges slowly with maturation reaching a peak height of no more than 9±10 mm in early adulthood. There are two periods when the gland may transiently enlarge beyond its normal dimensions, those being adolescence and pregnancy. This is principally due to physiological hypertrophy of prolactinsecreting cells. The pituitary gland is made up of three distinct lobes: anterior, intermediate, and posterior. The intermediate lobe is a vestigial remnant in humans and serves no function; however, it may be the site of an incidental cyst. The anterior lobe, or adenohypophysis, is a true endocrine organ responsible for synthesizing many hormones. Its function is regulated by stimulatory and inhibitory hormones arising from the hypothalamus. The posterior lobe of the pituitary gland, also called the neurohypophysis, is actually a downward extension of the hypothalamus. The hormones of the posterior pituitary gland, vasopressin and oxytocin, are synthesized within the hypothalamus and axonally transported to the posterior pituitary gland where they are stored and from which they are ultimately released. The anterior and posterior lobes of the pituitary gland are therefore two functionally distinct entities, although they reside in very close association within the sella turcica. The posterior lobe is much smaller than the anterior lobe and takes up only 10±20% of the sella turcica. The anterior lobe of the pituitary gland generally takes up the anterior, central, and lateral aspects of the sella turcica. The posterior lobe resides centrally in the midline just anterior to the dorsum sella. The lateral wings of the adenohypophysis usually extend laterally around the posterior pituitary gland. The anterior and posterior lobes of the pituitary gland are easily distinguishable on MRI. The anterior lobe of the pituitary gland is normally isointense with cerebral white matter on all pulse sequences, whereas the posterior lobe of the pituitary gland is easily distinguishable by its characteristic hyperintense signal on T1-weighted images. Following the administration of intravenous contrast material the anterior and posterior lobes, as well as the pituitary stalk, enhance intensely. The pituitary gland is enclosed within the sella turcica by a dural diaphragm (the diaphragma sella), the center of which

1

has a defect in it allowing for passage of the pituitary stalk. The superior surface of the diaphragma sella is lined with arachnoid, and the suprasellar cistern lies above the diaphragma sella. The suprasellar cistern contains all the vascular structures of the circle of Willis. The optic chiasm is immediately anterior to the pituitary stalk (infundibulum) within the suprasellar cistern. On either side of the sella turcica lie the cavernous sinuses. These are complex dural venous sinuses passing from the anteromedial aspect of the posterior fossa (Meckel's cave) anteriorly to the superior orbital ®ssure, transmitting cranial nerves III, IV, V1, and V2 within the lateral walls of the cavernous sinus. Cranial nerve VI is the only cranial nerve that actually passes within the cavernous sinus. The appearance of the cavernous sinuses on MRI is somewhat variable in signal intensity; however they are usually symmetric in dimensions. Following contrast administration the cavernous sinuses enhance intensely. The normal MRI anatomy is illustrated in Figure 1. Imaging of the pituitary gland and sella turcica for possible structural abnormalities is a frequent indication for MRI. Exact speci®cations of the technique are not provided here because the constantly improving capabilities of modern scanners require that technique be continually revised. However high spatial detail is very important in this area and should be achieved through the use of thin slices, a ®ne matrix size, and a relatively small ®eld-of-view. Requirements for spatial detail must be balanced against the need for adequate signal-to-noise ratio as well as imaging time. The pulse sequence for best tissue contrast is still a matter of opinion. Several groups have shown that short TR, short TE spin echo images (i.e. T1-weighted) generate very good contrast for visualizing pituitary pathology.1,2,3 Because fast spin echo (FSE) methods are now readily available, and do not incur the time penalty of conventional T2-weighted spin echo sequences, coronal T2-weighted FSE imaging is being utilized with increasing frequency as a supplementary sequence for investigation of pituitary adenomas. Several other pulse sequences have been implemented for parasellar imaging and have met with varying degrees of success.4 Paramagnetic contrast-enhanced images are widely used and are very useful. Although most adenomas are visible without the injection of intravenous contrast, several studies have shown that small adenomas may become visible only after contrast injection.5,6 To take better advantage of the differential rates of contrast enhancement between adenomas and normal pituitary gland, dynamic pituitary scanning immediately after bolus contrast injection is also useful.7

2 2.1

CONGENITAL ABNORMALITIES Pituitary Gland Hypoplasia

Congenital abnormalities of the pituitary gland and hypothalamus are usually present in association with anomalies of other midline cranial, orbital, and facial structures. Pituitary gland hypoplasia is usually accompanied by a small sella turcica, and is commonly clinically associated with growth failure and other endocrine abnormalities. A curious form of pituitary hypoplasia has recently been recognized to occur with a

2 PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

Figure 1 Normal MR anatomy of the pituitary region. T1-weighted sagittal (a), and coronal (b), images. A. Anterior pituitary gland, B. optic chiasm, C. cavernous sinus, D. pituitary stalk, E. right internal carotid artery (supraclinoid portion), F. left internal carotid artery (cavernous portion), G. medial temporal lobe, H. sphenoid sinus, I. hypothalamus, J. trigeminal nerve, K. pterygoid canal, L. sylvian ®ssure, M. mammillary body, N. frontal lobe, O. midbrain, P. pons, Q. medulla, R. interpeduncular cistern, S. genu of corpus callosum, T. tuber cinereum, U. cerebral aqueduct, V. third ventricle, W. posterior pituitary gland

slightly higher frequency in patients with a history of breech deliveries.8±10 These patients have short stature and growth hormone de®ciency, as well as the MRI ®ndings of pituitary hypoplasia, hypoplasia of the distal pituitary stalk, and absence of the normal hyperintensity within the posterior pituitary gland. 2.2 Ectopic Posterior Pituitary Tissue Occasionally, the normal hyperintensity on T1-weighted images within the posterior pituitary gland and in the posterior aspect of the sella turcica is not visualized. Instead there is a small area of hyperintensity seen along the course of the pituitary stalk. This may be associated with discontinuity of the pituitary stalk. These patients may present with hormonal abnormalities, particularly of the anterior pituitary gland, due to transection of the hypothalamo-pituitary portal anastomosis. It is rare for these patients to present with posterior pituitary gland hormonal de®ciencies. `Ectopic posterior pituitary tissue' is thought to occur by one of two mechanisms, either incomplete embryologic descent of hypothalamic neurons, or possibly due to traumatic transection of the pituitary stalk above the diaphragma sella. In either case, the neurosecretory granules descending down toward the posterior pituitary gland from the hypothalamus accumulate within the proximal stump of the amputated pituitary stalk. This simply becomes the new site or location of posterior pitu-

itary tissue thus accounting for the increased signal intensity on T1-weighted images as well as the normal hormonal pro®le of the posterior pituitary gland. 2.3

The Empty Sella Turcica

The terms `empty sella syndrome' or `empty sella turcica' refer to the roentgenographic, pneumoencephalographic, or computed tomographic appearance of the sella turcica, which is ®lled predominantly with cerebrospinal ¯uid (CSF), with the normal posterior pituitary gland and stalk being crowded posteriorly and inferiorly within the sella turcica. The cause of this displacement is the presence of a large defect in the diaphragma sella allowing the arachnoid membrane to herniate through the diaphragma sella adjacent to the stalk and thus allowing CSF pulsations to enter the sella turcica and result in the eventual displacement of the pituitary gland and stalk posteriorly. It is actually quite rare for patients with an `empty sella' to have symptoms referable to the area of the sella turcica. `The empty sella' is most commonly seen as an incidental ®nding of little or no clinical signi®cance. 2.4

Cephaloceles

Cephaloceles are herniations of the meninges (meningocele) or of the meninges and the brain (meningoencephalocele) through a congenital cranial defect. Cephaloceles in the sellar

PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

region (transphenoidal encephaloceles) are very rare. Most of these are associated with other midline anomalies, particularly agenesis of the corpus callosum.

tumor. This must also be interpreted in conjunction with endocrinologic markers. 3.2

3 TUMORS AND TUMOR-LIKE CONDITIONS 3.1 Pituitary Adenoma The most common tumors affecting the pituitary gland are pituitary adenomas. These are benign, slow growing, epithelial adenomas originating in the adenohypophysis. Pituitary adenomas are usually well demarcated lesions that are separated from the normal pituitary gland by a pseudocapsule of compressed tissue. Pituitary adenomas are commonly classi®ed based on size and hormonal activity. Adenomas measuring greater than 1 cm in diameter are referred to as macroadenomas; those less than 10 mm are microadenomas. Those adenomas that are hormonally active may also be referred to by the hormone that they secrete. For example, the most common hormonally active adenoma is the `prolactin-secreting microadenoma', or simply prolactinoma. The various hormonal types of adenomas are indistinguishable from one another by MRI imaging. On MRI, pituitary microadenomas are usually small, focal hypointensities within the pituitary gland on T1-weighted images. On T2-weighted images the corresponding lesion is seen as hyperintense to the surrounding pituitary tissue.3 Approximately 80±95% of pituitary adenomas present with these characteristic signal intensities (Figure 2). Intratumoral hemorrhage occurs in 20±30% of adenomas. These are usually macroadenomas (Figure 3). The incidence of intratumoral bleeding is higher in patients receiving bromocryptine therapy.11 We have found that tumors that are hyperintense on T1-weighted images are always cystic with the cyst containing elements of previous hemorrhage. Tissue contrast represented by a differential signal intensity between the tumor and normal pituitary gland is the most sensitive and reliable indicator of the presence of a microadenoma (Figure 2). Indirect signs previously utilized for computed tomography imaging of the pituitary gland such as tilt of the pituitary gland, contour of the superior aspect of the pituitary gland, and deviation of the pituitary stalk, all appear to be quite insensitive to the presence of microadenomas and indeed may be misleading (Figure 2). It is dif®cult to determine the accuracy of MRI for the diagnosis of pituitary microadenomas. It is estimated that about 90% of microadenomas are detected and accurately localized with MRI.3 The detection rate of macroadenomas approaches 100%. Their signal intensities are qualitatively similar to their smaller counterparts; however they more commonly demonstrate cystic degeneration or hemorrhage. An area of continued dif®culty in pituitary and parasellar imaging is that of the postoperative examination in search of residual or recurrent pituitary adenoma. In these cases, it may be very dif®cult to distinguish postoperative scarring, or graft material, from the normal gland or adenomatous tissue. This is especially true in the ®rst 6 months after surgery. In these cases, progressive growth of a soft tissue mass on sequential postoperative MR scans is the best imaging sign of recurrent

3

Craniopharyngioma

Craniopharyngiomas are epithelial-derived neoplasms thought to arise from remnants of Rathke's cleft in the region of the pars tuberalis. They account for 3% of all intracranial tumors and show equal incidence in males and females. Tumors can vary greatly in size from several millimeters to several centimeters in diameter with the epicenter of the tumour usually located in the suprasellar cistern. Craniopharyngiomas typically have both solid and cystic components (Figure 4). Calci®cation is commonly seen in the solid portion of the tumor. The cystic contents of the tumour can vary in color and viscosity. On MRI craniopharyngiomas are typically lobulated with heterogeneous signal intensities on T1- and T2-weighted images.12 The cystic component of the tumor is uniform and commonly hyperintense on both T1- and T2-weighted images, and contains ¯uid that has been likened to `machine oil'. 3.3

Rathke's Cleft Cyst

Rathke's cleft cysts share a common origin with craniopharyngiomas in that they originate from remnants of squamous epithelium of Rathke's cleft. The cysts are common incidental ®ndings at autopsy; however the larger cysts can be symptomatic. The contents of the cysts are typically mucoid. Less commonly they are ®lled with serous or desquamated cellular debris.13 The mucoid-containing cysts can be hyperintense on T1 and T2-weighted images; the serous-containing cysts have signal intensities that closely match CSF (Figure 5). Rathke's cleft cysts do not enhance following contrast injection, except for perhaps marginal enhancement around the cyst wall. If nodular enhancement or calci®cation is seen, craniopharyngioma should be suspected. 3.4

Meningioma

Approximately 10% of meningiomas occur in the parasellar region. These tumors arise from a variety of locations around the sella including the tuberculum sella, clinoid processes, medial sphenoid wing, and cavernous sinus. Meningiomas are most frequently isointense relative to gray matter on unenhanced T1-weighted images. Approximately 50% remain isointense on T2-weighted images.14,15 Meningiomas, as well as other extraaxial tumors, can demonstrate a typical CSF cleft between the tumor and the adjacent brain parenchyma. Also, meningiomas have been noted to demonstrate a peripheral black rim around their margins, thought to represent veins surrounding the tumor. Meningiomas commonly encase arteries of the parasellar region, particularly the cavernous and supraclinoid portions of the internal carotid artery. Meningiomas will typically narrow the vessels they encase (Figure 6). Adenomas, though they commonly encase vessels, usually do not narrow these vessels. Other distinguishing features of meningiomas include intense enhancement with intravenous contrast, tumor calci®cation, and hyperostosis of adjacent bone.

4 PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

Figure 2 Pituitary microadenoma. (a) T2-weighted, (b) T1-weighted, and (c) enhanced T1-weighted images. This 40-year-old female was noted to have marked elevation of cortisol in the blood and urine due to this hormonally active tumor secreting adrenocorticotropic hormone. This tumor demonstrates the typical features of a microadenoma including hyperintensity on T2-weighted images and hypointensity on T1-weighted images. Note the differential signal intensity between the intensely enhancing normal, anterior pituitary gland and the less enhancing tumor. Note also how the pituitary stalk deviates toward the tumor, thus stressing the unreliability of contralateral stalk deviation as a lateralizing sign of tumor

3.5 Germinoma and Teratoma Germinomas account for less than 2% of primary intracranial tumors. Most of these tumors occur in the pineal region. However, approximately 20% of these tumors occur in the suprasellar cistern or pituitary fossa. Suprasellar germinomas occur either as metastatic deposits from a pineal region tumor or arise primarily in the suprasellar cistern. Most patients present between 5 and 25 years of age. Although pineal region germinomas are most commonly seen in males, there is no sex

predilection for the primary suprasellar germinoma. Other tumors of germ cell origin occurring in this region, albeit rarely, include yolk sac tumors, choriocarcinoma, embryonal cell carcinoma, and teratoma. Suprasellar germinomas are usually large midline tumors with a propensity to in®ltrate and spread by the CSF pathways. In contrast to craniopharyngiomas, germinomas are homogeneous and only rarely have cystic components. Germinomas are mildly hypointense on T1weighted images and hyperintense on T2-weighted images. Marked uniform contrast enhancement is the rule (Figure 7).

PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

5

dermoid tumors show evidence of fatty component as well as small areas of calci®cation. 3.7

Metastases

Metastases to the pituitary gland are a common autopsy ®nding but only about 1±2% of cancer patients have symptomatic metastases to the pituitary gland. These are usually metastases from breast or bronchogenic carcinoma. 3.8

Figure 3 Pituitary macroadenoma. Nonenhanced T1-weighted image demonstrating a large, lobulated tumor occupying the sella and extending superiorly into the suprasellar cistern displacing the optic chiasm (A) superiorly. A focal area of increased signal intensity represents a small hemorrhage within the tumor. The tumor only abuts and does not clearly invade the cavernous sinus regions bilaterally

Chordoma

Chordomas are rare neoplasms arising from remnants of the primitive notochord. Most cranial chordomas are found in the midline in relation to the clivus. Chordomas are locally invasive and destructive and commonly result in large, lobulated, destructive lesions extending in an extradural fashion. Portions of the sella and the parasellar regions are commonly involved. Chordomas are characteristically isointense or hypointense on T1-weighted images and hyperintense on T2-weighted images17 with moderate enhancement following intravenous contrast administration.

Teratomas have mixed heterogeneous signal intensities, often demonstrating evidence of fat or calci®cation. 3.6 Epidermoid and Dermoid Epidermoids and dermoids are benign, slow-growing `inclusion tumors' that can occur intracranially or intraspinally resulting from inclusions of epithelium during the time of neural tube closure. Dermoids are approximately one-®fth as common as epidermoids. Epidermoids account for approximately 1% of all intracranial neoplasms. The cerebellopontine angle is the most frequent site for epidermoids with the parasellar region being the second most common site. Epidermoids are generally cystic tumors with the walls composed of simple, strati®ed, squamous epithelium resting on a layer of connective tissue. The interior of the cyst is composed of desquamated keratin products of the cyst wall. These tumors are benign and grow slowly with expansile characteristics. They also typically insinuate within and around structures, or conform to the space within which they are situated. Rarely the cyst can rupture producing a chemical granulomatous meningitis. Epidermoids are typically only slightly hyperintense to CSF on T1- and T2-weighted images, making the proton density image the most valuable sequence for identi®cation of the epidermoid tumor. Epidermoids are characteristically hyperintense to brain and CSF on the proton density sequence.16 Unfortunately epidermoids can also retain signal intensities identical to CSF making them nearly indistinguishable from arachnoid cysts. Enhancement, if it occurs, occurs only along the periphery of the tumor. Dermoids are more heterogeneous. Most

Figure 4 Craniopharyngioma. T1-weighted, enhanced, midline sagittal image of the parasellar region in this 3-year-old male demonstrating the typical ®ndings of a craniopharyngioma. Note the enhancing solid tumor A, and the large cystic portion of the tumor B, extending supero-posteriorly, draping over the dorsum sella into the posterior fossa displacing the pons, C

6 PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI 4

4.1

INFECTIONS AND INFLAMMATORY CONDITIONS Infections

Infections of the pituitary gland are a rare occurrence. Previous viral infection has been proposed as an etiology for diabetes insipidus. Leptomeningeal tuberculosis at the base of the brain can involve the pituitary gland. Bacterial infection of the pituitary gland is usually unnoticed until it manifests as a pituitary abscess, an entity which is also exceedingly rare. Pituitary abscesses are most frequently seen in association with preexisting conditions such as craniopharyngioma or pituitary adenoma. Transphenoidal surgery is rarely complicated by infection. Infection of the parasellar regions, in particular the cavernous sinus is also very rare. Cavernous sinus thrombophlebitis is thought to be secondary to venous spread of a septic embolus from a periorbital or perinasal source.

Figure 5 Rathke's cleft cyst. T1-weighted, enhanced coronal image of the sella region demonstrating a somewhat bilobed cystic mass A, extending from the sella into the suprasellar cistern through the diaphragma sella B, displacing the optic chiasm superiorly C. This 28year-old male presented with bitemporal hemianopsia and normal pituitary function. This typical appearance of the Rathke's cleft cyst has been likened to the appearance of a snowman

4.2

Lymphocytic Hypophysitis

Lymphocytic hypophysitis is characterized by diffuse in®ltration of the adenohypophysis by lymphocytes. It is most

3.9 Arachnoid Cysts Approximately 15% of arachnoid cysts occur in the juxtasellar region.18 Suprasellar arachnoid cysts are thought to be developmental in origin and arise from an imperforate membrane of Liliequist. They have MRI characteristics identical to normal CSF. 3.10 Hamartoma of the Tuber Cinereum Hamartomas of the tuber cinereum are sessile or pedunculated masses extending inferiorly from the hypothalamus into the suprasellar cistern between the pituitary stalk and mammillary bodies. Histologically they resemble cerebral cortex with little histologic similarity to the normal hypothalamus. Hamartomas of the tuber cinereum can be associated with precocious puberty. On MRI hamartomas are isointense to gray matter on all imaging sequences and should not enhance with contrast material (Figure 8).19 3.11 Eosinophilic Granulomatosis (Histiocytosis X, Langerhan's Cell Granulomatosis) Histiocytosis can involve the pituitary gland, pituitary stalk, or hypothalamus. Twenty-®ve per cent of cases develop the classical clinical triad of diabetes insipidus, exophthalmos, and lytic bone lesions (the Hand±Schuller±Christian syndrome). In these cases granulomas can be found in the hypothalamus or the pituitary stalk. With MRI there is thickening of the pituitary stalk, hyperintensity on T2-weighted images, and intense enhancement following contrast administration.20,21

Figure 6 Meningioma. This T1-weighted, enhanced coronal image demonstrates a homogeneously enhancing meningioma arising from the left anterior clinoid. Note the encasement of the left supraclinoid internal carotid artery, which is slightly narrowed when compared with the cavernous portion of the artery

PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI

4.3

7

Sarcoidosis

Sarcoidosis most commonly occurs in an intrathoracic location. Any part of the central nervous system however can also be involved and 5% of patients with sarcoidosis will have neurologic complaints. The most typical ®ndings for neurosarcoidosis include a basal meningitis with an abnormal, thick, enhancement pattern involving particularly the leptomeninges at the base of the brain characteristically in the midline. Abnormal parenchymal lesions are also noted within the brain. Neurosarcoidosis can be indistinguishable from other causes of leptomeningeal disease, in particular tuberculosis and leptomeningeal carcinomatosis. 4.4

Tolosa±Hunt Syndrome

This syndrome consists of painful ophthalmoplegia associated with a lesion within the cavernous sinus which is responsive to steroid therapy. Pathologically the lesion is similar to orbital pseudotumor. Figure 7 Germinoma. T1-weighted, midline, sagittal, enhanced image demonstrating homogeneously enhancing masses within the pineal region A, and suprasellar region B. The presence of tumors in both the pineal and suprasellar regions is highly characteristic of germinoma

commonly seen in late pregnancy or the postpartum period, and has been associated with other autoimmune disorders such as Hashimoto's thyroiditis. Clinically the woman complains of headache, visual loss, and failure to resume normal menstrual cycle. MRI demonstrates diffuse enlargement of the anterior pituitary gland without focal abnormality.

5 5.1

VASCULAR LESIONS AND INFARCTION Aneurysms

The large arteries in the suprasellar region are a common site of aneurysms. Aneurysms are for obvious reasons extremely important lesions to identify correctly. Fortunately their MRI appearance is distinctive and easily appreciated.22,23 Aneurysms centrally contain signal void created by rapidly ¯owing blood. This ¯owing blood may also cause a substantial ghosting artifact in the phase-encoding direction, which is a useful diagnostic sign. Thrombus within the aneurysm is laid down against the wall in a sequential fashion over a long period of time and results in a typical laminated appearance. Hemosiderin staining of the adjacent brain can also be seen. If confusion persists regarding the diagnosis, conventional angiography remains the gold standard for diagnosis of intracranial aneurysms. 5.2

Carotid Cavernous Fistulas

Carotid cavernous ®stulas (direct or indirect type) can be spontaneous, posttraumatic, or atherosclerotic. Dural arteriovenous malformations of the cavernous sinus are another form of abnormal venous communication in this region. On MRI, dilatation of the venous structures, in particular the ophthalmic vein and cavernous sinus, is usually clearly visible. The carotid artery can also be somewhat dilated.

Figure 8 Hamartoma of the tuber cinereum. T1-weighted, enhanced, midline sagittal image demonstrating a homogeneous mass located posterior to the pituitary stalk and anterior to the mammillary bodies, extending from the region of the tuber cinereum into the posterior aspect of the suprasellar cistern. This mass did not enhance and remained isointense to gray matter on all sequences, demonstrating the typical ®ndings of hamartoma of the tuber cinereum

5.3

Postpartum Pituitary Necrosis

During pregnancy the pituitary gland increases in size thus making it vulnerable to circulatory disturbances. Postpartum ischemic necrosis of the pituitary gland is a well-known entity, which can follow complicated deliveries associated with hemorrhage and shock.

8 PITUITARY GLAND AND PARASELLAR REGION STUDIED BY MRI 6 MISCELLANEOUS CONDITIONS 6.1 Diabetes Insipidus Diabetes insipidus results from the lack of appropriate release of antidiuretic hormone (vasopressin) from the posterior pituitary gland. This results in the inability to concentrate urine and causes polydipsia and polyuria. The inability to release vasopressin in response to normal stimuli can have several possible etiologies: (1) primary failure to synthesize the hormone in the periventricular nuclei of the hypothalamus, (2) destruction of these nuclei or proximal transport pathways in the hypothalamus and proximal pituitary stalk, (3) dysfunction of hypothalamic osmoreceptors such that despite adequate vasopressin reserves, the hormone is not released appropriately. The causes of diabetes insipidus include neoplasms, in®ltrative disorders, surgery, and head trauma. Approximately 30% of cases are idiopathic. MRI is valuable in demonstrating not only the causative lesions, but also the status of the posterior pituitary gland, the storage site for the neurosecretory vesicles containing vasopressin. It is a unique feature of this gland that it is hyperintense on T1-weighted images and regardless of the etiology this signal disappears in cases of diabetes insipidus. 6.2 Hemochromatosis Hemochromatosis is a metabolic disorder in which excess iron is deposited throughout the body particularly in the solid abdominal viscera. For unknown reasons the pituitary gland is frequently involved. There is relative preferential deposition of iron within the gonadotrophic cells of the adenohypophysis. Dysfunction commonly manifests as loss of libido and hypogonadism. The characteristic ®nding on MRI is marked hypointensity emanating from the anterior pituitary gland on T2-weighted images. This is due to the high levels of iron deposited within the region resulting in susceptibility effects and loss of signal.24

7 RELATED ARTICLES Cranial Nerves Investigated by MRI; Eye, Orbit, Ear, Nose, and Throat Studies Using MRI; Head and Neck Investigations by MRI; Intracranial Infections; Temperomandibular Joint MRI.

8 REFERENCES 1. D. W. Chakeres, A. Curtin, and G. Ford, Radiol. Clin. North Am., 1989, 27, 265. 2. W. W. Peck, W. P. Dillon, D. Norman, T. H. Newton, and C. B. Wilson, Am. J. Neuroradiol., 1988, 9, 1085. 3. W. Kucharczyk, D. O. Davis, W. M. Kelly, G. Sze, D. Norman, and T. H. Newton, Radiology, 1986, 161, 761. 4. T. Stadnik, A. Stevenaert, A. Beckers, R. Luypaert, T. Buisseret, and M. Osteaux, Radiology, 1990, 176, 419.

5. J. L. Doppman, J. A. Frank, A. J. Dwyer, E. H. Old®eld, D. L. Millor, L. R. Nieman, G. P. Chrousos, G. B. Cutler, Jr., and D. L. Loriaux. J. Comput. Assist. Tomogr., 1988, 12, 728. 6. D. R. Newton, W. P. Dillon, D. Norman, T. H. Newton, and C. B. Wilson. Am. J. Neuroradiol., 1989, 10(5), 949. 7. W. Kucharczyk, J. E. Bishop, D. B. Plewes, M. A. Keller, and S. George. Am. J. Roentgenol., 1994, 163, 671. 8. W. M. Kelly, W. Kucharczyk, J. Kucharczyk, B. Kjos, W. W. Peck, D. Norman, and T. H. Newton. Am. J. Neuroradiol. 1988, 9, 453. 9. I. Fujisawa, K. Kikuchi, K. Nishimura, K. Togashi, K. Itoh, S. Noma, S. Minami, T. Sagoh, T. Hiraoka, and T. Nomaoi, Radiology, 1987, 165, 487. 10. M. Maghnie, F. Triulzi, D. Larizza, G. Scotti, G. Beluf®, A. Cecchini, and F. Severi. Pediatr. Radiol., 1990, 20, 229. 11. D. M. Yousem, J. A. Arrington, S. J. Zinreich, A. J. Kumar, and R. N. Bryan, Radiology, 1989, 170, 239. 12. E. Pusey, K. E. Kortman, B. D. Flannigan, J. Tsuruda, and W. G. Bradley, Am. J. Neuroradiol., 1987, 8, 439. 13. W. Kucharczyk, W. W. Peck, W. M. Kelly, D. Norman, and T. H. Newton, Radiology, 1987, 165, 491. 14. M. Castillo, P. C. Davis, W. K. Ross, and J. C. Hoffman, Jr., J. Comput. Assist. Tomogr., 1989, 13, 679. 15. J. W. Yeakley, M. V. Kulkarni, C. B. McArdle, F. L. Hoar, and R. A. Tang, Am. J. Neuroradiol., 1988, 9, 279. 16. D. Tampieri, D. Melanson, and R. Ethier, Am. J. Neuroradiol., 1989, 10, 351. 17. G. Sze, L. S. Uichanco, M. N. Brant-Zawadzki, R. L. Davis, P. M. Gutin, C. B. Wilson, D. Norman, and T. H. Newton, Radiology, 1988, 166, 187. 18. S. N. Wiener, A. E. Pearlstein, and A. Eiber, J. Comput. Assist. Tomogr., 1987, 11, 236. 19. E. M. Burton, W. S. Ball, Jr., K. Crane, and L. M. Dolan, Am. J. Neuroradiol., 1989, 10, 497. 20. J. B. Moore, R. Kulkarn, D. C. Crutcher, and S. Bhimani, Am. J. Pediatr. Hematol. Oncol., 1989, 11, 174. 21. R. D. Tien, T. H. Newton, M. W. McDermott, W. P. Dillon, and J. Kucharczyk, Am. J. Neuroradiol., 1990, 11, 703. 22. S. W. Atlas, A. S. Mark, E. K. Fram, and R. Grossman, Radiology, 1988, 169, 455. 23. A. Biondi, G. Scialfa, and G. Scotti, Neuroradiology, 1988, 30, 214. 24. I. Fujisawa, M. Morikawa, Y. Nakano, and J. Konishi, Radiology, 1988, 168, 213.

Biographical Sketches Richard I. Farb. b 1959. B.Sc., 1981, University of Toronto. MD, 1985, University of Toronto, Canada. Radiology residency, Wayne State University Medical Center, Michigan 1991. Certi®cation by examination by the Royal College of Physicians (Canada) and the American Board of Radiology 1991. Two year Neuroradiology Fellowship at University of Toronto completed 1993. Currently staff neuroradiologist at Sunnybrook Health Science Centre and lecturer at University of Toronto. Walter Kucharczyk. b 1955. MD, 1979, University of Toronto, Canada. Radiology residency, University of Toronto, completing in 1983. Certi®ed by examination as a Fellow of the Royal College of Physicians (Canada) in 1983, followed by two years of Neuroradiology MRI post residency training at UCSF and University of Toronto. Successively, visiting assistant professor (UCSF), assistant professor, full professor (University of Toronto). Currently Professor and Chair of the Department of Medical Imaging at the University of Toronto. Main interests: MRI of pituitary gland and parasellar area, tissue characterization using MR techniques, interventional methods with MRI.

SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS

1

Sodium-23 Magnetic Resonance of Human Subjects

sence of an electric ®eld gradient, which is described by a second-rank tensor of strength eq. In this case there is another energy term in the Hamiltonian given by

Peter M. Joseph

where  is the angle between the major axis of the quadrupolar tensor and the z axis. From this Hamiltonian comes a quadrupolar energy EQ and frequency !Q = 2EQ/h depending on the quantum state. Equation (1) has important consequences for the energy levels of the nuclei, the relaxation rates, and signal strength. First, we see that H^Q depends quadratically on m; this changes the energy level differences of the outer levels but not the inner. Hence the quadrupolar interaction does not change the Larmor frequency corresponding to the inner transition. Both the magnitude and the sign of the energy level shift of the external transitions depend on the angle . If the local environment is isotropic, then averaging over angles gives zero average energy displacement. However, in situations of exceptional order (such as liquid crystals) the orientations are not random and a net shifting of the energy levels can occur. The resulting spectrum will show three lines as illustrated in Figure 1. Note that whereas the central line, corresponding to the inner transition, is narrow, the external spectral lines may be broadened by an amount dependent on the details of the microscopic electrostatic environment. In this situation, quantum theory predicts that only 40% of the total radiated energy will lie in the unshifted inner line, while the remaining 60% may be distributed over the broad outer lines.2 Depending on the experimental conditions, it is possible that the outer transitions will be so broadened as to be virtually undetectable with most NMR spectrometers. This leads to the widely discussed phenomenon of `NMR invisibility' of sodium; i.e. several experiments have suggested that the observed Na NMR signal strength is less than what should be observed based on measured Na concentration levels. For example, Shporer and Civan proposed that this would explain previously reported discrepancies between the strength of the Na NMR signal observed and that expected from tissue measurements.3 In particular, they demonstrated this effect for the relatively simple system of sodium linoleate in water.

Hospital of the University of Pennsylvania, Philadelphia, PA, USA

1 INTRODUCTION Sodium-23 is a nucleus with spin quantum number 32 whose gyromagnetic ratio, , corresponds to a frequency of 11.262 MHz teslaÿ1, which is approximately one-quarter that of protons. It is widely but inhomogeneously distributed in body tissues. It is present in highest concentrations in extracellular water in a concentration of about 150 mM. Hence in vivo sodium NMR signals are less than one-thousandth those of hydrogen so that sodium MR images can never hope to compete with proton images in quality. Only little preliminary research effort has been directed toward the clinical utility of sodium MRI. However, Na MRI may be able to contribute unique medical/physiological information that is not available in proton MRI. Intracellular Na is typically at the level of about 10±20 mM in most healthy cells. The enormous difference between intracellular (IC) and extracellular (EC) concentrations is maintained by the Na±K pump which actively transports the two kinds of ions across the cell membrane. Certain pathologies alter the function of that pump resulting in a signi®cant rise in IC Na. There is evidence from in vitro experiments that this occurs in certain disease states, especially ischemia, infarction, and cancer. Thus one aim of Na MRI is to demonstrate a rise in the IC sodium concentrations. While there is still debate as to whether that goal has been achieved with in vivo MRI, it is one of the factors driving research interest in this ®eld. There are various subtle aspects of the Na NMR signal that are thought to differ in the IC and EC environments, and much research has been directed toward clarifying this question.

2 BIOPHYSICS OF

e2 qQ‰3m2 ÿ I…I ‡ 1†Š‰3 cos2 …† ÿ 1Š H^Q ˆ 8I…2I ÿ 1†

Inner transition

23

Na NMR

The major difference between Na and 1H NMR is that Na has spin quantum number I = 32. This means there will be four energy levels, corresponding to the quantum numbers m = Iz = ÿ32, ÿ12, 12, and 32. Since normal NMR excitation and radiation permits only transitions whose m values differ by 1, there will in general be three possible Larmor frequencies; these are classi®ed as `inner', meaning 12 to ÿ12, and `outer', meaning either 32 to 12 or ÿ12 to ÿ32. If there are no other interactions in the system, the energy levels are equally spaced and the three Larmor frequencies are equal. However, the Na nucleus also has an electric quadrupole moment, denoted by eQ.1 (e = electron charge.) This implies that the energy of the nucleus will be in¯uenced by the pre-

…1†

Outer transition

Outer transition

Frequency

Figure 1 Schematic diagram of the spectrum from 23Na in an ordered environment with electric quadrupolar interaction in a ®rst-order approximation. The width of the outer transition lines depends on the strength and correlation time of the quadrupolar interaction. The inner transition carries 40% of the total signal

2 SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS Joseph and Summers4 noted that if the loss of Na NMR visibility were in fact due to a nonoverlapping displacement of the energy levels, then quantum theory predicts that the rate of nutation of the inner (unshifted) spectral component is twice that predicted by the usual classical vector model of the NMR excitation. In other words, if an NMR spectrometer is adjusted on a homogeneous phantom (such as aqueous NaCl) so that a particular radiofrequency pulse gives 90 of spin ¯ip, then the same pulse applied to the split Na system will rotate the unsplit component by 180 . They demonstrated the vanishing of the sodium linoleate signal under conditions in which a NaCl phantom gave optimal signal.4 This suggests that this `¯ip angle effect' could be used to test for the presence of quadrupolar splitting in vivo without the need actually to detect the invisible Na spectral lines. To date, however, no one has demonstrated this effect using realistic in vivo biological material. The more common interpretation of the quadrupolar interaction is that the orientation angle is isotropically distributed and is rapidly changing with a correlation time,  c, on the order of microseconds or less. In this situation the effect of H^Q is not to induce energy level splitting but to broaden the lines, i.e., to induce relaxation. The physics of this situation is rather complex.2,5,6 The result depends both on the whether the product !0 c is large or small compared with unity as well as the product !Q c. If !0 c << 1, the signal will show only one spectral line. If !0 c > 1 (but with !Q c << 1) this line may have two distinct decay rates (i.e., two rates for T1 and two for T2). The inner and outer transitions give the same spectral frequency but differ in their relaxation rates. The outer transitions produce 60% of the signal and the inner 40%. This phenomenon of biexponential relaxation (BR) is widely observed in nature. For example, most biological tissue shows a short T2 on the order of a few milliseconds and a longer one on the order of 20±30 ms. These results are interpreted in terms of a correlation time on the order of microseconds. In contrast, the relaxation times of NaCl±water are usually 50 ms or less for both T1 and T2 (at 2 tesla); this is due to a very short  c on the order of 10ÿ11 s. The phenomenon of BR is related to the question of Na visibility, i.e., if T2f << TE then the fast component will have died out before the signal is detected in a spin echo experiment. Obviously, this de®nition of `invisibility' is highly dependent on the NMR equipment, and especially the shortest TE used. The use of a pulse spectrometer with very short TE is equivalent to being able to detect a very broad spectral component with a CW spectrometer. Most recent experiments have reported high visibility in biological materials. Joseph and Summers4 found essentially 100% visibility in excised cat brain and porcine skeletal muscle. Kohler et al.7 found only 80% visibility for Na in the vitreous humor of enucleated eyes, but it increased to 100% after digestion with collagenase. The actual environment of biological Na ions is considerably more complex than that of a simple solution. Even in the EC space there is always a multitude of proteins, fatty acids, sugars, and other hydrophilic molecules with which the Na ions can interact. One expects short-term `bonding' and exchange of the Na ions among the various bound and free pools. Such exchange processes are known also to in¯uence the relaxation rates.6,8 It is often assumed that the degree of IC Na bonding is much greater than the EC, so that IC Na will

have a shorter T2 than EC. This conclusion is subject to considerable doubt however, since even homogeneous solutions can produce BR very similar to that seen in vivo. For example, the relatively simple system consisting of NaCl ions in 4% agarose gel shows T2 values of 5.5 and 23 ms.4 Since one motivation for doing Na MRI is the hope of imaging IC versus EC Na, we will review what is currently known about the scienti®c basis for this. All of this work relies on the use of a hyper®ne shift reagent (SR),9 such as dysprosium tripolyphosphate (Dy-TPP), which shifts the Larmor frequency of the EC Na only; this means that IC and EC Na have separate spectral lines. Such experiments show that IC Na often displays BR. Shinar and Navon10 observed BR in red blood cells (RBC) with T2 values of about 6.6 and 23.6 ms. However, the precision with which one can measure two T2 components is limited and somewhat dependent on the assumptions made in modeling the decay. For example, Shinar and Navon assumed a 40/60 distribution between the slow and fast components (T2s/T2f) as predicted by theory for a single population of spins with BR. From their data one can probably not rule out a somewhat shorter T2f with a different short/fast distribution ratio. Later work by Shinar and Navon11 in the nuclei of chicken RBCs showed T2f values in the 2±8 ms range, where the decay rate had a strong dependence on the IC concentration of DNA. The BR of the IC Na is only one-half of the question, however. One wants to know if the BR phenomenon is found only in the IC Na. In 1986, Shinar and Navon12 found that none of blood plasma, serum nor various solutions of proteins in water gave BR. This would suggest that the BR phenomenon is limited to IC Na. However, from a detailed analysis of the T1 and T2 decay rates they concluded that there was a signi®cant contribution to quadrupolar relaxation from a pool of bound Na even in this noncellular environment. That is, one assumed that the physics of BR was operative but the effect was too weak to be resolvable into two distinct T2 values. This experiment was done without the use of shift reagents. Since blood is a relatively small fraction of the ¯uid in most tissues, one wonders whether the results on blood plasma can be applied to interstitial Na in living tissues. A very careful study of this question was conducted by Foy and Burstein13 using Na spectroscopy of perfused hearts in both frog and rat models. They eliminated the contribution of blood by a selective inversion±recovery technique. In part of their experiment they used a shift reagent to distinguish IC from EC Na, with the result that 85% of the signal was EC in this model. They analyzed the transverse decay of the parenchymal signal (i.e., both IC and EC but excluding the blood) without SR and found BR with T2s & 25 ms and T2f & 2 ms in both models. Since the fast component was about 40% of the total signal it could not have come solely from the IC compartment. This is clearly a counter example to the hypothesis that short T2 components are found only in the IC compartment of living tissues. However, it is still possible that the average IC T2 could be less than that in the EC space. Pekar et al. proposed using double quantum (DQ) ®ltration to distinguish IC from EC sodium.14 The main point is that DQ, and in general multiple quantum (MQ), signals can emerge only if the physical parameters (strength of the quadrupolar constant and correlation time) are such that BR occurs. Pekar et al. used a SR and found the DQ sodium signal to be

SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS

limited to the IC space. This encouraged the hope that IC could be distinguished from EC solely by the physical characteristics of the NMR signal. However, subsequent work by Jelicks and Gupta15 demonstrated that the presence of the SR in the EC would itself quench an MQ NMR signal. This means that the absence of EC MQ signal as observed by Pekar et al. did not contradict the ®nding of Foy and Burstein that BR does occur in the EC space. However, by detecting an MQ signal in the presence of SR it does provide an alternative method for obtaining a pure IC signal.16,17

(90)

3

(Hard 180)

B1 ZPE

Gz

YPE

3 IMAGING TECHNIQUES The principles of imaging sodium do not differ from those used to image protons; basically, in addition to exciting the Na spins and reading out the signal, some means must be provided for spatial encoding.18 The main differences are due to the characteristics of the Na signal already discussed, namely, the low S/N and the short values of T1 and T2. Short T1, on the order of 50 ms in tissue, is an advantage since it implies rapid recovery of longitudinal magnetization after excitation so that short TR values can be used with little loss of signal. The low S/N forces one to employ techniques with relatively poor spatial resolution and long imaging times; typical values are 3  3 mm in-plane resolution and 10 mm slice thickness. Imaging times are usually a large fraction of 1 h. The most important aspect of Na imaging is the need to achieve short echo times. One wants quantitation of the short versus long T2 components, and the goal of imaging is to demonstrate these differences anatomically. A common technique is to use gradient echo (GE), which is equivalent to imaging the free induction decay (FID). The fact that GE is more sensitive than spin echo (SE) to background ®eld gradients is rarely of importance for sodium, partly because the low

implies less sensitivity of phase shift to the background ®elds and partly because the TE values achieved are usually less than 3 ms. There are several ways in which spatial encoding can be applied. The single slice two-dimensional Fourier transform (2DFT) technique widely used in proton images can be used. More commonly, however, for multislice imaging, three-dimensional Fourier transform (3DFT) is used (Figure 2). The initial excitation may be a hard, nonselective pulse or a `slab' which consists of a set of slices to be imaged. The slice de®nition is obtained by phase encoding the z axis. To reduce aliasing one can saturate spins outside of the desired region.19 As shown in Figure 2, it is possible to obtain both a FID and SE signals. Some workers have employed a prephasing Gx pulse during the phase-encoding interval so that the kx = 0 echo is delayed; this is called a GE. The advantage of the GE is that both positive and negative kx values are obtained in the same data acquisition (DA), which leads to fewer imaging artifacts due to background gradients, eddy currents, etc. However, the resulting delay in the minimum echo time is disadvantageous for detecting the short T2 component. Both of the FT techniques require the use of phase-encoding gradient pulses applied after excitation and before signal read out. This is a major disadvantage because they delay the time at which signal acquisition can begin and will therefore severely attenuate the short T2 component. A technique that

Gy

Gx (FID)

(SE)

DA

Figure 2 Pulse sequence diagram for 3DFT with both a FID and a spin echo (SE) detected. The 90 pulse is meant to excite a slab of slices in the z direction. The dotted lines on the Gz and Gy pulses indicate phase encoding which varies from one excitation to another (ZPE, z phase encoding; YPE, y phase encoding)

avoids that delay is projection±reconstruction imaging (PRI). PRI works by reconstructing an object from measurements of its projections along many lines or planes over a wide range of angles. The two dimensional (2D) form of PRI is the basis of all clinical X-ray computerized tomography (CT) scanners, and was also the ®rst technique used for MRI.20 Shepp21 showed how the 2D PRI method could be generalized to three dimensions. In this method, whose pulse sequence is illustrated in Figure 3, each excitation is completely nonselective and is followed immediately by readout of signal in the presence of all three gradients. The signal S(t) will be the Fourier transform of the spin density along the direction de®ned by the gradient vector, g: ……… S…t† ˆ

M…x; y; z† exp…i g
rt† dx dy dz

…2†

where r = (x, y, z) is the position vector and M(x, y, z) is the spin density at point r, weighted by the relevant relaxation factor. If we can neglect T2 decay during the readout period, Fourier transform of the signal yields a frequency distribution Sf (!) given by … ……… Sf …!† ˆ S…t† exp…ÿi  !t† dt ˆ M…x; y; z† …! ÿ g
r† dx dy dz …3†

where () is the Dirac delta function. Equation (3) indicates that the signal corresponding to any given frequency ! comes from a plane perpendicular to the g vector and displaced by

4 SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS (90)

B1

Gz

Gy

Gx

DA

Figure 3 Pulse sequence diagram for 3D PRI. The 90 pulse is hard. The relative magnitudes of the Gx, Gy, and Gz pulses are varied from pulse to pulse such that the gradient vector sweeps over a spherical surface in k-space

distance !/( g) from the origin. To reconstruct the density M(x, y, z) it is necessary to collect a set of signals in which the gradient vector ranges, more or less uniformly, over the full range of the 4 solid angle of a sphere.22 In the language of kspace, each gradient vector de®nes a vector in three-dimensional (3D) k-space, which varies with time t as … k…t† ˆ g dt ˆ tg

…for g ˆ const†

…4†

Hence, the 3D PRI method essentially scans k-space along a series of spokes starting from the origin and extending out to some maximum value. To reconstruct the function M(x, y, z) the data in k-space are ®rst multiplied by the factor k2 then Fourier transformed back to the image domain to yield a set of modi®ed planar data.23 These planar (`®ltered') data are then directly projected onto the 3D image matrix. The method is easily generalized to include multiple spin echoes.24 The major advantage of 3D PRI for Na MRI is that data acquisition can begin immediately after excitation, without the need to wait for gradient pulses to play out. This is especially advantageous when eddy currents and gradient ampli®er limitations add &1 ms to the nominal time duration of gradient pulses. The method is intrinsically three dimensional, so one is committed to a rather large number of pulses to sample k-space adequately. However, like 3DFT, the method is very ef®cient for signal to noise ratio (S/N) because each excitation includes the entire volume imaged. However, the method is sensitive to frequency offset and B0 inhomogeneity errors, but for Na MRI these are rarely a signi®cant problem. This method has no dephasing gradient prior to signal readout; this means that on any excitation only one-half of the k region is covered; i.e. k > 0 on one excitation and k < 0 on another. This technique is called `half echo', and results in the

earliest possible time for the k = 0 part of phase space. However, if T2f is much less than the DA time window, high k values will be attenuated and the image will suffer loss of spatial resolution. This is a consequence of imaging objects with very short T2 using frequency encoding. An interesting variant of this technique combines Fourier encoding in the slice direction with PRI in the transverse plane; a Hahn SE time of 3.6 ms has been obtained on a clinical MRI machine.25 Another approach to the problem of achieving rapid data acquisition after excitation is to use gradients in the rf B1 ®eld. This method has been applied to Na MRI, but the image quality obtained was less than that obtained by workers using the more conventional B0 gradient methods.26 When shift reagents are used, it is necessary somehow to image separately the two spectral lines produced. This can be done using various methods well known from proton chemical shift imaging.27 For example, Kohler et al.28 used a 2D phaseencoding technique equivalent to that originally proposed by Brown et al.29 This technique uses a standard spatially selective excitation produced by the application of the z gradient, followed by phase encoding in both the x and y planes. The signal is read out in the absence of a gradient, so that Fourier transform will yield the spectrum. This sequence commits the experimenter to a relatively large number of excitations (Nx
Ny). However, for Na imaging this is not a problem since one is usually forced to use a long scan time to achieve acceptable S/N. This technique can be generalized to use multiple echoes (on the long T2 component) and a matched ®lter can be used to optimize S/N.30 The previous techniques are classed as single quantum (SQ). There is interest in obtaining MQ signals and using them for imaging. Wimperis and co-workers have investigated this, and the details are discussed elsewhere.31 Triple quantum (TQ) is a better choice than DQ for this purpose. The minimum pulse sequence to generate MQ coherence consists of three rf pulses: 1-t1-2-t2-3-t3-DA, where the optimal values for the ¯ip angles i depend on the type of experiment, and the optimal interpulse time intervals ti depend on the relaxation properties of the Na. This pulse sequence gives a mixture of coherence orders, and it is necessary to `®lter' the signal to obtain the desired coherence. This is done by cycling the phase ( in Figure 4) of the last rf pulse; the choice of ¯ip angle = 109 maximizes the TQ signal component for this sequence.32 The simplest technique for MQ MRI is simply to append one of the previous SQ imaging pulse sequences to the previous MQ preparation sequence. However, a more ingenious imaging pulse sequence has been suggested by Wimperis and Wood31 which uses the unique properties of the TQ coherence to select the TQ signal without phase cycling. The pulse sequence is shown in Figure 4. The ®rst 90-180-90 set of pulses refocuses dephasing due to B0 inhomogeneity and/or chemical shift effects. During time interval t2 there is TQ coherence. This implies that the phases of the relevant matrix elements evolve at three times the SQ Larmor frequency. Thus, the gradient pulses applied during that time need only be onethird as strong (or long) as in SQ techniques. This is an advantage in systems whose gradients are designed for proton imaging and in which it is dif®cult to achieve the factor of four increase needed to obtain high-resolution Na images. Furthermore, the TQ signal decays during that time at a rate

SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS 180 (x) 90 (y) 90 (x) t1/2 t1/2

be roughly categorized as concerning edema, infarction, and neoplasia.

109 (f) t2

5

t3

B1

4.1 Gz PE Gy

Gx DA

Figure 4 Pulse sequence diagram for triple quantum Na imaging. The dotted lines for the Gz and Gy pulses indicate phase encoding (PE). The x axis is frequency encoded during the data acquisition period (DA)

comparable to T2s, so there is less need to minimize the time duration.31 Application of the frequency-encoding prephasing pulse (on Gx) during this time sets up a typical echo during the readout time as shown. However, in this case the area under the prephase pulse needs to be only one-third that of the readout gradient pulse; this unusual condition acts to select only the TQ desired coherence component so that phase cycling is in principle not necessary. (However, in practice the SQ signals are so strongly excited that phase cycling is recommended.) The strength of the TQ signal developed, S, depends on both t1 and t3 according to the function S…t1 ; t3 † / f …t1 † f …t3 †

…5†

f …t† ˆ exp…ÿt=T2s † ÿ exp…ÿt=T2f †

…6†

where

When T2s >> T2f, f(t) starts at zero, builds to a maximum in time &T2f, and decays with time constant T2s. Hence it is not necessary to achieve ultrashort echo or DA times with this sequence. The drawback to MQ imaging is the reduction in signal amplitude relative to SQ.15 In particular, the use of gradients to ®lter the MQ coherence reduces signal by a factor of two.32 Wimperis et al.33 have modi®ed the technique in an attempt to increase S/N. However, they demonstrated only modest image quality in a phantom of 200 mM Na, which is 10 times the amount present IC. Keltner et al.34 have used a technique similar to Figure 4 to image a phantom containing 12 mM Na, obtaining an excellent S/N in voxels of 1.5  1.5  1.5 cm size. To date, no one has published in vivo MQ images of Na with an image quality acceptable for biological application. 4 BIOLOGICAL APPLICATIONS To date, most applications of Na MRI to biomedical problems have been experimental. These applications can

Imaging Studies of Edema

Edema means the pathological increase in water content of tissues. When the edema is due to leaking blood vessels, the excess water is EC and its Na content should be about 150 mM, i.e., much greater than the IC level. This will result in increased Na signal. Furthermore, since the IC [Na] is so low, the amount of signal increase will largely re¯ect the increase in the size of the EC space. However, these same factors lead to increased signal in conventional T2-weighted proton images, so Na MRI has not found widespread use as a clinical tool for this purpose. One application of this principle was described by Lancaster et al.35 who imaged experimental lung edema in rats. Proton MRI is especially problematic in this case because of severe suppression of signal due to the inhomogeneous magnetic susceptibility of normal lung tissue. Sodium MRI is less sensitive to this effect because of its reduced . They demonstrated much more dramatic increases in Na signal than proton signal in their model of permeability edema. They also attempted to isolate the vascular component of Na by using an iron oxide superparamagnetic particle suspension to suppress the vascular Na. 4.2

Imaging of Stroke and Infarction

The rationale for this application is that if a cell dies, its Na±K pump will shut down and the IC [Na] will increase. However, to demonstrate the effect experimentally sometimes involves reperfusing the organ following vascular occlusion in order to provide a source of the excess Na. Working with such a dog heart model, Cannon et al.36 demonstrated increased Na signals in regions of ischemic damage. Some early work by Hilal et al.37 demonstrated increased Na signal in MRI of the brain in one patient; both an old and a recent (2 h) stroke were shown. However, visualization of old strokes is easily accomplished with T2-weighted proton MRI. The demonstration of early stroke would be very valuable clinically if it could be accomplished in a rapid and routine fashion. There has been little subsequent work in this area, and current research on MRI of stroke is more focused on the possible utility of diffusion-weighted proton MRI. 4.3

Imaging Studies of Cancer

There is much work in the area of distinguishing IC Na levels in normal and malignant cells. For example, Liebling and Gupta38 found that IC Na of a benign uterine leiomyoma was relatively depressed at only 5.1 mM, while a malignant leiomyosarcoma had the elevated value of 34.6 mM. There have been many in vitro nonimaging studies of tumors in animal models which will not be reviewed here. Imaging is useful when it is desirable to see the spatial distribution of Na in more detail. Summers et al.39 used Na MRI to study the development of necrosis in an animal model of human neuroblastoma; they found that the total Na signal (not distinguishing IC from EC) tended to rise with increasing tumor necrosis. Lin et al.40 used Na MRI to distinguish two different

6 SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS grades of tumor in a rat prostate tumor model; they found the two differed in the magnitude of T2s in a way that could not be seen using proton MRI. Several groups have demonstrated increased Na signal in human brain cancer. However, since most tumors are visualized using proton MRI, the interest is whether Na MRI can make more speci®c diagnoses. In some cases, the technique uses a relatively short gradient echo as well as several spin echoes with longer TE. In the normal brain, if TE > 10 ms the only structures visible are those containing cerebrospinal ¯uid and the vitreous humor of the eye. Looking only at echo times greater than 13 ms, Turski et al.41 found that a brain stem glioma showed less enhancement than an astrocytoma in the pons. Hashimoto et al.42 used a short TE (1.9 ms) gradient echo as well as longer spin echoes. However, they reported that the most useful differences between the Na and proton images of a meningioma were derived from the long TE images; the short TE images were obviously blurred. Schuierer et al.43 looked at a variety of supratentorial lesions with Na MRI at 4.0 tesla using only a TE = 11 ms echo; they found that Na MRI did not contribute any information that was not provided by proton MRI or X-ray CT. Hilal and co-workers44 have energetically applied Na MRI to brain cancer. Skirting the recognized dif®culties in differentiating IC from EC Na, they derive an image of `putative intracellular' Na from a gradient echo starting at TE = 0.2 ms and two spin echoes (TE = 12 and 24 ms). The long T2 signal, assumed to be purely EC, is used to estimate the EC volume fraction by comparison with the vitreous humor. The short T2 component is computed by subtracting the value of the EC signal extrapolated to TE = 0 (equivalent to what is called `curve stripping'). From these numbers an estimate of the putative IC concentration {[Na]/(IC volume)} is obtained for each pixel in the image. From their preliminary sample of 23 patients, they found a correlation between the IC [Na] and the histological grade of gliomas and astrocytomas. Like Hashimoto et al., they also found that Na MRI often reveals a larger tumor than proton MRI. Thus, there is ample evidence that Na levels are increased in various types of tumor. However, whether this is due to increases in the IC or EC space is a complex question, and the answer probably will depend on the type of tumor and which organ it is in. In particular, many tumors have a loosely organized cellular system and so have a relatively large EC space compared with normal tissues. In these cases, the increase in EC Na would probably correlate with the signal increase seen in proton MRI.

5 RELATED ARTICLES Chemical Shift Imaging; Image Formation Methods; Projection±Reconstruction in MRI.

6 REFERENCES 1. C. P. Slichter, `Principles of Magnetic Resonance', 3rd edn., Springer-Verlag, New York, 1990, pp. 494±500. 2. W. D. Rooney and C. S. Springer, NMR Biomed., 1991, 4, 209.

3. M. Shporer and M. M. Civan, Biophys. J., 1972, 12, 114. 4. P. M. Joseph and R. M. Summers, Magn. Reson. Med., 1987, 4, 67. 5. T. E. Bull, J. Magn. Reson., 1972, 8, 344. 6. S. Forsen and B. Lindman, Methods Biochem. Anal., 1981, 27, 289. 7. S. J. Kohler, N. H. Kolodny, D. J. D'Amico, S. Balasubramanian, P. Mainardi, and E. Gragoudas, J. Magn. Reson., 1989, 82, 505. 8. U. Eliav and G. Navon, J. Magn. Reson., 1990, 88, 223. 9. M. S. Albert, W. Huang, J. H. Lee, J. A. Balschi, and C. S. Springer, NMR Biomed., 1993, 6, 7. 10. H. Shinar and G. Navon, Biophys. Chem., 1984, 20, 275. 11. H. Shinar and G. Navon, Biophys. J., 1991, 59, 203. 12. H. Shinar and G. Navon, Magn. Reson. Med., 1986, 3, 927. 13. B. D. Foy and D. Burstein, Biophys. J., 1990, 58, 127. 14. J. Pekar, P. F. Renshaw, and J. S. Leigh, J. Magn. Reson., 1987, 72, 159. 15. L. A. Jelicks and R. K. Gupta, J. Magn. Reson., 1989, 83, 146. 16. J. L. Allis, A. M. L. Seymour, and G. K. Radda, J. Magn. Reson., 1991, 93, 71. 17. H. Shinar, T. Knubovets, U. Eliav, and G. Navon, Biophys. J., 1993, 64, 1273. 18. S. J. Kohler and N. H. Kolodny, Prog. NMR Spectrosc., 1992, 24, 411. 19. W. H. Perman, D. M. Thomasson, M. A. Bernstein, and P. A. Turski, Magn. Reson. Med., 1989, 9, 153. 20. P. C. Lauterbur, Nature (London), 1973, 242, 190. 21. L. A. Shepp, J. Comput. Assist. Tomogr., 1980, 4, 94. 22. A. K. Louis, J. Comput. Assist. Tomogr., 1982, 6, 334. 23. O. Nacioglu and Z. H. Cho, IEEE Trans. Nucl. Sci., 1984, 31, 553. 24. J. B. Ra, S. K. Hilal, and C. H. Oh, J. Comput. Assist. Tomogr., 1989, 13, 302. 25. J. B. Ra, S. K. Hilal, and Z. H. Cho, Magn. Reson. Med., 1986, 3, 296. 26. J. P. Boehmer, K. R. Metz, J. Mao, and R. W. Briggs, Magn. Reson. Med., 1990, 16, 335. 27. D. Burstein and M. Mattingly, J. Magn. Reson., 1989, 83, 197. 28. S. J. Kohler, E. K. Smith, and N. H. Kolodny, J. Magn. Reson., 1989, 83, 423. 29. T. R. Brown, B. M. Kincaid, and K. Ugurbil, Proc. Natl. Acad. Sci. U.S.A., 1982, 79, 3523. 30. D. Lu and P. M. Joseph, Magn. Reson. Imaging, 1995, 13, in press. 31. S. Wimperis and B. Wood, J. Magn. Reson., 1991, 95, 428. 32. J. W. C. VanderVeen, S. Slegt, J. H. N. Creyghton, A. F. Mehlkopf, and W. M. M. J. Bovee, J. Magn. Reson., Ser. B, 1993, 101, 87. 33. S. Wimperis, P. Cole, and P. Styles, J. Magn. Reson., 1992, 98, 628. 34. J. R. Keltner, S. T. S. Wong, and M. S. Roos, J. Magn. Reson., Ser. B, 1994, 104, 219. 35. L. Lancaster, A. R. Bogdan, H. L. Kundel, and B. McAffee, Magn. Reson. Med., 1991, 19, 96. 36. P. J. Cannon, A. A. Maudsley, S. K. Hilal, H. E. Simon, and F. Cassidy, J. Am. Coll. Cardiol., 1986, 7, 573. 37. S. K. Hilal, A. A. Maudsley, J. B. Ra, H. E. Simon, P. Roschmann, S. Wittekoek, Z. H. Cho, and S. K. Mun, J. Comput. Assist. Tomogr., 1985, 9, 1. 38. M. S. Liebling, R. K. Gupta, and H. L. Kundel, Ann. N.Y. Acad. Sci., 1987, 508, 149. 39. R. M. Summers, P. M. Joseph, and H. L. Kundel, Invest. Radiol., 1991, 26, 233. 40. R. Lin, A. R. Bogdan, L. Lancaster, K. Meyer, H. L. Kundel, E. Kassab, V. Liuolsi, M. Salscheider, and P. M. Joseph, Proc. IXth Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, Vol. 1, p. 722.

SODIUM-23 MAGNETIC RESONANCE OF HUMAN SUBJECTS 41. P. A. Turski, L. W. Houston, W. H. Perman, J. K. Hald, D. Turski, C. M. Strother, and J. F. Sackett, Radiology, 1987, 163, 245. 42. T. Hashimoto, H. Ikehira, H. Fukuda, A. Yamaura, O. Watanabe, Y. Tateno, R. Tanaka, and H. E. Simon, Am. J. Physiol. Imaging, 1991, 6, 74. 43. G. Schuierer, R. Ladebeck, H. Barfuss, D. Hentschel, and W. J. Huk, Magn. Reson. Med., 1991, 22, 1. 44. S. K. Hilal, C. H. Oh, I. K. Mun, and A. J. Silver, in `Magnetic Resonance Imaging' eds. D. Stark and W. Bradley, Mosby Year Book, St. Louis, MO, 1992, pp. 1091±1112.

7

Biographical Sketch Peter M. Joseph. b 1939. B.S., 1959, Ph.D., 1967, physics, Harvard University, USA; assistant professor of high energy physics, Carnegie Mellon University, 1970±72. NIH postdoctoral fellow in medical physics, Memorial Sloan Kettering Inst., New York, 1972±73. Instructor and assistant professor of clinical radiology, Columbia-Presbyterian Medical Center, New York, 1973±80. Associate Professor of Diagnostic Imaging Physics, University of Maryland, Baltimore, 1980±82. Associate and full professor, University of Pennsylvania, 1983. Approx. 120 publications in high energy physics, X-ray and computerized tomography scanning, and MRI. Current research specialty: electron beam tomography, sodium and ¯uorine MRI.

STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY

Structural and Functional MR in Epilepsy Graeme D. Jackson Brain Imaging Research Institute, Austin and Repatriation Medical Centre, Heidelberg, West Australia and Howard Florey Institute, University of Melbourne, Australia

1

This abundance of new MR techniques that allows so many aspects of brain structure, function and biochemistry to be investigated in the clinical setting has revolutionized the ability to detect brain abnormalities which underlie the epilepsy condition. In surgical programs for the treatment of epilepsy, MR has become at least as important as the EEG. The information from MR will have a major impact on the classi®cation and understanding of epilepsy syndromes.

and Alan Connelly Institute of Child Health and Great Ormond Street Hospital for Children, NHS Trust, London, UK

1 SUMMARY Recently, developments in magnetic resonance imaging (MRI), magnetic resonance spectroscopy (MRS), and functional magnetic resonance imaging (fMRI) have opened up new opportunities for the noninvasive investigation of the brain. In epilepsy, these noninvasive techniques play a major role in the clinical investigation of patients with epilepsy. MRI can noninvasively detect virtually all foreign tissue lesions (tumors) such as hamartomas, gliomas, oligodentrocytomas, dysembryoplastic neuroepithelial tumors and other developmental lesions. It has been able to de®ne these lesions with a great deal of anatomical accuracy. This in itself is a tremendous advance which can now easily be taken for granted. Perhaps even more impressive has been the ability of optimized structural imaging techniques to detect smaller abnormalities of gray matter, in particular lesions like subtle cortical dysplasias, minor abnormalities of gray matter, and especially hippocampal sclerosis. In the important lesion of hippocampal sclerosis, quantitative measures of both the abnormal morphology (volume) and abnormal signal (T2 relaxation measurement) have allowed this diagnosis to be made objectively. The clinical impact of this noninvasive information cannot be overstated. A major challenge now is to be able to reliably detect subtle areas of dysgenesis in the cortical gray matter. MR spectroscopy provides information about brain metabolites that appears to provide objective information about regional damage in the temporal lobe which is not readily apparent with conventional imaging. Both these methods have enabled the detection of bilateral temporal lobe abnormalities, and the consequences of this bilateral damage on cognitive and seizure outcome is being explored. These techniques show ®xed brain pathology. With the development of functional MRI it is possible to see signal change associated with activated neurons. As well as detecting normal activation (such as hand movement), it has been possible to image seizures. This gives information about both the spatial and the temporal location of signal changes during seizures. It appears that interictal activations may also be detectable.

2

EPILEPSY, THE CLINICAL PROBLEM

Epilepsy is a common problem occurring in up to 1% of the population.1,2 Intractable epilepsy which is a debilitating disorder may occur in up to 0.25% of the population. It is an extremely important neurological condition because there are severe social and medical consequences of the disorder3,4 (up to 1% of patients die per year and many more are severely affected). Individuals otherwise often have the capacity to live normal lives, and complete cure is possible in a large number of these if a seizure focus exists, can be identi®ed, and can be surgically removed. Epilepsy is a disorder predominantly of the gray matter. This includes a wide variety of pathologies such as tumors (some of which may be small), subtle lesions such as cortical dysplasia, and often minor abnormalities of development of subtle degrees of brain injury such as hippocampal sclerosis or cortical gliosis. The predominant abnormality in these patients varies from the macroscopic to the cellular level, and may be characterized by predominantly biochemical or metabolic abnormalities.

3

CLASSIFICATION OF EPILEPSY CONDITIONS

In general terms epilepsy can be divided into two main groups: generalized and partial epilepsy.5 In generalized epilepsy the seizure arises almost simultaneously in all parts of the brain. In partial epilepsy, by comparison, the seizure begins in one distinct part of the brain, and may then spread to other parts of the brain. There is an implication that generalized epilepsy involves an abnormality in multiple, or all, parts of the brain, while partial epilepsy implies that the abnormality is con®ned to a limited portion of the brain. The international classi®cation which is most widely accepted6 further divides the epilepsy conditions into those that are `symptomatic' (that is have a de®ned pathological cause such as a tumor) and those that are `idiopathic' (have no de®ned lesional cause). The recent development of MR tools for the investigation of brain structure, biochemistry, and function have had a major impact on the thinking in clinical epilepsy. The ability to de®ne lesions (such as hippocampal sclerosis and cortical dysplasia), which were previously `cryptogenic' and only detectable when pathological studies had been performed, means that the classi®cation and principles upon which clinical epilepsy is based are undergoing major revision at the present time as a direct consequence of the information provided by MR studies.

2 STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY 4 THE CHALLENGE FOR MR STUDIES: THE INFORMATION THAT IS SOUGHT IN EPILEPSY PATIENTS The problem of epilepsy necessitates understanding of brain structure, function, and biochemistry in normal and pathological states. The central problems in the understanding and management of intractable epilepsy are as follows. 1. To determine whether the epilepsy syndrome is generalized (no de®ned site of seizure onset, i.e. onset almost simultaneous in all or many parts of the brain) or partial (focal or localized seizure onset, with or without subsequent generalized spread). 2. To de®ne whether a structural abnormality of the brain exists which may give rise to the epilepsy disorder (if so, known as symptomatic epilepsy; if not, then de®ned as idiopathic epilepsy). 3. If partial, to de®ne the location, and extent, of the region (or regions) responsible for the generation of the seizures, and how functional events relate to underlying structural abnormality. 4. To understand which lesions are epileptogenic, and what abnormalities of structure and function de®ne such areas. 5. To determine the effects of seizures on the brain. Do seizures cause damage (e.g. cellular damage, neuronal loss, hippocampal sclerosis) or is it the disease or condition that gives rise to the epilepsy disorder which causes all of the damage to the brain? 6. To identify important functional areas of cortex (movement, speech, memory) which must be preserved if a neurosurgical procedure is to be performed for treatment of intractable seizures. Therefore, noninvasive investigations contribute to the solution of these problems by identifying gray matter lesions such as hippocampal sclerosis, replacing the use of invasive methods used to localize the site of seizure origin, de®ning the nature and extent of the structural, functional and metabolic abnormalities of the seizure focus, and determining preoperatively factors which in¯uence the likely seizure and functional outcome from surgical treatment.

5 STRUCTURAL ABNORMALITIES SHOWN BY MR IMAGING 5.1 Tumors Approximately 20% of all patients with intractable epilepsy will have a relatively large lesion (tumor) as the basis of their

Table 1

epilepsy. Before MRI only about 50% of these lesions were detected preoperatively (with CT) when located in the temporal lobes.7 As these patients generally have an excellent outcome after surgery it is essential to identify them. Many series have now established that MR detects virtually all tumors, including dysembryoblastic lesions, hamartomas, and gliomas. 5.2

Hippocampal Sclerosis

Hippocampal sclerosis is one of the most common lesions found in the brains of patients with intractable epilepsy.8±10 It is important for several reasons. It is a highly epileptogenic lesion. The side of the most affected hippocampus is almost always the side from which the majority of temporal lobe seizures originate. The detection of hippocampal sclerosis by MRI may obviate the need for invasive EEG monitoring with its attendant morbidity in patients being considered for surgery. Until new MR techniques arose, hippocampal sclerosis was considered as nonlesional epilepsy: the reliable detection of hippocampal sclerosis changes the clinical perspective of these patients. The ®rst reports of the MR detection of hippocampal sclerosis were rather confusing with encouraging results reported in some small studies, confusion with artifacts in others, and inadequate pathological veri®cation in many. There were even studies that failed to detect any abnormality. In 1987, Kuzniecky and colleagues, from the Montreal Neurologic Institute, published the ®rst major series of papers in which hippocampal sclerosis was detected in a systematic way, and shown to be related to pathological ®ndings.11 This study relied almost entirely on T2-weighted signal changes. Using optimally oriented imaging planes, and heavily T1-weighted inversion recovery sequences in addition to T2-weighted sequences, hippocampal sclerosis was shown to be reliably detected by visual inspection of images acquired at 0.3 tesla using the criteria of hippocampal atrophy and T2-weighted signal change in hippocampal gray matter.7,12 Since then optimized imaging at 1.5 tesla has shown that there are four main features of hippocampal sclerosis visible in MRI13,14 (Table 1). The criteria for assessing these optimized images include both morphological and signal intensity changes. The morphological features are atrophy and disruption of internal hippocampal structure. Abnormal signal in the hippocampus can be seen on both inversion±recovery (T1) and T2-weighted imaging. The imaging of the ®ne anatomical structure of the hippocampus, to a level of detail previously possible only with microscopic examination (Figure 1 shows details of this) may become an even more important method of detecting hippocampal sclerosis with improvements in imaging resolution.

Features of hippocampal sclerosis

MRI feature of hippocampal sclerosis

Suggested histopathological correlate of the MR imaging abnormality

. . . .

. . . .

Unilateral atrophy (right cf. left) Loss of internal morphological structure on IR images Increased signal on T2-weighted images Decreased signal on T1-weighted images (IR)

hippocampal atrophy loss of neurones in CA1, CA2, and CA4 and replacement gliosis [Figure 6(d)] gliosis gliosis

STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY

3

Figure 1 The features of hippocampal sclerosis on optimized imaging are shown here in the imaging plane which transects the hippocampus at right angles (a). These images show atrophy, and reduced T1 signal (b) and increased T2 signal (c)

5.3 Visual Analysis of the Hippocampus The visual diagnosis of hippocampal sclerosis can be highly subjective, and even experienced neuroradiologists may have dif®culty detecting this subtle lesion. It must be emphasized that not all features are seen in all cases. Unequivocal signal change (as long as the hippocampus is not enlarged) or atrophy may be accepted as diagnostic of hippocampal sclerosis, with the certainty increased if both are present. The addition of internal structure loss can be very helpful if the abnormality is subtle. Several series which compare blinded visual reports to pathological material have shown beyond doubt that hippocampal sclerosis can be reliably diagnosed by visual analysis. In

experienced hands, and with optimized images, high sensitivity and speci®city can be achieved. In order to do this optimized images must be performed13 and all features of hippocampal sclerosis must be appreciated and searched for. It is our experience that no single feature (such as atrophy) on its own is sensitive enough for reliable routine visual diagnosis, although some expert centers achieve an accurate diagnosis in about 80% of cases.15,16 5.3.1

Atrophy

The assessment of the cross-sectional size of the hippocampus must be made in images obtained in the modi®ed (tilted) coronal axis that transects the hippocampus at right

4 STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY

Figure 2

Internal structure of the hippocampus as seen on optimized MR images [normal (a),(b), hippocampal sclerosis(c),(d)]

angles. A smaller hippocampus as detected in this plane either qualitatively, or by quantitative methods, reliably predicts the side of the epileptogenic focus in the case of temporal lobe epilepsy but absolute measures of hippocampal size must be interpreted with caution.17 Quantitation of atrophy (hippocampal volume measurement) is slightly more sensitive than visual assessment of hippocampal size. But with the addition of other visual features of hippocampal sclerosis this difference is less marked, or even reversed. 5.3.2

Loss of the Normal Internal Morphological Structure of the Hippocampus

Normal internal morphological structure of the hippocampus is produced by the alveus, the molecular cell layer of the dentate gyrus, and the pyramidal cell layer of the cornu ammonis, and can be seen on optimized coronal MR images [Figure 2(a) and (b)]. In hippocampal sclerosis, the loss of this normal internal structure is a consequence of neuronal cell loss and replacement of normal anatomical layers with gliotic tissue [Figure 2(c) and (d)]. This feature of hippocampal sclerosis is potentially very important as, with increasing spatial resolution, thinning of the CA1 region of the cornu ammonis may prove to be the most sensitive and speci®c means of diagnosing hippocampal sclerosis. Attempts have been made to de®ne this with specially designed coils; however, increased resolution

which will routinely show this microanatomy is becoming possible using standard equipment. 5.3.3

Signal Hyperintensity on T2-Weighted Images

Increased T2-weighted signal when localized imprecisely to the `mesial temporal region' may be due to foreign tissue such as a glioma or hamartoma, to gliotic tissue in the hippocampus, to increased cerebrospinal ¯uid in the atrophied region, to ¯ow artifacts and occasionally from a developmental cyst in the hippocampal head stemming from failure of closure of the lateral aspect of the hippocampal ®ssure. Careful determination of the exact location of this signal change by detailed examination of the anatomical features enables the correct diagnosis to be made. It is important to have suf®cient knowledge of both the hippocampal anatomy and the easily recognizable artifacts that can occur in this region so that artifacts are not confused with signi®cant signal abnormalities in the hippocampal gray matter. 5.3.4

Signal Hypointensity on T1-Weighted IR Images

At 1.5 tesla, using a TR of 3500 and a TI of 300 ms, a sclerotic hippocampus appears small and dark, and the internal features are obscured (Figure 1). The presence of three features in a single coronal image makes the visual diagnosis of hippo-

STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY Table 2

340 Patients, pathological veri®cation of MRI diagnosis in 149

MRI diagnosis Hippocampal sclerosis Foreign tissue lesion (glioma, astrocytoma or dysembryoplastic tumor) Cortical dysplasia Vascular malformations (12 cavernous hemangiomas, two high ¯ow lesions) Cystic lesions Miscellaneous No lesion demonstrated Total

5

%

Number/location

57 13.5

194 46 (36 temporal, 10 extratemporal)

10.5 4

35 (12 temporal, 23 extratemporal) 14 (6 temporal, 8 extratemporal)

1.5 5

5 (4 temporal, 1 extratemporal) 17 (5 trauma, 1 tuberous sclerosis, 2 epidermoid, 4 extensive white matter lesions, 1 cerebellar atrophy, 4 uncertain) 29

8.5 100

campal sclerosis much easier, and makes it possible to detect mild degrees of abnormality. An abnormal signal on T1- or T2-weighted images arising from an atrophic hippocampus almost always represents hippocampal sclerosis. An abnormal signal arising from an apparently enlarged hippocampus may represent a hamartoma or glioma. If one relied only on a single feature such as atrophy, then these cases of the larger hippocampus being abnormal would be incorrectly lateralized. 5.4 Visual Analysis: Findings Other than Hippocampal Sclerosis Detailed analysis of MRI images reveals a high percentage of abnormalities which can be detected in the brains of patients with intractable partial epilepsy. As MR techniques improve, it is becoming clear that most patients with intractable epilepsy have detectable imaging abnormalities. Table 2 shows the range of abnormalities which were detected in a combined series of patients from several centers. While often, in the past, no cause could be identi®ed for many cases of intractable partial epilepsy, it is now becoming clear that most adults and children with partial epilepsy will have de®ned brain abnormalities visualized on appropriate optimized imaging.7,13,16,18±26 6 QUANTITATIVE DIAGNOSIS OF HIPPOCAMPAL SCLEROSIS There has been a widespread problem in achieving the best results using visual analysis (and routine reporting of MR studies) in the clinical environment. It is clear that the results of visual inspection can be replicated in different centers, and that the ®ndings are speci®c and sensitive when optimized images are interpreted in expert hands. It is equally clear that this expertise is hard to come by. This is probably because the lesion of hippocampal sclerosis differs from a normal hippocampus by a degree that might have previously been attributed to artifact or normal variation. Therefore considerable experience is required to gain expertise in this diagnosis. For this reason, for research purposes, and for the detection of abnormalities beyond the sensitivity of the eye, quantitative diagnosis of hippocampal abnormalities has been essential. The features

340

of hippocampal sclerosis are the same as when assessed visually. Much initial attention has been given to the quantitation of hippocampal atrophy but quantitation of signal characteristics is also possible. 6.1

Volumetric Analysis

In 1990 Jack and co-workers17,27,28 published their method of quantifying hippocampal atrophy which for the ®rst time enabled an objective measurement of hippocampal pathology. The use of volumetric measurement to assess hippocampal size had been successfully adopted (and adapted) by many centers.7,13,16,22±26,29±33 Despite different protocols, this has proven to be a reliable means of determining the lateralization of hippocampal pathology in up to 90% of cases with hippocampal sclerosis. It is our impression that visual analysis in the most expert centers detects virtually all cases that are detected by volumetric analysis, although volume techniques are more sensitive than visual analysis for the detection of the single feature of hippocampal atrophy. There are some cases, because features in addition to atrophy are considered, which can be detected by visual analysis and not by volumetric analysis.34 Because of the range of normal variation, and measurement error, the most reliable use of the volume measurement technique (and indeed visual analysis) has been in lateralization. It has not been easy to determine bilateral hippocampal abnormalities or to determine abnormality without comparison between sides. The strength of the volumetric technique is that the anterior±posterior distribution of the atrophy can be determined35 and the quantitative analysis removes the sometimes subjective nature of the analysis. The greatest weakness is that the only well veri®ed and reliable technique (because of normal variation) is the use of side to side ratios. This means that bilateral disease is not usually detectable. 6.2

T2 Relaxometry

As well as quantifying the hippocampal atrophy, one can quantify the T2 signal in the hippocampus by measuring the T2 relaxation time in the hippocampal gray matter (Figure 3).36 The T2 relaxation time can be measured quantitatively by measuring the decay in signal intensity at different echo times in a series of T2-weighted images acquired in the same slice.

6 STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY cise in normal tissue. This enables the detection of pathology without requiring comparison between two hippocampi. Therefore, as well as being sensitive, it permits the detection of pathology in the contralateral hippocampus. In our experience, the T2 relaxation time within the hippocampus is a robust and reliable objective measurement of hippocampal pathology, providing a means of assessing the hippocampus which is as good as our most skilled visual interpretation of hippocampal abnormality in optimized scans. In contrast to both visual interpretation and volumetric analysis of hippocampal atrophy, the de®nition of a normal hippocampal T2 relaxation time is very precise. Therefore, T2 quanti®cation has the ability to detect very mild, bilateral and progressive hippocampal abnormalities. Moreover, T2 values can be interpreted in terms of hippocampal pathology even when the other hippocampus is incomplete or distorted, such as when a lesion is present or following temporal lobe surgery. In these dif®cult cases, pathology of the residual ipsilateral hippocampus and the contralateral hippocampus may still be diagnosed. It has recently been shown that there is a very close correlation between hippocampal atrophy, hippocampal T2 abnormality and pathological ®ndings. Therefore, ®ndings from hippocampal volume studies (such as outcome, and pathology correlations) should apply to T2 abnormalities, while the latter has the advantage of detecting bilateral disease.

10 (a) Control 8 6 4

No. of temporal lobes

2

7

0 8

(b) Ipsilateral

6 4 2 0 10

(c) Contralateral

8 6 4 2 0 100 Normal £ 106

110

120

130

≥ 116 Hippocampal sclerosis

140

>146

T2 (ms)

Figure 3 The T2 map (a) is constructed from the T2 relaxation times measured for each pixel. The relaxation time is presented as intensity, and can be measured directly for any region of interest (shown for the hippocampus). The histogram (b) shows the distribution of measured T2 relaxation times in those patients with hippocampal sclerosis (labeled ipsilateral and contralateral to the side of seizure onset) and controls

Each pixel of the resulting T2 map is derived from the intensity in each of 16 images in that same slice. This objective measurement has a small range of values in normal subjects. The T2 relaxation time appears to be very pre-

MR SPECTROSCOPY

As discussed elsewhere in this volume, several lines of evidence suggest that almost all the N-acetylaspartate (NAA) within the brain is neuronal,37±40 and so a reduction in the NAA signal is commonly interpreted in terms of neuronal loss or damage. Such a case is shown for the temporal lobe in Figure 4. Here the NAA signal is reduced and the signals from choline containing compounds (Cho) and creatine + phosphocreatine (Cr) are increased. While interest focuses on the NAA signal as a marker of neuronal damage or loss, it is often impractical to quantitate this as an absolute quantity. Usually a marker of abnormality is being sought, so the ratio of NAA to Cho or Cr is often used as such a marker. In the temporal lobe it may often be dif®cult to resolve the Cho and Cr signals unequivocally, therefore we recommend that the ratio NAA/(Cho + Cr) is an appropriate marker of pathology for clinical use. MR spectroscopy studies have shown overall abnormalities in the NAA, Cr and Cho signals in the temporal lobes of patients with well localized intractable temporal lobe epilepsy.41,42 In comparison with controls, the temporal lobes ipsilateral to the seizure focus show a 22% reduction in NAA signal intensity, a 15% increase in the Cr signal, and a 25% increase in the Cho signal. The NAA change is interpreted in terms of neuronal loss (or damage). The interpretation of the increase in Cho and Cr signals is not yet clearly de®ned. However, studies of neural cells show that the concentrations of choline-containing compounds and of creatine + phosphocreatine are much higher in astrocyte and oligodendrocyte preparations than in cerebellar granule neurons.37 Thus, increases in Cr and Cho may re¯ect reactive astrocytosis suggesting that both neuronal loss and astrocytosis may be identi®ed by the noninvasive measurement of metabolites by MRS.37,43±45

STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY NAA Cho Cr (a)

(b)

4

3 2 Chemical shift (ppm)

1

Figure 4 Spectra from (a) the temporal lobe contralateral to the seizure focus in a patient with temporal lobe epilepsy and (b) from the ipsilateral temporal lobe. Note the decrease in NAA in the ipsilateral temporal lobe

Using the ratio NAA/(Cho + Cr), MRS has been used to lateralize up to 70 patients with intractable temporal lobe epilepsy. In about 40 of these cases the abnormality was bilateral, and like T2 relaxation time measurements could be used to detect bilateral abnormalities. MRS is sensitive to bilateral and diffuse pathology and it is an objective measure of metabolite abnormalities which cannot be visualized directly with MR

7

imaging. We view these MR spectroscopic abnormalities as a marker of regional abnormalities of the temporal lobe in these patients. We do not believe that it is a marker simply of hippocampal sclerosis, but it provides additional information about pathology of the temporal lobes which is not available by other MR methods. Chemical shift imaging (CSI) or magnetic resonance spectroscopic imaging has the advantages of being able to determine the regional distribution of metabolites and to identify areas of maximal abnormality. In a study of ten patients with temporal lobe epilepsy and ®ve controls the left±right asymmetry of NAA/Cr ratios was found to be signi®cantly different from controls in all cases.46 The use of such an asymmetry index alone precludes the detection of bilateral abnormalities. However, comparison of NAA/Cr ratios in patients and controls indicated that two patients had bilateral reduction in the NAA/Cr ratio in the posterior temporal region, and the greatest reduction was ipsilateral to the maximum EEG disturbance. A further CSI study, using a 2T Philips system with a 4 mL effective voxel size, was performed on eight patients with unilateral complex partial seizures and eight controls.47 Signi®cant asymmetry in the intensity of the NAA signal was found in all patients. In each case the lower NAA corresponded to the side of seizure focus as determined by EEG. No signi®cant changes in Cho or Cr were observed. It is apparent that CSI has distinct advantages over single voxel techniques in terms of coverage of the brain, and it is becoming the method of choice in a number of centers for the study of epilepsy. However, it is technically more demanding than single voxel MRS, particularly with respect to magnetic ®eld homogeneity (shimming), water suppression, and in®ltration of subcutaneous fat signal into voxels other than just those adjacent to the scalp. Cendes et al.46 noted that anterior temporal lobe structures were more accessible to single voxel methods, and reported only posterior and midtemporal results from their CSI study and Xue et al.48 have reported problems with suboptimal shimming when performing CSI in a large

Figure 5 Functional MR image showing the area of increased signal associated with leg movement near a tumor thought to involve the motor strip: (a) baseline image; (b) activation image; (c) superimposition image

8 STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY

Figure 6 (a) Signal change associated with the onset of a focal motor seizure in the same region in a 4 year old boy with intractable epilepsy. This activation was seen 20 s before the onset of the seizure. (b) The upper two rows (A±D and E±H) show activation seen using fMRI during two clinical seizures. After activity is ®rst seen, images 20, 60 and 100 s later are shown (all rows). In the ®rst row a clinical motor seizure began at the time of the third image. The lower row is a period of activation not associated with a seizure. The surface EEG suggested seizure onset in this region, and ictal single photon emission computed tomography (SPECT) localized it to this same region. The base anatomy image is the same for all these ®gures

region including both temporal lobes. They have therefore adopted the strategy of acquiring CSI volumes from each temporal lobe separately.

8 FUNCTIONAL MRI (fMRI) The clinical potential of fMRI is enormous and will be dealt with in many other sections of this Encyclopedia. For clinical epilepsy the following are major applications.

8.1

Mapping of Eloquent Areas of Cortex

Epilepsy surgery entails resection of abnormal areas of cortex in order to relieve the epilepsy condition. In many cases it is important to determine the location of important functions which must not be affected by this surgery such as movement, speech, and memory. The ability of fMRI to demonstrate this eloquent cortex helps in the preoperative assessment of these patients. At present this is largely limited to the motor strip49 (Figure 5), but demonstration of speech areas will also be important.

STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY

At the moment, many groups have produced images of signal changes during `speech activation'. While some of these are compeling, the problem of speech activation and localization is complex, and a great deal of validation and careful interpretation of these signal changes will be required before they assume their potential importance in clinical practice. 8.2 Functional Imaging of Seizures Functional MRI can detect cortical activation which occurs during partial motor seizures.50 Activation can be seen in the region which is activated during seizures even when no clinical seizure occurs. Also, quite remarkably, activation could ®rst be seen up to 1 min before the onset of clinical or EEG changes during similar seizures [Figure 6(a) and (b)]. The implication of these observations is that the vascular and oxygenation changes may precede, or at least be detectable earlier than, the EEG or clinical events which are associated with seizures. It also provides compeling evidence that subclinical activation can be identi®ed using fMRI, and this may enable precise localization of the seizure focus in some cases. These observations allow structural and dynamic functional information to be obtained in a single, integrated, totally noninvasive, MR examination, and point the way to the future role of MR as a means of imaging neurophysiology.

9 CONCLUSION The impact of MR and its application to clinical epilepsy is akin to the impact of the EEG in the 1940s. It is enabling abnormalities of the brain to be demonstrated by noninvasive techniques in many of these patients with epilepsy. The impact has been great in the ®eld of epilepsy surgery where, already, patients who had previously required invasive depth electrodes, are now able to go to resective surgery directly based on noninvasive studies which include noninvasive EEG, routine clinical evaluation, and these new MR techniques. In the broader ®eld of clinical epilepsy problems, MR ®ndings are beginning to have a large impact on the view of epilepsy, and of the syndromes that can be de®ned in individual patients. This will ultimately affect the classi®cation of epilepsy, with new syndromes such as temporal lobe epilepsy with hippocampal sclerosis being de®ned. The consequence of these new MR techniques will be to bene®t those who are most affected by this disease: patients with epilepsy.

10

RELATED ARTICLES

Brain MRS of Human Subjects; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions.

11

REFERENCES

1. W. A. Hauser, J. F. Annegers, and L. T. Kurland, Epilepsia, 1991, 32, 429.

9

2. W. A. Hauser, `Epilepsy Surgery'; ed. H. A. Luders, Raven Press, New York, 1992, pp. 133±141. 3. P. Klenerman, J. W. Sander, and S. D. Shorvon, J. Neurol. Neurosurg. Psychiatry 1993, 56, 149. 4. D. C. Taylor `Surgical treatment of the epilepsies', ed. J. Engel, Jr. 2nd edn., Raven Press, New York, 1993. 5. J. J. Engel `Seizures and epilepsy' Contemporary Neurology series, Davis, Philadelphia, 1989. 6. Commission on classi®cation and terminology of the international league against epilepsy, Epilepsia, 1989, 30, 389. 7. G. D. Jackson, S. F. Berkovic, B. M. Tress, R. M. Kalnins, G. Fabinyi, and P. F. Bladin, Neurology, 1990, 40(12), 1869. 8. J. H. Margerison and J. A. N. Corsellis, Brain, 1966, 89, 499. 9. M. A. Falconer, E. A. Serafetinides, and J. A. N. Corsellis, Arch. Neurol., 1964, 10, 233. 10. M. A. Falconer, Lancet, 1974, 2, 767. 11. R. Kuzniecky, V. De La Sayette, R. Ethier, D. Melanson, F. Andermann, S. Berkovic, Y. Robitaille, A. Olivier, T. Peters, and W. Feinder, Ann. Neurol., 1987, 22(3), 341. 12. S. F. Berkovic, F. Andermann, A. Olivier, R. Ethier, D. Melanson, Y. Robitaille, R. Kuzniecky, T. Peters, and W. Feindel, Ann. Neurol., 1991, 29, 175. 13. G. D. Jackson, S. F. Berkovic, J. S. Duncan, and A. Connelly, Am. J. Neurorad., 1993, 14, 753. 14. R. Kuzniecky, E. Faught, and R. Morawetz, Epilepsia, 1993, 34(6), 141. 15. C. R. J. Jack, F. W. Sharbrough, G. D. Cascino, K. A. Hirschorn, P. C. O'Brien, and W. R. Marsh, Ann. Neurol., 1992, 31(2), 138. 16. G. D. Cascino, C. R. Jack, Jr., K. A. Hirschorn, and F. W. Sharbrough, Epilepsy Res., 1992, (suppl. 5), 95. 17. C. J. Jack, M. D. Bentley, C. K. Twomey, and A. R. Zinsmeister, Radiology, 1990, 176(1), 205. 18. J. H. Cross, G. D. Jackson, B. G. R. Neville, A. Connelly, F. J. Kirkham, S. G. Boyd, M. C. Pitt, and D. G. Gadian, Arch. Dis. Child., 1993, 69, 104. 19. G. D. Cascino, C. R. Jack, Jr., J. E. Parisi, W. R. Marsh, P. J. Kelly, F. W. Sharbrough, K. A. Hirschorn, and M. R. Trenerty, Epilepsy Res., 1992, 11(1), 51. 20 R. Duncan, J. Patterson, D. M. Hadley, P. Macpherson, M. J. Brodie, I. Bone, A. P. McGeorge, and D. J. Wyper, J. Neurol. Neurosurg. Psychiatry, 1990, 53(1), 11. 21. P. Gulati, A. Jena, R. P. Tripathi, and A. K. Gupta, Indian Pediatr., 1991, 28(7), 761. 22. B. Jabbari, A. D. Huott, G. DiChiro, A. N. Martins, and S. B. Coker, Surg. Neurol., 1978, 10, 319. 23. R. Kuzniecky, A. Murro, D. King, R. Morawetz, J. Smith, K. Powers, F. Yaghmai, E. Faught, B. Gallagher, and O. C. Snead, Neurology, 1993, 43, 681. 24. C. J. Kilpatrick, B. M. Tress, C. O'Donnell, S. C. Rossiter, and J. L. Hopper, Epilepsia 1991, 32(3), 358. 25. K. Miura, M. Kito, F. Hayakawa, M. Maehara, T. Negoro, and K. Watanabe, J. Jpn. Epilepsy Soc., 1990, 8(2), 159. 26. S. S. Spencer, G. McCarthy, and D. D. Spencer, Neurology, 1993. 27. C. J. Jack, C. K. Twomey, A. R. Zinsmeister, F. W. Sharbrough, R. C. Petersen, and G. D. Cascino, Radiology, 1989, 172(2), 549. 28. C. J. Jack, F. W. Sharbrough, C. K. Twomey, G. D. Cascino, K. A. Hirschorn, W. R. Marsh, A. R. Zinsmeister, and B. Scheithauer, Radiology 1990, 175(2), 423. 29. M. Ashtari, W. B. Barr, N. Schaul, and B. Bogerts, Am. J. Neurorad., 1991, 12, 941. 30. M. Baulac, O. Granat, X. Gao, and D. Laplane, Epilepsia, 1991, 3(2), 2. 31. G. Castorina and D. L. McRae, Acta Radiol., 1963, 1, 541. 32. T. Lencz, G. McCarthy, R. A. Bronen, T. M. Scott, J. A. Inserni, K. J. Sars, R. A. Novelly, J. H. Kim, and D. D. Spencer, Ann. Neurol., 1992, 31(6), 629.

10 STRUCTURAL AND FUNCTIONAL MR IN EPILEPSY 33. K. Matsuda, K. Yagi, T. Mihara, T. Tottori, Y. Watanabe, and M. Seino, Jpn. J. Psychiatry Neurol., 1989, 43(3), 393. 34. G. D. Jackson, R. I. Kuzniecky, and G. D. Cascino, Neurology, 1994, 44, 42. 35. M. J. Cook, D. R. Fish, S. D. Shorvon, K. Straughan, and J. M. Stevens, Brain, 1992, 115, 1001. 36. G. D. Jackson, A. Connelly, J. S. Duncan, R. A. GruÈnewald, and D. G. Gadian, Neurology, 1993, 43, 1793. 37. J. Urenjak, S. R. Williams, D. G. Gadian, and M. Noble, J. Neurochem., 1992, 59, 55. 38. J. Urenjak, S. R. Williams, D. G. Gadian, and M. Noble, Neurosci., 1993, 13, 981. 39. J. W. Hugg, K. D. Laxer, G. B. Matson, A. A. Maudsley, C. A. Husted, and M. W. Weiner, Neurology, 1992, 42, 2011. 40. J. K. Joller, R. Zaczek, and J. T. Coyle, J. Neurochem., 1984, 43, 1136. 41. J. W. Hugg, K. D. Laxer, G. B. Matson, A. A. Maudsley, and M. W. Weiner, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, 1913. 42. J. Peeling, G. Sutherland, Neurology, 1993, 43, 589. 43. D. G. Gadian, A. Connelly, J. S. Duncan, J. M. Cross, F. J. Kirkham, C. L. Johnson, F. Vargha-Khadom, B. G. Nevik, and G. D. Jackson, Acta Neurol. Scand., 1994, suppl. 152, 116. 44. D. G. Gadian, A. Connelly, J. H. Cross, G. D. Jackson, F. J. Kirkham, J. V. Leonard, B. G. R. Neville and F. Vargha-Khadom, `New Trends in Pediatric Neurology', eds. N. Fejerman and N.A. Chamoles, Amsterdam 1993, pp. 23-32. 45. A. Connelly, D. G. Gadian, D. G. Jackson, J. H. Cross, M. D. King, J. S. Duncan, and F. J. Kirkham, `Proton Spectroscopy in the Investigation of Intractable Temporal Lobe Epilepsy', Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 234. 46. F. Cendes, F. Andermann, P. C. Preul, and D. L. Arnold, Ann. Neurol., 1994, 35, 211. 47. J. W. Hugg, K. D. Laxer, G. B. Matson, G. B. Maudsley, and M. W. Weiner, Ann. Neurol., 1993, 34, 788. 48. M. Xue, T. C. Ng, M. Modic, Y. Comair, and H. Kolem, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, 435. 49. C. R. Jack, R. M. Thompson, R. C. Botts, R. R. Butts, F. W. Sharbrough, P. J. Kelly, D. P. Hanson, S. J. Rieslerer, R. L.

Ehman, N. J. Hangrandrean, and G. D. Cascino, Radiology, 1994, 190(1), 85. 50. G. D. Jackson, A. Connelly, J. H. Cross, I. Gordon, and D. G. Gadian, Neurology, 1994, 44, 850.

Acknowledgements Dr Jackson was supported, in part, by grants from the Wellcome Trust and Action Research. We would like to thank David Gadian, Cheryl Johnson, Brian Neville, John Duncan, Richard GruÈnewald, Wim Van Paeschen, and Simon Robinson.

Biographical Sketches Graeme D. Jackson. b 1956. B.Sc. (Hons) Psychology, 1977; MB.BS. Monash University, Melbourne, Australia, 1982.; FRACP (Neurology), 1992; M.D. thesis 'Magnetic Resonance in Intractable Epilepsy', Monash University, 1995. Introduced to epileptology by Drs. Peter Bladin and Sam Berkovic, Austin Hospital, Melbourne, 1988±90. Then a research registrar, National Hospital for Neurology and Neurosurgery, 1990±92. Subsequently lecturer then senior lecturer and Honorary Consultant in Paediatric Neurology at the Institute of Child Health, and Great Ormond Street Hospital and NHS Trust, London, 1992± 1996. Director of the Brain Imaging Research Institute, Austin and Repatriation Medical Centre, Heidelberg, West Australia 1996±current. Howard Florey Institute, University of Melbourne 1998±current. Current research specialities: MR applications in epilepsy, neurotoxicity. Alan Connelly. b 1955. B.Sc. (Hons) Chemistry, University of Glasgow, Scotland, 1977; Ph.D., University of East Anglia, Norwich, UK. Ph.D. in high resolution NMR under the supervision of Prof. Robin Harris. Began work on in vivo spectroscopy and imaging as NMR development scientist at Oxford Research Systems, UK, 1983±88. Subsequently lecturer (1988±93) then senior lecturer (1993±present) at the Institute of Child Health, London, UK. Current research specialties: imaging and spectroscopy applications in stroke and epilepsy.

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

Systemically Induced Encephalopathies: Newer Clinical Applications of MRS Brian D. Ross, Stefan Bluml, Kay J. Seymour, Jeannie Tan, Jong-Hee Hwang and Alexander Lin Huntington Medical Research Institutes, Pasadena, CA, USA

1 SYSTEMICALLY INDUCED ENCEPHALOPATHIES

1

glutamate and of -aminobutyrate (GABA) from glutamine (both vital neurotransmitter amino acids) occurs principally in neurons.10 Lest it be thought that ammonia `toxicity' accounts for all of the clinical syndromes covered by the term HE, the interested reader is referred to Butterworth11 for an extensive review of several other well-documented alternatives. Metabolic theories abound: failure of oxidative energy metabolism (a corollary of glutamate and 2-oxoglutarate depletion from the Krebs cycle), tryptophan and serotonin (5-hydroxytryptamine) accumulation, branched-chain amino acid de®cits, endogenous benzodiazepine agonists which modify access of GABA to its inhibitory receptors, and neurotoxic fatty acids, octanoate in particular, have been proposed. An attractive unifying theory proposed by Zieve12 is that multiple neurotoxins derived from liver, blood, or from the diet, gain access via PSSs to a previously `sensitized' brain. No mechanism of sensitization is known, but with the advent of MRS, a candidate has been proposed in the form of cerebral myo-inositol (mI) depletion.13

1.1 Background: Biochemistry of Coma Metabolic disturbances of liver, kidney, endocrine or other systems have remote effects; those on the brain result in a variety of well-de®ned encephalopathies. When severe, these disorders present as coma. Posner and Plum published a comprehensive account of human coma.1,2 Many were the result of presumed metabolic events with normal brain anatomy, setting the stage for noninvasive elucidation by means of biochemically based techniques. Among these are MRA, diffusion imaging, positron emission tomography (PET), single-photon emission tomography (SPECT), and multinuclear magnetic resonance spectroscopy (MRS). MRS has increasingly been used to identify speci®c biochemical changes in the brain, from which information on diagnosis and pathogenesis of these poorly understood disorders is beginning to emerge. As a model for this group of disorders, we discuss hepatic encephalopathy, with additional remarks about diabetic, hyperosmolar, and hypoxia-induced encephalopathies. Renewed interest in the use of MRS for differential diagnosis of focal brain pathologies is re¯ected in studies of stroke and adrenoleukodystrophy. 1.2 Neurochemical Pathology of Hepatic Encephalopathy Hepatic encephalopathy (HE) is an excellent example of metabolic encephalopathy,3 with identi®able neurotoxins originating in a systemic disease. In animal studies, several distinct neurotoxins have been identi®ed. The earliest of these was ammonia;4,5 ammonia is normally fully removed from portal blood by hepatic urea synthesis.6,7 The combination of the loss of biosynthetic liver function and the diversion of nondetoxi®ed blood to the brain by so-called portal systemic shunts (PSSs) accounts for the frequently demonstrated excess of cerebral and cerebrospinal ¯uid glutamine8 [Equation (1)]. NH‡ 4 ‡ glutamate

GS

„

glutamine

…1†

glutaminase

The enzyme glutamine synthetase (GS) responsible for this reaction is located exclusively in astrocytes.9 Resynthesis of

1.3

Animal Studies in Hepatic Encephalopathy Using Multinuclear MRS

Three groups14±16 independently perfected methods for the noninvasive determination of cerebral glutamine (including a variable contribution from glutamate in each assay) with 1H MRS and con®rmed the elevation of this metabolite in a variety of animal models of acute liver `failure' with HE. One of these groups14 also demonstrated a hitherto unrecognized abnormality, the signi®cant reduction of choline (Cho; cerebral choline-containing compounds) in the 1H spectra of rats with acute HE. More recently, 15N NMR17±19 and 1H±15N heteronuclear multiple quantum coherence (HMQC)20,21 identi®ed cerebral glutamine unequivocally in vivo in HE produced by ammonia infusion in the normal and portocaval shunted rat, respectively. The glutamine protons coupled to amide nitrogen (termed HZ and HE) provide additional information regarding the compartmentalized pH adaptation during severe ammonia-induced coma in rats.22 Alkalinization of the astrocyte (glial) cell cytoplasm can be inferred from alterations in 1H±15N HMQC spectral line-widths in vivo, whereas whole brain pH, determined from chemical shift of inorganic phosphate (31P MRS, pH Measurement In Vivo in Whole Body Systems) is apparently unaltered.23 In extracts of portacaval shunt rat brain, high-performance liquid chromatography (HPLC) con®rms the accumulation of glutamine and the depletion of glycerophosphorylcholine (GPC). The latter has also been demonstrated in human subjects using a technique of quantitative proton-decoupled 31P MRS.24 It partly explains the reduced Cho in the prior 1H NMR study as well as demonstrating the expected depletion by 50% or more of the cerebral mI and scyllo-inositol (sI) content.25 This result in rats con®rmed the observations that ®rst emerged from human studies with short-echo time 1H MRS13,26,27 and establishes the validity of the portacaval shunt rat for future experimental studies.

2 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS 1.4 Pathogenesis, Diagnosis, and Therapeutic Management of Hepatic Encephalopathy in Humans: The Emerging Role of Proton MRS Studies using stimulated echo acquisition mode (STEAM) localized, short echo time 1H MRS de®ned the changes in 10 patients with clinically con®rmed chronic HE. The average increase in cerebral glutamine was estimated as +50%, Cho decreased 14% and mI decreased by 45%21 (Figure 1). Very similar ®ndings have now been reported at 2 T by Bruhn (Figure 2)26,29 and by McConnell30 using PRESS at short or long echo times. An early study which used long echo time PRESS-localized 1H MRS failed to identify the depletion of mI but was the ®rst clearly to show `Cho' depletion in human brain.31

NAA Cho

Cr

mI Liver disease

SCHE

a-Glx b, g-Glx HE

Severe HE

4

3

2

1

0 ppm

Figure 1 Development of hepatic encephalopathy in human subjects (1.5 T spectra). A series of spectra of parietal cortex (white matter) acquired under closely similar conditions (GE Signa 1.5 T, STEAM localization TR 1.5 s, TE 30 ms, NEX 128) from different patients is presented. A normal spectrum for comparison is that in Figure 7(b). In liver disease (top) there is a relatively normal spectrum with a slight decrease in choline (Cho). In subclinical hepatic encephalopathy (SCHE) there is a de®nite decrease in myo-inositol (mI) with a minor increase in the glutamine (Gln) regions (glutamine plus glutamate, Glx). There is a very signi®cant increase in the Glx regions in HE (grade 1) and mI is further depleted. The spectrum of grade 3 HE shows more severe changes in the biochemical markers of this disease, most notably glutamine. Cr, creatine

Control

Hepatic encephalopathy

(a)

(b) ml Cho

Gln↑

Cho↓

Gln

ml↓

4.0

3.6

3.2

2.8

(c)

3.0

2.4

2.0

1.6

Gln

4.0

3.6

3.2

2.8 2.4 Gln↑

2.0

1.6

(d)

2.0

Standard solution Glutamate (+ NAA)

3.0

2.0

Standard solution Glutamine + glutamate (+ NAA)

Figure 2 Identi®cation of glutamine in the proton spectra at 2.0 T. A 51-year-old patient with HE resulting from surgical portacaval shunt (b) is compared with a normal control (a). Spectra were obtained from an occipital gray matter location and show the expected changes of HE: increased glutamine, decreased Cho/Cr and mI/Cr. With the improved resolution at 2 T, separate analysis of glutamine and glutamate is possible. Inset are spectra from model solutions (c) 5 mM glutamate + 5 mM N-acetylaspartate (NAA); (d) 5 mM glutamine + 5 mM glutamate + 5 mM NAA, indicating that in this patient the increase in glutamine occurs without obvious depletion of cerebral glutamate. Spectra were acquired on a Siemens 2.0 T clinical spectrometer, with STEAM localization; TR 3 s, TE 20 ms. Abbreviations as in Figure 1. (Modi®ed from Bruhn et al.,26 with permission from J. Frahm, T. Michaelis and colleagues)

This information considerably alters our understanding of the pathogenesis of HE, emphasizing an underlying defect in cerebral osmoregulation in addition to the clear relationship to ammonia toxicity and glutamine accumulation by the cerebral astrocytes. Proton-decoupled 31P MRS revealed additional osmotic and metabolic disturbances in patients with HE.24 Quantitative analyses in 16 patients with liver disease, ten with and six without chronic hepatic encephalopathy, in four patients with hyponatremia, and in 20 age-matched normal subjects were reported (Figure 3). Patients with HE were distinguished from controls by signi®cant reduction in cerebral nucleoside triphosphate (NTP = ATP) (2.45 ‹ 0.20 versus 2.91 ‹ 0.21 mmol kgÿ1 brain; P < 0.0003), inorganic phosphate (P < 0.03) and phosphocreatine (P < 0.04). In addition to increased cerebral levels of glutamate plus glutamine (Glx) and decreased concentrations of mI, patients with HE showed reduction of total visible Cho (in 1H MRS), GPC (0.67 ‹ 0.13 versus 0.92 ‹ 0.20 mmol kgÿ1 brain in controls; P<0.005), and glycerophosphorylethanolamine (GPE) (0.40 ‹ 0.12 versus 0.68 ‹ 0.12 mmol kgÿ1 brain in controls; P < 0.0003) (in proton-decoupled 31P MRS). Of the reduction of `total Cho', 61% was accounted for by GPC, a cerebral osmolyte. Similar metabolic abnormalities were seen in hyponatremic patients. The results are consistent with disturbances of cerebral osmoregulation and energy metabolism in patients with chronic HE.24 1.5

Energetics of the Human Brain in Hepatic Encephalopathy

The predicted cerebral energy de®cit (reduction in [ATP]) has previously only been convincingly shown in mice and tree

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

3

Figure 3 Osmotic and metabolic disturbances in hepatic encephalopathy (HE). Averaged proton-decoupled 31P MRS from patients with (a) HE, (b) liver disease without HE, (c) elderly controls and (d) young controls. The averaged spectra calculated from all patients with HE can readily be distinguished from averaged spectra obtained from liver-diseased patients without HE and controls by reduced concentrations of phosphoethanolamine (PE), inorganic phosphate (Pi), glycerophosphorylethanolamine (GPE), glycerophosphorylcholine (GPC), and ATP. Furthermore, phosphocreatine (PCr) content (not obvious here owing to the expanded scale) was signi®cantly decreased in HE. Phosphorylcholine (PC) remained normal

shrews.32 Phosphorus-31 MRS should be the easiest tool with which to establish any energy de®cit in HE (as predicted) but Ross,33 Tropp,34 Chamuleau,35 Barbiroli,36 and Morgan37 have obtained con¯icting and hence unconvincing data with 31P MRS on such effects in humans. Using quantitative 1H MRS, however, Geissler and colleagues showed a small but signi®cant increase in cerebral creatine concentration [Cr] following liver transplantation in humans.38 Since [Cr] is the sum of phosphocreatine (PCr) and creatine, this may indicate that patients with HE have an `energy de®cit'. A 10% reduction of PCr and a 16% reduction of NTP (of which perhaps up to 80% of the 31P signal arises from ATP) is reported by Bluml et al. in patients with HE24 and is hypothesized to be a contributing factor in the complex pathogenesis of the disorder. Because ATP depletion occurred without a marked increase in glutamine accumulation, the depletion of TCA cycle intermediates,

hypothesized to limit energy metabolism,5 is deemed an unlikely explanation. Instead, the hypothesis proposed is that ATP depletion may occur in the astrocytes as part of a failure of astrocyte volume regulation, which is re¯ected in reduced concentrations of several cerebral osmolytes. In the astrocyte, a possible rate-limiting role of ATP as a co-factor for GS would be most relevant. The dif®culty which has been confronted in demonstrating such an effect by 31P MRS, where sensitivity is 10% that of 1H MRS, becomes obvious. Since reduced cerebral oxygen consumption and blood ¯ow are recognized abnormalities in HE, the 13C NMR techniques pioneered by Shulman may be best suited to demonstrate the anticipated reduction in the overall rate of the Krebs cycle.39 From these clinical studies, it appears that 1H MRS, particularly if applied with effective water suppression to reveal mI, is currently most ef®cient for the elucidation of pathogenesis of

4 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS HE, and also, as will be shown, for its early clinical diagnosis. Direct measurement of mI by 13C NMR may become a valuable adjunct.40±42

2 SUBCLINICAL HEPATIC ENCEPHALOPATHY As indicated, HE is a group of diseases presumed to have identical etiology but presenting a great variety of distinct clinical pictures. Underlying all of them is believed to be the entity of `subclinical HE' (SCHE), although it is by no means clear that the HE and coma of fulminant (very acute) liver failure, goes through any truly `subclinical' phase. Proton MRS accurately re¯ects the entity of SCHE,43,44 and in preliminary studies also appears to mirror the progressive and increasingly severe syndromes of overt HE de®ned by Parsons-Smith as grades 1±4 (Figure 1). Elevation of glutamine is rather more obvious at 2 T (Figure 3) but presents little dif®culty at 1.5 T. Figure 1 should not be taken as proof, however, that in the individual patient such orderly progression of neurochemical dysfunction occurs. Longitudinal studies have not been performed in suf®cient numbers to be certain of this. Nevertheless, it is tempting to suggest, as Figure 1 appears to indicate, that in the brain exposed to liver `toxins', Cho depletion (principally GPC) precedes the loss of mI and sI, with later accumulation of glutamine. If this sequence is correct, then perhaps `sensitization' of the brain is the result of mI or Cho depletion, or both. In keeping with the theory of many years standing, increasing cerebral glutamine underlies the neurological syndromes of grades 1±4, severe, overt chronic HE, as well as, perhaps, that of acute, fulminant HE and coma. Figure 4 shows the similar but more severe neurochemical changes of Reye's syndrome, giving an effective indication of what may be seen in acute HE.45 Unfortunately, published spectra from patients with fulminant HE are limited and dif®cult to interpret.46 2.1 Depletion of myo-Inositol and the Induction of Hepatic Encephalopathy A human `experiment' which goes some way towards verifying this sequence is the new interventional procedure known as TIPS (transjugular intrahepatic portal systemic shunt), which is used as a life-saving procedure in cirrhosis-induced hematemesis. Not surprisingly, TIPS induces clinical HE in up to 90% of survivors.47±49 Proton MRS performed both before and after TIPS in 10 patients showed a universal increase in mean intracerebral glutamine following the procedure. More importantly, in a small number of individuals in whom mI was normal prior to TIPS (these patients often show SCHE), a progressive reduction in mI/Cr and development of subclinical and clinical HE follows the introduction of the shunt (Figure 5). 2.2 Restoration of Cerebral myo-Inositol and Choline Accompanies Reversal of Hepatic Encephalopathy Yet another common human `experiment', that of orthotopic liver transplantation, allows the reverse process to be unequivocally demonstrated, thereby establishing a ®rm, albeit

Figure 4 Proton MRS in acute hepatic encephalopathy caused by Reye's syndrome. In vivo proton MRS spectra acquired from a parietal white matter region in infant brain from: (a) 10-month-old normal subject; (b) patient, day 2 after admission; (c) patient, day 8 after admission; (d) difference (b) ÿ (c) between days 2 and 8 in patient; (e) solution with 15 mM glutamine. All spectra are scaled. Spectral assignments: Lac, lactate (1.3 ppm); NAA, N-acetylaspartate (2.02 ppm); Gln, glutamine (2.10±2.50 ppm, and 3.65±3.90 ppm); Cr + PCr, creatine + phosphocreatine (3.03 ppm); Cho, choline-containing compounds (3.23 ppm); mI, myo-inositol (3.56 ppm). U, Unassigned (3.62 ppm). Notable abnormalities concern a huge accumulation of cerebral Gln, reduced Cho, and, later, reduced mI, all of which re¯ect liver failure. In addition there is a decrease in NAA and [Cr] and appearance of the unassigned peak. Reye's syndrome, a toxic viral disease associated with aspirin intake, is known to produce severe neuronal damage, and may not, therefore, completely correspond to the picture of acute liver failure. An occipital gray matter region gave almost identical results. (Reproduced with permission from Ernst et al.)45

circumstantial link between mI depletion and the syndromes of SCHE and overt HE (Figure 6). There was also a recovery in Glx and Cho also recovered. Indeed, an overshoot of cerebral Cho is consistently observed, perhaps linking the earlier Cho

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

5

cerebral ammonia metabolism, be it by ammonia toxicity or glutamine synthesis (as a newer variant of the theory would have it).52 The concept of an underlying brain `sensitization' receives substantial new impetus deserving of further research in animal models. Either Cho (GPC) depletion, caused perhaps by failure of hepatic synthesis of a necessary precursor, or cerebral mI depletion could ful®ll the role of `sensitizer'. In neither case is there a precedent; as a result basic research is urgently required. It is likely that NMR will play a crucial role in such investigations, and the portacaval shunt rat is a convenient model. Carbon-13 NMR is the only method of unequivocally determining mI as distinct from the lower concentrations of inositol-1-phosphate (Inos-1-P) and glycine with which mI coresonates in the 1H MR spectrum. 3.2

Figure 5 Effect of a transjugular intrahepatic portal systemic shunt (TIPS) on the proton MR spectrum. Spectra are from a 30-year-old female, 5 days after a hematemesis caused by esophageal varices and chronic alcoholic liver disease. Spectra acquired from a parietal white matter location (GE Signa 1.5 T, volume 12.5 cm3, STEAM TR 1.5 s TE 30 ms, NEX 128) before a TIPS procedure shows no abnormalities apart from a small but signi®cant reduction in Cho/Cr, attributable to liver disease (top). Three weeks after TIPS, changes are seen in the Glx and mI regions and a further reduction on Cho/Cr (middle). The bottom spectrum shows the progression of changes seen (13 weeks post-TIPS); Glx is markedly increased and mI is signi®cantly reduced. Abbreviations as in Figure 1. (With permission from Shonk et al.48)

Osmoregulation in Human Hepatic Encephalopathy

A new concept in HE can justi®ably be attributed to these endeavors with in vivo MRS. This is related to cell-volume regulation in the brain and is conveniently termed osmolyte disorder. GPC was ®rst identi®ed as an osmolyte in the kidney, while mI, GPC and taurine (in rats) were proposed for that role in brain,53 Haussinger et al.54 ®rst used the term hypo-osmolarity to link mI with HE. The importance of the concept may lie in the hitherto obscure connection between hepatic coma (in fulminant liver failure), a disease treated as an emergency because of the massive edema that occurs (dysregulation of cerebral

depletion with de®cient hepatic biosynthesis of some relevant precursor of GPC. 2.3 Ornithine Transcarbamylase De®ciency A description of MRS in HE would be incomplete without consideration of a rare but informative inborn error of hepatic urea synthesis, ornithine transcarbamylase (OTC) de®ciency. The single known biochemical consequence is hyperammonemia. Gadian and colleagues50 demonstrated the inevitable elevation of cerebral glutamine in two such patients, while Ross51 showed the extraordinary parallel with HE, in depletion of cerebral mI (Figure 7). 3 CONTRIBUTION OF NMR TO ASSESSMENT OF CLINICAL HEPATIC ENCEPHALOPATHY 3.1 Pathogenesis Both experimentally and clinically, NMR (particularly 1H MRS) supports the classical concept of HE as a disorder of

Figure 6 Restoration of biochemical abnormalities post-liver transplant. The patient is a 30-year-old male with acute-on-chronic hepatic encephalopathy secondary to hepatitis, and subsequently successfully treated by liver transplantation. Spectra were acquired 6 months apart from the same parietal white matter location, (15.0 cm3 STEAM TR 1.5 s, TR 30 ms, NEX 128) and scaled to the same Cr intensity for comparison. The obvious abnormalities before liver transplantation, increased -, -, and -glutamine, reduced Cho/Cr and mI/Cr (upper spectrum) were completely reversed 3 months after transplantation, and Cho/Cr exceeded normal (lower spectrum). Abbreviations as in Figure 1

6 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS 3.4

Unanswered Questions in Neurology of Hepatic Encephalopathy

Wilson's disease, caused by excess copper deposition, is believed to result in an encephalopathy analogous to HE. However, the 1H MRS ®ndings are not surprisingly rather different, lacking either mI depletion or glutamine accumulation (unpublished study from this laboratory). Myelopathy is an unusual form of chronic HE. It presents with paraplegia. Athough neurological considerations would suggest cord involvement, the 1H MRS ®ndings in the parietal cortex are typical of other patients with the more classical clinical presentation (see Fig. 4 in Ross et al.).51 Often noted in MRI, the basal ganglia may be `bright' in inversion recovery (IR) images of patients with known HE. There is no consistent relationship with 1H MRS ®ndings, and increasingly this MRI ®nding is recognized as nonspeci®c. Nevertheless, the extrapyramidal signs, the changes on postmortem, and these albeit inconsistent MRI ®ndings continue to suggest that there may be as yet unrecognized underlying neurochemical changes in the basal ganglia in HE.

4

Figure 7 Treated ornithine transcarbamylase (OTC) de®ciency versus healthy age-matched subject. Both spectra were acquired from a similar location in parietal white matter (STEAM TR 1.5 s, TE 30 ms, NEX 128) and processed as described in the literature.47 Single-enzyme defects in the urea cycle result in severe hyperammonemia and `hepatic encephalopathy'. Because the patient was receiving treatment with sodium benzoate, no excess of cerebral glutamine is present. However, as anticipated by studies in the commoner condition of hepatic encephalopathy caused by portosystemic shunting, a decrease in mI was noted in the proton spectrum of a 14-year-old with OTC de®ciency (a) when compared with a normal age-matched control (b). NAA, N-acetylaspartate; other abbreviations as in Figure 1

osmolytes perhaps?), and the more mundane and relatively slow-evolving biochemical disturbance outlined above as chronic HE and SCHE. 3.3 Proton MRS for Diagnosis Early in these investigations, it became apparent that mI depletion and Glx accumulation occurred when clinical HE was absent. Other systemic or metabolic diseases (apart from OTC de®ciency already discussed) did not result in mI depletion; consequently 1H MRS offers an unique opportunity for early, speci®c diagnosis of this still perplexing condition. Paradoxically, there is at present little enthusiasm for this, most probably because prevention (with lactulose or neomycin) and treatment (by liver transplantation) of overt or severe HE is relatively straightforward (albeit rather costly in the case of orthotopic liver transplantation (OLT) at $150 000±200 000 per patient in USA). If a ready medical means of restoring cerebral mI were to be discovered, the value of 1H MRS in diagnosis might increase. This point is covered in a recent report in which treatment with lactulose for 7 days in overt HE signi®cantly increased mI/Cr.55,56

4.1

OTHER SYSTEMIC ENCEPHALOPATHIES Diabetes Mellitus and Ketogenesis

Like HE, diabetic encephalopathy is common and obviously `metabolic' in origin. Three principal syndromes are recognized: diabetic ketoacidosis, lactacidosis, and nonketotic hyperosmolar coma. Elevated cerebral glucose,57 signi®cant excess of mI, a reversible accumulation of `Cho', and the presence of ketone bodies have been detected in various patients with varying severity of diabetic encephalopathy.58 The hyperosmolar state, which is discussed in more detail below, is well known in diabetes mellitus as so-called nonketotic hyperglycemia (NKH). NKH is responsible for coma and neurological symptoms more frequently even than diabetic ketoacidosis. MRS ®ndings in a young adult with NKH-induced coma59 are consistent with many years of experimental work in demonstrating altered cerebral osmolytes, Cho, mI and even N-acetylaspartate (NAA) (Figure 8). The most surprising ®nding of proton MRS, which requires further study, is the identi®cation of acetone (rather than the more widely anticipated -hydroxybutyrate and acetoacetate) as the ketone of human diabetic ketoacidotic encephalopathy (Figure 9).58 A therapeutic idea dating from the 1920s has reemerged in the use of a so-called `ketogenic diet' to provide excellent control of seizures in drug-resistant patients with epilepsy. Perhaps not surprisingly, the cerebral ketone body that accumulates in these patients is also acetone, rather than acetoacetate or hydroxybutyrate (Figure 10).60 In epilepsy, we suggest that this ®nding is more than of academic importance and may ultimately lead to more focused dietary or pharmacological treatments. At the very minimum, 1H MRS should be considered for clinical monitoring in iatrogenic ketosis. Long-term neurological and cerebral `complications' of diabetes contribute to the much increased mortality. The biochemical basis of these conditions may lie in those changes

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

7

Unexpected reduction of mI/Cr or Cho/Cr, in, for example, screening MRS examinations for dementia might be an indicator of hyponatremia.59 Central pontine myelinolysis (CPM) is a rare clinical disorder believed to follow inappropriate corrections of severe hyponatremia in patients. This link led Lien and colleagues to investigate the value of in vitro MRS of the newly discovered cerebral osmolytes in rats.68 The correction of cerebral osmolytes disturbed by hyponatremia is even more striking than the initial discovery of their depletion because of its relevance to clinical management. Slow restoration of plasma sodium is advocated and is accompanied by even slower restoration of cerebral osmolytes. In one human study, a mean interval of 10.5 weeks elapsed before cerebral osmolyte correction in three of four patients.67 Now we have identi®ed the MRS pattern of CPM in a patient in whom hyponatremia was not apparent, and in whom cerebral osmolytes were normal (Figure 12). An illuminating study of

Figure 8 Proton MRS of occipital gray matter in recovering nonketotic hyperglycemia. Excess Cho/Cr, NAA/Cr, mI/Cr (/) as well as abnormal intensities for glucose (0), lactate and (possibly) acetone (*) are indicated. Abbreviations as in Figure 1

in Cho, NAA, mI, ketones, and glucose, now recognized by 13 C and 1H MRS to be present in the acutely and chronically diabetic brain. 4.2 The Hyperosmolar State: Identi®cation of Idiogenic Organic Osmolytes by Proton MRS Lien et al. ®rst used 1H MRS to investigate a `new' family of cerebral metabolites collectively known as organic osmolytes because of their believed role in the maintenance of cerebral osmotic equilibrium.61 Such molecules were also ®rst thoroughly researched in the papilla of the kidney with the help of in vitro NMR.62 First in sporadic cases of diabetic hyperosmolar coma58 and in a patient after closed head injury,63,64 and then most convincingly in a single infant with holoprosencephaly and de®cient thirst mechanisms, these concepts were con®rmed as contributing to human encephalopathy by the use of 1H MRS (Figure 11). mI was three times normal, and other resonances, also markedly affected, returned toward normal with treatment. Difference spectroscopy is particularly helpful in identifying these changes.65,66 The converse, or hypoosmolar state of hyponatremia has been identi®ed by 1H MRS in signi®cant numbers of patients.67 4.3 The Hypoosmolar State: Hyponatremia and Pituitary Failure Hyponatremia is so common that its impact on the human brain spectrum needs to be included in all discussion of clinical MRS interpretations. The characteristic features are reduction of mI/Cr and Cho/Cr in the 1H MRS examination67 and reduced GPE and GPC in the proton-decoupled 31P spectrum.24 Similarities to the ®ndings in HE have already been discussed.

Figure 9 Diabetic ketoacidosis (DKA): cerebral 1H spectra during two DKA episodes and recovery. (a) Episode 1, acquired 3 days after admission to hospital, at a time when the patient had supposedly totally recovered and was ready to be discharged. A peak characteristic for the presence of ketone (K) bodies was noted at 2.22 ppm, obtained from an 18 cm3 volume in the left parietal lobe. The patient relapsed into DKA 2 days later. (b) Episode 2, acquired 5 months later from an occipital gray matter location (10.3 cm3) during a second episode of DKA. In addition to the ketone peak, peaks for glucose (G) were seen. (c) Recovery, 6 days after episode 2 (from the same occipital location), when no more ketone bodies were detected in the urine. (d) Occipital cortex of an age- and sex-matched healthy subject. Acquisition conditions: GE Signa 1.5 T, 4X software; STEAM TR 1.5 s, TE 30 ms, NEX 128. More detailed analysis of peak K indicates the resonance frequency to be that of acetone, rather than acetoacetate, the ketone body more commonly identi®ed in blood and urine of diabetics in coma. Abbreviations as in Figure 1. (Reproduced by permission of the Radiological Society of North America from R. Kreis and B. D. Ross, Radiology, 1992, 184, 123)

8 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS ment other previously unsuspected changes, notably the loss of the neuronal marker NAA. The very common occurrence of hypoxic encephalopathy in humans has given ample opportunity for veri®cation of these events in the human brain. `Recovery' from nonlethal hypoxic encephalopathy gives yet another view of the metabolic process. From 31P MRS in neonates,69 the expected changes emerged. More recently, 1H MRS has been used to quantify and plot the time course of changes which occur after oxygen deprivation, applying the information to de®ning the degree of irreversible hypoxic neuronal damageÐand hence prognosis. `Near-drowning' is the term applied when virtually total oxygen deprivation occurs owing to submersion in water. It must be presumed that all ATP and PCr is lost, and all glucose converts to lactate in this period of anoxia. Yet at the ®rst clinical examinations, 24 hours after rescue and arti®cial life support,

Figure 10 Identi®cation of the ketone body observed in patients on the ketogenic diet. Top trace is the summed 1H MRS of three examinations of a patient on ketogenic diet. Lower four traces show 1H MRS of model solutions containing 5 mmol lÿ1 of acetone, hydroxybutyric acid and hydroxymethyl glutaric acid, and 10 mmol lÿ1 acetoacetate. Each model solution also contained 10 mmol lÿ1 creatine as a chemical shift reference at 3.02 and 3.94. Abbreviations as in Figure 1. (Reproduced with permission from Seymour et al.)60

hyponatremia secondary to pituitary failure suggests that some cerebral osmolytes may also be markers of hormone activity. A clearer understanding of the neuroendocrine axis may emerge from systematic use of MRS (either 1H or 1H±31P) in these very common disorders (Figure 13). 5 HYPOXIC ENCEPHALOPATHY AND `BRAIN ATTACK' The prime example of a systemically induced encephalopathy is that resulting from insuf®cient oxygenation of blood, with resultant failure of cerebral oxygen delivery. Hypoxic encephalopathy is best understood in the context of energy failure and altered redox state through all cerebral metabolic pathways. Loss of ATP and PCr, accumulation of ADP, AMP, inorganic phosphate, and Cr are obvious consequences; H+ accumulates, both from failure to remove CO2 and from formation of lactate. Reduced partners of the equilibrium enzymes lactate and glutamate dehydrogenase accumulate. In practice, glutamate is probably equally rapidly converted to glutamine. Innumerable animal studies, using principally 31P NMR, but also 1H NMR, con®rm these principles and docu-

Figure 11 Time course of changes in principal cerebral organic osmolytes during correction of severe dehydration. A series of spectra were obtained from the same (occipital gray matter) brain location, in a 14-month-old child, recovering from severe dehydration and hypernatremia (plasma sodium 195 mmol lÿ1; normal range 135±142 mmol lÿ1) using identical acquisition conditions. Spectra were processed and scaled identically to permit subtraction of sequential spectra (difference spectroscopy). The day of examination refers to interval since admission to hospital. Examinations are numbered sequentially, from the ®rst (Exam. 1) to last (Exam. 5) on day 36. Compared with a relevant normal [spectrum (b), Figure 7], the principal abnormality appears to be a reversal of the intensities of Nacetylaspartate (NAA) (reduced) and mI (increased); as a result mI dominates the spectrum. These changes slowly reverse to be nearly normal on day 36. The resultant difference spectra more clearly identify the progressively falling concentrations of several metabolites, with a possible elevation in the resonance peak assigned to the neuronal marker NAA. Quantitative MRS de®nes the principal abnormality as a threefold increase in the concentration of the cerebral osmolyte mI. Abbreviations as in Figure 1

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

9

Figure 12 Proton MRS of the pons in central pontine myelinolysis (CPM). (a) A 29-year-old female with characteristic MRI changes that developed in the absence of hyponatremia. The spectra [9 weeks post-insult (b); 1 month later (c)] shows marked elevation in lipid, which appropriately re¯ects the histology of this lesion in autopsy-proven cases of CPM. (Illustration courtesy of David Dubowitz M.D.)

MRS demonstrates more often than not the absence of lactate, normal Cr (plus PCr), and NAA at nearly full concentration. Only subsequently do NAA and total Cr fall and lactate appears. The severity of the changes gives a fairly good guide to outcome.70 The classic events of anoxia described by Lowry71 are probably reversible, but secondary damage results in progressive cell death, the consequences of which are loss of NAA (in the case of dying neurons), loss of PCr and oxidative function and reaccumulation of lactate. It is unclear why the large accumulations of glutamine occur (Figure 14). A systematic analysis of the MRS changes and their impact on outcome after global hypoxia72 should go some way to clarifying the MRS literature in focal ischemia of the human brain in stroke and provide much needed noninvasive guidance in treatment of hypoxic encephalopathy. This includes tissue plasminogen activator

(TPA), neuroprotectives and other innovations in human `brain attack'. 5.1

Chemical Shift Imaging in Acute Stroke

Most clinical questions can be quickly answered using single-voxel technique discussed in the previous section, particularly since the arrival of complete automation of voxel selection, localization, shimming, water suppression, data acquisition, phasing and semiquantitative print-out of the salient metabolite ratios.73 Nevertheless, chemical shift imaging (CSI), a much more ef®cient use of MR signal,74 is re-emerging for clinical questions in which multivoxel information is essential. In our opinion, only stroke (in adults) and adrenoleukodystrophy (ALD) in children completely ful®ll this criterion. In

10 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

Figure 14 Near-drowning with fatal outcome. Brain spectrum from occipital cortex of a severe near-drowning victim (3 years old), examined 48 hours post-immersion (top), and the glutamine standard (bottom). The patient died on the 5th day post-injury. The 1H spectrum was acquired from a volume of 12 cm3 (STEAM TR 3.0 s, TE 30 ms, data processing including correction for residual water, line ®tting, and quanti®cation as described by Kreis et al.).63 The spectrum of the glutamine standard (10 mmol lÿ1) was acquired in the same way and scaled appropriately. Notable abnormalities in the patient spectrum are the very much reduced N-acetylaspartate (NAA) concentration and NAA/Cr ratio, excess of lactate (doublet at 1.3 ppm), and of the strongly coupled resonances of , , and glutamine protons. A 25% reduction in [Cr] was apparent on quantitative examination. Abbreviations as in Figure 1. (Thanks to R. Kreis, T. Ernst, and E. Arcinue)

Figure 13 Endocrine-responsive hyponatremia. Proton-decoupled 31P MRS shows almost complete depletion of GPE and GPC [(a) pretreatment] that recovered rapidly after initiation of pituitary replacement therapy [after 1(b) and 3(c) weeks of treatment]. Inset are expanded portions of the spectra. Abbreviations as in Figure 3. (Adapted from J. Tan et al., unpublished work in this laboratory)

tumor MRS, for example, single voxel75 matches CSI with 96± 97% sensitivity for diagnosis,76 making it generally unnecessary to use the more time-consuming CSI. The MRS changes in chronic stroke are well documented. More recent interest in reversal therapies for acute stroke has For

emphasized the need for very early differentiation of penumbra, infarcts and transient ischemic attacks. When cerebral artery ¯ow falls below 100 ml minÿ1 (normal ~130 ml minÿ1), lactate was observed in the symptomatic cerebral cortex with increasing frequency.77 This single-voxel study was excellent in determining differences between the affected and unaffected hemisphere and in distinguishing borderzone infarcts, territory infarcts and no infarct.77 However, obvious regional heterogeneity could not be determined. The power of a multivoxel (single-slice) 1H MRS acquisition in distinguishing three degrees of neurochemistry after stroke is illustrated with spectra obtained for a patient who had suffered two strokes 7 days apart (Figure 15). These are seen in T2 MRI and in diffusionweighted images (DWI) as bright areas. While it was readily apparent which of the two infarcts occurred ®rst, DWI could not immediately separate the acute from the subacute lesion.

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

11

(a) New infarct

Cho

59.42

60

Signal amplitude

NAA

Cr 56.92

40

20

0

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

ppm (b) Normal 120

Signal amplitude

80 69.84 71.63

60 40 20 0

4.0 3.5 3.0 2.5 2.0 (c) Older infarct (1 week)

1.5

1.0

0.5

ppm

42.70 79.31

Signal amplitude

30

20

10

0

–10 0

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

ppm

Figure 15 Chemical shift image (CSI) of acute stroke. The patient suffered a series of focal infarcts in the left internal capsule, seen in diffusionweighted images (DWI) in left of ®gure. Three representative spectra (CSI, STEAM TE 135 ms) illustrates a new infarct (a), normal contralateral brain region (b) and a slightly older infarct, possibly 1 week earlier (c). Note that reduced N-acetylaspartate (NAA) and increased choline (Cho) are present in the early and later stages of evolving stroke. However, lactate (inverted doublet at 1.33 ppm) was present only in the more recent stroke. In this case, DWI and CSI give similar information. (Reproduced with kind permission of Stephen Holmes M.D., Queens Medical Center, Honolulu, HI)

12 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

Figure 16 Diagnostic 1H MRS of brain abscess. Computed tomographic scan (CT) was misleadingly diagnosed as malignant tumor in a young man. MRI and 1H MRS con®rms the presence of an abscess, based upon high concentrations of bacterial breakdown products within a ringenhancing cyst. (Reproduced with kind permission of Else R. Danielsen, University of Copenhagen, Copenhagen, Denmark)

The synchronous acquisition of spectra from each shows subtle differences between them, and from normal as determined in the unaffected hemisphere. Intuitively, it is attractive to consider lactate (present as an inverted doublet at 1.3 ppm only in the more recent `infarct') to be the much needed marker of viable tissue. This example is merely a pilot to the much larger studies now in progress in which CSI will be performed rapidly and as part of a broader MRI protocol of DWI, MRS and perfusion imaging.

5.2 Adrenoleukodystrophy The most common of a very rare group of inherited brain disorders of white matter, the peroxisomal disease adrenoleukodystrophy, has now been studied exhaustively with singlevoxel MRS. MRS is consistently more sensitive than MRI in detection of neurologic abnormality,78 which leads to the next

question. If MRI is normal in a leukodystrophy, where is the single voxel to be sited to achieve reliable preclinical diagnosis in this disorder? Currently we use the posterior horn of the lateral ventricle as a landmark to future disease. It would be safer to apply a robust multivoxel technique.79 6

OTHER FOCAL BRAIN DISORDERS

There is now a huge published database on which to found the clinical practice of MRS (neurospectroscopy). In papers where original data are provided (i.e. actual spectra with details of the processing applied after acquisition), this can be reliably assumed to be transferable from one instrument to another, and even across ®eld-strength (1.0±4.02).59 In the USA, a CPT code (76390) signals release of MRS into the day-to-day clinical environment by virtue of its ef®cacy in a number of common diagnoses. Limited use has been made of the discov-

Figure 17 MRI and 1H MRS in human neurotransplantation. MRI and MRS of an individual patient with Huntington's disease showing CNS immune response to resuming immunosuppression after halting for 6 months. Signi®cant decrease of N-acetylaspartate (NAA/Cr) ratio and presence of lactate strongly suggests deterioration of graft growth without immunosuppression (middle). Abbreviations as in Figure 1

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

Cyclosporine Maintenance Cr Cho

NAA

mI

4

3

2

1

0

ppm

Cyclosporine Discontinued

Lactate

4

3

2

1

0

ppm Cyclosporine Resumed

4

3

2

ppm

1

0

13

14 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS ondary tumor, stroke versus venous infarction, tumor versus stroke, abcess versus tumor in the immunocompromised patient.80 Single-voxel 1H MRS readily distinguishes between lymphoma, toxoplasmosis, progressive multifocal leukoencephalopathy and coccidiomycosis in patients with AIDS.81 Bacterial abcesses can no longer be confused with tumor, stroke, or multiple sclerosis plaque, since 1H MRS identi®es the abcess from its unique bacterial metabolic products (E. R. Danielsen, personal communication; Figure 16). 6.1

Central Pontine Myelenolysis

A rare focal lesion known pathologically to consist of lipid®lled macrophages similarly yields an unmistakable 1H MR spectrum of triglyceride (see Figure 12). Lipid pathologies in the brain can be reliably studied only with short echo times. One example of a useful clinical MRS examination, prognosis after non accidental trauma (shaken-baby syndrome) hinges on identi®cation of increasing concentrations of free lipid or `macromolecules'82 in 1H MRS.83,84 6.2

Neurotransplantation in Humans

Among the neurodegenerative diseases regularly examined by 1H MRS, only Alzheimer disease appears to have a diagnostic pattern;85±87 Parkinson's disease and Huntington's disease do not. In particular, earlier demonstrations of lactate in brain spectra of patients with Huntington's disease, or of excess glutamate in basal ganglia of patients with Parkinson's disease, have not been con®rmed.88 A recent study of patients who had received fetal cell transplants into the diseased putamen and caudate demonstrates a new clinical role of 1H MRS in focal `lesions'. The maturation, viability, partial rejection and subsequent recovery of the grafts can all be tracked in repeated 1H MRS examinations (Figure 17).

7

Figure 18 Identi®cation of abnormal or delayed myelination with proton-decoupled 31P MRS. MRS of a child with an unidenti®ed dysmyelination syndrome who appears to lack signi®cant amounts of white matter. (a) 1H MRS of parietal `white matter' shows signi®cant reduction of Cho. An unidenti®ed peak at 1.4 ppm is possibly lipid. (b) 31 P MRS with low±normal phosphodiester (PDE) for age. (c) Protondecoupled 31P MRS is markedly abnormal for a 2-year-old. GPE and GPC are reduced to the neonatal MRS level. PE and PC, however, are normal for age. Abbreviations as in Figures 1 and 3

eries reviewed in Sections 1±4, perhaps because MRI is often entirely normal! By and large, radiologists have used MRS to add detail in the differential diagnosis of lesions identi®ed ®rst by MRI: tumor recurrence versus necrosis, primary versus sec-

ADVANCES IN MULTINUCLEAR CLINICAL NEUROSPECTROSCOPY

Human brain research with 31P MRS and 13C MRS are dealt with elsewhere in this series (MRI and MRS of Neuropsychiatry; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Central Nervous System Degenerative Disease Observed by MRI; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Brain Infection and Degenerative Disease Studied by Proton MRS and Brain MRS of Human Subjects). In the purely clinical setting, 31P MRS remains disappointing. With the exception of the discovery of a new syndrome characterized by absence of PCr from the brain spectrum89 and the characterization of brain death as spectra lacking both PCr and ATP, few diagnostic 31P MR brain spectra have been identi®ed. However, proton-decoupling of 31P spectra has brought signi®cant clinical advantages and several diagnostic uses to this modality. In particular, proton-decoupled 31P MRS now permits accurate evaluation of the state of myelination in the

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

Glycerol

mI

15

Cr Cho

(a) Gln

NAA

Glu

Cr

(b)

(c)

(d)

80

70

60

50

40

ppm Figure 19 Clinical neurospectroscopy with 13C MRS. (a) A 1H±13C spectrum acquired in Canavan disease. Metabolite peak areas were ®tted and subtracted from the spectrum (b). From the ®tted curves (c) the peak ratios of Glu, Gln, and N-acetylaspartate (NAA) relative to mI were calculated and compared with the peak ratios obtained from a model solution containing equal concentrations of NAA, Cr, mI, glutamate, and glutamine (d). Absolute concentrations in mmol per kilogram brain tissue were calculated using [mI] quanti®ed with 1H MRS in the occipital region of the brain as an internal reference. Abbreviations as in Figure 1

normally developing and dysmyelinating infant brain (Figure 18).90 The identi®cation of other membrane or myelin disorders in adults is also gaining some speci®city through the application of proton-decoupled 31P MRS, since GPC, GPE, phosphorylcholine (PC) and phosphoethanolamine apparently vary almost independently in different pathological settings.91 New insights into osmolyte disturbances of hepatic encephalopathy and hyponatremia were discussed above. From being a research tool in the 1980s, natural abundance 13 C MRS has achieved clinical status through identi®cation of mI, glutamate and several other useful neurometabolites in individual patients (Figure 19).41

FDA-IND approval of [1-13C]-glucose (FDA 56,510; Figure 20)91 infusion allows exploration in a clinical setting of the very promising techniques developed in the 1980s at Yale.92 This is a dynamic technique, with many parallels to functional MRI. Carbon-13 MRS has particular relevance to the diagnosis and elucidation of the encephalopathies discussed at the beginning of this article.93±95 8

CONCLUSIONS

Diffuse metabolic changes in brain biochemistry are the result of complex interactions of disordered biochemistry in

16 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS

(a) Glu C-4 Glu C-2 Glu C-3

Gln C-4 Asp

Gln C-2

50

40

(b)

36

30

(c)

Glu C-4

35

ppm

34

ppm

Glu C-4

15 - 20 (min)

60 - 80 (min)

10 - 15

40 - 60

5 - 10

20 - 40

0-5

0 - 20

33

32

36

35

ppm

34

33

Figure 20 Dynamic 13C MRS after [1-13C]-glucose infusion. Using intravenous 13C glucose infusion to enrich cerebral metabolite pools several fold higher than the 1% natural abundance (illustrated in Figure 19) provides a measure of glutamate (Glu) synthesis and metabolism. (a) Difference spectrum from a fed adult acquired 50±70 min after infusion start. Incorporation of the 13C label into Glu C-4 in fasted mature (b) and immature brain (c) is clearly different and has diagnostic value in the premature infant (S. Bluml, J.-H. Wang, and B. D. Ross91)

many other organs and tissues. Hepatic encephalopathies, diabetic coma, hypo- and hyperosmolar states, endocrine and hypoxic encephalopathy are examples of such conditions in which accurate application of quantitative NMR spectroscopy sheds new light. A thorough knowledge of the clinical MRS of coma will almost certainly be of crucial importance to health care in these dif®cult cases. Clinical uses of 1H MRS has focused on lesions already identi®ed by MRI, but its uses are likely to expand. Routine clinical MRI scanners can now provide excellent informative proton-decoupled 31P spectroscopy and 13C brain spectra; consequently these tools too are now entering clinical use.

9

RELATED ARTICLES

Animal Methods in MRS; Brain MRS of Human Subjects; In Vivo Hepatic MRS of Humans; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Water Suppression in Proton MRS of Humans and Animals.

10 REFERENCES 1. F. Plum and J. B. Posner, `The Diagnosis of Stupor and Coma', Davis, Philadelphia, 1986. 2. G. B. Young, A. H. Ropper, and C. F. Bolton, `Coma and Impaired Consciousness', McGraw-Hill, New York, 1998. 3. S. Sherlock, W. H. J. Summerskill, L. P. White, and E. A. Phear, Lancet, 1954, 2, 453. 4. S. Bessman and A. Bessman, J. Clin. Invest., 1955, 34, 622. 5. S Bessman and N. Pal, in `The Urea Cycle', eds. S. Grisolia, R. Baguena, and F. Mayor, Wiley, Chichester, 1976, p. 83. 6. H. A. Krebs and K. Henseleit, Hoppe-Seyler's Z. Physiol. Chem., 1932, 210, 33. 7. A. Geissler, K. Kanamori, and B. D. Ross, Biochem. J., 1992, 287, 813. 8. B. T. Hourani, E. M. Hamlin, and T. B. Reynolds, Arch. Intern. Med., 1971, 127, 1033. 9. M. D. Norenberg and A. Martinez-Hernandez, Brain Res., 1979, 161, 303. 10. J-C Reubi, C. van den Berg, and M. Cuenod, Neurosci. Lett., 1978, 10, 171. 11. R. F. Butterworth and G. P. Layrargues, `Hepatic Encephalopathy: Pathophysiology and Treatment', Humana Press, Clifton, 1989.

SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS 12. L. Zieve, in `Diseases of the Liver', eds. L. Schiff and E. R. Schiff, Lippincott, Philadelphia, 1987, p. 925. 13. R. Kreis, N. A Farrow, and B. D. Ross, Lancet, 1990, 336, 635. 14. N. E. P. Deutz, A. A. de Graaf, J. B. de Haan, W. M. M. J. Bovee and R. A. F. M. Chamuleau, J. Hepatol., 1987, 4(1), S13. 15. T. E. Bates, S. R. Williams, R. A. Kauppinen, and D. G. Gadian, J. Neurochem., 1989, 53, 102. 16. S. M. Fitzpatrick, H. P. Hetherington, K. L. Behar, and R. G. Shulman, J. Cereb. Blood Flow Metab., 1990, 10, 170. 17. K. Kanamori and B. D. Ross, Biochem. J., 1993, 293, 461. 18. K. Kanamori, F. Parivar, and B. D. Ross, NMR Biomed., 1993, 6, 21. 19. J. Shen, N. R. Sibson, G. Cline, K. L. Behar, D. L. Rothman, and R. G. Shulman, Devel. Neurosci., 1998, 20, 434. 20. K. Kanamori, B. D. Ross, and E. L. Kuo, Biochem. J., 1995, 311, 681. 21. K. Kanamori, B. D. Ross, and J. Tropp, J. Magn. Reson. B, 1995, 107, 107. 22. K. Kanamori and B. D. Ross, J. Neurochem., 1997, 68, 1209. 23. S. H. Fitzpatrick, H. P. Hetherington, K. L. Behar, and R. G. Shulman, J. Neurochem., 1989, 52, 741. 24. S. Bluml, E. Zuckerman, J. Tan, and B. D. Ross, J. Neurochem., 1998, 71, 1564. 25. R. A. Moats, Y-H. H. Lien, D. Filippi, and B. D. Ross, Biochem. J., 1993, 295, 15. 26. H. Bruhn, J. Frahm, T. Michaelis, K.-D Merboldt, W. HaÈnicke, M. L. Gyngell, P. Brunner, J. Frohlich, D. Haussinger, P. Schauder, and B.D. Ross, Hepatology, 1991, 14, 121. 27. R. Kreis, N. A. Farrow, and B. D. Ross, NMR Biomed., 1991, 4, 109. 28. R. Kreis, B. D. Ross, N. A. Farrow, and Z. Ackerman, Radiology, 1992, 182, 19. 29. H. Bruhn, K.-D. Merboldt, T. Michaelis, M. L. Gyngell, W. Hanicke, J. Frahm, P. Schauder, K. Held, G. Brunner, J. Frolich, D. Haussinger, and B. D. Ross, Proc. Xth Annu. Mtg Soc. Magn. Reson. Med., San Francisco, 1991, p. 400. 30. J. R. McConnell, C. S. Ong, W. K. Chu, M. F. Sorrell, B. W. Shaw, and R. K. Zetterman, Proc. XIth Annu. Mtg Soc. Magn. Reson. Med., Berlin, 1992, p. 1957. 31. R. A. F. M. Chamuleau, D. K. Bosman, W. M. M. J. Bovee, P. R. Luyten, and J.A. den Hollander, NMR Biomed., 1991, 4, 103. 32. S. Schenker, K. J. Breen, and A. M. Hoyumpa, Gastroenterology, 1974, 66 121. 33. B. D. Ross, M. R. Morgan, I. J. Cox, K. E. Hawley, and I. R. Young, J. Cereb. Blood Flow Metab., 1987, 7, 5396. 34. B. D. Ross, J. P. Roberts, J. Tropp, K. Derby, N. Bass, and C. Hawryszko, Magn. Reson. Imag., 1989, 7, 82. 35. P. R. Luyten, J. A. den Hollander, W. M. M. J. BoveÂe, B. D. Ross, D. K. Bosman, and R. A. F. M. Chamuleau, Proc. VIIIth Annu. Mtg Soc. Magn. Reson. Med., Amsterdam, 1989, p. 375. 36. L. Barbara, B. Barbiroli, S. Gaiani, L. Bolondi, S. So®a, G. Zironi, R. Lodi, S. Iotti, P. Zaniol, C. Sama, and S. Brillanti. Eur. J. Hepatol., 1993, 2, 60. 37. S. Taylor-Robinson, R. J. Mallalieu, J. Sargentoni, J. D. Bell, D. J. Bryant, G. A. Coutts, and M. Y. Morgan, Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, p. 89. 38. A. Geissler, N. Farrow, F. Villamil, L. Makowka, T. Ernst, R. Kreis, and B. Ross, Proc. XIth Annu. Mtg Soc. Magn. Reson. Med., Berlin, 1992, p. 647. 39. R. Gruetter, E. J. Novotny, S. D. Boulware, G. F. Mason, D. L. Rothman, G. I. Shulman, J. W. Prichard, and R. G. Shulman, J. Neurochem., 1994, 63, 1377. 40. R. Gruetter, D. L. Rothman, E. J. Novotny, and R. G. Shulman, Magn. Reson. Med., 1992, 25, 204. 41. S. Bluml, J. Magn. Reson., 1999, 136, 219.

17

42. S. Bluml and B. D. Ross, in `Impact of Molecular Biology and New Technical Developments in Diagnostic Imaging', eds. W. Semmler and M. Schwaiger, Springer-Verlag, Heidelberg, 1998, p. 43. 43. B. D. Ross, S. Jacobson, F. G. Villamil, R. A. Moats, T. Shonk, and J. Draguesku, Hepatology, 1993, 18, 105A. 44. B. D. Ross, S. Jacobson, F. Villamil, J. Korula, R. Kreis, T. Ernst, T. Shonk, and R.A. Moats, Radiology, 1994, 193, 457. 45. T. Ernst, B. D. Ross, and R. Flores, Lancet, 1992, 340, 486. 46. R. K Gupta, V. A Saraswat, H. Poptani, R. K. Dhiman, A. Kohli, R. B Gujral, and S. R. Naik, Am. J. Gastroenterol., 1993, 88, 670. 47. B. D. Ross, T. Shonk, R. A. Moats, S. Jacobson, J. Draguesku, T. Ernst, J. H. Lee, and R. Kreis, Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, p. 131. 48. T. Shonk, R. Moats, J. H. Lee, J. Korula, T. Ernst, R. Kreis, J. Draguesku, and B. D. Ross, Gastroenterology, 1993, 104, A-449, 1793 49. J. Korula, D. Kravetz, M. Katz, T. Shonk, S. Hanks, and B. D. Ross, Hepatology, 1993, 18, 282A, 903. 50. D. G. Gadian, A. Connelly, J. H. Cross, S. Burns, R. A Iles, and J. V. Leonard, Proc. Xth Annu. Mtg Soc. Magn. Reson. Med., San Francisco, 1991, p. 193. 51. B. D. Ross, R. Kreis, and T. Ernst, Eur. J. Radiol., 1992, 14, 128. 52. R. A. Hawkins, J. Jessy, A. M. Mans, and M. R. De Joseph, J. Neurochem., 1993, 60, 1000. 53. B. D. Ross and S. Bluml, NMR Biomed., 1996, 9, 279. 54. D. Haussinger, J. Laubenberger, S. V. Dahl, T. Ernst, S. Bayer, M. Langer, W. Gerok, and J. Hennig, Gastroenterology, 1994, 107, 1475. 55. L. J. Haseler, W. L. Sibbit Jr, H. N. Mojtahedzadeh, S. Reddy, V. P. Agarwal, and D. M. McCarthy, Am. J. Neuroradiol, 1998, 19, 1681. 56. R. N. Bryan and P. Barker, Am. J. Neuroradiology, 1998, 19, 1593. 57. H. Bruhn, T. Michaelis, K-D. Merboldt, W. Hanicke, M. L. Gyngell, and J. Frahm, Lancet, 1991, 337, 745. 58. R. Kreis, and B. D. Ross, Radiology, 1992, 184, 123. 59. E. R. Danielsen and B. D. Ross, `Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases', Marcel Dekker, New York, 1999. 60. K. J. Seymour, S. Bluml, J. Sutherling, W. Sutherling, and B. D. Ross, MAGMA, 1999, 8, 33. 61. Y-H. H. Lien, J. I. Shapiro, and L. Chan, J. Clin. Invest., 1990, 85, 1427. 62. G. G. Wong D.Phil. Thesis, University of Oxford, 1981. 63. R. Kreis, T. Ernst, and B. D. Ross, J. Magn. Reson., 1993, 102, 9. 64. B. D. Ross, T. Ernst, R. Kreis, L. J. Haseler, S. Bayer, E. Danielsen, S. Bluml, T. K. Shonk, J. C. Mandigo, W. Caton, C. Clark, S. Jensen, N. Lehman, E. Arcinue, R. Pudenz, and C. H. Shelden, JMRI, 1998, 8, 829. 65. J. H. Lee and B. D. Ross, Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, p. 1553. 66. J. H. Lee, E. Arcinue, and B. D. Ross, N. Engl. J. Med., 1994, 331, 439. 67. J. S. Videen, T. Michaelis, P. Pinto, and B. D. Ross, J. Clin. Invest., 1995, 95, 788. 68. Y.-H. H. Lien, J. I. Shapiro, and L. Chan, J. Clin. Invest., 1991, 88, 303. 69. P. L. Hope, E. B. Cady, P. S. Tofts, P. A. Hamilton, A. M. de Costello, D. T. Delpy, A. Chu, E. O. R. Reynolds, and D. R. Wilkie, Lancet, 1984, 366. 70. R. Kreis, T. Ernst, E. Arcinue, R. Flores, and B. D. Ross, Proc. XIth Annu. Mtg Soc. Magn. Reson. Med., Berlin, 1992, p. 237. 71. O. H. Lowry in `Neurology of the Newborn,' ed. J. J. Volpe, Saunders, Philadelphia, 1987, p. 33.

18 SYSTEMICALLY INDUCED ENCEPHALOPATHIES: NEWER CLINICAL APPLICATIONS OF MRS 72. R. Kreis, E. Arcinue, T. Ernst, T. Shonk, R. Flores and B. D. Ross, J. Clin. Invest., 1996, 97, 1. 73. G. A. Webb (ed.), `Annual Reports on NMR Spectroscopy', Vol. 25, Academic Press, London, 1993, p. 480. 74. D. R. Bailes, D. J. Bryant, H. A. Case, A. G. Collins, I. J. Cox, A. S. Hall, R. R. Harman, S. Khenia, P. McArthur, B. D. Ross, and I. R. Young, J. Magn. Reson., 1988, 77, 460. 75. A. Lin, S. Bluml, W. Caton, C. Duma, A. Mamelak, R. Rand, S. Wiseman, and B. D. Ross, Proc. VIIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Philadelphia, 1999, p. 1394. 76. M. C. Preul, Z. Caramanos, D. L. Collins, J.-G. Villemure, R. Leblanc, A. Olivier, R. Pokrupa, and D. L. Arnold, Nature Med., 1996, 2, 323. 77. J. van der Grond, K. J. van Everdingen, B. C. Eikelboom, J. Kenez, and W. P. T. M. Mali, JMRI, 1999, 9, 1. 78. P. Dechent, P. J. W. Pouwels, F. Hane®eld, and J. Frahm, Proc. VIth Annu. Mtg Soc. Magn. Reson. Med., Sydney, 1998, 3, p. 1754. 79. B. Kruse, P. B. Barker, P. C. M. van Zigl, J. H. Duyn, C. T. W. Moonen, and H. W. Moser, Ann. Neurol., 1994, 36, 595. 80. M. Castillo (ed.), `Proton MR Spectroscopy of the Brain', Vol. 8, Saunders, Philadelphia, PA, 1998, p. 926. 81. L. Chang, B. L. Miller, D. McBride, M. Cornford, G. Oropilla, S. Buchthal, F. Chiang, H. Aronow, and T. Ernst, Radiology, 1995, 197, 525. 82. K. L. Behar and T. Ogino, Magn. Reson. Med., 1993, 30, 38. 83. L. J. Haseler, E. Arcinue, E. R. Danielsen, S. Bluml, and B. D. Ross, Pediatrics, 1997, 99, 4. 84. T. Chen, N. Farrow, S. Bluml, C. Huang, and B. D. Ross, Proc. VIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Sydney, 1998, 1, 537. 85. B. L. Miller, R. Moats, T. Shonk, T. Ernst, S. Woolley, and B. D. Ross, Radiology, 1993, 187, 433. 86. R. A. Moats and T. Shonk, Am. J. Neuroradiol., 1995, 16, 1779. 87. T. K. Shonk, R. A. Moats, P. Gifford, T. Michaelis, J. C. Mandigo, J. Izumi, and B. D. Ross, Radiology, 1995, 195, 65. 88. T. Hoang, D. Dubowitz, S. Bluml, O. V. Kopyov, D. Jacques, and B. D. Ross, Proc. 27th Meeting of the Society for Neuroscience, New Orleans, FL, 1997, 23(2), 1682.

89. S. Stockler, F. Hanefeld, and J. Frahm, Lancet, 1996, 348, 789. 90. K. Seymour, S. Bluml, G. McComb, and B. D. Ross, Proc. VIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Sydney, 1998, 3, p. 1807. 91. S. Bluml, J. H. Hwang, and B. D. Ross, Proc. VIIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Philadelphia, 1999, p. 335. 92. G. F. Mason, R. Gruetter, D. L. Rothman, K. L. Behar, R. G. Shulman, and E. J. Novotny, J. Cereb. Blood Flow Metab., 1995, 15, 12. 93. S. Bluml, J. Magn. Reson., in press. 94. S. Bluml, J. H. Wang, A. Moreno, L. Lim, J. Tam, and B. D. Ross, Radiology, 1999, 213, 330P. 95. S. Bluml, A. Moreno, and B. D. Ross, Lancet submitted.

Acknowledgements Work reported was largely funded by the Rudi Schulte Research Institute, Santa Barbara, the Jameson Foundation and by funds from the HMRI MRS Program. BDR is grateful to the following colleagues: K. Kanamori, E. Rubaek-Danielsen, J.D. Roberts, N. Kumar, M. Linsey and C. Sharp.

Biographical Sketch Brian D. Ross. b 1938; B.Sc., 1958, University College, London. D.Phil., 1966, University of Oxford. M.B., 1961, University College Hospital, London. F.R.C.S., 1973, Royal College of Surgeons, London. M.R.C.Path., 1976, Royal College of Pathologists, London. 1989, F.R.C.Path. University of Oxford lecturer, Metabolic Medicine. Director, Renal Metabolism Unit and Consultant Chemical Pathologist, Radcliffe In®rmary, Oxford, 1976±84. Director of Clinical Spectroscopy Programs at Radcliffe In®rmary, Oxford, 1981±84, Hammersmith Hospital, London, 1986±88, and Huntington Medical Research Institutes, Pasadena, CA, 1986±present. Visiting Associate, California Institute of Technology, 1986±present.

TEMPOROMANDIBULAR JOINT MRI

Temporomandibular Joint MRI Steven E. Harms Baylor University Medical Center, Dallas, TX, USA

1

The disk position is primarily determined by weight bearing, with the thin zone being the point of maximum load. During mastication, weight bearing is transferred to the food between the teeth and the TMJ is unloaded. Without weight bearing to hold the thin zone in position, there is a tendency for the elastic retrodiskal tissues to pull the disk posteriorly. This force is opposed by a contraction of the anterior belly of the lateral pterygoid muscle. Disk displacement is most commonly anterior due to laxity or disruption of the elastic retrodiskal tissues that hold the disk posteriorly.1,2

1 INTRODUCTION

3

The value of medical imaging is determined by its ability to de®ne treatment decisions. If those decisions are separated by treatments that greatly differ in cost or morbidity, the value of the diagnostic test is high. On the other hand, if the treatment is safe and inexpensive, then it may be more cost-effective to treat without diagnosis. As the treatment of temporomandibular joint (TMJ) disease evolves, the role of MR imaging as a determinant for clinical management decisions becomes better de®ned. In this chapter, the MR imaging of the TMJ is reviewed with its relationship to the current concepts for clinical management of TMJ disorders.

Currently, most treatment teams accept the goal of TMJ treatment as relief of pain and/or restoration of function. This philosophy is a departure from the early years of TMJ treatment when restoration of normal anatomical relationships was thought to be the treatment goal. This distinction is important since relief of symptoms may be achieved in the face of abnormal anatomical relationships. Conservative TMJ therapy may consist of approaches such as: biofeedback, nonsteroidal antiin¯ammatory agents, muscle relaxers, bedtime sedatives, splints, and/or any combination of these approaches. Conservative therapy generally refers to the fact these treatments are generally noninvasive and reversible. In terms of cost, however, some conservative therapy may be more expensive than surgery. It is generally accepted that a trial of some form of conservative therapy should be attempted before invasive therapy is considered. Since the decision for conservative therapy approach is based upon the symptoms and clinical settings, MRI is not typically indicated in this management decision.4,5 If conservative therapy fails, then surgery may be required to relieve pain and/or restore function. There are a number of possible surgical treatments that are designed to address a particular physical defect. Joint adhesions may be treated by therapies as simple as joint lavage or as complex as arthro-

2 ANATOMY AND PHYSIOLOGY The TMJ is a compound joint with the three components being the temporal bone, the mandibular condyle, and the articular disk (Figure 1). There are two joint spaces: the superior joint space lying between the disk and the temporal bone and the inferior joint space lying between the disk and the mandibular condyle.1,2 The articular disk is a lens-shaped structure divided into three parts: the posterior band, the anterior band, and the thin zone. In the young asymptomatic population without previous orthodontic manipulation, the posterior band normally lies between 11 o'clock and 1 o'clock relative to the superior aspect of the mandibular condyle. As opposed to what was originally thought, the disk can be quite variable in location without producing symptoms.3 The articular disk of the TMJ is held in place by the discal ligaments which extend from the medial and the lateral poles of the condyle to the medial and lateral margins of the disk. As a result of the strong attachment to the condyle by the diskal ligaments, the disk and condyle move together as a unit called the disk±condyle complex. The posterior band of the disk is connected to the posterior aspect of the temporal fossa by elastic tissues called the retrodiskal attachments. Since the retrodiskal attachment is composed of two major elastic structures, this area is sometimes called the bilaminar zone. The anterior band of the disk is attached to the superior belly of the lateral pterygoid muscle.1,2 There are two movements that occur with opening: translation and rotation. The ®rst movement is rotation which is the swinging of the condyle relative to the disk. The next movement is the translation of disk±condyle complex anteriorly from the articular fossa to a position beneath the articular eminence.1,2

TREATMENT

Figure 1 Normal temporomandibular joint. The bone landmarks are clearly visible on the oblique sagittal SE 15/2000 image of the temporomandibular joint. The mandibular condyle, c, and eminence, e, are labeled. Normal disk architecture is demonstrated with good demonstration of the anterior band, a, posterior band, p, and the thin zone, t

2 TEMPOROMANDIBULAR JOINT MRI scopy.6 Although attempts to correct disk displacement are made by TMJ arthroscopists, it is generally recognized that the major treatment effect of arthroscopy is the lysis of joint adhesions.7 The lysis of adhesions is often highly effective in relieving joint symptomology without any correction in disk displacement. This discovery is a signi®cant change from the original concept of TMJ disorders which attributed TMJ pain and dysfunction to disk displacement alone. The success of adhesion lysis therapy has made the imaging diagnosis of adhesions an important part of the MR evaluation. Internal derangements of the articular disk can be approached with a variety of conventional surgical procedures. If the disk architecture is relatively well preserved, then disk plication may be used. The plication procedure pulls the usually anteriorly displaced disk back into position over the condyle and repairs the posterior attachment. Disk plication was once the most common surgical procedure but has fallen out of favor with many surgeons due to the common complication of adhesions in the postoperative period. In addition, abnormal disk architecture is often seen in association with displacement. Repositioning of a torn or otherwise deformed disk by plication is not possible. Diskectomy is often accompanied by replacement with a dermal graft, elastic cartilage graft, or alloplastic material.8 The common complication of giant cell foreign body reaction to Proplast TMJ implants has greatly reduced the use of alloplastic disk replacements.9,10 The poor outcome experienced with alloplastic implants has increased the popularity of autologous tissues as disk replacements. Postoperative adhesions remain a problem and despite attempts the ideal replacement material for severely degenerated or deformed disks has not yet been found. For end-stage TMJ disorders where degeneration changes in the condyle and temporal bone predominate, the surgical treatment is often total joint replacement. A variety of appliances have been designed for total TMJ replacement. All of the devices comprise some form of arti®cial fossa and condyle. Because metallic components are used, MRI evaluations of the TMJ in postoperative total joint patients is not possible. As with total joint replacement in other joints, the long-term durability of the appliance is a problem. As a result, surgeons tend to restrict this approach to older individuals or to patients with severe degenerative change.11 Avascular necrosis of the mandibular condyle has been described by some workers as a common cause of TMJ disorders.12,13 These workers favor corrections of the problem by condylar decompression or replacement. This diagnosis and treatment is highly controversial. Many in the TMJ treatment ®eld refute this theory.

4 TECHNIQUE The TMJ examination should consist of the following components: 1. static sagittal high-resolution images in the closed mouth position; 2. static coronal views in the closed mouth position; 3. kinematic or real time images during opening; 4. at least one sagittal T2-weighted series.

A dedicated TMJ surface coil is now considered an essential feature for adequate TMJ imaging. The coil may be as simple as a single 3-inch-diameter loop.14 Both sides must be evaluated simultaneously in order to demonstrate asymmetry. About 80% of TMJ disorders are bilateral. These coils can be combined either with a special combiner box or with signals separately acquired with phased array coils and receivers.15,16 The phased array approach will result in a theoretical square root of 2 S/N improvement but the technology is considerably more costly. The anatomical diagnosis of most TMJ disorders is made from the high-resolution sagittal images.14,17±23 A short TE proton density or T1-weighted pulse sequence, similar to the technique used for knee imaging, is required for optimal S/N and resolution. Because volume averaging effects are an important cause of image degradation, the slice thickness should be around 3 mm. An oblique image plane is performed using the angle of the mandibular body and condyle as determined on axial scout views as a reference. The short echo anatomical images are most important in TMJ diagnosis. As a result, traditional spin echo (SE) pulse sequences are preferred over fast SE or RARE sequences due to the blurring of short, effective TE echo images. T2 weighting can be achieved in both a dual echo SE sequence and a separate, long, effective TE fast SE (FSE) sequence. The FSE will result in improved image resolution but the fat signal will be more intense. The fat signal can be addressed with the addition of a fat saturation prepulse which lengthens the TR. Since the objective of T2 weighting is the identi®cation of ¯uid and not anatomical de®nition, there is little advantage in the use of FSE over a dual echo traditional SE.14,17±23 Coronal T1-weighted imaging is used to localize the disk in the medial-lateral direction. A faster sequence with diminished resolution compared with the sagittal images will usually be satisfactory.24,25 Kinematic or real-time motion images are needed to demonstrate TMJ physiology (Figure 2). A fast T1-weighted SE is usually satisfactory. Gradient echo sequences may be employed if the timing parameters are adjusted to demonstrate good diskal anatomy. Often gradient echo images for cine TMJ imaging are inadequate to demonstrate this anatomy and are not the favored method for this reason. Because multiple slices are needed, there is no time saving associated with the use of gradient echo sequences. Scan time on the order of one minute per view are satisfactory. At least two slices per side should be performed in case the disk moves medially or laterally out of the view of one slice. Since the condyle is the reference for disk reduction, the slices are oriented sagittally, not oblique sagittally, as in the anatomical images. The condyle is a rigid structure that must travel in a straight line in the sagittal plane. The disk, however, will tend to move anteromedially. If oblique images are performed, the late opening images will display a disk and no condyle for reference. The determination of disk reduction will require the identi®cation of the disk relative to the condyle. For kinematic imaging, the TMJ will need to be incrementally opened at predetermined intervals. We prefer to open in 3 mm increments until the patient can no longer proceed either due to pain or inability to open. One of several commercially available patient-controlled positioning imaging devices may be employed for establishing consistent incremen-

TEMPOROMANDIBULAR JOINT MRI

3

evaluated for anatomic defects. Osteophytes and beaking of the mandibular condyle are common. A steep articular eminence is associated with an increased incidence of TMJ disorders. Displacement of the condyle in the fossa may be a clue as to more subtle soft tissue defects. Hypointensity of the condylar marrow may indicate avascular necrosis.12,13 End-stage TMJ disorders result in severe degenerative changes. Erosions and large osteophytes are common. The eminence and condyle may be ¯attened. The joint space may be almost totally obliterated. If these ®ndings are detected, a total joint replacement may be indicated. In patients with a history of alloplastic implants, destruction of bone should be recognized as an effect associated with giant-cell foreign-body reaction. If large erosions are seen in these patients, look for an associated soft tissue mass that is hypointense on T2-weighted images which is the hallmark of giant-cell reaction. Fragments of implant material that are hypointense on all sequences may be identi®ed. Giant-cell foreign-body reactions may be very severe and on occasion can penetrate through the temporal bone into the middle cranial fossa.8±10 5.2

Figure 2 Kinematic imaging. Two images from a kinematic display are shown: the closed mouth (a) and the maximum open mouth (b) views. The closed mouth view demonstrates an anteriorly displaced disk, d, relative to the mandibular condyle, c. With opening, there is poor translation of the condyle which only translates to mid-eminence, e. The disk remains anteriorly displaced throughout opening

tation. A reliable device is needed in order to quantitate the interincisor opening at the time of disk reduction.26±29 Most commercially available instruments have this kinematic display capability. Unfortunately, some manufacturers offer this capability only at additional cost. The lack of kinematic display will greatly limit the ability to make clinically signi®cant TMJ diagnoses. A more sophisticated and more physiologic approach is the gating of the MR imaging acquisition. Gating similar to cardiac gating is used during repetitive opening of the TMJ. It remains to be seen if this approach provides a clinically signi®cant improvement in examination quality. Echo planar methods may be used in the future for realtime TMJ imaging. Some believe that this approach would be more physiologic. Imaging during mastication would be possible with echo planar imaging. 5 ELEMENTS OF A TMJ INTERPRETATION 5.1 Bone The landmarks for TMJ anatomy are the mandibular condyle and temporal bone. These structures should themselves be

Disk Position and Morphology

The common method for recording disk position in the closed mouth sagittal views is the relationship to the condyle. Imagine the round mandibular condyle as a clock face and then locate the posterior band of the disk. The posterior band will usually be the thickest part of the posterior aspect of the disk. Sometimes the posterior band will be ¯attened or the disk may be deformed with an unidenti®able posterior band. In these cases record position as the posterior aspect of the disk rather than the posterior band. The normal posterior band position is 11 o'clock to 1 o'clock; anything else is displaced. If the disk is displaced greater than the length of the disk from the fossa, then this severe displacement should be noted. Anterior displacements are the most common. Posterior displacements are rare.3±5 The position of the disk in the medial-lateral direction is demonstrated on the coronal views. If the disk is dif®cult to identify on the sagittal view, look on the coronal view where a pure medial or lateral displacement may be seen. Antero-medial displacements are common. Antero-lateral displacements are less commonly seen.24,25 Disk morphology should be described. Are the anterior band, posterior band, and thin zone well demonstrated? If not, the disk morphology is abnormal. The disk may be thickened or distorted. A thin posterior band is a common ®nding in early TMJ disease. End-stage TMJ disease results in fragmentation and degeneration. The disk may not be identi®able. Fragments of disk-like intensity may be seen. These areas commonly represent a combination of disk fragments, ®brosis, and metaplasia of the retrodiscal tissues. Tears or perforations are ®ndings described in TMJ arthrography as a communication of contrast from one joint space to another. Most of these perforations actually involve the retrodiscal tissues, not the disk itself. Perforations of the disk itself are rare and are almost always associated with other, more signi®cant, abnormalities. Temporomandibular joint perforations are dif®cult to diagnose with MRI. With more experience, MR

4 TEMPOROMANDIBULAR JOINT MRI imaging has demonstrated more detail of the defects that better de®ne appropriate treatment requirements.17

whereas a closed locked joint, due to disk displacement, usually requires surgery (an open surgical procedure).26±29

5.3 Hyperintense Signal on T2-Weighted Images

6

Fluid in either the superior or interior joint spaces is best seen on the T2-weighted images. As with other joints, ¯uid in the TMJ is a nonspeci®c sign of in¯ammation. Fluid may also be seen around the pterygoid muscle as a sign of possible myofascial disease. Edema in the retrodiskal area indicates possible retrodiskitis. Retrodiskitis is acutely painful and often dif®cult to separate from an internal derangement on physical examination. The MR diagnosis of retrodiskitis is important since this disorder may be effectively treated with antiin¯ammatory agents. Hyperintensity in retrodiskitis begins at the posterior margin of the disk and extends posteriorly in the fossa. Often the condyle is displaced antero-inferior by the soft tissue mass. This appearance should be distinguished from normal retrodiskal veins that are hyperintense on T2-weighted images. These veins can be distinguished by their location, posterior and inferior compared with retrodiskitis. In contrast to retrodiskitis which is con¯uent, retrodiskal veins are seen as multiple, small spots typical of vessels in cross-section.17

There is a multifaceted spectrum of causes for TMJ-associated pain and dysfunction. As the understanding of TMJ disorders increases, the methods for treatment become more focused and more effective. Accurate diagnosis of these entities is essential in designing treatment protocol that is appropriate for a particular patient. Modern MRI methods can make the most of the diagnostic determinations needed for these treatment decisions.

7

Kinematic imaging is needed to describe accurately abnormalities in joint function. Disk rotation and disk±condyle complex translation should be observed. The right and left sides should be compared side-by-side using a kinematic display.26±29 What is the relationship of the disk to the condyle during the opening cycle? If the disk is displaced in the closed mouth view, does it reduce upon opening? When does it reduce? Early or late? By knowing the interincisor distance associated with each step, the extent of opening at the point of disk reduction can be determined. Kinematic imaging is used for the diagnosis of joint adhesions. Adhesions between the disk and the condyle are demonstrated by ®xation of the disk to the condyle during the entire opening cycle. Adhesions to the eminence are shown as lack of separation between the disk and eminence as well as ®xation to the eminence during opening. Asymmetric translation is indirect evidence of adhesions. Very poor translation on physical examination is often called a closed locked joint. The classic description of closed locked physiology is an anteriorly displaced disk without reduction. Theoretically the displaced disk impairs condylar translation and locks the joint in the closed mouth position. MR studies, however, reveal that an anteriorly displaced disk, without reduction on opening, uncommonly produces locking. Other causes for a closed locked joint on physical examination should be considered. Severe adhesions can more commonly cause the clinical picture of a locked joint. When these adhesions totally encompass the condyle and prevent any translation, the condition is called ®brous ankylosis. Previous surgery with dislocated or adhesed implant is another common cause of poor translation in the postoperative patient. As mentioned previously, adhesions can be managed with arthroscopy,

RELATED ARTICLES

Cardiac Gating Practice; Eye, Orbit, Ear, Nose, and Throat Studies Using MRI; Surface and Other Local Coils for In Vivo Studies; Whole Body Machines: NMR Phased Array Coil Systems.

8 5.4 Kinematic Evaluation

SUMMARY

REFERENCES

1. J. P. Okeson, `Fundamentals of Occlusion and Temporomandibular Disorders', Mosby Year Book, St. Louis, MO, 1985. 2. W. E. Bell, `Temporomandibular Disorders: Classi®cation, Diagnosis, Management', 2nd edn., Mosby Year Book, Chicago, 1986. 3. L. T. Kircos, D. A. Ortendahl, A. S. Mark, and M. Arakawa, J. Oral Maxillofac. Surg., 1987, 45, 852. 4. F. M. Bush, J. H. Butler, and D. M. Abbott, in `Advances in Occlusion: Diagnosis and Treatment', eds. H. C. Lundeen and C. H. Gibbs, Publishing Sciences Group, Littleton, MA, 1982. 5. M. F. Dolwick, in `Internal Derangements of the Temporomandibular Joints,' eds. C. A. Helms, R. W. Katzberg, and M. F. Dolwick, Radiology Research and Education Foundation, San Francisco, 1983. 6. J. J. Moses, D. Sartoris, R. Glass, T. Tanaka, and I. Toker, J. Oral Maxillofac. Surg., 1989, 47, 674. 7. M. T. Montgomery, J. E. Van Sickels, S. E. Harms, and W. J. Thrash, J. Oral Maxillofac. Surg., 1989, 47, 1263. 8. D. P. Timmis, S. B. Aragon, J. E. Van Sickels, and T. B. Aufdemorte, J. Oral Maxillofac. Surg., 1986, 44, 541. 9. P. A. Kaplan, J. D. Ruskin, H. K. Tu, and M. A. Knibbe, Am. J. Roentgenol., 1988, 151, 337. 10. K. P. Schellhas, C. H. Wilkes, M. el Deeb, L. B. Lagrotteria, and M. R. Omlie, Am. J. Roentgenol., 1988, 151, 731. 11. R. W. Bessette, R. Katzberg, J. R. Natrella, and M. J. Rose, Plast. Reconstr. Surg., 1985, 75, 192. 12. K. P. Schellhas, C. H. Wilkes, H. M. Fritts, M. R. Omlie, and L. B. Lagrotteria, Am. J. Roentgenol., 1989, 152, 551. 13. C. H. Wilkes, Arch. Otolaryngol. Head Neck Surg., 1989, 115, 469. 14. S. E. Harms and R. W. Wilk, Radiographics, 1987, 7, 521. 15. C. J. Hardy, R. W. Katzberg, R. L. Frey, J. Szumouski, S. Totterman, and O. M. Mueller, Radiology, 1988, 167, 835. 16. J. S. Hyde, A. Jesmanowicz, W. Froncisz, J. B. Kneeland, T. M. Grist, and N. F. Campagna, J. Magn. Reson., 1986, 70, 512. 17. S. E. Harms, R. M. Wilk, L. M. Wolford, D. G. Chiles, and S. B. Milan, Radiology, 1985, 157, 133. 18. C. A. Helms, L. B. Kaban, C. McNeill, and, T. Dodson, Radiology, 1989, 172, 817.

TEMPOROMANDIBULAR JOINT MRI 19. C. A. Helms, T. Gillespy III, R. E. Sims, and M. L. Richardson, Radiol. Clin. North Am., 1986, 24, 189. 20. T. L. Miller, R. W. Katzberg, R. H. Tallents, R. W. Bessette, and K. Hayakawa, Radiology, 1985, 154, 121. 21. R. W. Katzberg, R. W. Bessette, R. H. Tallents, D. B. Plewes, and J. V. Manzione, Radiology, 1986, 158, 183. 22. R. W. Katzberg, Radiology, 1989, 170, 297. 23. M. T. Cirbus, M. S. Smilack, J. Beltran, and D. C. Simon, J. Prosthet. Dent., 1987, 57, 488. 24. R. W. Katzberg, P.-L. Westesson, R. H. Tallents, R. Anderson, K. Kurita, J. V. Manzione Jr., and S. Totterman, Radiology, 1988, 169, 741. 25. M. B. Khoury and E. Dolan, American Journal of Neuroroentgenography, 1986, 7, 869. 26. K. R. Burnett, C. L. Davis, and J. Read, Am. J. Roentgenol., 1987, 149, 959. 27. W. F. Conway, C. W. Hayes, and R. L. Campbell, J. Oral Maxillofac. Surg., 1988, 46, 930.

5

28. W. F. Conway, C. W. Hayes, R. L. Campbell, and D. M. Laskin, Radiology, 1989, 172, 821. 29. J. M. Fulmer and S. E. Harms, Top. Magn. Reson. Imag., 1989, 1, 75.

Biographical Sketch Steven E. Harms. M.D., 1978, University of Arkansas for Medical Sciences. Postdoctoral research associate, physical chemistry, Stony Brook, 1977±78; assistant radiologist, Depts. of Diagnostic Radiology and Biophysics; M.D., Anderson Cancer Center, Houston, 1982±83 TX. Director of Magnetic Resonance, Dept. of Radiology, Baylor University Medical Center, Dallas, 1983±present. Approx. 80 papers. Current research interests: three-dimensional imaging applications, development of new fast scanning sequence, and musculoskeletal, temperomandibular joint, breast, body, and ophthalmological applications of MRI.

APPLICATIONS OF 19 F-NMR TO ONCOLOGY

Applications of Oncology

19

F-NMR to

Paul M. J. McSheehy St George's Hospital Medical School, London, UK

Laurent P. Lemaire Laboratoire de Biophysique MeÂdicale Faculte de MeÂdecine, Angers, France

and John R. Grif®ths St George's Hospital Medical School, London, UK

1 INTRODUCTION The ®rst reported 19F MRS experiments that applied directly to chemotherapy date from 1984, in which the metabolism of an anticancer ¯uoropyrimidine, 5-¯uorouracil (5FU), was studied in mouse liver and a subcutaneous mouse tumor.1 Subsequently, most preclinical 19F MRS studies have indeed been concerned with 5FU and its metabolism in liver, and a variety of tumors in different animal models. This work, and other applications of 19F MRS, have been extensively reviewed since the late 1980s.2±5 The most recent review6 demonstrates that 19F MRS is of increasing use in the clinic and could, in the future, be used to individualize and optimize chemotherapy. This review will brie¯y describe the rationale for the development of ¯uoropyrimidines as anticancer drugs, the key 19F MRS experiments that have led to the exciting developments in the clinic, and the future mechanistic research required to realize fully the potential of this technique. Reference will also be made to other types of ¯uorinated compound which have been studied by 19F MRS to investigate tumor hypoxia, blood ¯ow and pH, and drug pharmacokinetics in animal models. These other studies show how 19F MRS can be used at an earlier stage in drug development as a research tool.

include capecitabine and the FPs, Flox (5-¯uorouridine), Tegafur (tetrahydro-2-furanyl-5-¯uorouracil), and Dox¯uoridine (5-deoxy-5-¯uorouridine)9; however, 5FU remains the FP of choice, predominantly for the treatment of breast, head and neck, and gastrointestinal tumors. Despite problems with toxicity, 5FU remains the only drug with signi®cant activity against colorectal adenocarcinoma, which is a tumor frequently metastasizing to the liver. At least 50% of 5FU is catabolized in the liver, i.e. deactivated, with probably small contributions from some types of tumors (Figure 1).10,11 The metabolism of 5FU can be modulated to increase cytotoxicity, e.g. by increasing Fnuct formation using methotrexate, interferon, or N(phosphonacetyl)-L-aspartate (N-PALA), or by increasing binding of FdUMP to TS using leucovorin, or decreasing catabolism by coadministration of thymidine. Because of the relatively wide chemical shift range of 19F MRS, many of the metabolites of 5FU can be resolved, especially in vitro, where pH-dependent differences in chemical shift can be exploited. It is this property, along with a relatively high sensitivity (~80% of 1H), and the absence of a background signal, that makes 19F MRS ideally suited to monitoring the pharmacokinetics of 5FU and other ¯uorine-containing drugs. That said, in models in vivo, and especially in the clinic, where ®elds of only 1.5 T are employed, the 5-¯uoronucleosides (Fnucs) and Fnuct, and the ¯uorocatabolites (Fcat) of ¯uoro- -ureidoproprionic acid (FUPA) and -¯uoro- -alanine(FBal) are not easily distinguishable. Nevertheless, 5FU, Fcat, and often Fnuct, can be detected in both liver and tumor in vivo at doses near or equivalent to those in clinical use, and this has allowed real-time pharmacokinetics to be used, which in some cases can predict tumor response. Figure 2 shows these signals, in this case those observed in a rat mammary tumor following treatment with 5FU.

Anabolism Uridine phosphorylase

The 5-¯uoropyrimidine (FP) 5FU was synthesized in 1957, and is an example of rational synthesis in which the substitution of a ¯uorine atom for hydrogen in position 5 of the pyrimidine uracil was intended to lead to misincorporation of the drug into RNA, and inhibition of DNA synthesis.7 Anticancer activity is now known to be mediated via conversion to cytotoxic 5-¯uoronucleotides (Fnuct) (see Figure 1). Speci®cally, FdUMP inhibits the key DNA synthesis enzyme thymidylate synthase (TS), FdUTP becomes misincorporated into DNA, and misincorporation of FUTP interferes with RNA maturation.8 Effective prodrugs of 5FU have also been synthesized which ®rst require conversion to 5FU in the liver, or tumor, and can have a better therapeutic ratio. These prodrugs

R-I-P

Catabolism 5-fluorouracil dR-I-P

FUrd PPRTase FUMP

2 5-FLUOROPYRIMIDINES: ANIMAL MODELS AND CLINICAL APPLICATIONS

5

PRPP Pyrimidine phosphoribosyl transferase

NADPH

Thymidine phosphorylase

Dihydropyrimidine dehydrogenase DHFU

FdUrd Thymidine kinase FdUMP

FUPA urea

FUDP Ribonucleotide reductase FUTP

FU-RNA

FdUDP

FBal

FdUTP

FU-DNA

Figure 1 Anabolic and catabolic pathways of 5FU. DHFU, dihydro¯uorouracil; FUPA, ¯uoro- -ureidopropionic acid; FBal, ¯uoro- -alanine; FUrd, 5-¯uorouridine; FUMP, 5-¯uorouridine monophosphate; FUDP, 5-¯uorouridine diphosphate; FUTP, 5¯uorouridine triphosphate; FdUrd, 5-¯uorodeoxyuridine; FdUMP, 5¯uorodeoxyuridine monophosphate; FdUDP, 5-¯uorodeoxyuridine diphosphate; FdUTP, 5-¯uorodeoxyuridine triphosphate; pyrimidine phosphoribosyl transferase

2 APPLICATIONS OF 19F-NMR TO ONCOLOGY Fcat

Fnuct 5FU 170–190 150–170

130–150

110–130

90–110 70–90 50–70 30–50 15

10

5

0

–5

–10

–15

–20

–25

–30

–35

ppm

Figure 2 Fluorine-19 MRS spectra of 5FU metabolism by a chemically induced primary rat mammary tumor. Spectra (20 min blocks) were acquired at 4.7 T using a one-turn surface coil on a 5 g tumor following injection of 5FU (150 mg kgÿ1 intraperitoneally)

After the demonstration that 19F MRS could detect 5FU and metabolites in vivo,1 research was initiated to relate the amount of signal to tumor response. Decreasing doses of 5FU administered as an intravenous bolus (120, 50, and 25 mg kgÿ1) to rats bearing the Walker carcinosarcoma led to decreased 5FU signal and Fnuct formed in the tumors. Tumors showed a signi®cant response to the 50 mg kgÿ1 dose, but not to the 25 mg kgÿ1 dose.12 Other studies13,14 showed a signi®cant correlation between shrinkage of mouse (RIF-1) tumors with the amount of Fnuct formed in the tumor. Furthermore, there was a negative correlation between Fcat in the mouse liver and Fnuct formed in the tumor, indicating liver deactivation and tumor activation of 5FU were competing processes. This latter observation was important for clinical studies, where the tumor signal-to-noise ratio (S/N) may be very low, but analysis of liver signals may still be of value for prognosis. However, in these animal studies the correlation between 5FU retention in the tumor and response to treatment was weaker or not signi®cant.13±15 Detection of increased Fnuct formation by 19F NMS, signifying increased cytotoxicity, has also been recorded in ascites tumor cells.16 Modulation of 5FU metabolism by a range of substances, including allopurinol,12 methotrexate,17±20 interferon20,21, NPALA and leucovorin,20 and thymidine,21,22 has been observed in animal tumor models. In some cases this has been shown to lead to a change in the hal¯ife (t1/2) for 5FU clearance,12,13,18,21,22 and, more signi®cantly, to an increase in the rate18,19 and ®nal amount of Fnuct formed;17±19,21,22 results that have also been con®rmed in extracts.12,18±21 In general, the

biochemical basis for these modulations is fairly well established. Fluorine-19 MRS demonstrated this could be observed in situ, noninvasively, and in real time, providing additional data on the rate of elimination of 5FU from the tumor, which normally can only be obtained by biopsy, indirectly via blood or urine samples, or by sacri®ce of many animals. Furthermore, in some cases, increased Fnuct formation was also correlated with tumor response.19,22 Besides biochemical modulation of 5FU metabolism, the tumor environment can be transiently altered to favor increased drug uptake, an approach with potentially important applications in the clinic where effective drug delivery to solid tumors can be a major problem. For example, breathing of carbogen gas by rodents can lead to increased tumor blood ¯ow and oxygenation,23,24,25 and in RIF-1 tumors this led to increased 5FU uptake and FNuct formation, with a corresponding increase in tumor growth inhibition.14 In the clinic, the SNR is lower, predominantly because of the lower ®eld strengths employed: 1.5±2 T. This means that in contrast to the experimental tumors in animals, generally only the 5FU and Fcat signals are observed, and rarely the cytotoxic species Fnuct. However, results accumulating in the clinic have shown correlations between 5FU retention in the tumor and patient response, whether partial, or complete.26±29 In three independent studies,26±28 5FU signals have been observed in 78±100% of responders, but in only 8±23% of nonresponders. Furthermore, there is clearly a slower clearance of 5FU from tumors (t1/2 of 20±60 min) compared with blood (8±20 min). This phenonomon is known as `tumor trapping', and its signi®-

APPLICATIONS OF 19 F-NMR TO ONCOLOGY

cance is described below. Some studies on the modulation of 5FU metabolism have also begun in the clinic, and changes have been detected using methotrexate, leucovorin, and interferon.6 Previously undetected, or more intense and prolonged, 5FU signals were observed, and, furthermore, when using combination chemotherapy, Fnuct signals were also detected. The clinical utility of this type of information would be the detection of nonresponders during the early phases of treatment, and the optimization of scheduling for others. A number of preclinical studies have been performed to identify the mechanism(s) for the trapping of 5FU that is observed clinically. It is already clear that the tumor energy status is likely to play a major role, since a high-energy status (high NTP/Pi ratio) would enable a more effective activation of 5FU to FNuct and would be likely to re¯ect a well-vascularized tumor, making 5FU access to the tumor easier. In two multinuclear MRS studies,15,30 there were signi®cant correlations between the pretreatment tumor Pi, PCr, or NTP, the ®nal amount of FNuct formed, and the response of the RIF-115 or primary rat30 tumors to 5FU treatment. Consistent with these observations, treatment with hydralazine (a vasodilator that deenergizes tumors31) favored the conversion of FNuct to 5FU.6 In contrast, in both the mouse15 and rat model,30 no signi®cant correlation was observed between the pretreatment pH measured by MRS (i.e. intracellular pH, pHi) and treatment response. However, in a rat ®brosarcoma model, a 2.5-fold increase in the 5FU t1/2 (increased 5FU trapping) was observed when the pHi was reduced by increased tumor glycolysis,32 suggesting tumor pH could have a role in tumor retention of 5FU. What may be of signi®cance is the low extracellular pH (pHe) of solid tumors,33 which in combination with a neutral pHi leads to a reverse, or negative, pH gradient across the tumor plasma membrane (i.e.ÿ
pH).34 One study using isolated tumor cells showed that 5FU uptake was increased when the ÿ
pH was increased;35 a similar relationship was observed in vivo.21 An explanation of these apparently con¯icting different studies is that in the rat ®brosarcoma model, the induced decrease in pHi probably casued a larger decrease in pHe,35 thereby increasing the ÿ
pH. Consequently, it may be that pHe and/or ÿ
pH measurements in vivo will correlate with 5FU trapping. These observations on tumor pH and energy status provide a clear direction for further studies that could lead to protocols with a potentially improved therapuetic index for 5FU. For chemotherapy, it is also bene®cial to study the pharmacology of a drug in other tissues besides tumor. With respect to the FPs, this normally involves liver studies to determine the rate of drug detoxi®cation and/or liver damage.11,36,37 Measurement of the rate of drug disappearance in the liver, formation or destruction of 5FU, and the appearance of catabolites, with or without rescue schedules, have been performed. In principle, these studies may also relate to tumor cytotoxicity.13 Similar studies have been performed clinically, the ®rst in 1987 by Wolf et al.,38 and most recently to provide parameters such as the maximal velocity of metabolic conversion, using only the proportional peak areas of the signals from 5FU and its metabolites.39 The relatively high doses of 5FU used (0.6±1 g mÿ2 dÿ1 over 5 days) can also lead to a number of adverse clinical reactions including myelosuppression, diarrhea, and mucositis. The cardiotoxicity of 5FU was ®rst reported in 1975,40 and inci-

5

dences of about 8% have been reported.41 Various hypotheses were proposed to explain the cardiotoxicity of this drug, including ischemia secondary to coronary artery spasms, interaction of 5FU with the coagulation system, immunoallergic phenomena, and direct toxicity of 5FU on the myocardium.42 The precise biochemical mechanism underlying this cardiotoxicity was unclear. It has been suggested that the major catabolite of 5FU, Fbal, might be transformed into ¯uoroacetate (FAC), a highly cardiotoxic and neurotoxic poison.43 Fluorine-19 MRS demonstrated, using an isolated perfused rat liver model, that although Fbal is indeed metabolized into FAC, commercial solutions of 5FU also contain `impurities'. Degradation compounds are gradually formed in the basic medium, which is indispensable to the solubilization of 5FU. Two cardiotoxic compounds were identi®ed: ¯uoroacetaldehyde, and ¯uorosemialdehyde malonic acid, the former being metabolized into FAC, a violent cardiotoxic compound.44,45

3

OTHER FLUORINATED COMPOUNDS

Apart from the FP work described that has led directly to similar clinical studies, 19F MRS has been applied to other ¯uorinated compounds as a research tool to investigate tumor hypoxia and pH and the pharmacokinetics of drugs in development. The useful clinical activity of 5FU as a TS inhibitor encouraged the development of other TS inhibitors, such as the antifolate CB 3717. This drug had clinical anticancer activity against breast and hepatoma, but dose-limiting nephrotoxicity. Animal studies revealed renal and hepatic toxicity that was partly due to long-term retention of the drug. A ¯uorinated version of the drug, CB 3988, was synthesized to determine tissue distribution and concentration by NMR. Pharmacokinetics were studied in vivo in the mouse and rat using both surface and solenoid coils for ¯uorine imaging and spectroscopy.46 These studies demonstrated, noninvasively, both hepatic and renal clearance and, ultimately, localization of the drug to the abdominal cavity. The t1/2 values determined were similar to those measured by high-performance liquid chromatography of body ¯uids. Recently, another ¯uorinated TS inhibitor, ZD9331, has been developed and is currently in phase II/III clinical trials. Attempts to study a ¯uorinated alkylating agent, , -di¯uorochlorambucil, were limited because of binding of the drug to proteins in blood plasma and tumors and excessive toxicity.47 A di¯uorinated nucleoside analog, gemcitabine, developed in the 1980s is now showing considerable success in the clinic against a range of solid tumors. The high doses that are used (approximately 1.5 g mÿ2) have permitted 19Fimaging in rat tumors;48 19F MRS showed increased uptake in gemcitabine-sensitive murine human xenografts compared with less sensitive xenografts.49 The energy state of a tumor is critical in determining chemo- and radiosensitivity, and a major regulator of this is the blood ¯ow to the tumor. A number of methods have been applied to measure directly or indirectly tumor hypoxia, or blood supply. A number of studies have been performed using 19 F-labeled nitromidazoles, such as ¯uoromisonidazole, Ro070741,47 and SR-4554.50 These are bioreductive agents that in hypoxic conditions undergo nitroreduction to reactive intermediates such as hydroxylamines, which bind to

4 APPLICATIONS OF 19F-NMR TO ONCOLOGY macromolecules. Selective retention was seen in tumor compared with normal tissue, and increased retention was seen in mouse tumors that were known to have a greater hypoxic fraction.14,47,50 Per¯uorocarbons (PFCs) are ¯uorinated organic molecules which provide a high S/N for 19F MRS studies. The 19 F spin±lattice relaxation rates of the PFC resonances are sensitive to the oxygen tension (PO2), and this has been exploited to measure tumor oxygenation. The principle has been reviewed,4 and new methods,51 and PFC types,25,52 including one with 12 equivalent ¯uorine atoms,53 are continually developed. One group has used a PFC to demonstrate potential chemotherapeutic applications.24,54 A PFC emulsion was injected intraperitoneally, and 3 days later 19F MRS was performed prior to and immediately following administration of the radiosensitizer nicotinamide. Highly signi®cant increases in PO2 were recorded in the mouse RIF-1 tumor with the maximum effect at 60±70 min posttreatment.54 Tumor pH is another important determinant of chemotherapeutic response, as described for 5FU. Although in vivo pH is generally measured by 31P MRS, a number of speci®c ¯uorinated probes have been developed by Deutsch and colleagues to measure intracellular pH.55,56 While these studies provided useful information on cell function, there is some doubt about their applicability to studying tumors in vivo; other probes, while con®rming 31P measurements of pH, have suffered from poor solubility.57 The Fnuct formed from FPs is an unequivocal intracellular pH marker,12,15 although the S/N is generally low, and the composite nature of the peak (many Fnucs each with a slightly different pKa) precludes absolute pH measurements. Nevertheless, dynamic 5FU-induced increases in tumor pHi were recorded using FNuct in RIF-1 tumors.14 Recently, new pH probes have been described that are suf®ciently soluble and sensitive to allow studies in vivo.58,59 One, ZK 150471, is a di¯uorinated extracellular probe that has been shown to give comparable measurements of tumor pHe when compared with a 31P-pH probe.60 The other, F6POL, is a mono¯uorinated probe that has been shown to measure both pHi and pHe in perfused rat heart59 and may permit measurements in situ of tumors. Finally, the ¯uorinated 2-deoxyglucose analog 2-¯uoro-2deoxyglucose (FDG) has been used to study glucose uptake in ascites tumors.61,62 Contrary to dogma, FDG metabolism does not cease with phosphorylation to FDG 6-phosphate. By 19F MRS, it has been shown that FDG 6-phosphate may be epimerized to the mannose 6-phosphate derivative; this undergoes further conversions by ascites cells as well as in brain, heart, and muscle to a number of FDG metabolites.62 Signi®cantly, one of these metabolites, NDP-FDM, persists for 24±48 h in ascites cells at suf®ciently high concentrations to permit 19F imaging, allowing the possibility of using 19F imaging of FDG metabolites for the detection of tumors in the body.

4 SUMMARY The high sensitivity of ¯uorine, absence of a background signal, and wide chemical shift range make the ¯uorine atom ideal for noninvasive measurements of drug pharmacokinetics. In some cases the ¯uorine is already an integral part of the drug, which allows direct applications of animal models to the clinic, with the increasingly realizable potential of optimizing

drug schedules. Other ¯uorinated compounds can be used as probes of tumor glycolysis, pH, and hypoxia, which are important determinants of chemo- and radiosensitivity. 5

RELATED ARTICLES

Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; In Vivo Hepatic MRS of Humans; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Spectroscopic Studies of Animal Tumor Models. 6

REFERENCES

1. A. N. Stevens, P. G. Morris, R. A. Iles, P. W. Sheldon, and J. R. Grif®ths, Br. J. Cancer, 1984, 50, 113. 2. P. M. J. McSheehy and J. R. Grif®ths, NMR Biomed., 1989, 2, 133. 3. M. C. Malet-Martino, R. Martino, and J. P. Armand, Bull. Cancer Paris, 1990, 77, 1223. 4. M. J. W. Prior, R. J. Maxwell, and J. R. Grif®ths, in `NMR Basic Principles and Progress', ed. Springer-Verlag, Berlin, 1992, Vol. 28, Chap. 3 5. R. J. Maxwell, Cancer Surv., 1993, 17, 415. 6. M. P. N. Findlay and M. O. Leach, Anticancer Drugs, 1994, 5, 260. 7. C. Heidelberger, P. V. Danenberg, and R. G. Moran, Adv. Enzymol., 1983, 54, 57. 8. H. M. Pinedo, and G. F. J. Peters, J. Clin. Oncol., 1988, 6, 1653. 9. Y. O. Rustum, A. Harstrick, S. Cao, U. Vanhoefer, M.-B. Yin, H. Wilke, and S. Seeber, J. Clin. Oncol., 1997, 15, 389. 10. F. N. M. Naguib, M. H. el Kouni, and S. Cha, Cancer Res., 1985, 45, 5405. 11. M. J. W. Prior, R. J. Maxwell, and J. R. Grif®ths, Biochem. Pharm., 1990, 39, 857. 12. P. M. J. McSheehy, M. J. W. Prior, and J. R. Grif®ths, Br. J. Cancer, 1989, 60, 303. 13. P. E. Sijens, Y. Huang, N. J. Baldwin, and T. C. Ng, Cancer Res., 1991, 51, 1384. 14. P. M. J. McSheehy, S. P. Robinson, A. S. E. Ojugo, E. O. Aboagye, M. B. Cannell, M. O. Leach, I. R. Judson, and J. R. Grif®ths, Cancer Res., 1998, 58, 1185. 15. P. E. Sijens, N. J. Baldwin, and T. C. Ng, Magn. Reson. Med., 1991, 19, 373. 16. P. M. J. McSheehy, R. J. Maxwell, and J. R. Grif®ths, NMR Biomed., 1991, 4, 274. 17. J. A. Koutcher, R. C. Sawyer, A. B. Kornblith, R. L. Stol®, D. S. Martin, M. L. Devitt, D. Cowburn, and C. W. Young, Magn. Reson. Med., 1991, 19, 113. 18. A. El-Tahtawy and W. Wolf, Cancer Res., 1991, 51, 5806. 19. P. M. J. McSheehy, M. J. W. Prior, and J. R. Grif®ths, Br. J. Cancer, 1992, 65, 309. 20. Y. J. L. Kamm, I. M. C. M. Rietjens, J. Vervoort, A. Heerachap, G. Rosenbusch, H. Hofs, and D. J. T. Wagner, Cancer Res., 1994, 54, 4321. 21. P. M. J. McSheehy, M. T. Seymour, A. S. E. Ojugo, L. M. Rodrigues, M. O. Leach, I. R. Judson, and J. R. Grif®ths, Eur. J. Cancer, 1997, 33, 2418. 22. P. E. Sijens and T. C. Ng, Magn. Reson. Imaging, 1992, 10, 385. 23. S. P. Robinson, L. M. Rodrigues, A. S. E. Ojugo, P. M. J. McSheehy, F. A. Howe, and J. R. Grif®ths, Br. J. Cancer, 1997, 75, 1000. 24. K. G. Helmer, S. Han, and C. H. Sotak, NMR Biomed., 1998, 11, 120.

APPLICATIONS OF 19 F-NMR TO ONCOLOGY 25. S. Hunjan, R. P. Mason, A. Constantinescu, P. Peschke, E. W. Hahn, and P. P. Antich, Int. J. Radiat. Oncol. Biol. Phys., 1998, 41, 161. 26. C. A. Presant, W. Wolf, M. J. Albright, K. L. Servis, R. Ring, D. Atkinson, R. L. Ong, C. Wiseman, M. King, D. Blayney, P. Kennedy, A. El-Tahtawy, M. Singh, and J. Shani, J. Clin. Oncol., 1990, 8, 1868. 27. M. P. N. Findlay, M. O. Leach, D. Cunningham, D. J. Collins, G. S. Payne, J. Glaholm, J. L. Mansi, and V. R. McCready, Ann. Oncol., 1993, 4, 597. 28. C. A. Presant, W. Wolf, V. Waluch, C. Wiseman, P. Kennedy, D. Blayney, and R. R. Brechner, The Lancet, 1994, 343, 1184. 29. H.-P. Schlemmer, P. Bachert, W. Semmler, P. Hohenberger, P. Schlag, W. J. Lorenz, and G. Van Kaick, Magn. Reson. Imaging, 1994, 12, 497. 30. L. P. Lemaire, P. M. J. McSheehy, and J. R. Grif®ths, Cancer Chemother. Pharmacol., 1998, 42, 201. 31. G. M. Tozer, R. J. Maxwell, J. R. Grif®ths, and P. Pham, Br. J. Cancer, 1990, 62, 553. 32. J.-L. Guerquin-Kern, F. Leteurtre, A. Croisy, and J.-M. Lhoste, Cancer Res., 1991, 51, 5770. 33. J. R. Grif®ths, Br. J. Cancer, 1991, 64, 425. 34. M. Stubbs, L. M. Rodrigues, F. A. Howe, J. Wang, K. S. Jeong, R. L. Veech, and J. R. Grif®ths, Cancer Res., 1994, 54, 4011. 35. A. S. E. Ojugo, P. M. J. McSheehy, M. Stubbs, G. Alder, R. J. Maxwell, C. L. Bashford, M. O. Leach, I. R. Judson, and J. R. Grif®ths, Br. J. Cancer, 1998, 77, 873. 36. M. Harada, K. Koga, I. Miura, and H. Nishitani, Magn. Reson. Med., 1991, 22, 499. 37. Y. Kanazawa, S. Kuragi, S. Shinahara, Y. Noda, and K. Masuda, Chem. Pharm. Bull., 1994, 42, 774. 38. W. Wolf, M. J. Albright, M. S. Silver, H. Weber, U. Reichardt, and R. Sauer, Magn. Reson. Imaging, 1987, 5, 165. 39. R. E. Port, H.-P. Schlemmer, and P. Bachert, Eur. J. Clin. Pharmacol., 1994, 47, 187. 40. R. G. Dent and I. McColl, The Lancet, 1975, 1, 347. 41. W. J. Gradishar and E. E. Vokes, Ann. Oncol., 1990, 1, 409. 42. N. J. Freeman and M. E. Costanza, Cancer, 1988, 61, 36. 43. I. Matsubara, J. Kamiya, and S. Imai, Jpn. J. Pharmacol., 1980, 30, 871. 44. L. Lemaire, M. C. Malet-Martino, M. de Forni, R. Martino, and B. Lasserre, Br. J. Cancer, 1992, 66, 119. 45. L. Lemaire, M. C. Malet-Martino, M. de Forni, R. Martino, and B. Lasserre, Oncol. Rep., 1994, 1, 173. 46. D. R. Newell, R. J. Maxwell, and B. T. Golding, NMR Biomed., 1992, 5, 273. 47. P. Workman, R. J. Maxwell, and J. R. Grif®ths, NMR Biomed., 1992, 5, 270.

5

48. M. E. Belleman, G. Brix, U. Haberkorn, J. Blatter, L. Gerlach, F. Oberdorfer, and W. J. Lorenz, Proc. 3rd Ann Mtg. Int. Soc. Magn. Reson. Med., Nice, 1995, 3, 1685. 49. P. E. G. Kristjansen, B. Quistorff, M. Spang-Thomsen, and H. H. Hansen, Ann. Oncol., 1993, 4, 157. 50. E. O. Aboagye, R. J. Maxwell, A. D. Lewis, P. Workman, M. Tracey, and J. R. Grif®ths, Br. J. Cancer, 1998, 77, 65. 51. R. P. Mason, H. Shukla, and P. P. Antich, Magn. Reson. Med., 1993, 29, 296. 52. B. J. Dardzinski and C. H. Sotak, Magn. Reson. Med., 1994, 32, 88. 53. C. H. Sotak, P. S. Hees, H. N. Huang, M. H. Hung, C. G. Krespan, and S. Raynolds, Magn. Reson. Med., 1993, 29, 188. 54. P. S. Hees and C. H. Sotak, Magn. Reson. Med., 1993, 29, 303. 55. C. J. Deutsch and J. S. Taylor, Biophys. J., 1989, 55, 799. 56. M. Bental and C. J. Deutsch, Am. J. Physiol., 1994, 266, C541. 57. J. S. Beech and R. A. Iles, Magn. Reson. Med., 1991, 19, 386. 58. T. Frenzel, S. Kossler, H. Bauer, U. Niedballa, and H. J. Wienmann, Invest. Radiol., 1994, 29, S220. 59. S. Hunjan, R. P. Mason, V. D. Mehta, V. Padmakar, P. V. Kulkarni, V. Arora, and P. P. Antich, Mag. Res. Med., 1998, 39, 551. 60. A. S. E. Ojugo, P. M. J. McSheehy, D. J. O. McIntyre, C. McCoy, M. Stubbs, M. O. Leach, I. R. Judson, and J. R. Grif®ths, NMR Biomed., 1999, 12, 000. 61. M. Kojima, S. Kuribayashi, Y. Kanazawa, T. Hardahira, Y. Maehara, and H. Endo, Chem. Pharm. Bull., 1988, 36, 1194. 62. Y. Kanazawa, K. Umayahara, T. Shimmura, and T. Yamashita, NMR Biomed., 1997, 10, 35.

Biographical Sketches Paul M. J. McSheehy. b 1959. B.Sc.(Hons), 1980, Biochemistry, University of Sussex, Ph.D., 1984, Biochemistry, University of London, UK. Introduced to in vivo MRI by Professor J. R. Grif®ths in 1986. Bone endocrinologist, currently Research Fellow at St. George's Hospital Medical School UK, studying the action mechanisms, and pharmacokinetics of chemotherapeutic drugs. Laurent P. Lemaire. b 1968. B.S. 1989, Biochemistry±Biophysics, Ph.D., 1993, Pharmacology, University of Toulouse, France. Introduced to NMR by P. Cozzone in 1989. Current research speciality: metabolism of anticancer drugs by NMR. John R. Grif®ths. b 1945. M.B. B.S., 1969, London, D. Phil., 1974, Biochemistry, Oxford, Studied ESR with George Radda, 1970±74 and (with David Gadian and Richard Isles) 31P MRS of livers, 1979±80. Approx. 100 publications. Including (with E. R. Andrews et al.,) `Clinical Magnetic Resonance: Imaging and Spectroscopy', publ. Wiley, Chichester, 1990. Since 1980 research interests have been in MRS as applied to cancer, latterly also MRI of nerves.

BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy Wolfhard Semmler Institut fuÈr Diagnostikforschung (IDF) an der Freien UniversitaÈt, D-14050 Berlin, Germany

and Peter Bachert Forschungsschwerpunkt Radiologische Diagnostik und Therapie, Deutsches Krebsforschungszentrum (DKFZ), D-69120 Heidelberg, Germany

1 INTRODUCTION 1.1 Value of 31P NMR Spectroscopy for Studies of the Human Brain In vivo 31P NMR spectroscopy in humans was ®rst applied in studies of muscle diseases and soft tissue tumors. Subsequently, metabolic, ischemic, and neoplastic disorders of the human brain have been studied by means of this technique. In vivo 31P NMR spectroscopy offers a window to study highenergy phosphate and membrane phospholipid turnover in human neurometabolism. It also permits the noninvasive measurement of the intracellular pH in the tissue.

1.2 Problems of Tumor Diagnosis The major diagnostic problems in oncology are: (i) early tumor detection, (ii) tumor grading, and (iii) tumor staging. After diagnosis of a malignant tumorous disease and selection of an adequate therapy, monitoring is important for the optimization of the treatment and for the prediction of the response to treatment. The value of 31P NMR spectroscopy for solving these clinical problems is not yet established, although animal experiments could clearly demonstrate the high potential of this noninvasive technique for monitoring tumor biochemistry and the response of tumors to radiation therapy and chemotherapy.1±6

1.3 Heterogeneity of Neoplastic Tissue Tumor tissue is heterogeneous on both macroscopic and microscopic scales. Vital and hypoxic tissue as well as necrotic cell masses and hemorrhage may coexist in tumors, re¯ecting regional differences in oxygen supply, vascularization, and nutrition of the cells.1,5 Because of the poor spatial resolution,

1

31

P NMR spectroscopy can provide only an overall picture of the metabolic state of the tumor, being completely insensitive to tissue heterogeneity below the centimeter range. This is a particular problem when high-grade gliomas are examined in vivo by 31P NMR spectroscopy. Histopathological studies show heterogeneities in these tumors, but often they are also heterogeneous macroscopically as can be detected by MR imaging. 31P NMR spectra will therefore comprise different signal contributions from vital, hypoxic, and necrotic tissue within the selected voxel rather than give a metabolic ®ngerprint of this tumor entity. The problem is less serious in the case of the relatively homogeneous meningioma and pituitary adenoma. In any case, however, MR imaging is mandatory before each NMR spectroscopic examination in order to detect gross inhomogeneities within the tumor mass.

1.4

NMR Spectroscopic Techniques

Adequate performance of the method requires the acquisition of localized NMR spectra with high signal-to-noise ratio (SNR) and high frequency resolution. A variety of localization techniques has been developed for 31P NMR spectroscopy in humans. Localization means that NMR signals are obtained from a volume of interest (VOI) in the examined organ, while signal contributions of regions outside the VOI are suppressed. Single-voxel localization methods, e.g. ISIS,7 and STEAM,8 are available that employ selective rf pulses in the presence of orthogonal magnetic ®eld gradients. A frequently used localization technique for in vivo 31P NMR spectroscopy studies of the human brain is chemical shift imaging (CSI), also called spectroscopic or metabolic imaging, which employs phaseencoding gradients in two or three spatial directions to generate a 2D or 3D array of localized spectra in one measurement process.9±14 These techniques permit the selective acquisition of 31 P NMR signals from regions with volumes >30 cm3 within the human brain. Frequency resolution and SNR of 31P NMR spectra can be improved signi®cantly using 31P-{1H} double resonance. By this means multiplet splittings resulting from scalar 31P±1H spin±spin couplings are removed (proton-decoupling15) and 31P NMR signal intensities are enhanced owing to dipolar relaxation of phosphorus nuclei interacting with close protons (31P± {1H} NOE16,17). Luyten and co-workers obtained localized (ISIS with 200 cm3 voxel) proton-decoupled 31P±{1H} NMR spectra with resolved phosphomonoester (PME) and phosphodiester (PDE) resonance bands from the human brain.18 The 2D 31P CSI combined with 1H irradiation to induce NOE signal enhancements yields in vivo brain spectra of good quality from 88 voxels with 3 cm  3 cm  4 cm volume each in a measurement time of 19 min.19 In general, comparison of in vivo NMR spectra without quantitative analysis is of limited value. Quanti®cation must include the ®t of the signals in the time or frequency domain to obtain chemical shifts, line widths, and a measure of signal intensity. Determination of absolute metabolite concentrations in tissue from in vivo 31P NMR data would be of great value. However, this is a complicated problem and, therefore, ratios of integrated peak areas are often used for data analysis.

2 BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY 2 ASSESSMENT OF NEOPLASTIC BRAIN DISEASE WITH THE USE OF 31P NMR SPECTROSCOPY

PCr

2.1 Detection and Differentiation of Brain Tumors Tumor diagnosis was one of the primary issues to which P NMR spectroscopy has been directed. The central diagnostic problem is the detection, localization, and differentiation of a tumor in its early stage when the neoplastic tissue mass is still very small. Because of its poor spatial resolution, 31P NMR can hardly contribute to the solution of this problem. To our knowledge there is no case reported which clearly demonstrates the value of 31P NMR spectroscopy for early cancer detection and where this modality was indispensable for clinical decisions. On the other hand, the potential of 31P NMR spectroscopy for tumor grading19 is believed to be high, because changes of signal intensities and linewidths of the various resonances, and of the inorganic phosphate (Pi) chemical shift (a measure of intracellular pH) are observed in tumor spectra in comparison to spectra from normal brain tissue. These effects re¯ect metabolic differences of tumor and surrounding unaffected tissue as well as physiological conditions related to tumor oxygenation, perfusion, microcirculation, and angiogenic activity.1,5,20,21 Spectral differences are also expected when human brain tumors are examined at different growth stages. Animal experiments show strong changes of 31P NMR spectra during growth (Figure 1). A progressive decrease of high-energy phosphate levels is attributed to an increase of the fraction of hypoxic cells,22 while an increase of PME and Pi intensities may result from the emergence of necrotic cells.1 The frequently observed alkalinic shift of pH in tumor tissue20 is explained by necrosis.5 For tumor characterization, the presence of an NMR-visible compound unique to a speci®c type of tumor cells or tumor species would be highly important. In fact, no tumor-speci®c resonance has been found in in vivo 31P NMR spectra of all tumors studied so far. Neuroepithelial and mesodermal tumors and cerebral metastases, for example, show the same resonances as spectra obtained from healthy brain tissue.23±25 The 31P NMR spectrum of a glioblastoma, a neuroepithelial tumor, shows a reduction of PME, Pi, PDE, and PCr relative to nucleoside 5'-triphosphate (NTP) signals [Figure 2(c)]. Heiss et al. found a similar spectral pattern in their 31P NMR study of glioblastomas except for a cystic glioblastoma which produced only poor spectral SNR.23 In contrast, Arnold et al. observed enhanced PME and barely reduced PDE levels in glioblastomas23,26 (Figure 3(e)). Resonance intensities from high- and low-grade astrocytomas, which were also examined in this study, did not differ signi®cantly. In pituitary adenomas, reduced PDE and exceedingly high PME signals were found (Figure 3(b)). Segebarth et al. observed elevated PME and reduced PCr intensities in the 31P NMR spectrum of a prolactinoma27 (Figure 4(a)) in comparison to the spectrum from the unaffected brain region of the same patient (Figure 4(b)). Hwang et al. found reduced PDE/NTP and PCr/NTP signal intensity ratios and elevated pH values in high-grade gliomas and in a meningioma.28 Similar ®ndings were reported from other studies.29,30 The localized in vivo 31P NMR spectra in Figure 2 show increased PME and decreased PDE and PCr signal intensities for the meningioma (Figure

PME

31

Pi

a-NTP g-NTP

b-NTP

(a)

(b)

(c)

(d)

20

10

0 –10 d (ppm)

–20

–30

Figure 1 Metabolic changes during tumor growth re¯ected in in vivo 31 P NMR spectra, showing progressive decline of high-energy phosphate signals (NTP, nucleoside 5'-triphosphate, mainly ATP, adenosine 5'-triphosphate; PCr, phosphocreatine) and an increase of inorganic phosphate (Pi) signals. The concentration of phosphomonoesters (PME), which are attributed to precursors of major membrane phospholipids, is elevated. (a) Base line spectrum and subsequent spectra recorded at (b) day 3, (c) day 7, and (d) day 11. The pH value of the tumor tissue dropped from 7.1 for (a) to 6.6 for (c). Experimental tumor: subcutaneously implanted MOPC 104E myeloma (modi®ed from Evanochko et al.2)

2(b)),24 a mesodermal tumor that originates from meningial tissue, compared with the spectrum from healthy brain tissue (Figure 2(a)). Negendank reviewed in vivo NMR data of human tumors.31 His evaluation shows that about 80% of all examined highgrade glioma (Kernohan III±IV) showed reduced PDE and about 50% reduced PCr signal intensities. In 96% of all examined meningioma, low PDE and low PCr signals were observed. Intracellular pH values were higher in high-grade glioma and meningioma than in normal brain tissue.

3

BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

PDE

PCr a-NTP

Pi g-NTP

PME

PDE Healthy brain

PCr

Pi

b-NTP

a-NTP

PME

g-NTP

Healthy brain b-NTP

(a) (a) Meningioma Pituitary adenoma (b)

(b)

Glioblastoma Meningioma

(c)

(c)

Astrocytoma II 10

0

–10

–20

d (ppm)

Figure 2 Localized 31P NMR spectra of (a) healthy human brain tissue, (b) meningioma, and (c) glioblastoma obtained with the ISIS localization technique (modi®ed from Heindel et al.24)

(d)

Glioblastoma

Benign pituitary adenomas showed reduced PDE levels in 75% and high PME and low PCr levels in all patients examined.31 In all cases the observed pH was comparable to the value measured in normal brain tissue. Low PDE/NTP signal intensity ratios and elevated pH values were found in ependymoma. In vivo 31P NMR spectra from neuroepithelial tumors (glioma) and mesodermal tumors (meningioma) are similar, so this technique does not allow the differentiation between these two entities. On the other hand, the pH, which was found in the normal range for low-grade and elevated in high-grade glial tumors, may help to grade gliomas. However, discrepant pH values have also been reported.23,25 High pH values in posttherapy stages of soft tissue sarcomas were found to be related to necrosis.5,32 In a recent study, Rutter et al. employed 1D 31P CSI in vivo and detected statistically signi®cant differences in NMR parameters of brain hemispheres of patients with untreated brain tumors (astrocytomas, glioblastomas, meningiomas).33 PME/NTP and PDE/NTP signal intensity ratios were higher in glioblastomas and astrocytomas than in healthy brain tissue. The Pi/NTP and PCr/NTP ratios of astrocytomas were higher compared with glioblastomas and normal brain.

(e) 15

0

–15 d (ppm)

Figure 3 Localized 31P NMR spectra in vivo of (a) normal human brain tissue and different tumors of the brain: (b) pituitary adenoma, (c) meningioma, (d) grade II astrocytoma, and (e) glioblastoma. Spectra were obtained by means of the ISIS localization technique with voxel sizes of 41±220 cm3 (modi®ed from Arnold et al.26)

2.2

Therapy Monitoring of Brain Tumors by Means of NMR

31

P

Studies of experimental tumors in animals by means of in vivo 31P NMR spectroscopy showed signi®cant changes of levels of phosphorus-containing metabolites during the course

4 BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

Prolactinoma (a)

Uninfiltrated tissue

(b)

20

10

0

–10

–20

–30

clinically during therapy. Similar results were obtained in the examination of a patient with a grade II astrocytoma treated with radiation therapy of 60 Gy total dose after subtotal surgical removal of the tumor. In their 31P NMR follow-up study of tumor radiation therapy, Heindel et al. examined patients with meningiomas, glioblastomas, astrocytomas, and other cerebral tumors.24 NMR signal intensity changes similar to those in Figure 7, in particular a PME signal reduction, were observed in the spectra of grade II astrocytoma upon radiation therapy (Figure 8). Arnold et al. examined four patients with malignant gliomas after intraarterial (Figure 9) and three patients after intravenous 1,3-bis(2-chloroethyl)-1-nitrosourea (BCNU) therapy.44,45 They found statistically signi®cant differences of pH values upon evaluation of 31P NMR spectra acquired before and after chemotherapy: i.v. application of BCNU resulted in a transient acidosis and i.a. administration of the same drug in an alkalosis in the tumor tissue. These pH changes were detected before

d (ppm)

Figure 4 Localized 31P NMR spectra in vivo of (a) a prolactinoma and (b) the unaffected brain tissue in the same patient (modi®ed from Segebarth et al.27) Pi

of radiation therapy and chemotherapy.2,3 A common feature is the decrease of high-energy phosphate signals (PCr, NTP) and of pH and the increase of Pi, PME, and PDE resonances; however, the opposite effects have also been observed.2,3,6 A strong increase in the Pi resonance is often accompanied by a collapse of high-energy phosphate signals, as demonstrated in animal studies (Figure 5). This effect has also been observed in a clinical study monitoring the hyperthermic regional perfusion therapy of the recurrence of a malignant melanoma.34 The majority of studies on therapy monitoring by means of in vivo 31P NMR spectroscopy have been performed in soft tissue tumors32,34±41 (for reviews, see Steen21 and Negendank31). Only a few studies focused on monitoring treatment of brain tumors, among them monitoring of radiation therapy and/or chemotherapy,23,24,26,27,29,42 immunotherapy,43 and embolization.18 One of the ®rst reports in this ®eld was published in 1987 by Segebarth et al.27 Figure 6 shows the 31P NMR spectra obtained in a patient with invasive prolactinoma before and one and ®ve weeks after pharmacotherapy with bromocriptine [the same patient as in Figure 4(a)]. Upon therapy the patient improved clinically while no morphological changes of the tumor were found by MR imaging. After ®ve weeks, a small decrease of the NTP level was detected, but no pH change. Elevated pH values were observed in the tumor tissue before and one and ®ve weeks after therapy when compared with the pH of the unaffected brain tissue. More pronounced 31P spectral effects were found after two weeks of radiation therapy in a patient with an intracranial lymphoma. The enhanced PME and the reduced PCr resonances before treatment changed to the intensities of normal tissue after a radiation dose of 24.2 Gy (Figure 7). The pH of the tumor tissue did not change. The patient improved

(d)

251 min after TNF

(c)

167 min after TNF

(b) PME

PCr g-NTP a-NTP PDE b-NTP

83 min after TNF

(a) Control 10

0

–10

–20

d (ppm)

Figure 5 Changes of in vivo 31P NMR spectra of an experimental tumor (murine methylcholanthrene-induced [Meth-A] sarcoma) upon application of recombinant human tumor necrosis factor  (rHuTNF). The series of spectra shows a progressive decline of high-energy phosphate signals (NTP, PCr) and a strong increase in the inorganic phosphate (Pi) level (modi®ed from Shine et al.6)

BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

PDE PME Pi PCr PME

PDE g-NTP PCr a-NTP

g-NTP a-NTP b-NTP

(a) b-NTP

Pi

5

Before radiation therapy

Before pharmacotherapy

(a) After radiation therapy

(b)

1 week after therapy 20

(b)

10

0 –10 d (ppm)

–20

–30

Figure 7 Localized 31P NMR spectra in vivo of intracranial lymphoma obtained (a) before and (b) after radiation therapy with a dose of 24.2 Gy (modi®ed from Segebarth et al.27) 5 weeks after therapy (c)

20

10

0 –10 d (ppm)

–20

–30

Figure 6 Localized 31P NMR spectra of prolactinoma obtained in a patient before and one and ®ve weeks after the beginning of bromocriptine pharmacotherapy. The spectra were rescaled to take into account differences of measured voxel sizes (modi®ed from Segebarth et al.27)

any effects were seen by imaging modalities. The authors discuss the alkalosis after i.a. administration as being a result of cell membrane damage due to the high local drug concentration in this route of application. In soft tissue sarcomas, alkalosis was found to be associated with necrosis32 which may also be true for brain tumors. Superselective catheter embolization of meningiomas with poly(vinyl alcohol) particles is performed to minimize bleeding during subsequent neurosurgery. This treatment form is suitable to validate 31P NMR spectroscopy for clinical application. Since the spectra re¯ect the strong ischemia caused by the embolization, in vivo NMR spectroscopy should help to assess response to therapy. Employing twodimensional CSI, increased Pi/ -NTP signal intensity ratio was detected in meningioma the day following embolization.18 A complete depletion of the NTP pool was not observed (Figure 10).

In studies of therapy monitoring of human brain tumors by means of 31P NMR spectroscopy, different tumor entities have been examined and also different treatment protocols applied. The comparability of results of independent studies can be compromised when different measurement techniques are used. Besides further progress in experimental techniques and quanti®cation, a generally applicable examination protocol must be established. Notwithstanding these current limitations, the clinical studies presented on cerebral neoplasms demonstrate the potential of 31P NMR spectroscopy for monitoring tumor therapy response in patients.

3

CONCLUSIONS

At present, clinical 31P NMR spectroscopy of human brain tumors can be rated as follows: 1. The method allows an assessment noninvasively of bioenergetic status, levels of intermediates of phospholipid metabolism, intracellular pH, and the extent of necrosis within macroscopic regions of proliferating tissue in the brain. 2. Limitations in spatial localization, sensitivity, and spectral resolution prevent differential diagnosis of human brain tumors by means of 31P NMR spectroscopy. 3. Tumor therapy monitoring is feasible with the use of 31P NMR due to the observation of cellular energy status and membrane turnover rate, and the quantitative comparison with spectral data obtained before the beginning of therapy and from unaffected tissue.

6 BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

PCr Pi

PCr g-NTP PDE

a-NTP

PME

PDE b-NTP

Before radiotherapy

g-NTP

Pi

a-NTP b-NTP

PME

(a)

Before BCNUtherapy

(a)

22 Gy

(b)

After radiotherapy 56 Gy

After BCNUtherapy (b)

(c)

10 20

10

0

–10 d (ppm)

–20

–30

Figure 8 Follow-up study in a patient with grade II astrocytoma. Localized 31P NMR spectra obtained (a) before, (b) after 22 Gy, and (c) after completion of the radiation therapy (56 Gy) show intensity changes of phosphomonoester (PME) and phosphocreatine (PCr) resonances (modi®ed from Heindel et al.24)

5

0

–5 –10 d (ppm)

–15

–20

Figure 9 Localized 31P NMR spectra of a glioblastoma in a patient (a) before and (b) after superselective intraarterial infusion of 1,3-bis(2chloroethyl)-1-nitrosourea (BCNU) over 3 h (modi®ed from Arnold et al.45)

BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY

6

7

REFERENCES

PCr

PDE

g-NTP

Pi

a-NTP

Healthy brain

PME b-NTP

(a)

Meningioma

(b)

8

0

–8

–16

d (ppm)

Figure 10 31P spectroscopic imaging of the brain of a patient with meningioma after presurgical superselective catheter embolization. Localized spectra (voxel size 3 cm  3 cm  3 cm) from (a) unaffected hemisphere and (b) meningioma. For quantitative evaluation, Lorentzian line-®ts are superimposed on the measured spectra (modi®ed from Knopp et al.18)

4 ABBREVIATIONS CSI chemical shift imaging ISIS image-selected in vivo spectroscopy NOE nuclear Overhauser effect NTP nucleoside 5'-triphosphate PCr phosphocreatine PDE phosphodiester Pi inorganic phosphate PME phosphomonoester SNR signal-to-noise ratio STEAM stimulated echo acquisition mode VOI volume of interest

5 RELATED ARTICLES Chemical Shift Imaging; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; Proton Decoupling During In Vivo Whole Body Phosphorus MRS; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Single Voxel Whole Body Phosphorus MRS; Whole Body Studies: Impact of MRS.

1. T. C. Ng, W. T. Evanochko, R. N. Hiramoto, V. K. Ganta, M. B. Lilly, A. J. Lawson, T. H. Corbett, J. R. Durant, and J. D. Glickson, J. Magn. Reson., 1982, 49, 271. 2. W. T. Evanochko, T. C. Ng, and J. D. Glickson, Magn. Reson. Med., 1984, 1, 508. 3. S. Naruse, K. Hirakawa, Y. Horikawa, C. Tanaka, T. Higuchi, S. Ueda, H. Nishikawa, and H. Watari, Cancer Res., 1985, 45, 2429. 4. P. F. Daly, R. C. Lyon, P. J. Faustino, and J. S. Cohen, J. Biol. Chem., 1987, 262, 14875. 5. P. Vaupel, F. Kallinowski, and P. Okunieff, Cancer Res., 1989, 49, 6449. 6. N. Shine, M. A. Palladino, J. S. Patton, A. Deisseroth, G. S. Karczmar, G. B. Matson, and M. W. Weiner, Cancer Res., 1989, 49, 2123. 7. R. J. Ordidge, A. Connelly, and J. A. B. Lohman, J. Magn. Reson., 1986, 66, 283. 8. K. D. Merboldt, D. Chien, W. HaÈnicke, M. L. Gyngell, H. Bruhn, and J. Frahm, J. Magn. Reson., 1990, 89, 343. 9. P. C. Lauterbur, D. M. Kramer, W. V. House, and C.-N. Chen, J. Am. Chem. Soc., 1975, 97, 6866. 10. T. R. Brown, B. M. Kincaid, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1982, 79, 3523. 11. A. A. Maudsley, S. K. Hilal, W. H. Perman, and H. E. Simon, J. Magn. Reson., 1983, 51, 147. 12. A. A. Maudsley, S. K. Hilal, H. E. Simon, and S. Wittekoek, Radiology, 1984, 153, 745. 13. D. B. Vigneron, S. J. Nelson, J. Murphy-Boesch, D. A. Kelly, H. B. Kessler, T. R. Brown, and J. S. Taylor, Radiology, 1990, 177, 643. 14. J. W. Hugg, G. B. Matson, D. B. Twieg, A. A. Maudsley, D. Sappey-Marinier, and M. W. Weiner, Magn. Reson. Imaging, 1992, 10, 227. 15. P. R. Luyten, G. Bruntink, F. M. Sloff, J. W. A. H. Vermeulen, J. I. van der Heijden, J. A. den Hollander, and A. Heerschap, NMR Biomed., 1989, 1, 177. 16. P. Bachert-Baumann, F. Ermark, H.-J. Zabel, R. Sauter, W. Semmler, and W. J. Lorenz, Magn. Reson. Med., 1990, 15, 165. 17. P. Bachert and M. E. Bellemann, J. Magn. Reson., 1992, 100, 146. 18. M. V. Knopp, P. Bachert, G. Ende, M. Blankenhorn, H. Kolem, W. Semmler, T. Hess, M. Forsting, K. Sartor, W. J. Lorenz, and G. van Kaick, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 1954. 19. D. L. Arnold, E. A. Shoubridge, J. G. Villemure, and W. Feindel, NMR Biomed., 1990, 3, 184. 20. R. D. OberhaÈnsli, D. Hilton-Jones, P. J. Bore, L. J. Hands, R. P. Rampling, and G. K. Radda, Lancet, 1986, ii, 8. 21. R. G. Steen, Cancer Res., 1989, 49, 4075. 22. W. T. Evanochko, T. C. Ng, M. B. Lilly, A. J. Lawson, T. H. Corbett, J. R. Durant, and J. D. Glickson, Proc. Natl. Acad. Sci. USA, 1983, 80, 334. 23. D. L. Arnold, E. A. Shoubridge, J. G. Villemure, and W. Feindel, Proc. 7th Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1988, p. 333. 24. W. Heindel, J. Bunke, S. Glathe, W. Steinbrich, and L. Mollevanger, J. Comput. Assist. Tomogr., 1989, 12, 907. 25. W. D. Heiss, W. Heindel, K. Herholz, J. Rudolf, J. Bunke, J. Jeske, and G. Friedmann, J. Nucl. Med., 1990, 31, 302. 26. D. L. Arnold, J. Emrich, E. A. Shoubridge, J.-G. Villemure, and W. Feindel, J. Neurosurg., 1991, 74, 447. 27. C. M. Segebarth, D. F. BaleÂriaux, D. L. Arnold, P. R. Luyten, and J. A. den Hollander, Radiology, 1987, 165, 215. 28. Y. C. Hwang, J. Mantil, M. D. Boska, D.-R. Hwang, W. Banks, M. Jacobs, and C. Peterson, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 3610.

8 BRAIN NEOPLASMS IN HUMANS STUDIED BY PHOSPHORUS-31 NMR SPECTROSCOPY 29. B. Hubesch, D. Sappey-Marinier, K. Roth, D. J. Meyerhoff, G. B. Matson, and M. W. Weiner, Radiology, 1990, 174, 401. 30. T. A. D. Cadoux-Hudson, M. J. Blackledge, B. Rajagopalan, D. J. Taylor, and G. K. Radda, Br. J. Cancer, 1989, 60, 430. 31. W. Negendank, NMR Biomed., 1992, 5, 303. 32. D. Sostman, M. Dewhirst, C. Charles, K. Leopold, D. Moore, R. Burn, A. Tucker, J. Harrelson, and J. Oleson, Proc. 9th Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 319. 33. A. Rutter, H. Hugenholtz, J. K. Saunders, and I. C. Smith, Invest. Radiol., 1995, 30, 359. 34. W. Semmler, G. Gademann, P. Schlag, P. Bachert-Baumann, H.-J. Zabel, W. J. Lorenz, and G. van Kaick, Magn. Reson, Imaging, 1988, 6, 335. 35. J. M. Maris, A. E. Evans, A. D. McLaughlin, G. J. D'Angio, L. Bolinger, H. Manos, and B. Chance, N. Engl. J. Med., 1985, 312, 1500. 36. T. C. Ng, S. Vijayakumar, A. W. Majors, F. J. Thomas, T. F. Meaney, and N. J. Baldwin, Int. J. Radiat. Oncol. Biol. Phys., 1987, 13, 1545. 37. B. D. Ross, J. T. Helper, I. J. Cox, I. R. Young, R. Kempf, A. Makepeace, and J. Pennock, Arch. Surg., 1987, 122, 1464. 38. W. Semmler, G. Gademann, P. Bachert-Baumann, H. J. Zabel, W. J. Lorenz, and G. van Kaick, Radiology, 1988, 166, 533. 39. O. M. Redmond, J. Stack, M. Scully, P. Dervan, D. Carney, and J. T. Ennis, Proc. 7th Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1988, p. 432. 40. G. S. Karczmar, D. J. Meyerhoff, M. D. Boska, B. Hubesch, J. Poole, G. B. Matson, F. Valone, and M. W. Weiner, Radiology, 1991, 179, 149. 41. O. M. Redmond, J. P. Stack, N. G. O'Connor, D. N. Carney, P. A. Dervan, B. J. Hurson, and J. T. Ennis, Magn. Reson. Med., 1992, 25, 30.

42. S. Naruse, Y. Horikawa, C. Tanaka, T. Higuchi, H. Sekimoto, S. Ueda, and K. Hirakawa, Radiology, 1986, 160, 827. 43. B. D. Ross, J. Tropp, K. A. Derby, S. Sugiura, C. Hawryszko, D. B. Jacques, and M. Ingram, J. Comput. Assist. Tomogr., 1989, 13, 189. 44. D. L. Arnold, E. A. Shoubridge, W. Feindel, and J.-G. Villemure, Can. J. Neurol. Sci., 1987, 14, 570. 45. D. L. Arnold, E. A. Shoubridge, J. Emrich, W. Feindel, and J.-G. Villemure, Invest. Radiol., 1989, 24, 958.

Biographical Sketches Wolfhard Semmler. b 1944. M.Sc. (Dipl.-Phys.), 1972, Ph.D. (Dr. rer. nat.), 1976, Free University of Berlin, M.D. (Dr. med.), 1990, University of Heidelberg. University of Aarhus (Denmark), 1976; Rutgers University and Bell Laboratories, New Jersey, USA, 1977±79; Hahn± Meitner-Institute, Berlin, 1979±83; Department of Radiology, Free University of Berlin, 1983±85; German Cancer Research Center (DKFZ), Heidelberg, 1985±91; Director, Institute of Diagnostic Research, Free University of Berlin, 1992±present. Approx. 120 publications. Research specialties: MR imaging and spectroscopy, development of contrast media for MRI, basic principles of contrast media for imaging modalities. Peter Bachert. b 1954. M.Sc., 1979, Ph.D. (Dr. rer. nat), 1983, University of Heidelberg. Introduced to NMR by K. H. Hausser and U. Haeberlen. Application scientist, Siemens, 1984±86; German Cancer Research Center (DKFZ), Heidelberg, 1986±present. Approx. 40 publications. Research specialties: applications of NMR to problems in biophysics and biomedicine.

BRAIN NEOPLASMS STUDIED BY MRI

Brain Neoplasms Studied by MRI Andrew P. Kelly and Michael N. Brant-Zawadzki Hoag Memorial Hospital Presbyterian, Newport Beach, CA, USA

1 INTRODUCTION Magnetic resonance imaging (MRI) has become the imaging modality of choice for the evaluation of brain neoplasms. There are two main reasons why MRI has supplanted computerized axial tomography (CT) scanning at many institutions in the USA. First and foremost is the superior sensitivity that MRI possesses in detecting alterations in brain tissue caused by neoplasms. As location of an intracranial neoplasm is a factor with important diagnostic as well as prognostic implications, the ability of MRI to image in multiple planes is the second reason. This advantage helps to determine whether a lesion is intraaxial (of parenchymal origin, such as a glioma) versus extraaxial (of dural origin, for example, such as a meningioma). A short article on MRI of brain neoplasms leaves important areas uncovered. For more extensive discussion of the topic, the reader is referred to the excellent recent reviews on brain neoplasms in the books edited by Stark and Bradley1 and Atlas.2±4 Also, cross references to pertinent related subjects discussed elsewhere in this volume are given at the end of this article. The initial part of this article focuses on the general features exhibited by many brain neoplasms and why MRI is sensitive to these changes. Technical considerations behind deciding appropriate imaging pulse sequences are also discussed. The second section discusses some of the most common brain neoplasms and their characteristics, as displayed by MRI.

2 GENERAL NEOPLASM FEATURES SEEN ON MRI MRI of the brain and its pathology takes advantage of the exponential time constants, T1 and T2, exhibited by normal brain tissue, and alterations in these values caused by brain neoplasms. The reader is referred to other chapters in this volume for discussions on the complex basis of signal intensity and magnetic relaxation characteristics of normal brain and brain pathology. Essentially, calculated T1 and T2 values of brain tumors, such as astrocytomas, are longer than normal gray and white matter, but widespread differences exist even within single histological classi®cations, and attempts at histologic strati®cation of brain neoplasms by quantitative T1 and T2 relaxation analysis have proven futile.5 The presence of edema, hemorrhage, necrosis, cyst development, and even calci®cation can help characterize neoplasms in the brain. Standard MRI sequences employed by most institutions include a sagittal short TR, short TE (T1-weighted) localizing sequence, axial long TR, long TE (T2-weighted) sequences, and axial T1-weighted sequences, usually obtained after adminis-

1

tration of a paramagnetic contrast agent such as gadoliniumDTPA. Gradient recalled sequences are sometimes added to help characterize lesions in regards to foci of hemorrhage or calci®cation, as these sequences optimize magnetic susceptibility effects. Additional planes are included when deemed necessary for further information on tumor location and extension. Because T2-weighted spin echo (SE) MRI sequences have the disadvantage of long acquisition times resulting in degradation of image quality secondary to patient motion, many scans are now performed using fast spin echo (FSE) techniques. The usefulness of FSE sequences already has been demonstrated in the imaging of pathologic intracranial conditions.6,7 Utilizing these techniques, a routine entire brain scan can now be completed in under 15 min. By analyzing tumor signal intensity on different pulse sequences, some insight can be gained as to the nature of the cells making up the tumor. For example, neoplasms containing a high nuclear/cytoplasm ratio, such as lymphoma and meningioma, may display this lack of water content as low signal intensity on T2-weighted sequences. 3 3.1

SPECIFIC TUMOR FEATURES SEEN ON MRI Location, Mass Effect, In®ltration and Hydrocephalus

The multiplanar capability of MRI allows the determination of the location of a tumor, i.e., intra-axial versus extraaxial, and thus helps to identify the potential cell of origin. The multiplanar capability of MRI aids in de®ning the extent of certain tumors that can be in®ltrative, such as gliomas and primary lymphomas, which can extend via white matter tracts such as the corpus callosum. By assessing ventricular size and shape as well as gray and white matter interfaces, the presence or absence of mass effect and hydrocephalus can be determined. Recently, mass effect and necrosis, as displayed by and graded on MRI, were found to be statistically signi®cant characteristics for grading astrocytic series gliomas, as compared with biopsy ®ndings, which sometimes can be subject to sampling error.8 3.2

Edema

One of the major diagnostic advantages of MRI over CT is in the detection of brain edema, which is one of the most striking features associated with tumors. Termed `vasogenic' edema, it is probably secondary to neovascularity devoid of the usual blood±brain barrier or stretching of normal vessels secondary to mass effect.9 Edema, by increasing bulk water, causes prolongation of T1 and T2, as most tumors do; thus, changes detected on T2-weighted images actually represent tumor in addition to edema. MRI is highly sensitive to these changes, but speci®cally de®ning areas representing actual tumor versus edema is dif®cult if using T2-weighted images alone. 3.3

Necrosis and Cyst Formation

In most cases, intratumoral necrosis is considered to be a sign of a more aggressive lesion, and necrosis is well demonstrated by MRI. Necrosis can be cystic or noncystic; therefore, a varied appearance can be seen on MRI. Cystic necrosis with increased bulk water content will prolong T1 and T2 relaxation

2 BRAIN NEOPLASMS STUDIED BY MRI which, when accompanied by vasogenic edema, can cause a confusing picture on MRI. The physics behind the appearance of blood on MRI is beyond the scope of this article, but the evolution of blood breakdown products from oxyhemoglobin through paramagnetic methemoglobin to hemosiderin is followed well by MRI and affords the potential for timing the event.4 Certain tumors, especially metastases from melanoma, lung and renal cell carcinoma, have a propensity to bleed, as do a small percentage of gliomas. Hemorrhage itself should not be considered speci®c for tumor, since most cerebral bleeds have other causes. 3.5

Figure 1 A 16-year-old female with a large juvenile pilocytic astrocytoma occupying much of the right frontal lobe. (a) Before gadolinium-DTPA T1-weighted sagittal image obtained on a 1.5-T GE Signa magnet shows the large cystic tumor with a mural nodule (white arrowhead). (b) Note the dense enhancement of the mural nodule after contrast administration

times, while nonnecrotic cysts can show shortened T1 and T2 relaxation times due to hemorrhage and accumulation of proteinaceous debris, leading to high signal intensity on T1weighted images.2 Benign or malignant lesions may have cystic components. Benign lesions, such as arachnoid cysts, will exhibit the behavior of `true' cysts, following CSF signal intensity on all pulse sequences. Other benign cysts, such as colloid cysts or craniopharyngiomas, show greater variability in cyst components which determine signal changes on MRI. Malignant lesions may contain cystic components for a number of reasons, ranging from true tumoral cysts, to hemorrhage into solid lesions with subsequent clot lysis, and other causes of necrosis. Certain tumors may present as mural nodules in the cyst wall, particularly childhood astrocytomas and juvenile pilocytic astrocytomas, the latter having a very low malignant potential (Figure 1). 3.4 Hemorrhage in Brain Neoplasms Many primary brain neoplasms may initially be discovered secondary to symptoms related to intratumoral hemorrhage,

Use of Paramagnetic Contrast Agents and Enhancement

A highly regulated and consistent internal milieu must be maintained for optimal brain and spinal cord function. Specialized capillaries, with endothelial tight junctions, provide this `blood±brain barrier' in the normal state, aided by foot processes from nearby astrocytes. By administration of paramagnetic contrast agents such as gadolinium-DTPA, areas of breakdown in the blood±brain barrier caused by tumor can be detected. This is most helpful in determining to a close approximation the extent of neoplasm versus edema by comparing post-gadolinium T1-weighted images with T2-weighted images. By enhancing the relaxation of nearby water protons, contrast agents can decrease T1 values in areas where a breakdown in the blood±brain barrier has occurred. The vast majority of glioblastoma multiforme tumors will enhance in a heterogeneous, thick, irregular pattern, but it is important to note that degree of enhancement does not correlate with aggressiveness of tumor.10 Metastases also enhance in almost all instances. Contrast is most useful in detecting lesions that are isointense on T1-weighted images, show little edema on T2weighted images, but which strongly enhance. This is seen in such tumors as meningiomas and acoustic neuromas. Postoperatively, contrast agents can help detect areas of suspected tumor recurrence or residual tumor, but these changes can be nonspeci®c, as discussed, below. 3.6

Postoperative Changes and Radiation

MRI with gadolinium is indicated following surgery, radiation or chemotherapy to follow tumor size, but recurrent tumor is not all that enhances. Local enhancement secondary to leptomeningeal scarring may persist for years after surgery, and only a size increase on sequential studies in a region of enhancement is likely to be de®nitive evidence of recurrent tumor.10 Many intracranial neoplasms now undergo radiotherapy as a mainstay for treatment, and most high-grade astrocytomas are now treated by gross resection followed by high energy local radiotherapy (radiation implants or radiosurgery). Sites of radiation necrosis can enhance and show edema, simulating tumor.11 3.7

Tumoral Calci®cation

Certain neoplasms, such as oligodendrogliomas and craniopharyngiomas, have a tendency to display areas of calci®cation, but identi®cation of tumoral calci®cation in itself seldom causes one to favor a particular neoplasm over another.

BRAIN NEOPLASMS STUDIED BY MRI

3

Secondary to the low resonant proton density of calci®ed tissue, calcium may be missed on MRI; when seen, it is generally noted to appear hypo- or isointense on T1- and T2weighted spin echo sequences, and with gradient echo sequences signal loss is more profound due to the sensitivity of this method to the heterogeneous magnetic susceptibility found in calci®ed tissue. Recent articles have described occasional calci®ed brain lesion as appearing bright on T1-weighted images due to shortening of T1 relaxation times by a surface relaxation mechanism.12

4 SPECIFIC NEOPLASMS AND CHARACTERISTICS DISPLAYED ON MRI Since location as well as histologic subtype are important features that help determine the clinical presentation and prognosis of most brain neoplasms, an attempt has been made to categorize the most common neoplasms according to site of origin, with subcategorization according to cell of origin. 4.1 Intra-axial Lesions 4.1.1

Gliomas and Tumor of Glial Cell Origin

Almost 50% of primary brain tumors are gliomas, and three major tumor types are recognized, corresponding to types of glial cell: astrocytes, oligodendrocytes and ependymal cells. Since neoplastically transformed astrocytes give rise to 75± 95% of all gliomas, the discussion here will center on astrocytomas.11 The classi®cation system providing the greatest prognostic validity is a three-level system where: grade I refers to low grade benign astrocytomas, such as the juvenile pilocytic astrocytoma; grade II indicates anaplastic astrocytoma with intermediate grade of malignancy; and grade III refers to the highly malignant astrocytomas and glioblastoma multiforme. MRI features suggesting a more benign variety include absence of necrosis, well-de®ned margins, minimal edema and little mass effect. The hallmark of the glioblastoma multiforme, on the other hand, is necrosis, marked edema and mass effect, with prominent enhancement (Figure 2). Wide variability exists, however, since highly malignant and in®ltrative tumors may not show edema or demonstrate signi®cant enhancement. The reader is referred to other texts on brain neoplasms for discussions of oligodendrogliomas and ependymomas.1,2 4.1.2

Nonglial-Cell Intra-axial Tumors

Once considered rare, the frequency of primary intracranial lymphoma is increasing due to its occurrence in patients with immunode®ciencies, particularly patients with the acquired immune de®ciency syndrome (AIDS). Because of this and its fairly characteristic MRI appearance, lymphoma deserves mention. More than 50% of cases are multifocal, and can exist in supratentorial and infratentorial locations. Lymphomas can in®ltrate and cross the corpus callosum, a property shared with gliomas. Dense hypercellularity causes this tumor to appear isointense to hypointense on T2-weighted images. Most lymphomas enhance densely and homogeneously after contrast is administered, and tend to show less edema than gliomas of the

Figure 2 A 69-year-old man with a high-grade astrocytoma. (a) T2weighted axial image obtained on a 1.5-T Siemens Magneton magnet demonstrates a large necrotic tumor with extensive surrounding edema in the right temporal and parietal lobes. (b) Postgadolinium T1weighted image shows the ring-enhancing mass lesion, but the edema is not as evident on this sequence

same size, although wide variability in enhancement patterns and edema can exist (Figure 3). Medulloblastoma, a common primary intracranial neoplasm in children, is one of several tumors occurring more commonly in an infratentorial location. Other tumors commonly seen in this location include cerebellar astrocytomas, juvenile pilocytic astrocytomas, hemangioblastomas, and fourth ventricular ependymomas. Because of its usually cerebellar location, multiplanar MRI is important in diagnosis. Tending to be hypercellular like lymphoma, medulloblastoma enhances diffusely and is usually isointense to hypointense on T2-weighted images.13 An important use of MRI is in diagnosis of CSF dissemination of tumor. Termed leptomeningeal spread, this is a common pathway for this malignancy to metastasize or recur. Gadolinium-enhanced views of the brain and spinal cord are used to evaluate for CSF seeding.11 In searching for intracranial metastatic spread from extracranial primary tumors such as breast, lung, and colon carcinomas and melanoma, MRI with gadolinium is the opti-

4 BRAIN NEOPLASMS STUDIED BY MRI

Figure 3 A 29-year-old male with AIDS and primary intracranial lymphoma. (a) Note the low signal intensity (black arrowhead) on this T2-weighted axial sequence obtained on a 1.5-T GE Signa magnet. (b) Post-gadolinium T1-weighted axial image shows little enhancement in this case

mal screening test.14 Metastases incite greater edema as compared with primary tumors. The presence of multiple lesions strongly favors metastic disease over a primary tumor such as glioma, although a small percentage of gliomas are multicentric.

Figure 4 A 60-year-old male with a meningioma originating from the falx. (a) On this coronal T1-weighted precontrast image, the signal intensity of the tumor is similar to the brain parenchyma (black arrowhead). The images were obtained on a 1.5-T Siemens Magneton magnet. (b) After gadolinium administration, the borders of this densely enhancing meningioma are easily de®ned

and cystic foci. Usually little mass effect or signi®cant edema is seen, though these tumors can be quite large. A hallmark is intense enhancement after contrast administration, and a tapered extension of enhancement along the tumor base, called a `dural tail', may be present, (Figure 4).

4.2 Extra-axial Tumors

4.2.2

4.2.1

Being a relatively common site of tumor occurrence, the cerebellopontine angle (CPA) is well imaged by MRI. The prototype tumor occurring in this location is the acoustic neuroma arising from the eighth cranial nerve. On precontrast images, enlargement of the seventh and eighth cranial nerve complex is seen, and intense enhancement is demonstrated with contrast. Extension into the internal auditory canal highly suggests this tumor type.

Meningiomas

Comprising 10±20% of intracranial tumors, meningiomas are the most common extra-axial tumor, with an autopsy prevalence of 1±2%.3 Common sites of occurrence include the cerebral convexities, the falx, and the sphenoid wing. Meningiomas originate from the dural layer covering the brain parenchyma. Because most meningiomas show only a slight increase in T1 over white matter and a T2 within normal range for brain, they may appear mildly hypointense on T1-weighted images and isointense to hyperintense on T2-weighted images. Heterogeneous signal intensity, especially on T2-weighted images, may be seen secondary to vascular ¯ow voids, calci®cation,

4.2.3

Acoustic Neuroma: a Cerebellopontine Angle Tumor

Pituitary Gland Tumors

MRI has become the primary modality for diagnosis of hormone-secreting pituitary microadenomas as well as other tumors that may occur in the sella or suprasellar locations,

BRAIN NEOPLASMS STUDIED BY MRI

such as craniopharyngiomas. Coronal and sagittal pre- and post-gadolinium images are usually employed. Timing of the imaging with the administration of contrast agent is important, as enhancement of a microadenoma will be delayed compared with the normal enhancement of the rest of the gland. Larger pituitary tumors, such as macroadenomas, may displace the carotid artery or invade the cavernous sinus, both features being depicted well by MRI. 4.2.4

Other Extra-axial Tumors

Tumors may originate in the bones comprising the base of the skull, chordomas and chondrosarcoma being two such examples. A dif®cult area to image by CT, MRI demonstrates skull base masses well and may depict in®ltrative changes in the marrow-containing portions of the skull base, as can be seen with metastatic involvement of the clivus. A number of tumors may originate in the pineal gland, and multiplanar imaging is important in identifying the pineal gland as the origin. Ependymoma, a tumor which originates in the ependymal lining cells of the ventricular system, is the most common intraventricular brain neoplasm and can cause expansion of the ventricle at the site of origin.15 Extrusion through various ventricular foramina, such as the foramen of Magendie or Luschka, highly suggest this tumor type when it originates in the fourth ventricle. 5 SUMMARY MRI has proven its usefulness in the diagnosis and followup of brain neoplasms. As discussed, its major limitations are in the diagnosis of tumor recurrence after surgical excision and in distinguishing tumor recurrence from radiation necrosis. Recent developments in magnetic resonance spectroscopy (MRS) and positron emission tomography (PET) offer further help in such cases. Experimental work with proton MRS, for example, has shown that increased levels of lactate, choline and lipids may be associated with certain malignancies.16 In the future, the noninvasive assessment of brain neoplasm histology will probably combine the efforts of MRI, MRS and PET as the two latter technologies are further developed. 6 RELATED ARTICLES Brain MRS of Infants and Children; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Cranial Nerves Investigated by MRI; Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Gadolinium Chelates:

5

Chemistry, Safety, and Behavior; Hemorrhage in the Brain and Neck Observed by MRI; MRI in Clinical Medicine; Relaxation Measurements in Imaging Studies. 7

REFERENCES

1. A. N. Hasso, K. E. Kortman, and W. G. Bradley, in `Magnetic Resonance Imaging', 2nd edn., eds. D. D. Stark and W. G. Bradley, Mosby Year Book, St. Louis, 1992, Chap. 25. 2. S. W. Atlas, in `Magnetic Resonance Imaging of the Brain and Spine', ed. S. W. Atlas, Raven, New York, 1991, Chap. 10. 3. H. I. Goldberg, in `Magnetic Resonance Imaging of the Brain and Spine', ed. S. W. Atlas, Raven, New York, 1991, Chap. 11. 4. K. R. Thulborn and S. W. Atlas, in `Magnetic Resonance Imaging of the Brain and Spine', ed. S. W. Atlas, Raven, New York, 1991, Chap. 9. 5. M. Just and M. Thelen, Radiology, 1988, 169, 779. 6. S. W. Atlas, D. B. Hackney, D. M. Yousem, and J. Listerud, Radiology, 1991, 181, 165. 7. G. H. Zoarski, J. K. Maskey, Y. Anzai, W. N. Hanafee, P. S. Melki, R. V. Mulkern, F. A. Jolesz and R. B. Lufkin, Radiology, 1993 188, 323. 8. B. L. Dean, B. P. Drayer, C. R. Bird, R. A. Flom, et al., Radiology, 1990, 174, 411. 9. W. M. Kelly and M. Brant-Zawadzki, in `Radiology, Diagnosis, Imaging, Intervention', eds. J. M. Taveras and J. T. Ferrucci, J. B. Lippincott, Philadelphia, 1989, Chap. 53. 10. W. G. Bradley, Jr., W. T. C. Yuh, and G. M. Bydder, J. Magn. Reson. Imaging, 1993, 3, 199. 11. R. B. Schwartz and M. T. Mantello, Semin. Ultrasound CT MR, 1992, 13, 449. 12. R. M. Henkelman, J. F. Watts, and W. Kucharczyk, Radiology, 1991, 179, 199. 13. S. P. Meyers, S. S. Kemp, and R. W. Tarr, Am. J. Roentgenol., 1992, 158, 859. 14. J. H. Bisese, Semin. Ultrasound CT MR, 1992, 13, 473. 15. J. Jelinek, J. G. Smirniotopoulos, J. E. Parisi, and M. Kanzer, Am. J. Roentgenol., 1990, 155, 365. 16. P. R. Luyten, J. H. Marien, and W. Heindel, P. M. van Gerwen, K. Herholz, J. A. den Hollander, G. Friedmann, and W. D. Heiss, Radiology, 1990, 176, 791.

Biographical Sketches Andrew P. Kelly. b. 1960. B.A. Chemistry, 1983, California State University, Fullerton; M.D., 1988, University of California, Davis. Fellow in MRI, Hoag Memorial Hospital Presbyterian, Newport Beach, CA, 1993±present. Michael Brant-Zawadzki. Approx. 150 articles, including the textbook `Magnetic Resonance Imaging of the Central Nervous System'. A frequent lecturer on MR imaging applications and contrast agents.

BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY

Body Fat Metabolism: Observation by MR Imaging and Spectroscopy E. Louise Thomas and Jimmy D. Bell Imperial College School of Medicine, Hammersmith Hospital, London, UK

1 INTRODUCTION The importance of lipids in both health and disease is increasingly recognized. The major causes of morbidity and mortality in the western world include cancer, coronary heart disease (CHD), diabetes mellitus, and obesity. Lipids appear to have a major role in both treatment and prevention.1,2 There is, therefore, a growing need for in vivo methods for studying human lipid metabolism and investigating lipid depots in the body. The development and use of nuclear magnetic resonance (NMR) techniques of imaging (MRI) and spectroscopy (MRS) for lipid studies is a major recent advance. Since the late 1980s, interest has increased in the use of both in vivo and in vitro NMR for investigation of human and animal metabolism. Researchers are taking advantage of the nondestructive and noninvasive nature of the technique. These characteristics are particularly important for widespread, population-based studies, which often require serial examinations. The initial applications of in vivo NMR to the study of lipids were principally concerned with development of the NMR methodology, rather than focusing on speci®c biochemical problems.3,4 However, even at this early stage, the potential of MRI and MRS when applied to lipid metabolism could be envisaged. In this chapter, we will review the recent use of these techniques both in vivo and in vitro for investigation of human lipid metabolism and body fat composition and deposition.

2 USEFUL NUCLEI FOR NMR STUDIES OF LIPIDS MRS with 1H, 2H, 13C, and 31P nuclei has been applied to the study of lipid metabolism in animals and humans.3±9 The 1 H nucleus has the highest natural abundance in biological tissues and is the highest sensitivity in NMR compared with other useful nuclei. In the past, the application of in vivo 1H MRS to the study of lipids was not considered to be a useful technique. Problems arising from the intrinsically small chemical shift range resulted in severe signal overlap, which had limited its use. Recently, however, 1H NMR has been shown to be an extremely powerful noninvasive method to determine intramuscular lipid content.7,8 MRS using 13C in vivo has been successfully applied to studies of lipid metabolism and adipose tissue composition.3,4,9 Compared with studies using 1H, 13C MRS is relatively insensitive because of the low gyromagnetic ratio and low natural

1

abundance (1.1%) of the 13C nucleus. Its relative sensitivity is further reduced by the 1H±13C coupling, which splits the 13C signal, effectively reducing intensity. These problems are partially offset by the short relaxation times of most 13C resonances, the high adipose tissue content of 13C-containing compounds, and the use of decoupling techniques. Moreover, the very fact that 13C has 1.1% natural abundance has been exploited by using 13C-enriched metabolites in turnover studies, similar to classical 14C-tracer studies. MRS with 2H has only had limited use in in vivo metabolic studies. The deuterium nucleus is quadrupolar (I = 1). It has both a low gyromagnetic ratio and natural abundance (0.015%), which leads to a very low sensitivity relative to proton spectroscopy (1.4510ÿ6). These unfavorable properties are offset by its short relaxation times and high body content (12 mmol/l), allowing its detection by natural abundance in vivo MRS. Furthermore, the tissue content can be readily increased by use of deuterated water (D2O), a fact that Brereton et al. have utilized to great effect in lipid turnover studies.5 Limited information can be obtained from in vivo 31P MRS of lipids because 31P NMR detects only phosphate-containing compounds. Consequently, 31P MRS has principally been applied to lipid studies in vitro.10±12 One area of research where in vivo 31P MRS has been applied is to the study of membrane phospholipids. However, as the signals from the bound phospholipids tend to be very broad, the useful resonances come principally from phospholipid precursors (phosphocholine and phosphoethanolamine) or breakdown (glycerophosphocholine and glycerophosphoethanolamine) products and, therefore, will not be discussed further in this chapter.

3

MRI OF BODY FAT CONTENT AND DISTRIBUTION

The accurate determination of total body fat content and measurement of regional fat depots has become an important issue as the contribution of body fat to diseases such as noninsulin-dependent diabetes mellitus and CHD has become clearer. There are many techniques that have long been used to give estimates of total and peripheral body fat content with varying degrees of accuracy. However, until the development of techniques such as X-ray computed tomography (CT) and MRI, it was not possible to differentiate between subcutaneous and internal fat depots (Figure 1). Internal fat depots, in particular visceral fat, may be a key factor in disease development. CT gives an accurate direct measurement of visceral fat depots, though exposure to ionizing radiation makes whole body fat measurements, especially for serial studies, impractical; consequently only single slices tend to be acquired. Initial studies using MRI to study body fat have focused on the validation of the technique. MRI has been validated in phantoms, animals, and human cadavers and has been shown to measure both muscle and adipose tissue in vivo accurately, showing good agreement with values produced by dissection and chemical analysis.13±17 MRI has also been compared with other techniques such as underwater weighing, anthropometry, body water dilution, impedance, and dual-energy X-ray absorptiometry (DEXA).18±24 Generally there is good degree of

2 BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY (a)

CT scan obtained at the level of the umbilicus contains a substantial amount of retroperitoneal fat, which is less metabolically active than other visceral fat depots. It has been suggested that changes occurring in the entire visceral fat depot may be `diluted out' by the presence of the less-active retroperitoneal fat in the single slice. Factors such as this can have a profound effect on the ®nal results and their interpretation. It is, therefore, more appropriate to collect suf®cient data from the entire depot to be studied. One approach is to obtain multislice data from the whole body, as shown in Figure 2. The main applications of MRI to the measurement of body fat depots in human subjects have been following interventions such as diet and exercise.25±27 Ross et al., using whole body MRI, have shown signi®cant reductions in total and regional body fat content in obese subjects following a combination of diet and exercise (resistance or aerobic) and diet alone.25,26 They found similar changes in body composition in response

(b)

Figure 1 Typical abdominal transverse MR images showing distribution of internal and subcutaneous fat in a lean (a) and an obese (b) volunteer. Images were acquired using a rapid T1-weighted spin echo sequence

correlation between the different methods, though agreement between methods for individuals can be quite variable.24 A wide variety of methodologies have been used in the application of MRI to the study of body fat. Different parameters and methods of data collection ranging from extrapolation of single-slice or multiple-slice acquisitions over selected regions of the body to whole body fat measurements have been used. Many studies have been published with single-slice MR data from the abdomen, generally at the level of L4/L5; however, there are signi®cant drawbacks with this approach. For an accurate measurement of body fat content, it is important that suf®cient data be collected from the whole region of interest so that subtle changes are not missed or overinterpreted. A change or lack of change reported using single-slice CT or MRI scans from a selected region of the abdomen might not re¯ect the effect of the intervention on the entire adipose tissue depot. Indeed, it has previously been shown that a single

Figure 2 Transverse MR images showing distribution of internal and subcutaneous fat from a whole body MRI data set from a healthy female volunteer age 21 years, basal metabolic index 27.9 kg/mÿ2, waist to hips girth ratio 0.81, subcutaneous fat 29.6 l, and visceral fat 2.25 l. Images were acquired using a rapid T1-weighted spin echo sequence from the volunteer's ®ngertips to her toes by acquiring 10 mm thick transverse images with 30 mm gaps between the slices

3

BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY

to diet combined with resistance exercise and diet combined with aerobic exercise.26 Furthermore, signi®cantly more subcutaneous fat was lost from the abdomen compared with the lower body and there was a greater loss of visceral fat from the upper than from the lower abdomen. The combination of diet and exercise resulted in a greater fat loss than occurred with diet alone.25 Thomas et al., also using whole body MRI, have suggested that there is a preferential loss of visceral fat in lean women following moderate aerobic exercise without dietary restriction.27 Interestingly, the change in body composition was only detected using MRI; weight and body fat content measured by impedance and anthropometry were not signi®cantly different following exercise. A preferential loss of visceral fat has also been reported in obese individuals following dietary restriction and treatment with dexfen¯uramine.28±31 However, these studies evaluated regional body fat distribution by measuring the area on a single MRI scan, which, as discussed above, could give misleading results.

(a)

3

(b)

2 Chemical shift (ppm)

1

0

ET(-CH2-)n

4 IN VIVO MRS 4.1 Proton MRS The application of in vivo proton MRS to the study of lipids has until recently been rather limited because of the small chemical shift range of 1H resonances and the intense water signal. However, a number of researchers have shown that in vivo 1H MRS can be used to determine noninvasively the triglyceride content within muscle cells known as intramyocellular lipids (IMCL). This is particularly important as there is evidence to suggest that these `muscle triglycerides' may be implicated in the pathogenesis of insulin resistance. Schick et al., using 1H MRS to detect signals from human skeletal muscle, reported that the methyl and methylene lipid signals each consisted of two well-resolved peaks (Figure 3).7 They suggested that the two pairs of peaks corresponded to IMCL and triacylglycerols in adipocytes between muscle ®bers (extramyocellular lipids, EMCL). Evidence in support of this interpretation has come from other groups.32,33 Boesch et al. compared measurement of IMCL by in vivo 1H MRS with morphometry and chemical analysis of human biopsy samples and suggested that measurement of IMCL by 1H MRS had the best correlation with the estimation of the `true' level of IMCL.32 Szczepaniak et al., in a very elegant study of subjects with congenital lipodystrophy, a condition associated with almost complete absence of EMCL, showed that the in vivo 1H NMR spectra revealed only single methylene and methyl resonances, corresponding to IMCL.33 Using 1H MRS, Rico-Sanz et al. have shown diversity in the level of IMCL (as well as other muscle metabolites) in human skeletal muscle.34 Levels of IMCL were signi®cantly lower in the tibialis muscle than in the soleus and gastrocnemius muscles, possibly resulting from differences in ®ber type composition and deposition of metabolites owing to adaptation of the muscles for locomotion. To date, most studies have concentrated on looking at the effects of exercise on IMCL. Boesch et al. reported a signi®cant decrease (about 40%) in IMCL in tibialis anterior muscle

IT(-CH2-)n

CHo

TCr

ET(-CH3) IT(-CH3)

Figure 3 In vivo 1H MR spectrum from the soleus muscle of a healthy volunteer before (a) and after (b) line ®tting. Cho, choline and carnitine; TCr, creatine and phosphocreatine; ET(±CH2±)n, extracellular muscle triglycerides methylene; IT(±CH2±)n, intracellular muscle triglycerides methylene; ET(±CH3)n, extracellular muscle triglycerides methyl; IT(±CH3)n intracellular muscle triglycerides methyl

following 3 h of intensive cycling.8 However, it appears that the nature and intensity of the exercise may also be important, as Rico-Sanz et al. showed no changes in IMCL levels in the tibialis, gastrocnemius, or soleus muscles following two different 90 min moderate exercise protocols.35 Interestingly, there have been several reports in abstract form demonstrating that IMCL content assessed by 1H MRS is elevated in insulin-resistant individuals.36,37 However, this relationship was not found in subjects from all ethnic groups.37 NMR images of the leg also tend to be acquired during the spectroscopy examination. These provide additional information from which it is possible to measure subcutaneous fat, bone marrow, bone, and levels of EMCL. The role of EMCL in muscle metabolism is not fully understood, but it is thought that EMCL and IMCL may have different roles. The combination of MRI and 1H MRS will be excellent tools for increasing our knowledge of the metabolism of these two lipid depots.

4 BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY 4.2 Carbon-13 MRS High-resolution 13C MRS has found widespread use in the study of lipids in vitro.38,39 In particular, the clear distinction of multiple, different fatty acid groups allows this technique to be used, often quantitatively, in studies of dietary oils.40±43 This degree of resolution cannot be achieved in vivo, which limits the utility of this technique in noninvasive human research. However, major lipid groups can be distinguished and, within these constraints, important clinical and biochemical work has been carried out using proton-coupled 13C and proton-decoupled {1H} 13C MRS. A typical 13C NMR spectrum of human adipose tissue is shown in Figure 4. Canioni et al. ®rst showed that in vivo 13C {1H} MRS could detect differences in the lipid composition of adipose tissue and liver in rats fed a diet high in polyunsaturated fatty acids.3 Further work by Sillerud et al. examined 13C{1H} NMR of triacylglycerols in rat adipocytes in vitro and advanced our knowledge particularly of signal assignment in biological systems.44 Moonen et al. developed in vivo 13C{1H} NMR to characterize adipose tissue in human subjects.4 They con®rmed that linoleic acid (C18 : 2n±6), usually the principal polyunsaturated fatty acid present in human adipose tissue, dominated the polyunsaturated fatty acid carbon signal observed in vivo, allowing estimation of this stored essential fatty acid. In vivo 13C MRS has been used to study the fatty acid composition of adipose tissue in rats fed diets based on signi®cantly different fatty acid mixtures.3,45,46 In animals fed fats with different fatty acid content (butter/lard, olive oil, sun-

8

2

1

9 3

200

150

4

100

67

5

50

10 11 12

0 ppm

Figure 4 Natural abundance in vivo 13C{1H} NMR spectrum of human adipose tissue, dominated by signal from triglycerides. Peak assignment (referenced to ±CH3): 1. C=O (171.79 ppm); 2. ±CH=CH± (monounsaturated) and ±CH=CH±CH2±CH=CH± (polyunsaturated) (129.83 ppm); 3. ±CH=CH±CH2±CH=CH± (polyunsaturated) (128.13 ppm); 4. C2 glycerol (69.21 ppm); 5. C1, C3 glycerol (62.05 ppm); 6. ±CH2CH2±CO±O±R (33.93 ppm); 7. ±CH2±CH2±CH3 (32.14 ppm); 8. ±(CH2)n± (29.69 ppm); 9. ±CH2±CH=CH± (27.39 ppm); 10. ±CH2± CH2±CO±O±R (25.04 ppm); 11. ±CH2±CH2±CH3 (22.94 ppm); 12. ± CH2±CH3 (14.1 ppm)

¯ower oil, ®sh oil), clear differences were found in the in vivo spectra, consistent with a signi®cant effect of dietary fatty acid intake on the tissue fatty acid pro®le. Progressively increasing total unsaturated and polyunsaturated fatty acid content in the diet was re¯ected in the adipose tissue composition. However, data from both groups appeared to show discrepancy in the results obtained by 13C MRS compared with GLC, which worsened in adipose tissue from animals on a diet high in complex polyunsaturated fat, a ®nding later con®rmed in human subjects.47 In humans, the turnover of fatty acids in adipose tissue is slow (half-life 600 days).48 The fatty acid pro®le of human subcutaneous fat, therefore, provides an index of the habitual dietary fatty acid intake over the previous 2 to 3 years.49 This information is important for long-term epidemiological surveys but has previously been limited by the need for repeat tissue biopsies for fatty acid estimation. The application of in vivo 13 C MRS to the noninvasive analysis of adipose tissue composition is, therefore, of particular value for human nutritional studies. Beckmann et al. have used in vivo 13C MRS to show a signi®cant change in human adipose tissue fatty acid composition following a fat-reduced diet, with a correlation between diet and tissue monounsaturated fatty acids.9 No difference was found in the degree of polyunsaturation, and this may re¯ect the limited period (6 months) of dietary change. Thomas et al. studied vegan subjects as a de®ned population, with an established long-term diet high in polyunsaturated fatty acids.50 A signi®cantly increased adipose tissue total unsaturated and mono- and polyunsaturated fatty acid carbon content was shown in the 13C spectra from the vegan group compared with omnivore controls (Figure 5). Interestingly they found no signi®cant differences in adipose tissue composition between vegetarian subjects and the omnivore controls. MRS with 13C has also been used to investigate the in¯uence of maternal diet on infant adipose tissue composition. Thomas et al. compared the adipose tissue composition of breast-fed infants of women who maintained either an omnivore or a vegan diet.51 The adipose tissue composition of infants directly re¯ects that of their mothers, with the vegan infants having 70% more polyunsaturated fatty acid carbons than the omnivore infants. Although the consequences of essential fatty acid de®ciency in formula-fed infants are well documented, less is known about the effects of very high levels of long-chain polyunsaturated fatty acids in the infant diet. The long-term effects for vegan infants having such high levels of polyunsaturated fatty acids in their diet and adipose tissue are unknown. In the same study, Thomas et al. studied the adipose tissue composition of term and pre-term infants at birth, 6 weeks, and 6 months and compared them with their mothers. They found an increase in unsaturated fatty acids with increasing gestational age and maturity (Figure 6). The complex interactions between lipids in plasma and those in the body tissues are not fully de®ned and are becoming an area for NMR-based, dynamic studies of nutrition and lipid metabolism. In vivo 13C MRS has also been used in combination with GLC to study the adipose tissue composition in malnourished patients with cirrhosis of the liver.52 No signi®cant differences were found in overall adipose tissue composition in the patients compared with healthy volunteers by either technique, although

BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY

Monounsaturated + Polyunsaturated

Omnivore

Polyunsaturated

180

150

120

180

150 ppm

120

Vegan

Figure 5 Alkene region from natural abundance in vivo coupled 13C NMR spectra of human thigh adipose tissue in vegan and omnivore subjects. Spectra are scaled to the carboxyl signal. The vegan subjects, compared with omnivores, have an increased adipose tissue content of both total unsaturated and polyunsaturated fatty acid carbons (polyunsaturated: 3.45‹0.87% versus 2.39‹0.55% of total fatty acid carbons, p < 0.05). This re¯ects at least 3 years' adherence to the vegan diet, which, in the absence of all animal products, is high in polyunsaturated fats (polyunsaturates 33.13‹7.4% versus 25.2‹8.5% total dietary fat polyunsaturated, p < 0.05)

GLC did reveal signi®cant differences in individual fatty acids. Following liver transplantation and subsequent recovery, the patients were re-scanned by 13C MRS. The resulting increase in body fat mass was accompanied by a preferential increase in saturated fatty acids.52 This may be a dietary effect, as high levels of saturated fatty acids in patients' diets following liver transplantation would result in the deposition of saturated fatty acids in the adipose tissue. Alternatively, this may be secondary to a general repletion of membrane polyunsaturated fatty

acids or the use of essential fatty acids (polyunsaturated) for biosynthesis of eicosanoids in the postoperative period. Dimand et al. have shown that the composition of human adipose tissue may also be a useful marker of fatty acid status in diseases such as cystic ®brosis.53 They found levels of polyunsaturated fatty acids to be reduced and monounsaturated fatty acids elevated in patients compared with healthy controls. Several studies have investigated the in¯uence of exercise on adipose tissue composition.27,54 Intensive exercise was shown to have a signi®cant independent effect on adipose tissue composition, with a signi®cant decrease in polyunsaturated fatty acids following 10 weeks of basic military training.54 Interestingly, no changes were found in adipose tissue composition following more moderate exercise.27 The use of in vivo 13C MRS in humans and animals has not solely been applied to adipose tissue. Barnard et al. used in vivo 13C{1H} NMR spectroscopy to examine the hepatic fatty acid pro®le in rats fed fats with different fatty acid patterns.55 The liver has a pivotal role in lipid metabolism. Hepatic receptor-mediated low-density lipoprotein (LDL) clearance is a major process controlling plasma LDL concentrations. Dietary saturated fat may increase plasma LDL by suppression of hepatic LDL-receptor activity, whereas dietary substitution with polyunsaturated fatty acids may increase hepatic LDL clearance and lower plasma LDL, possibly by increasing membrane ¯uidity.56 A signi®cant dietary effect was shown with an increasing hepatic content of unsaturated fatty acids and an increasing degree of polyunsaturation as these fatty acids increased in the diet. Investigation of the metabolism of speci®c fatty acids is largely unexplored and a possible future application of 13C MRS. The use of 13C-labeled fatty acids to trace metabolic pathways and kinetic rates is an exciting prospect. These studies may be restricted by the limited availability of labeled fatty acids, their oxidation after administration, and the sensitivity of the system, given the pre-existing strong lipid signals. However, preliminary work has been performed by Cunnane et al. and this demonstrated the ability of in vitro 13C{1H} NMR to detect carbon-speci®c incorporation of injected [U-13C]-eicosapentaenoic acid in extracted rat liver lipids.57

9 Unsaturated fatty acid carbons (%)

Carboxyl

5

8 7 6 5 4 3 2 1 0 Preterm (n = 6)

Full-term at birth (n = 21)

Six-week-old infants (n = 10)

Figure 6 Fatty acid composition of adipose tissue from preterm infants and full-term infants at birth and after 6 weeks of development

6 BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY (a)

4.3 Deuterium Spectroscopy There is considerable interest in monitoring lipid turnover in both health and disease. A novel approach to the study of dynamic in vivo tissue lipid metabolism was developed by Brereton et al.5 They investigated the use of in vivo 2H MRS in mice. Administration of D2O (10% v/v) in the drinking water for 3±4 days allowed the detection of deuterium-enriched tissue lipid resonances in spectra acquired in less than 2 min, as shown in Figure 7a. The spectra consist of resonances from deuterated water (HOD) and the CHD-group of the tissue lipids. Removal of D2O from the drinking water led to clear changes in the intensity of both signals (Figure 7b). The loss of 2H from the water signal was signi®cantly faster than from tissue lipids. The loss of 2H from the lipid resonance was, therefore, proposed as a noninvasive measure of the rate of fat utilization. This method has since been applied to the study of fat utilization in obesity. Obesity is one of the most common medical disorders and is characterized by excess adipose tissue and elevated plasma nonesteri®ed fatty acids. Furthermore, obesity is known to lead to glucose intolerance and insulin resistance. Fat turnover, as re¯ected by the levels of plasma nonesteri®ed fatty acids, may be implicated in this process. However, the lack of in vivo techniques for measuring fat utilization has greatly hampered progress in this area. Body fat turnover in mouse models of obesity and diabetes mellitus have been studied in vivo using the 2H NMR technique.58 Brereton et al. showed that the rates of fat utilization in obese mice were signi®cantly lower than the rates for nonobese mice; the induction of diabetes did not affect utilization of fat as a metabolic fuel.58 These studies clearly suggest that 2H MRS will provide researchers with a powerful method for noninvasive assessment of fat turnover rates.

12

–12 ppm

(b)

5 IN VITRO MRS Plasma concentrations of LDL and high-density lipoproteins (HDL) are well-established markers for CHD risk assessment. Standard methods for quanti®cation and compositional analysis of plasma lipoproteins require laborious and time-consuming physical separation based on particle size, density, or apolipoprotein content. Bell et al. have shown that MRS can be readily applied to distinguish lipoprotein fractions as well as to study alterations in plasma lipoproteins associated with dietary manipulation, both in intact plasma and in the isolated lipoprotein classes.11,12 For example, sophisticated computer deconvolution methods (line-®tting analysis) are being developed to identify and quantify lipoprotein fractions in the 1H NMR spectra of intact plasma samples (500 l) within minutes.59,60 The effects of dietary manipulation on the 1H NMR pro®le of lipoproteins are illustrated in Figure 8. Changes in composition following ®sh oil supplementation are shown by the presence of a new lipid resonance (peak 2A) arising from ®sh oil n-3/!-3 fatty acids incorporated into plasma lipoproteins, while the methylene (±(CH2)n±) signal (peak 3) is markedly reduced. Furthermore, T2 analysis of the lipid resonances showed that these changes in lipid composition of the LDL

0

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

Figure 7 In vivo 2H MRS in mice. (a) Natural abundance spectrum of a mouse. The resonance linked to the CHD-group of lipids is to high ®eld at 1 ppm. Chemical shifts were reference to the deuterated water (HOD) resonance assigned to 4.8 ppm. (b) Spectra of a mouse following the removal of 10% (v/v) D2O from its drinking water. The spectra were recorded at 1, 4, 5, 8, 11, 15, 16 and 21 days after the resumption of normal drinking water, indicated by (a) to (h), respectively. Spectrum (i) was recorded prior to the administration of D2O to drinking water. (Adapted with permission of Heyden & Son Ltd from I. M. Brereton, D. M. Doddrell, S. M. Oakenfull, D. Moss, and M. G. Irving, NMR Biomed., 1989, 2, 55.)

particles are accompanied by alteration in the structural characteristics of these particles.11 6

SUMMARY

In this chapter, we have reviewed the current state of in vivo MRS as applied to the study of lipids. It is clear that this is a relatively new area of NMR research. However, it offers

BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY (a)

(b)

3

2 1

1.3 1.2 1.1 1.0 0.9 0.8 ppm

3

2A 1

1.3 1.2 1.1 1.0 0.9 0.8 ppm

Figure 8 Expansions of spin echo 1H NMR spectra of intact human plasma (a) before and (b) after 7 days of ®sh oil supplementation. Peaks 1, 2: terminal ±CH3 group from LDL/HDL and very-low-density lipoprotein (VLDL), respectively; peak 3: ±(CH2)n± of VLDL. Peak 2A arises from n-3/!-3 fatty acids incorporated into plasma lipoproteins

many exciting prospects for future research into human lipid metabolism and body composition. 7 RELATED ARTICLES Brain MRS of Human Subjects; Imaging and Spectroscopy of Muscle; Whole Body Studies: Impact of MRS. 8 REFERENCES 1. J. E. Kinsella, B. Lokesh, and R. A. Stone, Am. J. Clin. Nutr., 1990, 52, 1. 2. R. A. DeFronzo and E. Ferrannini, Diabetes Care, 1991, 14, 173. 3. P. Canioni, J. R. Alger, and R. G. Shulman, Biochemistry, 1983, 22, 4974. 4. C. T. W. Moonen, R. J. Dimand, and K. L. Cox, Magn. Reson. Med., 1988, 6, 140. 5. I. M. Brereton, M. G. Irving. J. Field, and D. M. Doddrell, Biochem. Biophys. Res. Commun., 1986, 137, 579. 6. P. C. Dagnelie, J. D. Bell, S. C. R. Williams, I. J. Cox, D. K. Menon, J. Sargentoni, and G. A. Coutts, NMR Biomed., 1993, 6, 2. 7. F. Schick, B. Eismann, W. I. Jung, H. Bongers, M. Bunse, and O. Lutz, Magn. Reson. Med., 1993, 29, 158. 8. C. Boesch, J. Slotboom, H. Hoppeler, and R. Kreis, Magn. Reson. Med., 1997, 37, 484. 9. N. Beckmann, J. J. Brocard, U. Keller, and J. Seelig, Magn. Reson. Med., 1992, 27, 97. 10. J. D. Bell, O. Lavender, V. C. Morris, and P. J. Sadler, Magn. Reson. Med., 1991, 17, 414. 11. J. D. Bell, J. C. C. Brown, R. E. Norman, and P. J. Sadler, NMR Biomed., 1988, 1, 90. 12. J. D. Bell, M. L. Barnard, H. G. Parkes, E. L. Thomas, C. H. Brennan, S. C. Cunnane, and P. C. Dagnelie, J. Lipid Res., 1996, 37, 1664. 13. R. Ross, L. LeÂger, R. Guardo, J. de Guise, and B. G. Pike, J. Appl. Physiol., 1991, 70, 2164. 14. P. A. Fowler, M. F. Fuller, C. A. Glasbey, G. G. Cameron, and M. A. Foster, Am. J. Clin. Nutr., 1992, 56, 7.

7

15. N. Abate, A. Garg, R. Coleman, S. M. Grundy, and R. M. Peshock, Am. J. Clin. Nutr., 1997, 65, 403. 16. M. L. Barnard, J. V. Hajnal, J. E. Schwieso, E. L. Thomas, J. D. Bell, N. Saeed, G. Frost, and S. R. Bloom, NMR Biomed., 1996, 9, 156. 17. N. Mitsiopoulos, R. N. Baumgartner, S. B. Heyms®eld, W. Lyons, D. Gallagher, and R. Ross, J. Appl. Physiol., 1998, 85, 115. 18. P. Tothill, T. S. Han, A. Avenell, G. McNeill, and D. M. Reid, Eur. J. Clin. Nutr., 1996, 50, 747. 19. A. SohlstroÈm, L. O. Wahlund, and E. Forsum, Am. J. Clin. Nutr., 1993, 58, 830. 20. M. A. Staten, W. G. Totty, and W. M. Kohrt, Invest. Radiol., 1989, 24, 345. 21. R. Ross, L. LeÂger, D. Morris, J. de Guise, and R. Guardo, J. Appl. Physiol., 1992, 72, 787. 22. P. A. Fowler, M. F. Fuller, C. A. Glasbey, M. A. Foster, G. G. Cameron, G. McNeill, and R. J. Maughan, Am. J. Clin. Nutr., 1991, 54, 18. 23. R. Ross, D. S. Kimberley, Y. Martel, J. de Guise, and L. Avruch, Am. J. Clin. Nutr., 1993, 57, 470. 24. E. L. Thomas, N. Saeed, J. V. Hajnal, A. E. Brynes, A.P. Goldstone, G. Frost, and J. D. Bell, J. Appl. Physiol., 1998, 85, 1778. 25. R. Ross, H. Pedwell, and J. Rissanen, Am. J. Clin. Nutr., 1995, 61, 1179. 26. R. Ross and J. Rissanen, Am. J. Clin. Nutr., 1994, 60, 695. 27. E. L. Thomas, A. Byrnes, J. V. Hajnal, N. Saeed, G. Frost, and J. D. Bell, Proc. VIth Ann Mtg. (Int) Soc. Magn. Reson. Med., Sydney, 1998, (Vol. 3), p. 1812. 28. D. S. Gray, K. Fujioka, P. M. Colletti, H. Kim, W. Devine, T. Cuyegkeng, and T. Pappas, Am. J. Clin. Nutr., 1991, 54, 623. 29. R. Leenen, K. van der Kooy, P. Deurenberg, J. C. Seidell, J. A. Weststrate, F. J. Schouten, and J. G. Hautvast, Am. J. Physiol., 1992, 263, E913. 30. K. van der Kooy, R. Leenen, J. C. Seidell, P. Deurenberg, and J. G. Hautvast, Am. J. Clin. Nutr., 1993, 58, 853. 31. S. J. Marks, N. R. Moore, M. L. Clark, B. J. Strauss, and T. D. Hockaday, Obes. Res., 1996, 4, 1. 32. C. Boesch, R. Kreis, H. Howald, S. Matter, R. Billeter, B. EssenGustavsson, and H. Hoppeler, Proc. VIth Ann Mtg. (Int) Soc. Magn. Reson. Med., Sydney, 1998, (Vol. 3), p. 1785. 33. L. S. Szczepaniak, D. T. Stein, F. Schick, A. Garg, and J. D. McGarry, Proc. Vth Ann Mtg. (Int) Soc. Magn. Reson. Med., Vancouver, 1997, (Vol. 2), p. 1334. 34. J. Rico-Sanz, E. L. Thomas, G. Jenkinson, S. Mierisova, R. Iles, and J. D. Bell, J. Appl. Physiol, 1999, 87 (in press). 35. J. Rico-Sanz, J. V. Hajnal, E. L. Thomas, S. Mierisova, M. AlaKorpela, and J. D. Bell, J. Physiol., 1998, 510, 615. 36. D. T. Stein, R. Dobbins, L. Szczepaniak, C. Malloy, and J. D. McGarry, Diabetes, 1997, 46, (Suppl 1), 23A. 37. N. G. Forouhi, G. Jenkinson, S. Mullick, E. L. Thomas, U. Bhonsle, P. M. McKeigue, and J. D. Bell, Proc. Br. Diabetic Assoc., 1998, S36. 38. J. G. Batchelor, R. J. Cushley, and J. H. Prestegard, J. Org. Chem., 1974, 39, 1698. 39. J. Bus, I. Sies, and M. S. F. Lie Ken Jie, Chem. Phys. Lipids, 1976, 17, 501. 40. F. D. Gunstone, M. R. Pollard, C. M. Scrimgeour, and L. Vedanayagam, Chem. Phys. Lipids, 1977, 18, 115. 41. J. N. Shoolery, Prog. NMR Spect., 1977, 11, 79. 42. N. G. Soon, Lipids, 1985, 20, 778. 43. F. D. Gunstone, Chem. Phys. Lipids, 1991, 59, 83. 44. L. O. Sillerud, C. H. Han, M. W. Bitensky, and A. A. Francendese, J. Biol. Chem., 1986, 261, 4380. 45. M. L. Barnard, J. D. Bell, S. C. R. Williams, T. A. B. Sanders, H. G. Parkes, K. K. Changani, J. S. Beech, M. L. Jackson, and S. R. Bloom, Proc. XIth Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 3339.

8 BODY FAT METABOLISM: OBSERVATION BY MR IMAGING AND SPECTROSCOPY 46. T. W. Fan, A. J. Clifford, and R. M. Higashi, J. Lipid Res., 1994, 35, 678. 47. E. L. Thomas, S. C. Cunnane, and J. D. Bell, NMR Biomed., 1998, 11, 290. 48. J. Hirsch, J. W. Farquhar, E. H. Ahrens, M. L. Peterson, and W. Stoffel, Am. J. Clin. Nutr., 1960, 8, 499. 49. A. C. Beynen, R. J. J. Hermus, and J. G. A. J. Hautvast, Am. J. Clin. Nutr., 1980, 33, 81. 50. E. L. Thomas, G. Frost, M. L. Barnard, D. J. Bryant, J. Simbrunner, S. D. Taylor-Robinson, G. A. Coutts, M. Burl, S. R. Bloom, K. D. Sales, and J. D. Bell, Lipids, 1996, 31, 145. 51. E. L. Thomas, J. D. Hanrahan, M. Ala-Korpela, G. Jenkinson, D. Azzopardi, R. A. Iles, and J. D. Bell, Lipids, 1997, 32, 645. 52. E. L. Thomas, S. D. Taylor-Robinson, M. L. Barnard, G. Frost, J. Sargentoni, B. R. Davidson, S. C. Cunnane, and J. D. Bell, Hepatology, 1997, 25, 178. 53. R. J. Dimand, C. T. W. Moonen, S. Chu, E. M. Bradbury, G. Kurland, and K. L. Cox, Pediatr. Res., 1988, 24, 243. 54. E. L. Thomas, A. Byrnes, G. Jenkinson, M. Jubb, G. Frost, and J. D. Bell, J. Magn. Reson. Anal. 1999, in press. 55. M. L. Barnard, J. D. Bell, S. C. R. Williams, T. A. B. Sanders, and S. R. Bloom, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 92. 56. J. Loscalzo, J. Freedman, A. Rudd, I. Barsky-Vasserman, and D. E. Vaughan, Arteriosclerosis, 1987, 7, 450.

57. S. C. Cunnane, R. J. McDonagh, S. Narayan, and D. J. Kyle, Lipids, 1993, 28, 273. 58. I. M. Brereton, D. M. Doddrell, S. M. Okenfull, D. Moss, and M. G. Irving, NMR Biomed., 1989, 2, 55. 59. M. Ala-Korpela, Y. Hiltunen, J. Jokisaari, S. Eskelinen, K. Kiviniitty, M. Savolainen, and Y. A. Kesaniemi, NMR Biomed., 1993, 6, 225. 60. J. D. Otvos, E. J. Jeyarajah, D. W. Bennett, and R. M. Krauss, Clin. Chem., 1993, 38, 1632.

Biographical Sketches E. Louise Thomas. b 1970; B.Sc. Biochemistry University of London, 1992; Ph.D. London, 1996. Ph.D. thesis on in vivo 13C NMR spectroscopy under the supervision of J. D. Bell and K. D. Sales. Currently a Senior Research Fellow at Imperial College School of Medicine. Over 20 publications. Research interests include application of MRI/MRS to the study of human fat and muscle metabolism. Jimmy D. Bell. b 1958; B.Sc. Biochemistry University of Warwick, 1982; Ph.D. London, 1987. Lecturer Hammersmith Hospital 1989±94. Senior lecturer Royal Postgraduate Medical School 1994±97. Senior lecturer at MRC Clinical Sciences Centre, Imperial College 1997±present. Approximately 90 publications. Research interests include application of MRI/MRS in clinical research, and MRS to the chemistry of tissue and body ¯uids.

Dietary Changes Studied by MRS Maria L. Barnard and Jimmy D. Bell Royal Postgraduate Medical School, Hammersmith Hospital, London, UK

1 2 3 4 5 6

Introduction Useful Nuclei for Dietary Studies In Vivo NMR Spectroscopy High-Resolution NMR Spectroscopy Related Articles References

1

INTRODUCTION

1 1 1 4 5 5

The importance of diet in both the preservation of health and treatment of disease is increasingly recognized. The major causes of morbidity and mortality in the Western world include cancer, coronary heart disease, diabetes mellitus, and obesity. Nutrition appears to have a major role both in their treatment and prevention.1,2 There is therefore a growing need for the application of current scientific techniques to nutritional studies, to delineate the response of human metabolism to dietary alterations. The development and use of spectroscopy for nutritional studies is a major recent advance. The last decade has seen increasing interest in the use of both in vivo and in vitro NMR spectroscopy for investigation of human and animal metabolism. Researchers are taking advantage of the nondestructive and noninvasive nature of the technique. These characteristics are particularly important for widespread, population-based studies of human nutrition, which often require serial examinations. The initial applications of NMR to dietary studies were principally concerned with development of the NMR methodology, rather than focusing on specific nutritional and biochemical problems.3,4 However, even at this early stage the potential of NMR spectroscopy when applied to lipid and glycogen metabolism could be envisaged. In this article we will review the recent use of in vivo and in vitro NMR spectroscopy for the investigation of the metabolic effects of diet.

2

USEFUL NUCLEI FOR DIETARY STUDIES

Carbon-13, 2 H, 1 H and 31 P NMR techniques have all been applied to study the effects of dietary manipulation in animals and humans.3 – 9 Limited information can be obtained with in vivo 31 P and 1 H NMR spectroscopy as applied to nutrition. Phosphorus-31 NMR detects only phosphatecontaining compounds; 1 H NMR suffers from a small chemical shift range, resulting in severe signal overlap and signals near the water resonance which cannot be readily detected. This has meant that 1 H and 31 P NMR techniques have principally been applied to dietary studies in vitro.10 – 12 However, 2 H and 13 C

NMR methods have been used in elegant in vivo metabolic studies. The deuterium nucleus is quadrupolar (I = 1). It has both a low gyromagnetic ratio and natural abundance (0.015%), which leads to a very low sensitivity relative to proton spectroscopy (1.45 × 10−6 ). These unfavourable properties are offset by its short relaxation times and high body content (12 mM), allowing its detection by natural abundance in vivo NMR spectroscopy. Furthermore, the tissue content can be readily increased by the use of deuterated water (D2 O), a fact that Brereton et al.5 have utilized to great effect in lipid turnover studies. Carbon-13 NMR spectroscopy in vivo has also been successfully applied in dietary studies.3,4,8,9 Compared with 1 H NMR, 13 C NMR spectroscopy is very insensitive due to the low gyromagnetic ratio and low natural abundance (1.1%) of the 13 C nucleus. Its relative sensitivity is further reduced by the 1 H– 13 C coupling which splits the 13 C signal, effectively reducing intensity. These problems are partially offset by the short relaxation times of most 13 C resonances, the use of decoupling techniques, and the high tissue content of important dietary-related, 13 C-containing compounds, including lipids and glycogen. Moreover, the very fact that 13 C has 1.1% natural abundance has been exploited by using 13 C-enriched metabolites in turnover studies, similar to classical 14 C tracer studies.

3 IN VIVO NMR SPECTROSCOPY 3.1 Lipid Metabolism 3.1.1 Carbon-13 NMR Spectroscopy

High-resolution 13 C NMR spectroscopy has found widespread use in the study of lipids in vitro.13,14 In particular, the clear distinction of multiple different fatty acid groups allows this technique to be used, often quantitatively, in studies of dietary oils.15 – 18 This degree of resolution cannot be achieved in vivo, which limits the utility of this technique in noninvasive human research. However, major lipid groups can be distinguished and, within these constraints, important clinical and biochemical work has been carried out using proton-coupled 13 C and proton-decoupled 13 C{1 H} NMR spectroscopy. Canioni et al.3 first showed that in vivo 13 C{1 H} NMR spectroscopy could detect differences in the lipid composition of adipose tissue and liver in rats fed a diet high in polyunsaturated fatty acids. Further work by Sillerud et al.19 examined 13 C{1 H} NMR of triacylglycerols in rat adipocytes in vivo, advancing our knowledge particularly of signal assignment in biological systems. Moonen et al.4 developed in vivo 13 C{1 H} NMR to characterize adipose tissue in human subjects. They confirmed that linoleic acid (C18:2n−6 ), usually the principal polyunsaturated fatty acid present in human adipose tissue, dominated the polyunsaturated fatty acid carbon signal observed in vivo, allowing estimation of this stored essential fatty acid. A typical 13 C NMR spectrum of human adipose tissue is shown in Figure 1. Barnard et al.9 demonstrated the capability of in vivo 13 1 C{ H} NMR spectroscopy to monitor the effect of diet on

2 DIETARY CHANGES STUDIED BY MRS (a)

8

(b)

(c)

(d) Polyunsaturated

Monounsaturated Polyunsaturated

2

1

9 3

200

150

4

100

67

5

50

10 11 12

0 ppm

Figure 1 Natural abundance in vivo 13 C{1 H} NMR spectrum of human adipose tissue, dominated by signals from triacylglycerols. Peak assignment (referenced to –CH3 ): 1, C=O (171.79 ppm); 2, –CH=CH–(monounsaturated) and (129.83 ppm); 3, –CH=CHCH2 CH=CH–(polyunsaturated) –CH=CHCH2 CH=CH–(polyunsaturated) (128.13 ppm); 4, C-2 glycerol (69.21 ppm); 5, C-1, C-3 glycerol (62.05 ppm); 6, –CH2 CH2 CO2 R (33.93 ppm); 7, –CH2 CH2 CH3 (32.14 ppm); 8, –(CH2 )n –(29.69 ppm); 9, –CH2 CH=CH–(27.39 ppm); 10, –CH2 CH2 CO2 R (25.04 ppm); 11, –CH2 CH2 CH3 (22.94 ppm); 12, –CH2 CH3 (14.1 ppm)

adipose tissue in rats. In animals fed fats with different fatty acid patterns (butter/lard, olive oil, sunflower oil, fish oil), clear differences were found in the in vivo spectra, consistent with a significant effect of dietary fatty acid intake on the tissue fatty acid profile. Progressively increasing the total unsaturated and polyunsaturated fatty acid content in the diet was reflected in the adipose tissue composition (Figure 2), and results were correlated with in vitro spectroscopy and gas chromatography analyses. In humans, the turnover of fatty acids in adipose tissue is slow (T 1 = 600 d). The fatty acid profile of human 2 subcutaneous fat therefore provides an index of the habitual dietary fatty acid intake over the previous 2–3 years.20 This is of importance in long-term epidemiological surveys, but has previously been limited by the need for repeat tissue biopsies for fatty acid estimation. The application of in vivo 13 C NMR spectroscopy to the noninvasive analysis of adipose tissue composition is therefore of particular value for human nutritional studies. Beckmann et al.8 have used in vivo 13 C NMR to show a significant change in human adipose tissue fatty acid composition following a fat-reduced diet, with a correlation between diet and tissue monounsaturated fatty acids. No difference was found in the degree of polyunsaturation, and this may reflect the limited 6 month period of dietary change. Bryant et al.21 chose to study vegan subjects as a defined population, with an established long-term diet, high in polyunsaturated fatty acids. A significantly increased adipose tissue total unsaturated and mono- and polyunsaturated fatty acid carbon content was shown in the 13 C spectra from the vegan group compared with omnivore controls (Figure 3). In addition, this was associated with a significant reduction in plasma concentrations of total cholesterol (4.55 ± 1.30 versus

134 130 126 ppm

134 130 126 ppm

134 130 126 ppm

134 130 126 ppm

Figure 2 Olefinic region of natural abundance in vivo 13 C{1 H} NMR spectra of epididymal fat pad in rats fed fats high in different fatty acids. Unsaturated fatty acid carbon resonances appear in this region: monounsaturated (–CH=CH–) and polyunsaturated (–CH=CHCH2 CH=CH–) at 129.8 ppm; polyunsaturated (–CH=CH–CH2 CH=CH–) at 128.2 ppm. (a) Butter/lard diet–saturated fatty acids; (b) olive oil diet–monounsaturated fatty acids [oleic acid (C18:1n−9 )]; (c) sunflower oil–polyunsaturated fatty acids [linoleic acid (C18:2n−6 )]; (d) fish oil–(n − 3) polyunsaturated fatty acids [(C20:5n−3 ) and (C22:6n−3 )]. In these rats, adipose tissue lipid composition measured by NMR correlates to dietary content. Increasing dietary unsaturated and polyunsaturated fatty acid content produces similar changes in the adipose tissue, shown by an increasing total signal from C=C carbons and an increasing signal from polyunsaturated carbons at 128.2 ppm. Further, there appears to be a novel peak (*) at 132 ppm from unsaturated carbons at the n − 3 position (–CH=*CHCH2 CH3 ), a double bond particularly found in the n − 3 fatty acids of fish oil

5.39 ± 1.42 mmol l−1 , p < 0.05) and low-density lipoprotein (LDL) cholesterol (2.72 ± 1.06 versus 3.37 ± 1.34 mmol l−1 , p < 0.05) in the vegan subjects. The complex interactions between lipids in plasma and the body tissues are not fully defined, but are likely to be the subject of future NMR-based, dynamic studies of nutrition and lipid metabolism. The interaction between diet and tissue lipids has been further studied by NMR in the liver. The liver has a pivotal role in lipid metabolism. Hepatic receptor-mediated LDL clearance is a major process controlling plasma LDL concentrations, and dietary saturated fat may increase plasma LDL levels by suppression of hepatic LDL receptor activity. The mechanisms underlying this effect are unclear, but it has been suggested that enrichment of hepatic membrane lipids with saturated fatty acids interferes with LDL receptor function.22 Dietary substitution with polyunsaturated fatty acids may then increase hepatic LDL clearance and lower plasma LDL levels, possibly by increasing membrane fluidity.23 Barnard et al.24 used in vivo 13 C{1 H} NMR to examine the hepatic fatty acid profile in rats fed fats with different fatty acid patterns (butter/lard, sunflower oil, fish oil). The resolution of hepatic 13 C spectra is limited. The authors therefore used analysis by integration of signals across chemical shift ranges, and suggested expressing results by calculation of unsaturation and polyunsaturation indices (Table 1). A significant dietary effect was shown, with an increasing hepatic content of unsaturated fatty acids and an increasing degree of polyunsaturation as these fatty acids increased in the diet. However, in vitro 13 C{1 H} NMR studies of the liver lipid extracts showed a complex relationship,

DIETARY CHANGES STUDIED BY MRS (a) Carboxyl

Monounsaturated Polyunsaturated

Polyunsaturated

3

availability of labeled fatty acids, their oxidation after administration, and the sensitivity of the system, given the preexisting strong lipid signals. However, preliminary work has been performed by Cunnane et al.25 and this demonstrated the ability of in vitro 13 C{1 H} NMR to detect carbon-specific incorporation of injected [U-13 C] eicosapentaenoic acid in extracted rat liver lipids. 3.1.2 Deuterium Spectroscopy

(b)

180 ppm

170

160

150

140

130

120

Total unsaturated carbons

Figure 3 Olefinic region from natural abundance in vivo coupled 13 C NMR spectra of human thigh adipose tissue in (a) vegan and (b) omnivore subjects. Spectra are scaled to the carboxyl signal. The vegan subjects, compared to omnivores, have an increased adipose tissue content of both total unsaturated and polyunsaturated fatty acid carbons (polyunsaturated: 3.45 ± 0.87 versus 2.39 ± 0.55% of total fatty acid carbons, p < 0.05). This reflects at least 3 years’ adherence to the vegan diet, which, in the absence of all animal products, is high in polyunsaturated fats (polyunsaturated: 33.13 ± 7.4 versus 25.2 ± 8.5% of total dietary fatty acids, p < 0.05)

with evidence of hepatic metabolism of the dietary fatty acids. Indeed, the liver is a major site of fatty acid uptake, desaturation, synthesis, storage, and secretion. Investigation of the metabolism of specific dietary fatty acids is a largely unexplored possible future application of 13 C NMR spectroscopy. The use of 13 C-labeled fatty acids to trace metabolic pathways and kinetic rates is an exciting prospect. These studies may be restricted by the limited Table 1

Hepatic Fatty Acid Profile by In Vivo

Dietary Fat Butter/lard Sunflower oil Fish oil

13

There is considerable interest in monitoring lipid turnover in both health and disease. A novel approach to the study of dynamic in vivo tissue lipid metabolism was developed by Brereton et al.5 They investigated the use of in vivo 2 H NMR spectroscopy in mice. Administration of D2 O (10% v/v) in the drinking water for 3–4 d allowed the detection of deuteriumenriched tissue lipid resonances in spectra acquired in less than 2 min, as shown in Figure 4(a). The spectra consist of resonances from deuterated water (HOD) and the CHD groups of tissue lipids. Removal of D2 O from the drinking water led to clear changes in the intensity of both signals [Figure 4(b)]. The loss of deuterium from the water signal was significantly faster than from tissue lipids. The loss of deuterium from the lipid resonance was therefore proposed as a noninvasive measure of the rate of fat utilization. This method has since been applied to the study of fat utilization in obesity. Obesity is one of the most common dietary disorders. It is characterized by excess adipose tissue and elevated levels of plasma nonesterified fatty acids. Furthermore, obesity is known to lead to glucose intolerance and insulin resistance. Fat turnover, as reflected by the levels of plasma nonesterified fatty acids, has been suggested as being implicated in this process. However, the lack of in vivo techniques for measuring fat utilization has greatly hampered progress in this area. Body fat turnover in mouse models of obesity and diabetes mellitus have been studied in vivo using the 2 H NMR technique.26 Brereton et al.26 showed that the rates of fat utilization in obese mice were significantly lower than the rates for nonobese mice (Figure 5) and the induction of diabetes did not affect utilization of fat as a metabolic fuel. These studies clearly suggest that deuterium NMR spectroscopy provides researchers with a powerful method for noninvasively assessing fat turnover rates. 3.1.3 Proton Spectroscopy

The application of in vivo proton NMR spectroscopy to dietary studies has been rather limited. Indeed, many of the

C{1 H} NMR Spectroscopy of Rats Fed Different Fats

Unsaturation Indexa

Polyunsaturation Indexb

Methylene–Methyl/%c

Olefinic/%c

0.19 ± 0.04 0.24 ± 0.01 0.27 ± 0.01

0.90 ± 0.04 1.04 ± 0.04 1.32 ± 0.03

84.1 ± 0.91 80.9 ± 0.31 78.9 ± 0.79

15.9 ± 0.91 19.1 ± 0.31 21.1 ± 0.79

p < 0.05 for all differences between dietary groups. a Unsaturation index = olefinic/methylene–methyl signal integration (measures the ratio of unsaturated to saturated fatty acid carbons). b Polyunsaturation index = signal intensity at 128.2 ppm (=C–C–C=)/129.8 ppm (–C=C–) (measures the ratio of polyunsaturated to monounsaturated + polyunsaturated carbons). c Olefinic and methylene–methyl signal area as the percentage of the total fatty acid signal integration (measures the percentage of fatty acid carbons that are unsaturated or saturated, respectively).

4 DIETARY CHANGES STUDIED BY MRS (a)

(a)

Relative intensity

100

50

0

12

0

–12 ppm (b)

100

Relative intensity

(b)

50

0 0

8

16 Time (d)

24

Figure 5 Plots of 2 H NMR signal intensity as a function of time showing (a) best fit single exponential decay curves for HOD and (b) best fit biexponential function for CHD intensities. The experimental data for control (solid square) and obese (solid circle) mice were used to calculate the body water and fat utilization rates. (Reproduced by permission of Heyden & Son Limited from I. M. Brereton, D. M. Doddrell, S. M. Oakenfull, D. Moss, and M. G. Irving, NMR in Biomed., 1989, 2, 55) (a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

Figure 4 (a) Natural abundance 2 H NMR spectrum of a mouse. The CHD resonance is to high field at 1 ppm. Chemical shifts were referenced to the HOD resonance assigned to 4.8 ppm. (b) Deuterium NMR spectra of a mouse following the removal of 10% (v/v) of D2 O from its drinking water. The spectra were recorded (a) 1, (b) 4, (c) 5, (d) 8, (e) 11, (f) 15, (g) 16 and (h) 21 days after the resumption of normal drinking water. Spectrum (i) was recorded prior to the administration of D2 O to the drinking water (Adapted with permission of Heyden & Son Limited from I. M. Brereton, D. M. Doddrell, S. M. Oakenfull, D. Moss, and M. G. Irving, NMR in Biomed., 1989, 2, 55)

metabolites observed by 1 H NMR spectroscopy do not appear to be affected by diet, with the exception of lipids.7 However, due to the small chemical shift range of 1 H resonances and the intense water signal, in vivo 1 H lipid studies have also been restricted. A potentially important alternative application of 1 H NMR is the use of fat imaging to determine quantitative fat distribution in human subjects.27 Regional fat distribution has been correlated to serum lipids and coronary heart disease. Existing methods for measurement of body fat distribution are limited, and fat imaging may in future become the modality of

choice in nutritional and metabolic studies evaluating the risk of coronary heart disease. 3.2 Glycogen Studies

Tissue glycogen values may be a marker of carbohydrate intake and could therefore be of value in dietary studies. Shulman and co-workers28 have developed a 13 C NMR technique to provide a direct noninvasive method to measure liver and muscle glycogen concentration in humans. The time course of liver glycogen concentration, following standard test meals, has been carried out by 13 C NMR spectroscopy. This opens the possibility of following quantitative changes in glycogen storage with varying diets.29,30

4 HIGH-RESOLUTION NMR SPECTROSCOPY

The unique exploratory capability of NMR spectroscopy has been utilized to investigate the effects of diet on the metabolic profile of body fluids.31 The 1 H NMR spectra of urine from

DIETARY CHANGES STUDIED BY MRS Cit

(a)

(a)

5

(b)

3

NMe Cr Cn

2

Suc

1

3

2A 1

Dmg Ac

(b)

Dma

Og 1.3 1.2 1.1 1.0 0.9 0.8 ppm

NAc Lac

Tau 4

3

2

1

d (ppm)

Figure 6 Proton NMR spectra at 500 MHz (aliphatic regions) of urine from 1-month-old rats fed either (a) chow or (b) casein diets. Assignments: Lac, lactate; Ac, acetate; NAc, N -acetyl groups of glycoproteins; Suc, succinate; Og, 2-oxoglutarate; Cit, citrate; Dma; dimethylamine; Dmg, dimethylglycine; Cr, creatinine; Cn, creatine; Tau, taurine; NMe, betaine plus trimethylamine N -oxide. (Reproduced by permission of Academic Press, Inc. from J. D. Bell, O. Lavender, V. C. Morris, and P. J. Sadler., Magn. Reson. Med. 1991, 17, 414)

rats fed different diets are shown in Figure 6. Clear changes in the metabolic profile of the urine can be observed.10 Rats fed casein diets for 1 month postweaning did not excrete 2oxoglutarate, and excreted lower levels of hippurate, succinate, and citrate compared with age/sex-matched rats fed standard chow diets. The acquisition of these spectra took ∼2 min. This suggests that high-resolution NMR of urine or other body fluids could be utilized as a fast, multicomponent technique in human studies, for example to screen rapidly for multiple metabolic effects or to check dietary compliance. Plasma concentrations of LDLs and high-density lipoproteins (HDLs) are well established markers for coronary heart disease risk assessment. Standard methods for quantification and compositional analysis of plasma lipoproteins require laborious and time-consuming physical separation based on particle size, density, or apolipoprotein content. Bell et al. have shown that NMR spectroscopy can be readily applied to study alterations in plasma lipoproteins associated with dietary manipulation, both in intact plasma or in the separated lipoprotein classes.11,12 The effects of fish oil supplementation on the 1 H NMR profile of lipoproteins in intact plasma are illustrated in Figure 7. Changes in composition are shown by a new lipid resonance (peak 2A) arising from fish oil n − 3/ω − 3 fatty acids incorporated into plasma lipoproteins, while the methylene [–(CH2 )n –] signal (peak 3) is markedly reduced. Furthermore, T 2 analysis of the lipid resonances showed that these changes in lipid composition of the LDL particles are accompanied by alteration in the structural characteristics of these particles.

1.3 1.2 1.1 1.0 0.9 0.8 ppm

Figure 7 Expansions of spin echo 1 H NMR spectra of intact human plasma (a) before and (b) after 7 d of fish oil supplementation. Assignments: peaks 1 and 2, terminal –CH3 group from LDLs/HDLs and very low-density lipoproteins (VLDLs), respectively; peak 3, –(CH2 )n –of VLDLs. Peak 2A arises from n − 3/ω − 3 fatty acids incorporated into plasma lipoproteins. (Reproduced by permission of Birkhauser Verlag from P. C. Dagnelie, J. D. Bell, M. L. Barnard, and S. C. R. Williams, in ‘Omega-3 Fatty Acids: Metabolism and Biological Effects’ ed. C. A. Drevon, I. Baksaas, and H. E. Krokan, Birkhauser Verlag, Basel/Switzerland, 1993, p. 27–34)

At present, sophisticated computer deconvolution methods (line-fitting analysis) are being developed to identify and quantify lipoprotein fractions in the 1 H NMR spectra of intact plasma samples (500 µl) within minutes.32,33 In this article we have reviewed the current state of NMR spectroscopy as applied to dietary studies. It is clear that this is a relatively new area of NMR research. However, it offers many exciting prospects for future research into diet and nutrition in humans.

5 RELATED ARTICLES

Analysis of High-Resolution Solution State Spectra; Brain Neoplasms Studied by MRI; Body Fluids; Chemical Shifts in Biochemical Systems; High-Field Whole Body Systems; Lipoproteins; Whole Body Studies: Impact of MRS.

6 REFERENCES 1. J. E. Kinsella, B. Lokesh, and R. A. Stone, Am. J. Clin. Nutr., 1990, 52, 1. 2. R. A. DeFronzo and E. Ferrannini, Diabetes Care., 1991, 14, 173. 3. P. Canioni, J. R. Alger, and R. G. Shulman, Biochemistry, 1983, 22, 4974. 4. C. T. W. Moonen, R. J. Dimand, and K. L. Cox, Magn. Reson. Med., 1988, 6, 140. 5. I. M. Brereton, M. G. Irving, J. Field J, and D. M. Doddrell, Biochem. Biophys. Res. Commun., 1986, 137, 579. 6. P. C. Dagnelie, J. D. Bell, S. C. R. Williams, I. J. Cox, D. K. Menon, J. Sargentoni, and G. A. Coutts, NMR Biomed., 1993, 6, 2.

6 DIETARY CHANGES STUDIED BY MRS 7. F. Schick, B. Eismann, W. I. Jung, H. Bongers, M. Bunse, and O. Lutz, Magn. Reson. Med., 1993, 29, 158. 8. N. Beckmann, J. J. Brocard, U. Keller, and J. Seelig, Magn. Reson. Med., 1992, 27, 97. 9. M. L. Barnard, J. D. Bell, S. C. R. Williams, T. A. B. Sanders, H. G. Parkes, K. K. Changani, J. S. Beech, M. L. Jackson, and S. R. Bloom, Proceedings of the 11th Annual Meeting of the Society of Magnetic Resonance in Medicine, 1992 , p. 3339. 10. J. D. Bell, O. Lavender, V. C. Morris, and P. J. Sadler., Magn. Reson. Med., 1991, 17, 414. 11. J. D. Bell, J. C. C. Brown, R. E. Norman, P. J. Sadler, and D. R. Newell, NMR Biomed., 1988, 1, 90. 12. P. C. Dagnelie, J. D. Bell, M. L. Barnard, and S. C. R. William, in Omega-3 Fatty Acids: Metabolism and Biological Effects, ed. C. A. Drevon, I. Baksaas, and H. E. Krokan, Birkhauser Verlag, Basel, 1993, p. 27. 13. J. G. Batchelor, R. J. Cushley, and J. H. Prestegard, J. Org. Chem., 1974, 39, 1698. 14. J. Bus, I. Sies, and M. S. F. Lie Ken Jie, Chem. Phys. Lipids, 1976, 18, 130. 15. F. D. Gunstone, M. R. Pollard, C. M. Scrimgeour, and H. S. Vedanayagam, Chem. Phys. Lipids, 1977, 18, 115. 16. J. N. Shoolery, Prog. NMR Spectrosc., 1977, 11, 79. 17. N. G. Soon, Lipids, 1985, 20, 778. 18. F. D. Gunstone, Chem. Phys. Lipids, 1991, 59, 83. 19. L. O. Sillerud, C. H. Han, M. W. Bitensky, and A. A. Francendese, J. Biol. Chem., 1986, 261, 4380. 20. A. C. Beynen, R. J. J. Hermus, and J. G. A. J. Hautvast, Am. J. Clin. Nutr., 1980, 33, 81. 21. D. J. Bryant, J. D. Bell, E. L. Thomas, S. D. Taylor-Robinson, J. Simbrunner, J. Sargentoni, M. Burl, G. A. Coutts, G. Frost, M. L. Barnard, S. Cunnane, and R. A. Iles, Proceedings of the 12th Annual Meeting of the Society of Magnetic Resonance in Medicine, 1993 , P. 1048. 22. S. M. Grundy, and M. A. Denke, J. Lipid Res., 1990, 31, 1149. 23. J. Loscalzo, J. Freedman, A. Rudd, I. Barsky-Vasserman, and D. E. Vaughan, Arteriosclerosis, 1987, 7, 450. 24. M. L. Barnard, J. D. Bell, S. C. R. Williams, T. A. B. Sanders, and S. R. Bloom, Proceedings of the 12th Annual Meeting of the Society of Magnetic Resonance in Medicine, 1993 , p. 92.

25. S. C. Cunnane, R. J. McDonagh, S. Narayan, and D. J. Kyle, Lipids, 1993, 28, 273. 26. I. M. Brereton, D. M. Doddrell, S. M. Oakenfull, D. Moss, and M. G. Irving, NMR Biomed., 1989, 2 55. 27. R. Ross, L. Leger, D. Morris, J. De Guise, and R. Guardo, J. Appl. Physiol., 1992, 72, 787. 28. M. J. Avisoz, D. L. Rothman, E. Nadel, and R. G. Shulman. Proc. Natl. Acad. Sci. USA, 1988, 85, 1634. 29. T. Jue, D. L. Rothman, B. A. Tavitian, and R. G. Shulman. Proc. Natl. Acad. Sci. USA, 1989, 86, 1439. 30. M. Ishihara, H. Ikehira, S. Nishikawa, T. Hashimoto, K. Yamada, F. Shishido, T. Ogino, K. Cho, S. Kobayashi, M. Kawana, T. Matumoto, T. A. Iinuma, N. Arimizu, and Y. Tateno, Am. J. Phys. Imag., 1992, 7, 32. 31. J. D. Bell, J. C. C. Brown, and P. J. Sadler, NMR Biomed., 1989, 2, 246. 32. M. Ala-Korpela, Y. Hiltunen, J. Jokisaari, S. Eskelinen, K. Kiviniity, M. J. Savolainen, and Y. A. Kesaniemi. NMR Biomed., 1993, 6, 225. 33. J. D. Otvos, E. J. Jeyarajah, D. W. Bennett, and R. M. Krauss. Clin. Chem., 1993, 38, 1632.

Biographical Sketches Maria L. Barnard. b 1960. B.Sc., 1981, M.B., Ch.B., 1984, University of Bristol, UK. Introduced to NMR by Graeme M. Bydder and Jimmy D. Bell, MRI Unit, Hammersmith Hospital during Medical Research Council (MRC) UK Training Fellowship with Stephen R. Bloom, Division of Endocrinology and Metabolism, Hammersmith Hospital, UK 1989–93. MRC (UK) Travelling Fellow to R.G. Shulman, Yale University, USA, 1992–93. Approx. 23 publications. Research interests: application of NMR spectroscopy to metabolic and nutritional studies. Jimmy D. Bell. b 1958. B.Sc. Biochemistry 1982, University of Warwick. Ph.D. (supervisor P. J. Sadler), 1987, London. Senior NMR research fellow (with I. Young and D. G. Gadian), Hammersmith Hospital, 1989–93. Faculty in Biochemistry, Royal Postgraduate Medical School, 1993–present. Approx. 60 publications. Research interests include application of MRI/MRS in clinical research, and NMR spectroscopy to the chemistry of tissues and body fluids

Hepatic and Other Systemically Induced Encephalopathies: Applications of MRS

originating in a systemic disease. In animal studies, several distinct neurotoxins have been identified. The earliest of these was ammonia;3,4 ammonia is normally fully removed from portal blood by hepatic urea synthesis.5,6 The combination of the loss of biosynthetic liver function and the diversion of nondetoxified blood to the brain by so-called portal systemic shunts (PSSs) accounts for the frequently demonstrated excess of cerebral and cerebrospinal fluid glutamine7 [equation (1)]. NH4+ + glutamate

Brian D. Ross Huntington Medical Research Institutes, Pasadena, CA, USA

1 Background: Biochemistry of Coma 2 Neurochemical Pathology of Hepatic Encephalopathy 3 Animal Studies in HE using Multinuclear MR Spectroscopy 4 Pathogenesis, Diagnosis, and Therapeutic Management of HE in Man: The Emerging Role of Proton MRS 5 Energetics of the Human Brain in HE 6 Subclinical HE 7 mI Depletion and the Induction of HE 8 Restoration of Cerebral mI and Cho Accompanies Reversal of HE 9 Ornithine Transcarbamylase (OTC) Deficiency 10 Contribution of NMR to Clinical HE 11 Other Systemic Encephalopathies 12 Conclusions 13 Related Articles 14 References

1

1 1 1 1 2 2 3 4 4 4 5 7 7 7

BACKGROUND: BIOCHEMISTRY OF COMA

Metabolic disturbances of liver, kidney, endocrine or other systems have remote effects; those on the brain result in a variety of well defined encephalopathies. When severe, these disorders present as coma. Posner and Plum published a comprehensive account of human coma.1 Many were the result of presumed metabolic events with normal brain anatomy, setting the stage for noninvasive elucidation by means of biochemically based techniques. Among these are MRA, diffusion imaging, positron emission tomography (PET), single-photon emission tomography (SPECT), and multinuclear magnetic resonance spectroscopy (MRS). MRS has increasingly been used to identify specific biochemical changes in the brain, from which information on diagnosis and pathogenesis of these poorly understood disorders is beginning to emerge. As a model for this group of disorders, we discuss hepatic encephalopathy, with additional remarks about diabetic, hyperosmolar, and hypoxia-induced encephalopathies.

2

NEUROCHEMICAL PATHOLOGY OF HEPATIC ENCEPHALOPATHY

Hepatic encephalopathy (HE) is an excellent example of metabolic encephalopathy,2 with identifiable neurotoxins

GS glutamine
glutaminase

(1)

The enzyme glutamine synthetase (GS) responsible for this reaction is located exclusively in astrocytes.8 Resynthesis of glutamate and of γ -aminobutyrate (GABA) from glutamine (both vital neurotransmitter amino acids) occurs principally in neurons.9 Lest it be thought that ammonia ‘toxicity’ accounts for all of the clinical syndromes covered by the term HE, the interested reader is referred to Butterworth10 for an extensive review of several other well-documented alternatives. Metabolic theories abound: failure of oxidative energy metabolism (a corollary of glutamate and 2-oxoglutarate depletion from the Krebs cycle), tryptophan and serotonin (5-hydroxytryptamine) accumulation, branched-chain amino acid deficits, endogenous benzodiazepine agonists which modify access of GABA to its inhibitory receptors, and neurotoxic fatty acids, octanoate in particular, have been proposed. An attractive unifying theory proposed by Zieve11 is that multiple neurotoxins derived from liver, blood, or from the diet, gain access via PSSs to a previously ‘sensitized’ brain. No mechanism of sensitization is known, but with the advent of MRS, a candidate has been proposed in the form of cerebral myo-inositol depletion.12

3 ANIMAL STUDIES IN HE USING MULTINUCLEAR MR SPECTROSCOPY

Three groups13 – 15 independently perfected methods for the noninvasive determination of cerebral glutamine (including a variable contribution from glutamate in each assay) with 1 H NMR spectroscopy (now referred to as MRS) and confirmed the elevation of this metabolite in a variety of animal models of acute liver ‘failure’ with HE. One of these groups13 also demonstrated a hitherto unrecognized abnormality, the significant reduction of choline (Cho; cerebral choline-containing compounds) in the 1 H spectra of rats with acute HE. More recently, 15 N NMR16,17 and 1 H– 15 N HMQC (unpublished work from this laboratory) identified cerebral glutamine unequivocally in vivo in HE produced by ammonia infusion in the normal and portocaval shunted rat, respectively. Finally, in extracts of portacaval shunt (PCS) rat brain, highperformance liquid chromatography confirms the accumulation of glutamine and the depletion of glycerophosphoryl choline (GPC, which is therefore the explanation for the reduced ‘Cho’ in the previous 1 H NMR study), as well as demonstrating the expected depletion by 50% or more, of the cerebral myoinositol (mI) and scyllo-inositol (sI) content.18 This result in rats confirmed the observations which first emerged from human studies with short echo time 1 H MRS,12,19,20 and establishes the validity of the portacaval shunt rat for future experimental studies.

2 HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS 4

PATHOGENESIS, DIAGNOSIS, AND THERAPEUTIC MANAGEMENT OF HE IN MAN: THE EMERGING ROLE OF PROTON MRS

Studies using STEAM localized, short echo time 1 H MRS defined the changes in 10 patients with clinically confirmed chronic HE. The average increase in cerebral glutamine was estimated as +50%, Cho decreased 14% and mI decreased by 45%21 (Figure 1). Very similar findings have now been reported at 2T by Bruhn;19,22 and by McConnell23 using PRESS at short or long echo times. An early study which used long echo time PRESS-localized 1 H MRS failed to identify the depletion of mI, but was the first clearly to show ‘Cho’ depletion in human brain.24

NAA Cho

Cr

mI Liver disease

5 ENERGETICS OF THE HUMAN BRAIN IN HE

The predicted cerebral energy deficit (reduction in [ATP]) has convincingly only been shown in mice and tree shrews.25 Phosphorus-31 MRS should be the easiest tool with which to establish any energy deficit in HE (as predicted). Ross,26 Tropp,27 Chamuleau,28 Barbiroli,29 and Morgan30 have obtained conflicting and hence unconvincing data with 31 P MRS on such effects in man. Using quantitative 1 H MRS, however, Geissler and colleagues showed a small but significant increase in cerebral creatine concentration [Cr] following liver transplantation in man.31 Since [Cr] is the sum of phosphocreatine (PCr) and creatine, this may indicate that patients with HE have an ‘energy deficit’. The difficulty which has been confronted in demonstrating such an effect by 31 P MRS, where sensitivity is 10% that of 1 H MRS, becomes obvious. Since reduced cerebral oxygen consumption and blood flow are recognized abnormalities in HE, the 13 C NMR techniques pioneered by Shulman may be best suited to demonstrate the anticipated reduction in the overall rate of the Krebs cycle.32 From these clinical studies, it appears that 1 H MRS, particularly if applied with effective water suppression to reveal mI, is currently most efficient for the elucidation of pathogenesis of HE, and also, as will be shown, for its early clinical diagnosis. Direct measurement of mI by 13 C NMR may become a valuable adjunct.33

6 SUBCLINICAL HE SCHE

As indicated, HE is a group of diseases presumed to have identical etiology, but presenting a great variety of distinct clinical pictures. Underlying all of them is believed to be the entity of ‘subclinical HE’, although it is by no means clear

a-Glx b, g-Glx

Control

HE

Hepatic encephalopathy

(a)

(b) ml Cho

Gln↑

Cho↓

Gln

ml↓

Severe HE

4.0

3.6

3.2

2.8

(c)

3.0

4

3

2

1

0 ppm

Figure 1 Development of hepatic encephalopathy in human subjects (1.5 T spectra). A series of spectra of parietal cortex (white matter) acquired under closely similar conditions (GE Signa 1.5 T, STEAM localization TR 1.5 s, TE 30 ms, NEX 128) from different patients is presented. A normal spectrum for comparison is that in Figure 6(b). In liver disease (top) there is a relatively normal spectrum with a slight decrease in choline. In subclinical hepatic encephalopathy (SCHE) there is a definite decrease in myo-inositol (mI) with a minor increase in the glutamine regions (glutamine plus glutamate = Glx). There is a very significant increase in the Glx regions in HE (grade 1) and mI is further depleted. The spectrum of grade 3 HE shows more severe changes in the biochemical markers of this disease, most notably glutamine

2.4

2.0

1.6

Gln

4.0

3.6

3.2

2.8 2.4 Gln↑

2.0

1.6

(d)

2.0

Standard solution Glutamate (+ NAA)

3.0

2.0

Standard solution Glutamine + glutamate (+ NAA)

Figure 2 Identification of glutamine in the proton spectra at 2.0 T. A 51-year-old patient with HE due to surgical portacaval shunt (b) is compared with a normal control (a). Spectra were obtained from an occipital GM location and show the expected changes of HE: increased glutamine, decreased Cho/Cr and mI/Cr. With the improved resolution at 2 T, separate analysis of glutamine and glutamate is possible. Inset are spectra from model solutions (c) 5 mM glutamate + 5 mM NAA; (d) 5 mM glutamine + 5 mM glutamate + 5 mM NAA, indicating that in this patient the increase in glutamine occurs without obvious depletion of cerebral glutamate. Spectra were acquired on a Siemens 2.0 T clinical spectrometer, with STEAM localization; TR 3 s, TE 20 ms. (Modified from Bruhn et al.)22

HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS

glutamine is rather more obvious at 2 T (Figure 2), but presents little difficulty at 1.5 T. Figure 1 should not be taken as proof, however, that in the individual patient such orderly progression of neurochemical dysfunction occurs. Longitudinal studies have not been performed in sufficient numbers to be certain of this. Nevertheless, it is tempting to suggest, as Figure 1 appears to indicate, that in the brain exposed to liver ‘toxins’, Cho depletion (possibly GPC) precedes the loss of mI and sI, with later accumulation of glutamine. If this sequence is correct, then perhaps ‘sensitization’ of the brain is the result of mI or Cho depletion, or both. In keeping with the theory of many years standing, increasing cerebral glutamine underlies the neurological syndromes of grades 1–4, severe, overt chronic HE, as well as, perhaps, that of acute, fulminant HE and coma. Figure 3 shows the similar but more severe neurochemical changes of Reye’s syndrome, giving an effective indication of what may be seen in acute HE.36 Unfortunately, published spectra from patients with fulminant HE are limited and difficult to interpret.37

PCr

(a)

mI

Gln

Gln

Lac

U

(b)

3

(c) Lipids

7 mI DEPLETION AND THE INDUCTION OF HE (d)

A human ‘experiment’ which goes some way towards verifying this sequence is the new interventional procedure known as NAA

Pre-TIPS mI (e)

4

3

2

1

0

Figure 3 Proton MRS in acute hepatic encephalopathy due to Reye’s syndrome. In vivo proton MRS spectra acquired from a parietal white matter region in infant brain from: (a) 10-month-old normal subject; (b) patient, day 2 after admission; (c) patient, day 8 after admission; (d) difference (b)–(c) between days 2 and 8 in patient; (e) solution with 15 mM glutamine. All spectra are scaled. Spectral assignments: Lac, lactate (1.3 ppm); NAA, N -acetylaspartate (2.02 ppm); Gln, glutamine (2.10–2.50 ppm, and 3.65–3.90 ppm); Cr + PCr, creatine + phosphocreatine (3.03 ppm); Cho, choline containing compounds (3.23 ppm); mI, myo-inositol (3.56 ppm). U, Unassigned (3.62 ppm). Notable abnormalities concern a huge accumulation of cerebral glutamine, reduced Cho, and, later, reduced mI, all of which reflect liver failure. In addition there is a decrease in NAA and [Cr] and appearance of the unassigned peak. Reye’s syndrome, a toxic viral disease associated with aspirin intake is known to produce severe neuronal damage, and may not therefore completely correspond to the picture of acute liver failure. An occipital grey matter region gave almost identical results. (Reproduced with permission from Ernst et al.)36

that the HE and coma of fulminant (very acute) liver failure, goes through any truly ‘subclinical’ phase. Proton MRS accurately reflects the entity of SCHE,34,35 and in preliminary studies also appears to mirror the progressive and increasingly severe syndromes of overt HE defined by Parsons-Smith as grades 1–4 (Figure 1). Elevation of

Glx

Glx

Post-TIPS (3 wks)

Post-TIPS (13 wks)

4

3

2

1

0 ppm

Figure 4 Effect of TIPS on the proton MR spectrum. Spectra are from a 30-year-old female, 5 days after a hematemesis due to esophageal varices and chronic alcoholic liver disease. Spectra acquired from a parietal WM location (GE Signa 1.5 T, volume 12.5 cm3 , STEAM TR 1.5 s TE 30 ms, NEX 128) before a TIPS procedure shows no abnormalities apart from a small but significant reduction in Cho/Cr, attributable to liver disease (top). Three weeks after TIPS, changes are seen in the Glx and mI regions and a further reduction on Cho/Cr (middle). The bottom spectrum shows the progression of changes seen (13 weeks post-TIPS); Glx is markedly increased and mI is significantly reduced

4 HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS TIPS (transjugular intrahepatic portal systemic shunt), which is used as a life-saving procedure in cirrhosis-induced hematemesis. Not surprisingly, TIPS induces clinical HE in up to 90% of survivors.38,39 Proton MRS performed both before and after TIPS in 10 patients showed a universal increase in mean intracerebral glutamine following the procedure. More importantly, in a small number of individuals in whom mI was normal prior to TIPS (these patients often show subclinical HE), a progressive reduction in mI/Cr and development of subclinical and clinical HE follows the introduction of the shunt (Figure 4).

8

(a)

mI

RESTORATION OF CEREBRAL mI AND CHO ACCOMPANIES REVERSAL OF HE

Yet another common human ‘experiment’, that of orthotopic liver transplantation, allows the reverse process to be unequivocally demonstrated, thereby establishing a firm, albeit circumstantial link between mI depletion and the syndromes of subclinical and overt HE (Figure 5). Glutamine plus glutamate and Cho also recovered. Indeed, an overshoot of cerebral Cho is consistently observed, perhaps linking the earlier Cho depletion with deficient hepatic biosynthesis of some relevant precursor of GPC.

NAA (b) Cr Cho Cr Glx

4

9

ORNITHINE TRANSCARBAMYLASE (OTC) DEFICIENCY

A description of MRS in HE would be incomplete without consideration of a rare but informative inborn error

Pre transplant

2

1

0 ppm

Figure 6 Treated OTC deficiency versus healthy age-matched subject. Both spectra were acquired from a similar location in parietal WM (STEAM TR 1.5 s, TE 30 ms, NEX 128) and processed as described in the literature46 . Single-enzyme defects in the urea cycle result in severe hyperammonemia and ‘hepatic encephalopathy’. Because the patient was receiving treatment with sodium benzoate, no excess of cerebral glutamine is present. However, as anticipated by studies in the commoner condition of hepatic encephalopathy due to portosystemic shunting, a decrease in mI was noted in the proton spectrum of a 14 year-old with OTC deficiency (a) when compared with a normal agematched control (b)

of hepatic urea synthesis, ornithine transcarbamylase (OTC) deficiency. The single known biochemical consequence is hyperammonemia. Gadian and colleagues?? demonstrated the inevitable elevation of cerebral glutamine in two such patients, while Ross40 showed the extraordinary parallel with HE, in depletion of cerebral mI (Figure 6).

Cho

mI

Post transplant Glx

4

3

3

10 CONTRIBUTION OF NMR TO CLINICAL HE

2

1

0 ppm

Figure 5 Restoration of biochemical abnormalities post-liver transplant. The patient is a 30-year-old male with acute-on-chronic hepatic encephalopathy secondary to hepatitis, and subsequently successfully treated by liver transplantation. Spectra were acquired 6 months apart from the same parietal WM location, (15.0 cm3 STEAM TR 1.5 s, TR 30 ms, NEX 128) and scaled to the same Cr intensity for comparison. The obvious abnormalities before liver transplantation, increased α, β, and γ glutamine, reduced Cho/Cr and mI/Cr (upper spectrum) were completely reversed 3 months after transplantation, and Cho/Cr exceeded normal (lower spectrum)

10.1 Pathogenesis of HE

Both experimentally and clinically, NMR (particularly 1 H MRS) supports the classical concept of HE as a disorder of cerebral ammonia metabolism, be it by ammonia toxicity or glutamine synthesis (as a newer variant of the theory would have it).41 The concept of an underlying brain ‘sensitization’ receives substantial new impetus deserving of further research in animal models. Either ‘Cho’ (GPC) depletion, due perhaps to failure of hepatic synthesis of a necessary precursor, or cerebral mI depletion could fulfill the role of ‘sensitizer’. In neither case is there a precedent, so that basic research is

HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS

urgently required. It is likely that NMR will play a crucial role in such investigations, and the PCS rat is a convenient model. Carbon-13 NMR is the only method of unequivocally determining mI as distinct from the lower concentrations of inositol-1-phosphate (Inos-1-P) and glycine with which mI coresonates in the 1 H MR spectrum. 10.2

Unanswered Questions in Neurology of HE

Wilson’s disease, caused by excess copper deposition, is believed to result in an encephalopathy analogous to HE. However, the 1 H MRS findings are not surprisingly rather different, lacking either mI depletion or glutamine accumulation (unpublished study from this laboratory). Myelopathy is an unusual form of chronic HE. It presents with myelopathy and paraplegia. Athough neurological considerations would suggest cord involvement, the 1 H MRS findings in the parietal cortex are typical of other patients with the more classical clinical presentation (see Fig. 4 in Ross et al.).40 Often noted in MRI, the basal ganglia may be ‘bright’ in inversion recovery (IR) images of patients with known HE. There is no consistent relationship with 1 H MRS findings, and increasingly this MRS finding is recognized as nonspecific. Nevertheless, the extrapyramidal signs, the changes on postmortem, and these albeit inconsistent MRI findings continue to suggest that there may be as yet unrecognized underlying neurochemical changes in the basal ganglia in HE.

11 OTHER SYSTEMIC ENCEPHALOPATHIES 11.1

Glx

(a)

(c) Cho

Cr

mI

Proton MRS for Diagnosis

Early in these investigations, it became apparent that mI depletion and Glx accumulation occurred when clinical HE was absent. Other systemic or metabolic diseases (apart from OTC deficiency already discussed) did not result in mI depletion, so that 1 H MRS offers an unique opportunity for early, specific diagnosis of this still perplexing condition. Paradoxically, there is at present little enthusiasm for this, most probably because prevention (with lactulose or neomycin) and treatment (by liver transplantation) of overt or severe HE is relatively straightforward (albeit rather costly in the case of orthotopic liver transplantation (OLT) at $150 000–200 000 per patient in USA). If a ready medical means of restoring cerebral mI were to be discovered, the value of 1 H MRS in diagnosis might increase. 10.3

NAA

5

Lipids (b)

(d)

G G K

4

3

2

1

4

3

2

1

Figure 7 Diabetic ketoacidosis (DKA): cerebral 1 H spectra during two DKA episodes and recovery. (a) Episode 1, acquired 3 days after admission to hospital, at a time when the patient had supposedly totally recovered and was ready to be discharged. A peak characteristic for the presence of ketone (K) bodies was noted at 2.22 ppm, obtained from an 18 cm3 volume in the left parietal lobe. The patient relapsed into DKA 2 days later. (b) Episode 2, acquired 5 months later from an occipital gray matter location (10.3 cm3 ) during a second episode of DKA. In addition to the ketone peak, peaks for glucose (G) were seen. (c) Recovery, 6 days after episode 2 (from the same occipital location), when no more ketone bodies were detected in the urine. (d) Occipital cortex of an age- and sex-matched healthy subject. Acquisition conditions: GE Signa 1.5 T, 4X software; STEAM TR 1.5 s, TE 30 ms, NEX 128. More detailed analysis of peak K indicates the resonance frequency to be that of acetone, rather than acetoacetate, the ketone body more commonly identified in blood and urine of diabetics in coma. (Reproduced by permission of the Radiological Society of North America from Kreis and Ross)43

The most surprising finding of proton MRS which requires further study is the identification of acetone (rather than the more widely anticipated β-hydroxybutyrate and acetoacetate) as the ketone of human diabetic ketoacidotic encephalopathy43 (Figure 7). Long-term neurological and cerebral ‘complications’ of diabetes contribute to the much increased mortality. The biochemical basis of these conditions may lie in those changes in Cho, N -acetylaspartate (NAA), mI, ketones, and glucose, now recognized by 13 C and 1 H MRS, to be present in the acutely and chronically diabetic brain.

Diabetes Mellitus

Like HE, diabetic encephalopathy is common and obviously ‘metabolic’ in origin. Three principal syndromes are recognized: diabetic ketoacidosis, lactacidosis, and nonketotic hyperosmolar coma. Elevated cerebral glucose,42 significant excess of mI, a reversible accumulation of ‘Cho’, and the presence of ketone bodies have been detected in various patients with varying severity of diabetic encephalopathy.43

11.2 The Hyperosmolar State: Identification of Idiogenic Organic Osmolytes by Proton MRS

Lien et al. first used 1 H MRS to investigate a ‘new’ family of cerebral metabolites collectively known as organic osmolytes, because of their believed role in the maintenance of cerebral osmotic equilibrium.44 Such molecules were also first thoroughly researched in the papilla of the kidney with the help of in vitro NMR.45

6 HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS Day 4 NA

D

Exam 5-3

Exam 5-1 mI

Day 7

mI Cho

Cr

ay 12

Cho Cr

NA

Exam 5

Exam 5

Exam 1

Exam 3

Exam 5-2

NA

Exam 5

Versus day 36

Exam 2

inorganic phosphate, and Cr are obvious consequences; H+ accumulates, both due to failure of removal of CO2 and formation of lactate. Reduced partners of the equilibrium enzymes, lactate and glutamate dehydrogenase, accumulate. In practice, glutamate is probably equally rapidly converted to glutamine. Innumerable animal studies, using principally 31 P NMR, but also 1 H NMR, confirm these principles and document other previously unsuspected changes, notably the loss of NAA. The very common occurrence of hypoxic encephalopathy in humans has given ample opportunity for verification of these events in the human brain. ‘Recovery’ from nonlethal hypoxic encephalopathy gives yet another view of the metabolic process. From 31 P MRS in neonates,50 the expected changes emerged. More recently, 1 H MRS has been used to quantify and plot the time course of changes which occur after oxygen deprivation, applying the information to defining the degree of irreversible hypoxic neuronal damage—and hence prognosis. ‘Near-drowning’ is the term applied when virtually total oxygen deprivation occurs, due to submersion in water. It must be presumed that all ATP and PCr is lost, and all glucose

Figure 8 Time course of changes in principal cerebral organic osmolytes during correction of severe dehydration. A series of spectra were obtained from the same (occipital gray matter) brain location, in a 14-month-old child, recovering from severe dehydration and hypernatremia (plasma sodium 195 mEq l−1 ; normal range 135–142 mEq l−1 ) using identical acquisition conditions. Spectra were processed and scaled identically to permit subtraction of sequential spectra (difference spectroscopy). The day of examination refers to interval since admission to hospital. Examinations are numbered sequentially, from the first (Exam. 1) to last (Exam. 5) on day 36. Compared with a relevant normal [spectrum (b), Figure 6], the principal abnormality appears to be a reversal of the intensities of NAA (reduced) and mI (increased), so that mI dominates the spectrum. These changes slowly reverse to be nearly normal on day 36. The resultant difference spectra more clearly identify the progressively falling concentrations of several metabolites, with a possible elevation in the resonance peak assigned to the neuronal marker NAA. Quantitative MRS defines the principal abnormality as a threefold increase in the concentration of the cerebral osmolyte

First in sporadic cases of diabetic hyperosmolar coma43 and in a patient after closed head injury,46 and then most convincingly in a single infant with holoprosencephally and deficient thirst mechanisms, these concepts were confirmed as contributing to human encephalopathy by the use of 1 H MRS (Figure 8). mI was three times normal, and other resonances, also markedly affected, returned toward normal with treatment. Difference spectroscopy is particularly helpful in identifying these changes.47,48 The converse, or hypoosmolar state of hyponatremia has been identified by 1 H MRS in significant numbers of patients.49 11.3

Hypoxic Encephalopathy

The prime example of a systemically induced encephalopathy is that due to insufficient oxygenation of blood, with resultant failure of cerebral oxygen delivery. Hypoxic encephalopathy is best understood in the context of energy failure and altered redox state through all cerebral metabolic pathways. Loss of ATP and PCr, accumulation of ADP, AMP,

Cr

Lactate NAA

Cho mI Glx

Glx

Neardrowning (3 years)

Glutamine standard

b, g

a 4

3

2

1

0 ppm

Figure 9 Near-drowning with fatal outcome. Brain spectrum from occipital cortex of a severe near-drowning victim (3 years old), examined 48 hours post-immersion (top), and the glutamine standard (bottom). The patient died on the 5th day post-injury. The 1 H spectrum was acquired from a volume of 12 cm3 (STEAM TR 3.0 s, TE 30 ms, data processing including correction for residual water, line fitting, and quantification as described by Kreis et al.).46 The spectrum of the glutamine standard (10 mM) was acquired in the same way, and scaled appropriately. Notable abnormalities in the patient spectrum are the very much reduced NAA/Cr ratio (and [NAA] concentration), excess of lactate (doublet at 1.3 ppm), and of the strongly coupled resonances of α, β, and γ glutamine protons. A 25% reduction in [Cr] was apparent on quantitative examination. (Thanks to Dr. Roland Kreis, Thomas Ernst, and Edgardo Arcinue MD)

HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS

converts to lactate in this period of anoxia. Yet at the first clinical examinations, 24 hours after rescue and artificial life support, MRS demonstrates more often than not the absence of lactate, normal Cr (plus PCr), and NAA at nearly full concentration. Only subsequently do NAA and total Cr fall and lactate appear. The severity of the changes gives a fairly good guide to outcome.51 The classic events of anoxia described by Lowry52 are probably reversible, but secondary damage results in progressive cell death, the consequences of which are loss of NAA (in the case of dying neurons), loss of PCr and oxidative function and reaccumulation of lactate. It is unclear why the large accumulations of glutamine occur, or what the consequences are for survival (Figure 9).

12 CONCLUSIONS

Diffuse metabolic changes in brain biochemistry are the result of complex interactions of disordered biochemistry in many other organs and tissues. Hepatic encephalopathies, diabetic coma and hyperosmolar states and hypoxic encephalopathy are examples of such conditions in which accurate application of quantitative NMR spectroscopy sheds new light. Diagnostic utility of MRS in metabolic encephalopathies will increase as new therapeutic options are developed.

13 RELATED ARTICLES

Animal Methods in MRS; Brain MRS of Human Subjects; In Vivo Hepatic MRS of Humans; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Water Suppression in Proton MRS of Humans and Animals.

14 REFERENCES 1. F. Plum and J. B. Posner, The Diagnosis of Stupor and Coma, Davis, Philadelphia, 1986. 2. S. Sherlock, W. H. J. Summerskill, L. P. White, and E. A. Phear, Lancet, 1954, 2, 453. 3. S. Bessman and A. Bessman, J. Clin. Invest., 1955, 34, 622. 4. S Bessman and N. Pal, in The Urea Cycle, eds. S. Grisolia, R. Baguena, and F. Mayor, Wiley, Chichester, 1976, p. 83. 5. H. A. Krebs and K. Henseleit, Hoppe-Seyler’s Z. Physiol. Chem., 1932, 210, 33. 6. A. Geissler, K. Kanamori, and B. D. Ross, Biochem. J., 1992, 287, 813. 7. B. T. Hourani, E. M. Hamlin, and T. B. Reynolds, Arch. Intern. Med., 1971, 127, 1033. 8. M. D. Norenberg and A. Martinez-Hernandez, Brain Res., 1979, 161, 303. 9. J-C Reubi, C. van den Berg, and M. Cuenod, Neurosci. Lett., 1978, 10, 171. 10. R. F. Butterworth and G. P. Layrargues, Hepatic Encephalopathy: Pathophysiology and Treatment, Humana Press, Clifton, 1989. 11. L. Zieve in Diseases of the Liver, eds. L. Schiff and E. R. Schiff, Lippincott, Philadelphia, 1987, p. 925. 12. R. Kreis, N. A Farrow, and B. D. Ross, Lancet, 1990, 336, 635.

7

13. N. E. P. Deutz, A. A. de Graaf, J. B. de Haan, W. M. M. J. Bovee and R. A. F. M. Chamuleau, J. Hepatol., 1987, 4(1), S13. 14. T. E. Bates, S. R. Williams, R. A. Kauppinen, and D. G. Gadian, J. Neurochem., 1989, 53, 102. 15. S. M. Fitzpatrick, H. P. Hetherington, K. L. Behar, and R. G. Shulman, J. Cereb. Blood Flow Metab., 1990, 10, 170. 16. K. Kanamori, F. Parivar, and B. D. Ross, NMR Biomed., 1993, 6, 21. 17. K. Kanamori and B. D. Ross, Biochem. J., 1993, 293, 461. 18. R. A. Moats, Y-H. H. Lien, D. Filippi, and B. D. Ross, Biochem. J., 1993, 295, 15. 19. H. Bruhn, J. Frahm, T. Michaelis, K.-D Merboldt, W. H¨anicke, M. L. Gyngell, P. Brunner, J. Frohlich, D. Haussinger, P. Schauder, and B.D. Ross, Hepatology, 1991, 14, 121. 20. R. Kreis, N. A. Farrow, and B. D. Ross, NMR Biomed., 1991, 4, 109. 21. R. Kreis, B. D. Ross, N. A. Farrow, and Z. Ackerman, Radiology, 1992, 182, 19. 22. H. Bruhn, K.-D. Merboldt, T. Michaelis, M. L. Gyngell, W. Hanicke, J. Frahm, P. Schauder, K. Held, G. Brunner, J. Frolich, D. Haussinger, and B. D. Ross. Soc. Magn. Reson. Med., 1991, 10, 400. 23. J. R. McConnell, C. S. Ong, W. K. Chu, M. F. Sorrell, B. W. Shaw, and R. K. Zetterman, ‘11th Annual Meeting of the Society of Magnetic Resonance in Medicine’, Berlin, WIP, 1992 , p. 1957. 24. R. A. F. M. Chamuleau, D. K. Bosman, W. M. M. J. Bovee, P. R. Luyten, and J.A. den Hollander, NMR Biomed., 1991, 4, 103. 25. S. Schenker, K. J. Breen, and A. M. Hoyumpa, Gastroenterology, 1974, 66 121. 26. B. D. Ross, M. R. Morgan, I. J. Cox, K. E. Hawley, and I. R. Young, J. Cereb. Blood Flow Metab., 1987, 7, 5396. 27. B. D. Ross, J. P. Roberts, J. Tropp, K. Derby, N. Bass, and C. Hawryszko, Magn. Reson. Imag., 1989, 7, 82. 28. P. R. Luyten, J. A. den Hollander, W. M. M. J. Bov´ee, B. D. Ross, D. K. Bosman, and R. A. F. M. Chamuleau, Proceedings, Society of Magnetic Resonance in Medicine, Amsterdam, 1989 , 8, 375. 29. L. Barbara, B. Barbiroli, S. Gaiani, L. Bolondi, S. Sofia, G. Zironi, R. Lodi, S. Iotti, P. Zaniol, C. Sama, and S. Brillanti. Eur. J. Hepatol., 1993 , 2, 60. 30. S. Taylor-Robinson, R. J. Mallalieu, J. Sargentoni, J. D. Bell, D. J. Bryant, G. A. Coutts, and M. Y. Morgan, Proceedings, Society of Magnetic Resonance in Medicine, New York, 1993 , 12, 89. 31. A. Geissler, N. Farrow, F. Villamil, L. Makowka, T. Ernst, R. Kreis, and B. Ross, Annual Meeting of Society of Magnetic Resonance in Medicine, Berlin, 1992 , 11, 647. 32. G. F. Mason, R. Gruetter, D. L. Rothman, K. L. Behar, R. G. Shulman, and E. J. Novotny J. Neurochem., 1994, in press. 33. R. Gruetter, D. L. Rothman, E. J. Novotny, and R. G. Shulman, Magn. Reson. Med., 1992, 25, 204. 34. B. D. Ross, S. Jacobson, F. G. Villamil, R. A. Moats, T. Shonk, and J. Draguesku, Hepatology, 1993, 18, 105A. 35. B. D. Ross, S. Jacobson, F. Villamil, J. Korula, R. Kreis, T. Ernst, T. Shonk, and R.A. Moats, Radiology, 1994, 193, 457. 36. T. Ernst, B. D. Ross, and R. Flores, Lancet, 1992, 340, 486. 37. R. K Gupta, V. A Saraswat, H. Poptani, R. K. Dhiman, A. Kohli, R. B Gujral, and S. R. Naik, Am. J. Gastroent., 1993, 88, 670. 38. B. D. Ross, T. Shonk, R. A. Moats, S. Jacobson, J. Draguesku, T. Ernst, J. H. Lee, and R. Kreis Proceedings, Society of Magnetic Resonance in Medicine, New York, 1993 , 12, 131. 39. T. Shonk, R. Moats, J. H. Lee, J. Korula, T. Ernst, R. Kreis, J. Draguesku, and B. D. Ross, Gastroenterology, 1993, 104, A-449, 1793.

8 HEPATIC AND OTHER SYSTEMICALLY INDUCED ENCEPHALOPATHIES: APPLICATIONS OF MRS 39a. J. Korula, D. Kravetz, M. Katz, T. Shonk, S. Hanks, and B. D. Ross, Hepatology, 1993, 18, 282A, 903. 40. D. G. Gadian, A. Connelly, J. H. Cross, S. Burns, R. A Iles, and J. V. Leonard, Annual Meeting of the Society of Magnetic Resonance in Medicine, San Francisco, 1991 , 10, 193. 41. B. D Ross, R. Kreis, and T. Ernst, Eur. J. Radiol., 1992, 14, 128. 42. R. A. Hawkins, J. Jessy, A. M. Mans, and M. R. De Joseph, J. Neurochem., 1993, 60, 1000. 43. H. Bruhn, T. Michaelis, K-D. Merboldt, W. Hanicke, M. L. Gyngell, and J. Frahm, Lancet, 1991, 337, 745. 44. R. Kreis, and B. D. Ross, Radiology, 1992, 184, 123. 45. Y-H. H. Lien, J. I. Shapiro, and L. Chan, J. Clin. Invest., 1990, 85, 1427. 46. G. G. Wong D.Phil. Thesis, University of Oxford, 1981. 47. R. Kreis, T. Ernst, and B. D. Ross, J. Magn. Reson., 1993, 102, 9. 48. J. H. Lee and B. D. Ross, Proceedings, Society of Magnetic Resonance in Medicine, New York, 1993 , 12, 1553. 49. J. H. Lee, E. Arcinue, and B. D. Ross, N. Eng. J. Med., 1994, 331, 439. 50. J. S. Videen, T. Michaelis, P. Pinto, and B. D. Ross, J. Clin. Invest., 1995, 95, 788. 51. P. L. Hope, E. B. Cady, P. S. Tofts, P. A. Hamilton, A. M. de Costello, D. T. Delpy, A. Chu, E. O. R. Reynolds, and D. R. Wilkie, Lancet, 1984, 366. 52. R. Kreis, T. Ernst, E. Arcinue, R. Flores, and B. D. Ross, Society of Magnetic Resonance in Medicine, Berlin, 1992 , 11, 237.

53. O. H. Lowry in Neurology of the Newborn, ed. J. J. Volpe, Saunders, Philadelphia, 1987, p 33.

Acknowledgments Work reported was largely funded by the L.K. Whittier Foundation of California, the Norris Foundation, the Richard M. Lucas Cancer Foundation (1, 2, 3 and 13) and by funds from the HMRI MRS Program. BDR is grateful to the following colleagues: T. Ernst, N. Farrow, A. Geissler, K. Kanamori, R. Kreis, J. Mandigo-Bellinger, T. Michaelis, R.A. Moats, E. Rubaek-Danielsen, T. Shonk, J. Videen, and J.D. Roberts.

Biographical Sketch Brian D. Ross. b 1938; B.Sc., 1958, University College, London. D.Phil., 1966, University of Oxford. M.B., 1961, University College Hospital, London. F.R.C.S., 1973, Royal College of Surgeons, London. M.R.C.Path., 1976, Royal College of Pathologists, London. 1989, F.R.C.Path. University of Oxford lecturer, Metabolic Medicine. Director, Renal Metabolism Unit and Consultant Chemical Pathologist, Radcliffe Infirmary, Oxford, 1976–84. Director of Clinical Spectroscopy Programs at Radcliffe Infirmary, Oxford, 1981–84, Hammersmith Hospital, London, 1986–88, and Huntington Medical Research Institutes, Pasadena, CA, 1986–present. Visiting Associate, California Institute of Technology, 1986–present.

IN VIVO HEPATIC MRS OF HUMANS

In Vivo Hepatic MRS of Humans Isobel Jane Cox Imperial College School of Medicine, London, UK

1 INTRODUCTION The liver is the largest organ in the human body. By virtue of its anatomical position, the liver is readily suited to study by in vivo magnetic resonance spectroscopy (MRS) (Whole Body Studies: Impact of MRS). The liver is the principal homoeostatic organ in the body, being responsible for the metabolism of carbohydrates, fats, and circulating proteins and the detoxi®cation of the body's waste products. It is the most important site for the metabolism of drugs and alcohol. Bile is produced in the liver, which is important for the digestion of fats in the gut and which also acts as a transport medium for the excretion of bilirubin and certain drugs. Information about a number of these processes can be obtained by multinuclear MRS study. For example, the phosphorus spectrum provides an assessment of hepatic energy state, membrane turnover, levels of endoplasmic reticulum and glycolytic/gluconeogenic intermediates. Signals from glycogen and lipids can be measured using natural abundance 13C MRS. Fluorine-19 MRS can be used to detect metabolites of ¯uorinated drugs. Proton MRS provides a measure of hepatic water and fat content, but it has proved very dif®cult to obtain a metabolite spectrum because of the effects of respiratory motion. Clinical hepatic MRS builds on experience from the study of cell lines (see Cells and Cell Systems MRS) and animal models (see Animal Methods in MRS). The majority of human applications to date have used 31P MRS as this is technically the easiest, and the data obtained have been used to aid diagnoses, monitor treatment effects, and to study hepatic biochemistry.

2.1.1

1

The Phosphomonoester Resonance

The peak assigned to phosphomonoesters (PMEs) contains at least 10 components,1 including: glucose 6-phosphate (G6P), glycerol 3-phosphate (G3P), glycerol 1-phosphate (PG), and ribose 5-phosphate, which are intermediates in carbohydrate metabolism; coenzyme A, which is important in the metabolism of fatty acyl moieties; phosphoethanolamine (PE) and phosphocholine (PCh), which are metabolites on the pathway of membrane synthesis; adenosine monophosphate (AMP), which is an intermediate of adenosine triphosphate (ATP) and ADP turnover; and 2,3-diphosphoglycerate (2,3-DPG) which is present in blood, although it is unclear how much 2,3-DPG in circulating blood contributes to the in vivo spectrum. 2.1.2

The Inorganic Phosphate Resonance

The Pi signal is thought to represent about 40% of intracellular levels of Pi.2 It is unclear why the remainder is invisible, but may be bound in the mitochondria. Pi, together with ATP and ADP, play a key role in energy metabolism, and changes in hepatic energy state may therefore be re¯ected by an alteration in the ATP/Pi ratio. The chemical shift of Pi in liver cells is dependent on pH, and cytosolic pH has been measured to be 7.18 ‹ 0.03 in perfused rat liver using microelectrodes.3 In humans, the pH has been found to be consistent with this, and examples of values from healthy adult volunteers are: 7.18 ‹ 0.08,4 7.32 ‹ 0.02,5 7.36 ‹ 0.23,6 and 7.23 ‹ 0.14.7

ATP Pi GPE GPC

g

a b DPDE

PCh

2 THE NORMAL LIVER SPECTRUM 20

2.1 Phosphorus-31 MRS The phosphorus liver spectrum from a healthy adult volunteer comprises six major peaks, assigned to phosphomonoesters (PMEs), inorganic phosphate (Pi), phosphodiesters (PDEs) and the three phosphate groups (, , and ) in nucleotide triphosphates (NTPs) (Figure 1). Notably absent from this spectrum are signals from membrane phospholipids such as phosphatidylcholine. The phosphorus nuclei in such large molecules have reduced mobility and the signals are NMR invisible. Similarly, the signal from adenosine diphosphate (ADP) bound to proteins is not detectable using NMR methods, and therefore the small ADP signal observed only represents free ADP.

15

10

5

0

–5

–10

–15

–20

–25 ppm

Figure 1 Proton decoupled localized 31P NMR spectrum of the human liver, obtained at 1.5 T, using a 14 cm diameter surface coil. The localization scheme was ISIS with frequency modulated inversion and excitation pulses. The spectrum was obtained in 280 scans with 3 s repetition time. Proton decoupling has improved resolution in the phosphodiester, phosphomonoester, and diphosphodiester regions. Well-resolved resonances are observed for glycerophosphoethanolamine (GPE), glycerophosphocholine (GPC), and phosphocholine (PCh). DPDE, diphosphodiester. (Reproduced by permission of Heydon & Son from P. R. Luyten, G. Bruntink, F. M. Sloff, J. W. A. H. Vermeulen, J. I. van der Heijden, J. A. den Hollander, and A. Heerschap, NMR Biomed., 1989, 4, 177)

2 IN VIVO HEPATIC MRS OF HUMANS 2.1.3

The Phosphodiester Resonance

The phosphodiester (PDE) signal contains contributions from at least three water-soluble metabolites,1 including glycerophosphoethanolamine (GPE) and glycerophosphocholine (GPC). GPE and GPC are intermediates in phospholipid catabolism. At the ®eld strengths used for clinical studies, a broad, ®eld-dependent signal has also been identi®ed, which has been assigned to phospholipid bilayer, including endoplasmic reticulum, with a small contribution from motionally averaged macromolecules.8,9 In addition, phosphoenolpyruvate (PEP), although not a PDE, resonates in this region. 2.1.4

The Nucleotide Triphosphate Resonances

The dominant contribution to the nucleotide triphosphate (NTP) peaks are from ATP, but 10±20% of the peak area may be due to guanosine triphosphate (GTP) and uridine triphosphate (UTP). Further signals can be identi®ed, overlapping with peaks from the three phosphate groups. For example, the NTP resonance overlaps with signals from ADP and nicotinamide adenosine dinucleotide (NAD) or its reduced form (NADH). The NTP signal overlaps with signals from ADP. It is interesting that the majority of ADP is NMR invisible, as discussed above. NAD and NADH play a pivotal role in electron transport in redox reactions, but it remains unclear what proportion is bound to proteins and is therefore not detected by in vivo MRS. Binding to magnesium affects the chemical shifts of ATP, and therefore 31P MRS provides information about the ratio of free ATP to magnesium-bound ATP.4 This, in turn, re¯ects the intracellular concentration of free magnesium.10 Oberhaensli et al.4 measured the separation of ATP and ATP to be 13.70 ‹ 0.07 ppm, and estimated that 86% of the cytosolic ATP pool was magnesium bound, corresponding to a free cytosolic magnesium concentration of approximately 300 mol Lÿ1. This result is consistent with the estimated free magnesium concentration of 450 mol Lÿ1 in the perfused rat liver.11 2.1.5

Age-Related Changes

The hepatic phosphorus spectrum from neonates and infants differs from the adult liver, in that the signal assigned to PMEs is elevated compared with that for ATP, and that for PDE/ATP is reduced.12 The chemical shift of the PME signal is 6.8 ‹ 0.1 ppm, suggesting that PE is a major component. This would be consistent with the ®ndings from neonatal brain, in which an elevated PME level has been attributed to PE. 2.2

1

H MRS

The 1H spectrum is dominated by signals from water and lipid. While this spectrum can be used to evaluate the degree of fatty in®ltration, for example, both resonances need to be suppressed in order to observe metabolite signals. There has been one early paper describing high-resolution in vivo hepatic MRS, tentatively assigning the minor peaks to carnitine, taurine, glutamate, and glutamine.13 Further in vivo studies have not been forthcoming, because of the problems of effective water and lipid suppression. However, the potential of 1H MRS is illustrated by in vitro MRS studies of human tissue,1 in which resonances from choline-containing groups, creatine, acetate, glutamine/glutamate, and glycogen have been assigned

to the main resonances in the aliphatic region. Less intense signals from threonine, succinate, alanine, citrate, glycine, aspartate, and taurine have also been observed. The aromatic region shows resonances from a number of metabolites including NTP, nucleotide diphosphate (NDP), AMP, NAD and NADH. 2.3

13

C MRS

The natural abundance of 13C is only 1.1%. Therefore the only endogenous metabolites detectable in the liver by natural abundance 13C MRS are storage compounds, such as triglycerides and glycogen, which can achieve high intracellular concentrations. Fats are insoluble in water and are transported in the plasma as protein±lipid complexes (lipoproteins). The liver plays a major role in the metabolism of lipoproteins, synthesizing very low density proteins (VLDLs) and high density lipoproteins (HDLs). The liver and kidney are the major sites of HDL catabolism; low density lipoproteins (LDLs) are degraded by the liver after uptake by speci®c cell receptors. Triglycerides may be of dietary origin, but are also formed in the liver from circulating free fatty acids and glycerol and incorporated into VLDLs. There have been no studies to date interpreting the liver lipid pro®le with different diet or in disease. This re¯ects the fact that the concentrations of bulk triglycerides do not change suf®ciently rapidly with time for short-term studies of fat metabolism to be carried out by this means. On the other hand, glucose homeostasis and the maintenance of the blood sugar is an important function of the liver. In the fasting state, blood glucose is maintained either by glucose released from the breakdown of glycogen (glycogenolysis) or by newly synthesized glucose (gluconeogenesis). The relative contributions of these processes to glucose production have been dif®cult to quantify in humans, and therefore a noninvasive measurement at multiple time points of hepatic glycogen levels using 13C MRS is of considerable importance. The chemical shift range of the 13C spectrum is 200 ppm, and the hepatic spectrum is dominated by resonances from saturated groups (±*CH3 and ±*CH2±) from the fatty acyl chain at 15±35 ppm, unsaturated groups (±H*C=CH±) from the fatty acyl chain at 128±131 ppm, and carbonyl groups (±*CO±) from the fatty acyl chain at 172 ppm.14 The signal from C-1 of glycogen resonates at 100.5 ppm, away from other resonances, and is therefore the most easily detected peak from glycogen. Smaller resonances from ethanolamine, choline, glucose, and carbonyl groups in proteins and phospholipids are observed in high-resolution spectra of rat liver.14 While signals from triacylglycerides dominate the in vivo spectrum, often obliterating signals from other compounds resonating nearby, the majority of applications of hepatic 13C MRS have concentrated on the detection and quantitation of glycogen levels.15±19 The relaxation times of glycogen are short; the T1 is 240 ms and the T2 is 30 ms,20 so that rapid repetition times can be used. The sensitivity of measurement of the glycogen signal can be improved by proton decoupling21 or polarization transfer techniques.22 Signals from super®cial lipids may need to be suppressed, for example by using shaped rf pulses, incorporation of an inversion pulse parallel to the coil axis at the sample surface,17 or by exploiting the characteristics of the surface transmitter±receiver coil.19

IN VIVO HEPATIC MRS OF HUMANS

2.4

19

F MRS

The body does not naturally contain 19F and therefore there are no signals in the hepatic 19F MRS spectrum of the normal liver. However, signals from 19F-containing drugs or their metabolites can be detected, for example 5-¯uorouracil (5FU), enabling information about the time course and metabolic pathways of the drugs to be obtained.23

3 METHODOLOGY AND QUANTIFICATION 3.1 Field Strength The majority of studies have been undertaken at ®eld strengths of around 1.5 T. Boska et al. have formally compared the quality of hepatic 31P MRS spectra at 1.5 T and 2.0 T.24 The signal-to-noise ratio (S/N) of a localized hepatic 31P MRS spectrum improved by 32 ‹ 10%, which is comparable with the improvement by 33 ‹ 13% in the S/N of a spectrum from a loading phantom. Hepatic spectra have been reported by two groups using 4 T systems, with the aim of illustrating the increased sensitivity and spectral dispersion obtained at higher ®eld strengths.25,26 The hepatic 31P MRS spectrum showed that resonances from PME, Pi, and PDE were well resolved, the two peaks from GPC and GPE in the PDE region could be distinguished, and resonances from nicotinamide adenine dinucleotide (NAD) and its phosphate (NADP) were seen as a shoulder on the NTP/ NDP peak.26 The C-1 resonance from glycogen was detected at 4 T with an acceptable S/N in an acquisition time of 12±1525 or 20 min.26

degrades spectral resolution30 and is a problem common to all nuclei. The presence of paramagnetics broadens the water linewidth, for example, compared with the linewidth in the brain and muscle. However, reasonable linewidths can be achieved in the hepatic phosphorus spectrum, and linewidths of 11 Hz have been measured at 1.6 T for Pi.31 Another problem is the varying contribution of signals from overlying tissues, for example, subcutaneous lipid and posterior and anterior muscle. In 31P MRS the contributions from muscle can be separated from the liver spectrum using the phosphocreatine signal as a marker for muscle. In both 13C and 1H MRS, contributions from subcutaneous lipid may be a serious problem, particularly if a surface coil is used, since these lipid signals are relatively intense and can mask the very much weaker metabolite signals. Other factors need to be considered for hepatic 31P MRS. Many of the signals are multicomponent, and the limited chemical shift dispersion of in vivo 31P MRS spectra is a major cause of poor spectral resolution. Interpretation of a change in signal intensity to a speci®c metabolite or metabolites is often dif®cult without recourse to animal models or biopsy studies. The spectral resolution in the 31P spectrum can be improved by 31 P±1H decoupling,5 since the three-bond coupling constant in, for example, PE, GPC, 2,3-DPG, and NAD, may become a dominant factor in determining the residual linewidth of these signals (see Proton Decoupling During In Vivo Whole Body Phosphorus MRS). Finally, the relaxation times T1 and T2 are dif®cult to characterize for each individual study because the T1 values can be long and there is limited time available for a clinical examination. The overall T1 of the multicomponent signals does vary in some pathologies,32 so it is not always possible to extrapolate from values obtained from healthy control subjects.

3.4 3.2 Localization In the majority of studies, localization to the liver has been achieved by a suitable choice of rf coils combined with localization techniques. By virtue of the position of the liver, a surface receiver coil has almost always been used. The transmitter coil has been the same or another surface coil (producing a nonuniform B1 ®eld)4,6 or a saddle-shaped transmit coil (producing a uniform B1 ®eld).7 Additional localization has generally been achieved using single volume methods, such as ISIS, or multivoxel techniques such as chemical shift imaging or rotating frame zeugmatography.27 Choice of technique will depend to some extent on the clinical problem and on the timescale of spectral acquisition.

3.3 Quanti®cation (see Quantitation in In Vivo MRS) It is dif®cult to realize the full potential of the technique because a number of factors limit the accuracy with which absolute quantitation can be achieved in hepatic MRS. Nevertheless, the majority of centers have provided some measure of absolute quantitation in both hepatic 31P and 13C MRS.6,17,28,29 Factors to be borne in mind include the effect of respiratory motion, which causes ghosting of signals and

3

Relaxation Parameters in Hepatic

31

P MRS

A number of studies have speci®cally addressed the issue of T1 measurement in healthy subjects. For example, Blackledge et al.33 used an inversion±recovery depth-selection sequence based on rotating frame zeugmatography methods to measure T1 values of the liver. They measured the T1 of ATP to be 0.33 ‹ 0.05 s, that of PME as 0.74 ‹ 0.10 s, and that of Pi as 0.44 ‹ 0.06 s. The PDE peak appeared to exhibit heterogeneity in T1, with a fast component (T1 < 200 ms) and a slowly recovering component (T1 > 1 s). Due to off-resonance effects, Blackledge et al. could not quote values for other peaks. Buchthal et al.34 used progressive saturation techniques, in combination with a surface coil and one-dimensional phase encoding. A uniform 90 pulse was achieved by using an adiabatic half passage pulse that compensated for the spatially nonuniform rf ®eld of the surface coil. The mean T1 values from seven individuals were: Pi, 0.41 ‹ 0.1 s; PDE, 1.4 ‹ 0.13 s; and ATP, 0.68 ‹ 0.09 s. Meyerhoff et al.6 used a fast inversion±recovery sequence, combined with ISIS, to measure T1 relaxation values from four healthy volunteers. The results were: PME, 0.84 ‹ 0.26 s; Pi, 0.97 ‹ 0.25 s; PDE, 1.36 ‹ 0.37 s; ATP, 0.35 ‹ 0.06 s; ATP, 0.46 ‹ 0.09 s; and ATP, 0.35 ‹ 0.05 s. Cox et al.32 estimated T1 values, from a comparison of signal intensities at TR 0.5 and 5 s, to be: PME, up to 3 s; PDE, 3±6 s; Pi, 1 s; and ATP, 0.7 s.

4 IN VIVO HEPATIC MRS OF HUMANS 4 MRS STUDIES IN PROTEIN, CARBOHYDRATE, AND LIPID METABOLISM IN THE NORMAL LIVER The complex interrelationships between protein, carbohydrate and fat metabolism are illustrated in Figure 2. Different aspects of these pathways can be probed measuring spectral changes after a metabolic challenge. The formation of glycogen has been followed under different conditions, for example after a long fast,17 an intravenous infusion of [1-13C]glucose under hyperglycemic and hyperinsulinemic clamp conditions19 and an oral intake of glucose in the form of bolus,18,19 Rothman et al.17 serially measured hepatic glycogen concentrations using 13C MRS at 3±12 h intervals during a 68 h fast. The liver volume was determined by MRI. Net hepatic glycogenolysis was calculated by multiplying the rate of glycogen breakdown by the liver volume. The net rate of gluconeogenesis was calculated by subtracting the rate of net hepatic glycogeneolysis from the rate of glucose production in the whole body measured with tritiated glucose. Even in the ®rst 22 h of fast they found that gluconeogenesis accounted for a substantial fraction of total glucose production (64 ‹ 5%) and this increased to 82 ‹ 5% (22±46 h of fasting) and 96 ‹ 1% (46±64 h of fasting). These important results suggest that hepatic gluconeogenesis is always operating at an appreciable rate in humans. Ishihara et al.18 measured glycogen and glucose resonances in one subject, after a bolus of glucose (100 g) following a 20 h fast, and also after administration of 13C enriched (99%) [1-13C]D-glucose (1 g) mixed with 75 g glucose. Beckmann et al.19 followed the formation of glycogen in the liver of normal volunteers after an intravenous infusion of

CO2

Urea cycle

Urea

From diet: Ketone bodies Krebs cycle Fatty acids

Amino acids

Acetyl-CoA

Fatty acids

Triglycerides Phospholipids

Cholesterol

Lipoprotein

Pyruvate

(Dietary and tissue protein) Lactate

Glucose

Glucose 6-phosphate

Glucose release

(Dietary carbohydrate) Glycogen

Figure 2 Interrelationships of protein, carbohydrate and lipid metabolism in the liver. (Reprinted by permission of BaillieÁre Tindall from P. J. Kumar and M. L. Clark, (eds) `Clinical Medicine', 1987, p. 212)

[1-13C]glucose under hyperglycemic and hyperinsulinemic clamp conditions and an oral intake of glucose in the form of bolus. They found that changes in the glycogen signal correlated well with the time course of insulin and glucagon during the spectral measurement. They showed that liver glycogen formation in man can be followed using nonlabeled glucose or [1-13C]glucose with a low level of enrichment (16.6%). The use of nonlabeled glucose has the advantage that quantitation of net liver glycogen synthesis is simpli®ed because label dilution through the various metabolic pathways of glucose is avoided. Measurements of glucose uptake, estimated from the increase in the glycogen signal, was consistent with ®ndings from more complex and invasive studies of glucose uptake in the liver. The average liver glycogen concentration after a 12 h overnight fast in 18 volunteers without any dietary preparation was estimated to be 229 ‹ 34 mM. Of the physiological precursors for hepatic gluconeogenesis, the amino acid L-alanine is of special importance since it is a key protein-derived gluconeogenic precursor. It is cleared rapidly by the liver and has a half-life of only 30 min in the plasma of healthy subjects. After an overnight fast the plasma concentration of glucose falls and glycolysis in the liver is inhibited while gluconeogenesis is stimulated. Infusion of Lalanine induces rapid and consistent changes in 31P MRS spectra (Figure 3).35 A marked change with a clear dose±response relationship was observed for PME/ATP (maximal change +98%) and Pi/ATP (ÿ33%), while smaller changes were demonstrated for PDE/ATP (+24%) which were independent of the alanine dose. Levels of ATP did not change, suggesting there was no change in phosphorylation status. In vitro MRS spectra obtained from animal models after alanine infusion showed marked increases in the gluconeogenic intermediates, 3-phosphoglycerate (PME region) and phosphoenolpyruvate (PDE region). Since the plasma glucose concentrations in humans were unaltered following the alanine infusion, MRS spectral changes have been interpreted as suggesting increased ¯ux through the gluconeogenic pathway, possibly with glucagon rather than glucose as the endpoint. Therefore this metabolic stress test constitutes a useful tool for use in studies of gluconeogenesis. Conventional liver function tests measure plasma clearance of the compounds metabolized by the liver, but these can be in¯uenced by extrahepatic factors such as renal excretion. Therefore a direct measure of liver function would be of value. One example studied in detail has been a measure of hepatic sugar metabolism following a rapid intravenous bolus injection of fructose,4,36,37 and the dose±response curves have been established.36 Fructose is a major source of carbohydrate in the Western diet and is metabolized mainly in the liver. It is rapidly metabolized to fructose 1-phosphate by fructokinase in the liver, causing a rapid decrease in ATP and Pi. Thus, following intravenous infusion of fructose (bolus injection of 200±250 mg kgÿ1) ®ndings using hepatic 31P MRS showed there was an approximately threefold increase in PME, predominantly fructose 1-phosphate, by 10 min and a return to basal values by 20±30 min. The hepatic concentration of Pi decreased by more than 80% immediately after fructose infusion, while the plasma Pi concentration dropped by only 20%. Pi levels rebounded and hepatic Pi concentration was increased to about three times the preinfusion values by 15 min. Thereafter, hepatic Pi levels gradually decreased, while plasma Pi

IN VIVO HEPATIC MRS OF HUMANS (a)

(e)

(b)

(f)

110

105

100

(g)

(c)

10 0 –10 –20 Chemical shift (ppm)

90 ppm

110

105

100

95

90 ppm

Figure 4 Typical 13C NMR spectrum of the C-1 position of liver glycogen from one control subject (left panel) and one type II diabetic patient (right panel) 4 h after the liquid meal. (Reproduced by permission of The American Society for Clinical Investigation, Inc., from I. Magnusson, D. L. Rothman, L. D. Katz, R. G. Shulman, and G. I. Shulman, J. Clin. Invest., 1992, 90, 1323)

(h)

(d)

95

5

10 0 –10 –20 Chemical shift (ppm)

Figure 3 Representative 31P NMR spectra of normal human liver before and after a bolus infusion of L-alanine 2.80 mmol kgÿ1 body weight. 31P NMR data were acquired using a two-dimensional pulse sequence with a repetition time of 1 s and a pulse angle of 45 (256 signal averages). Spectra shown are plotted on the same absolute scale (referenced to the highest ATP peak of the whole set of spectra), so that absolute peak areas of all spectra are comparable. Two baseline spectra were obtained (a and b). Time after start of L-alanine infusion (i.e. midtime-point of data collection referenced to start of infusion) for spectra (c±h): (c) 8, (d) 13, (e) 26, (f) 39, (g) 67, and (h) 103 min. (Reproduced by permission of the Biochemical Society and Portland Press from P. C. Dagnelie, D. K. Menon, I. J. Cox, J. D. Bell, J. Sargentoni, G. A. Coutts, J. Urenjak, and R. A. Iles, Clin. Sci., 1992, 83, 183)

increased. The changes in Pi were several times greater in liver cells than in blood plasma, suggesting that large concentration gradients had been generated across the liver cell membrane. Hepatic ATP levels showed a transient decrease by 60% after infusion of fructose, but after 60 min the ATP pool was still reduced by about 40%, and this was attributed to the breakdown of ATP to inosine monophosphate and, ultimately, to uric acid. These changes in PME, Pi, and ATP showed a linear dose dependence.36

5 DIFFUSE LIVER DISEASE 5.1 Diabetes Alterations in glucose±glycogen metabolism are important consequences for a number of diseases. For example, type II diabetes mellitus is characterized by fasting hyperglycemia and an excessive, prolonged rise in the plasma glucose concentration after glucose or meal ingestion. Magnusson et al.38 measured glycogenolysis and gluconeogenesis in seven type II diabetic subjects and ®ve control subjects during 23 h of fast-

ing. They found that increased gluconeogenesis was responsible for the increased whole body glucose production in type II diabetes mellitus after an overnight fast. In detail, 4 h after a meal liver glycogen concentration was lower in diabetics than in controls, 131 ‹ 20 versus 282 ‹ 60 mmol Lÿ1 liver (p < 0.05) (Figure 4). Net hepatic glycogenolysis was decreased in the diabetics, 1.3 ‹ 0.2 versus 2.8 ‹ 0.7 mol per (kg body wt  min), p < 0.05. Whole body glucose production was increased in the diabetics, 11.1 ‹ 0.6 versus 8.9 ‹ 0.5 mol per (kg body wt  min), p < 0.05. Gluconeogenesis was consequently increased in the diabetics, 9.8 ‹ 0.7 versus 6.1 ‹ 0.5 mol per (kg body wt  min), p < 0.01, and accounted for 88 ‹ 2% of total glucose production compared with 70 ‹ 6% in controls, p < 0.05. 5.2

Hereditary Fructose Intolerance

Hereditary fructose intolerance (HFI) is a rare autosomal recessive disorder. The effect of fructose on liver metabolism in patients with HFI and in heterozygotes for HFI has been studied by hepatic 31P MRS.39 In ®ve patients with HFI small amounts of fructose (1.5 g) were followed by an increase in sugar phosphates and a decrease in Pi, and it was suggested that hepatic 31P MRS could be used to diagnose fructose intolerance and to monitor patients' compliance with a fructoserestricted diet. In eight heterozygotes, ingestion of much larger amounts of fructose (50 g) led to an accumulation of sugar phosphates, a reduction of Pi, and a larger increase in plasma urate compared with control subjects. The effects were most pronounced in heterozygotes with gout, and it was suggested that heterozygosity for HFI may predispose to hyperuricemia. 5.3

Familial Gout

The hyperuricemia responsible for the development of gouty arthritis results from a wide range of environmental factors and, less commonly, from inborn errors of metabolism. As a continuation from the studies on homozygous and heterozygous patients with HFI, 11 volunteers with familial gout were examined with hepatic 31P MRS following a 50 g load of oral fructose.40 Spectral changes in response to the fructose load similar in magnitude to those observed in earlier studies of

6 IN VIVO HEPATIC MRS OF HUMANS obligate heterozygotes for HFI were found in 2 of the 11 patients with familial gout. In one family the index patient's three brothers and his mother all showed the fructose-induced abnormality of metabolism, in agreement with the maternal inheritance of the gout in this family group. The test dose of fructose used produced a signi®cantly larger increment in the concentration of serum urate in the patients showing changes on hepatic 31P MRS. The biochemical basis for the fructoseinduced increase in purine metabolism was discussed, and it was suggested that a lower activity of aldolase B prevailed in the two patients with hereditary gout who showed fructoseinduced abnormality of metabolism. It was suggested that in these two subjects the causal role for generating gout is the aberrant fructose metabolism, and indeed that a restriction of fructose ingestion is a possible approach to the clinical management of patients with this disorder. 5.4 Cirrhosis The clinical spectrum of cirrhosis varies widely, and adequate characterization of the functional state is therefore essential for making management decisions in these patients. At present, characterization usually depends on the assessment of indirect clinical and laboratory measurements of hepatic function. These parameters provide some measure of the functional grade of cirrhosis, but are subject to extrahepatic in¯uences that may reduce their value as markers of liver function. Quantitative estimates of liver function may also be obtained from various clearance studies, but individual tests may not provide a good picture of overall liver function. Several groups of workers have used hepatic 31P MRS to study subjects with alcoholic cirrhosis and chronic liver disease of other etiologies,28,29,41±45 and two studies have addressed hepatic 31P MRS in the context of varying functional severity of liver disease.43,45 The results from different centers are not consistent, re¯ecting differences in the acquisition parameters (and, therefore, the degree of quantitation) and also differences in clinical state of the patients. Patients with alcoholic liver disease were studied using image-guided hepatic 31P MRS methods, to yield an estimate of absolute molar concentrations of phosphorus metabolites in a 64±120 cm3 volume within the liver.41 The patient group included nine patients with alcoholic cirrhosis of varying severity. The pH value was more acidotic in the cirrhotic patient group (7.26 ‹ 0.2) compared with controls (7.44 ‹ 0.2) (p = 0.04). The chemical shifts of the other peaks were not signi®cantly different between the cirrhotic and control group. The metabolite ratios were similar in the control and cirrhotic group. However, absolute concentrations of metabolites were decreased by 13±50%. Except for the decrease in PME of 13%, all the decreases from values obtained in normal liver were signi®cant (p < 0.05). There was no signi®cant difference between the absolute values for cirrhotics and hepatitics. In a study of patients with alcohol liver disease, including ®ve patients with cirrhosis, Angus et al.42 found no change in any of the metabolite ratios in the patients with cirrhosis. There was no change in Pi/ATP, suggesting no evidence of impaired cellular energetics. The hepatic pH was also similar in patients with cirrhosis and in controls. This patient group had previously been found to have established cirrhosis, and had been abstinent from alcohol for at least 6 months.

Absolute concentrations were measured using one-dimensional chemical shift imaging (CSI) techniques in ®ve healthy volunteers and ®ve patients with alcoholic cirrhosis.29 Absolute metabolite levels were calculated with reference to an external standard. Metabolite ratios were not altered in cirrhosis compared with controls, although absolute concentrations of all hepatic metabolites tended to be lower. However, only the reduction in ATP, of 31%, was signi®cant. Histological evidence suggested that the reduction in ATP levels re¯ected fewer functioning liver cells per volume of liver, since functional cells had been destroyed and replaced by ®brosis. The amount of ATP per liver parenchymal cell was thought to be unchanged. The authors suggested that metabolite ratios were of limited diagnostic value in the assessment of alcoholic cirrhosis, if they were unsupported by quantitative analysis. Fourteen patients with liver cirrhosis of differing severity were examined using one-dimensional CSI methods.43 Patients were divided into two groups according to the severity of their liver disease using Child's classi®cation and the aminopyrine breath (AB) test. The PME/total phosphorus ratio was signi®cantly higher in patients with mild and severe cirrhosis. There was no change in the PME T1 value in cirrhotics. There was a signi®cant negative linear correlation of PME with percentage dose of 14CO2 excreted over 2 h in the AB test. The pH values were signi®cantly elevated in mild cirrhosis (pH 7.45), but not in severe cirrhosis (pH 7.36), compared with controls (pH 7.29). This was the ®rst paper highlighting the clinical potential of hepatic 31P MRS as a noninvasive means of assessing the severity of liver cirrhosis. In a study of a group of 86 patients with histologically proven cirrhosis of varying etiology and functional grade (Menon et al.45), patients with liver disease showed a signi®cantly higher median PME/ATP (p < 0.0001), PME/PDE (p < 0.0001), PME signal height ratio (SHR) (p < 0.0001), and Pi SHR (p < 0.02), and a lower median PDE/ATP (p < 0.001) and PDE SHR (p < 0.001) (the SHR was obtained by dividing the peak height at TR 5 s by that at TR 0.5 s to yield a T1related SHR). The magnitude of these changes signi®cantly and progressively increased with worsening functional state (Figure 5).

Child's C

PME

Child's A

PDE

PDE Healthy control

Child's B

+20 +10

0

–10 –20 ppm

PCr a

Pi PME

+20 +10

g

0

bATP

–10 –20 ppm

Figure 5 Hepatic 31P MR spectra collected using a two-dimensional CSI sequence with TR Tp-5 s, pulse angle 45 from a healthy volunteer, and patients with Child's A, B and C functional grades of alcoholic cirrhosis

IN VIVO HEPATIC MRS OF HUMANS

Etiological differences were noted in patients with compensated cirrhosis. Spectra from patient with cirrhosis secondary to viral hepatitis showed a signi®cantly higher PME/ATP, Pi/ ATP, and PME/PDE, and those with cirrhosis secondary to primary sclerosing cholangitis showed a signi®cantly lower Pi/ ATP than did other etiological groups.45 However, spectral appearances did not vary with etiology of cirrhosis in decompensated patients. In vitro 31P MRS of perchloric extracts of biopsy samples of liver tissue obtained from 10 patients with cirrhosis at transplant hepatectomy showed increases in levels of PE and PC, and a reduction in levels of GPE and GPC. These changes suggest regenerative activity in cirrhotic livers. The reduction in soluble PDE in the aqueous extracts did not quantitatively account for the reduced PDE resonance seen in vivo, and the changes seen in vivo may therefore be partly due to reduction in contributions from hepatic endoplasmic reticulum, resulting from replacement of hepatocytes by ®brous tissue. In order to determine the feasibility and utility of dynamic hepatic 31P MRS, Dufour et al. studied six healthy subjects and nine patients with nonalcoholic cirrhosis after an intravenous infusion of a fructose load (250 mg kgÿ1).44 The basal spectra between the healthy and cirrhotic subjects had comparable metabolite areas, except that the contribution of PDE was signi®cantly smaller in cirrhotic patients than in healthy subjects (33 ‹ 5% versus 38 ‹ 5%; p < 0.05). In both groups, the relative PME peak increased and then returned to basal values, Pi decreased then rapidly increased and overshot basal values and slowly returned to basal values, and ATP decreased and slowly returned to basal values. However, in the cirrhotic group the magnitude of these changes were reduced; PME increased to 9 ‹ 5% versus 20 ‹ 8% in controls, Pi decreased by 5 ‹ 4% versus 11 ‹ 3% of total area (p < 0.005). These measurements correlated with the severity of the impairment of liver function measured by galactose-elimination capacity. 5.5 Alcoholic Hepatitis As with studies on alcoholic cirrhosis the hepatic MRS ®ndings varied according to the MRS technique used, the degree of quantitation achieved and the clinical state of the patient. Meyerhoff et al.41 found that in 10 patients absolute concentrations of metabolites were decreased by 25±44%, but there was no change in metabolite ratios. Also, intracellular pH was more alkaline in the patient group compared with controls. Angus et al.42 measured the metabolite ratios to be abnormal in 16 patients with hepatitis, none of whom had consumed alcohol within the previous 72 h. PME/Pi and PME/ATP were elevated, and the PME level showed a signi®cant positive correlation with the severity of alcoholic hepatitis, assessed by histology. Pi/ATP and pH were similar to control values. These data were similar to ®ndings from four other patients previously studied by the same group.28 5.6 Viral Hepatitis A number of patients with viral hepatitis have been studied, generally as a subgroup of a study of a heterogeneous patient group. Elevated PME/ATP has been reported in ®ve patients with acute viral hepatitis,28 three patients with viral hepatitis B,41 and one patient with non-A non-B hepatitis.7 Oberhaensli

7

et al. found that the high phosphomonoesters returned to normal as liver function became normal, suggesting they were probably associated with liver regeneration 1±2 weeks after the onset of jaundice when the investigations were performed.28 5.7

Chronic Hepatitis

Serial changes in phosphorus metabolites were monitored after fructose infusion in ®ve patients with chronic hepatitis.46 Following 0.5 g kgÿ1 fructose the increase in PME at 15±20 min (151 ‹ 49% of the preadministration value) was signi®cantly less than in healthy volunteers (p < 0.05). Also, the rebound of Pi at 35±40 min (126 ‹ 42%) was signi®cantly less than that in healthy volunteers (p < 0.05). These ®ndings were interpreted as suggesting that reduced fructose utilization is caused by impaired fructose transport into liver cells, thus indicating that this is a promising method in the functional evaluation of certain diffuse liver diseases. 5.8

Glycogen Storage Disease

Over 10 forms of glycogen storage disease resulting from inherited enzyme defects have been characterized. The different diseases are characterized by a storage of glycogen in abnormal quantities and/or synthesis of glycogen with an abnormal structure. Beckmann et al.47 obtained hepatic 13C MRS spectra from one patient with type IIIA glycogen storage disease, which is characterized by an increased glycogen concentration of abnormal structure in liver and muscle, due to inactivation of the amylo-1,6-glucosidase debrancher enzyme. They found glycogen levels, measured from the C-1 resonance, to be increased by a factor of 2±3 compared with well-trained athletic normals. A hepatic 31P MRS study of two patients with glucose 6phosphatase de®ciency (glycogen storage disease type 1A) showed that, after an overnight fast, PME was increased and Pi was low±normal (Pi/PME was markedly reduced).48 The increase in PME was attributed to accumulation of sugar phosphates (mainly glycolytic intermediates), on the basis of chemical shift measurements. After 1 g kgÿ1 oral glucose, hepatic sugar phosphates decreased by 40±64% and reached normal levels, whereas Pi increased by 40±130%. Liver Pi levels remained elevated in both patients 30 min after ingestion of glucose. Liver PME and Pi levels did not change in four control subjects after a glucose load. These high levels of PME were interpreted as suggesting that the activity of residual glucose 6-phosphatase may be enhanced, thus increasing hepatic glucose production and reducing the degree of hypoglycemia during fasting. The ®nding of large ¯uctuation in hepatic Pi levels may be directly related to the increase in uric acid production typically seen in these patients. 5.9

Galactose Intolerance

Galactosemia is an autosomal recessive disorder caused by a de®ciency of the enzyme UDPglucose : -D-galactose-1-phosphate uridyltransferase (EC 1.7.7.12). Liver damage is thought to be caused by accumulation of metabolites of galactose, although the exact mechanism is unclear. In a study of two galactosemic patients,49 an oral load of 20 mg kgÿ1 produced signi®cant changes in the hepatic 31P MRS spectrum of one of

8 IN VIVO HEPATIC MRS OF HUMANS difference in T1 was not taken into account when comparing cancer and normal hepatic spectra. Despite these comments, a number of points can be made. The most obvious and most consistent abnormality is an elevation of PME. The change in PDE is less consistent, with some groups reporting an elevation and others reporting a reduction. Changes in the Pi signal showed no obvious pattern. The pH in tumors was either similar or slightly higher than that of normal tissue. Furthermore there were no obvious differences between different types of tumor, either between primary and secondary tumors or between secondary tumors from different sites. It is interesting that there appears to be no obvious compromise of energy metabolism, and effects of ischemia or hypoxia do not dominate the tumor spectrum. Thus, in distinguishing neoplasms from normal tissue, only PME can be considered a diagnostic discriminant. Since proton decoupling during acquisition of the 31P MRS data has not been performed in human tumor studies, there is no direct information on individual components in vivo. However, highresolution MRS of perchloric acid extracts of tissue obtained at the time of surgery show that phosphoethanolamine and phosphocholine were elevated in four hepatocellular carcinomas and four secondary tumors.1,56 It is hoped that MRS will provide a marker of treatment response, particularly in the early stages of therapy. For example, hepatic embolization combined with intraarterial administration of cytostatic changes (chemoembolization) may be used to treat primary and metastatic cancers to the liver. As an acute response to chemoembolization in three patients, ATP, PME, and/or PDE concentrations diminished, whereas Pi concentrations increased or stayed relatively constant.52 Long-term follow-up after chemoembolization showed elevated PME/ATP and increased ATP concentrations in the absence of changes by conventional imaging techniques.52 In another study of 10 patients with liver metastases from colorectal carcinoma and two patients with hepatocellular carcinoma a marked increase in Pi and a decrease in ATP were observed during the ®rst few hours after local chemotherapy of chemoembolization and later, PME increased and PDE decreased.55 In a study of two

the patients. The peak at 5.2 ppm increased on two occasions to about twice its original size 60 min after galactose administration. In vitro animal studies showed this increase was largely due to galactose 1-phosphate. 5.10 Miscellaneous Hypothyroidism is known to affect nearly every organ and organ system in the body. However, in seven patients with severe hypothyroidism there were no differences in the hepatic 31 P MRS spectra compared with controls, either before or after treatment with thyroid hormone substitution therapy.50 In a study of seven patients with iron overload28 (ferritin levels 600 to >3000 mg mLÿ1), the hepatic 31P MRS resonances were all broadened, resulting in part from susceptibility effects of Fe3+, which is stored as ferritin and hemosiderin in liver lysosomes. The line broadening correlated with the degree of iron overload. 6 FOCAL LIVER DISEASE The majority of applications of NMR to focal liver disease have involved tumors, and these have recently been reviewed by Negendank.51 Ideally, the tumor spectrum should be compared with the normal spectrum of cells from which the neoplasm arises, which is possible for primary liver cancers, but not for secondary liver tumors. A further problem arises because the normal liver spectrum has prominent PME and PDE signals, and as these are signals that are abnormal in tumors it can be dif®cult to de®ne the tumor spectrum. The 31P MRS characteristics of hepatic tumors in adults are summarized in Table 1 and Figure 6. A number of technical points must be noted: in the majority of studies peak intensities were reported as relative ratios of areas, and molar quantitation was only attempted in two studies; most spectra were reported to be contaminated with signals from background tissue, but few authors provided estimates of the extent of contamination; and data were partially saturated, so that the effect of any Table 1

31

P MRS Spectral Parameters for Patients with Hepatic Malignancy (n 5 2)

Diagnosis Primary malignancy Hepatocellular carcinoma52 Hepatocellular carcinoma53 Hepatocellular carcinoma54 Hepatocellular carcinoma55 Various56 Secondary malignancy Lymphoma57 Adenocarcinoma56 Carcinoid56 Various52 Various28,59 Breast carcinoma54 Various58 Colorectal carcinoma55 Various53 a

VOI, volume of interest.

MRS ®ndingsa

n 2 7 3 2 3

PME/ATP PME/ATP PME/Pi :, PME : PME/PDE

: :, if tumor occupied more than 50% of VOI PME/ATP :

22 11 14 3 3 2 19 10 30

PME/ATP PME/PDE PME/PDE PME/ATP PME : PME/ATP PME/ATP PME : PME/ATP

:, PME/Pi : : : :, ATP ;, Pi ;

:

:, PME/Pi :, PDE/Pi :, pH : :, PDE/ATP :, Pi/ATP : :, if tumor occupied more than 50% of VOI

IN VIVO HEPATIC MRS OF HUMANS

(a)

(d)

(b)

(e)

(c)

(f)

40

20

0

–20

–40 ppm

40

9

Liver phosphomonoester levels decreased following chemotherapy in all but one of the patients who had abnormally high ratios in their initial spectra; this decrease was seen as early as 1 day and as late as 2 weeks after commencing treatment. In patients with initially normal PME levels, treatment did not produce a fall in PMEs. These ®ndings suggest that detection of falling PME levels indicates that the drugs are reaching the target cells and affecting tumor cell metabolism, which may be of considerable clinical importance. A study of two children with neuroblastoma60 examined on a number of occasions from the ages of 1±9 months, showed the hepatic 31P MRS spectrum from the region of pathology to be abnormal with elevated PME levels in the initial spectroscopic examinations. Further treatment showed a reduction in PME levels, and the spectra ®nally became similar to the normal subject (aged 3 months). A small number of other focal liver lesions have been studied.54,58 The metabolite ratios from two patients with cavernous hemangiomas were apparently normal, although the S/N was reduced.54 The in vitro high-resolution MRS spectra from a multiloculated cyst showed elevated PME and reduced PDE, and this was similar to data from extracts of tumor tissue.1 20

0

–20

–40 ppm

Figure 6 Representative 31P hepatic MRS spectra obtained from a healthy adult volunteer and patients with hepatic malignancies of varying histology using CSI techniques, TR-1 s, pulse angle 45 . (a) Spectrum from a healthy 22-year-old female acquired using twodimensional CSI; nominal planar resolution, 30 mm; total number of data collections, 256. (b) Spectrum from a 61-year-old man with squamous cell carcinoma obtained from a voxel containing tumor acquired using four-dimensional CSI; nominal resolution, 30 mm  30 mm  30 mm; total number of data collections 2048. (c) Spectrum from a 59-year-old man with carcinoid liver metastases obtained from a voxel containing tumor acquired using four-dimensional CSI; nominal resolution, 20 mm  20 mm  20 mm; total number of data collections 2048. (d) Spectrum from a 56-year-old man with carcinoid liver metastases obtained from a voxel containing tumor acquired using four-dimensional CSI; nominal resolution, 30 mm  30 mm  30 mm; total number of data collections 2048. (e) Spectrum from a 79-year-old man with carcinoid liver metastases obtained from a voxel containing tumor acquired using four-dimensional CSI; nominal resolution, 30 mm  30 mm  30 mm; total number of data collections 2048. (f) Spectrum from a 53-year-old man with hepatocellular carcinoma obtained from a voxel containing tumor acquired using fourdimensional CSI; nominal resolution, 30 mm  30 mm  30 mm; total number of data collections 2048. (Reproduced by permission of Heydon & Son from I. J. Cox, J. D. Bell, C. J. Peden, R. A. Iles, C. S. Foster, P. Watanapa, and R. C. N. Williamson, NMR Biomed., 1992, 5, 114)

patients with carcinoid syndrome, successful arterial embolization was accompanied by a decrease in PME/PDE and an increase in Pi/ATP, whereas in a patient in whom the tumor blood supply was not effectively interrupted there was little change in metabolite ratios.7 In a study of 22 patients with lymphoma, 11 were re-examined after chemotherapy treatment.57 Before treatment, PME/ ATP and PME/Pi were elevated and approximately related to clinical stage, although there were some notable discrepancies.

7

DRUGS AND THE LIVER

The liver is the major site of drug metabolism. Lipid-soluble drugs are converted to more water-soluble forms by a group of hepatic mixed-function enzymes, including cytochrome P450. These processes facilitate excretion of the drugs in urine or bile. Drug metabolism ®rstly involves oxidation, reduction, or demethylation of the drug, followed by conjugation of the derivatives produced with glucrudine, sulfate, or glutathione. These conjugates are excreted in the urine and bile as they cannot be reabsorbed by renal tubular or bile ductular cells. A number of factors affect drug metabolism, including the microsomal enzyme system (which will in¯uence the speed of metabolism), the route of administration, the liver blood ¯ow, and competitive inhibition. The majority of studies involving alcohol abuse have reported the spectral characteristics of resultant liver damage, rather than the effects of alcohol per se.28,29,41±45 However, it may be important to distinguish between the two effects, particularly as the metabolic consequences of an alcohol load may persist for a few hours to a few weeks and may depend on the degree of liver injury. In a study of three healthy volunteers, acute alcohol ingestion (0.5±1.0 g kgÿ1 alcohol) was associated with a transient but signi®cant elevation in PME/ATP.61 In 14 patients with minimal liver injury, active drinking was associated with elevation in mean PME/ATP (p = 0.12) and in mean PDE/ATP (p < 0.0001); abstinence from alcohol was associated with a prompt (3±7 day) reduction in PME/ATP and with a reduction in PDE/ATP over a longer timescale (3±28 days). In eight patients with alcoholic cirrhosis, active drinking was associated with elevation in mean PME/ATP (p < 0.05) and mean PDE/ ATP (p = 0.4); abstinence from alcohol had no effect on the mean PME/ATP, although the mean PDE/ATP fell to within or below the reference range. The reversible elevation in PME/ ATP in healthy volunteers given an alcohol load and in chronic

10 IN VIVO HEPATIC MRS OF HUMANS 19

FBAL

Post administration 250 57 49 40

Ti m e( m in )

5Fu 31 22 13 4 0 10

0

–10

–20 ppm

Figure 7 Serial 19F spectra acquired with a surface coil over the liver of a patient (66 year old male) in a 1.5 T MRI system, operating at 59.8 MHz for 19F. Each spectrum is the result of 128 FIDs using a 250 s, 90 pulse with 512 complex points collected over a period of 8.5 min. Actual spectral width was 29 to ÿ54 ppm. A shifted sine-bell squared apodization was used for S/N enhancement. (Reproduced by permission of Pergamon Journals Ltd. from W. Wolf, M. J. Albright, M. S. Silver, H. Weber, U. Reichardt, and R. Sauer, Magn. Reson. Imag., 1987, 5, 165)

alcohol abusers while actively drinking was interpreted as re¯ecting changes in hepatic redox potential as a result of obligatory alcohol metabolism. The irreversible elevation in PME/ ATP, observed in patients with cirrhosis, probably re¯ects changes associated with hepatocyte regeneration. The reversible elevation in PDE/ATP observed in chronic alcohol abusers while actively drinking most likely re¯ects induction of hepatocyte endoplasmic reticulum. Self-poisoning with acetaminophen accounts for many emergency hospital admissions. Overdoses cause cell damage through increased oxidation of the drug as the conjugation pathways are saturated. Early prediction of outcome in patients with acetaminophen overdose is dif®cult; for example, aspartate aminotransferase levels peak late after the ingestion of the drug and correlate poorly with prognosis. While the prothrombin time gives some indication of synthetic ability, methods for directly assessing cell viability are sought. Dixon and colleagues62 studied 18 patients with acetaminophen poisoning and found that the concentrations of all metabolites fell in parallel with a decrease in the synthetic ability of the liver. ATP and PDE fell to about 20% of their normal concentrations in severely affected patients. The reduction in PDE was interpreted as a reduction in endoplasmic reticulum. The direct observation of 5FU and its metabolites in the human liver was ®rst reported in 1987.63 Signals from 5FU and one of its catabolites, -¯uoro- -alanine (FBAL), were observed in the human liver in patients undergoing cancer chemotherapy (Figure 7). The pharmacokinetics of 5FU in the tumors of 11 patients with carcinoma at various sites were studied, including three patients with tumor located in the liver.64 A long liver tumor pool of 5FU was observed in six of 11 tumors, including one of three patients with carcinoma in the liver. The hal¯ife of this `trapped' pool was 0.33±1.3 h, considerably longer than the hal¯ife of 5FU in blood (5±15 min). Neither the anabolites or catabolites of 5FU were detected by

F MRS. Patient response to chemotherapy appeared to correlate with the extent of trapping of free 5FU in tumors. Semmler et al.65 reported hepatic 19F MRS data from eight patients with liver tumors, receiving intra-arterial 5FU. Contributions from liver and tumor could not be distinguished in the spectrum. The time constants for the kinetics of 5FU ranged from 8 to 75 min, whereas the time constants for FBAL were either approximately 15 or 50 min. A broad peak comprising nucleoside and nucleotide anabolites was detected in one patient. More recently, Findlay et al.66 studied hepatic 19F MRS in 26 colorectal cancer patients treated with a continuous low dose intravenous infusion of 5FU, until the point of refractory disease, at which time interferon- was added with the objective of modulating 5FU activity. In patients observed by MRS during the ®rst 8 weeks of 5FU treatment, those with a visible 5FU signal were likely to respond to treatment (p = 0.017). At the time of interferon- addition, MRS showed that seven patients developed new or increased 5FU signals, and four patients showed a signal from the active metabolites of 5FU. The patients who exhibited a new or increased 5FU signal were more likely to show further response to interferon- (p = 0.007). The in vivo pharmacokinetics of a model drug, ¯eroxacin, was studied in healthy human subjects.67 After oral administration a signal was detected in the hepatic 19F MRS spectrum, and also muscle, and monitored over a period of 24 h. Pharmacokinetic data for the liver were obtained, combining MRS results with high-performance liquid chromatography (HPLC) analysis of plasma: tmax = 1.4 h; Cmax = 53 mol Lÿ1; t1/2 = 4.4 h (fast phase) and 10.8 h (slow phase).

8

FUTURE APPLICATIONS

Hepatic MRS can provide a direct measure of hepatic function, for example from the baseline spectrum or following a metabolic challenge. Since the indirect clinical and laboratory measures of hepatic function may be subject to extrahepatic in¯uences, direct noninvasive measurement is of importance. A range of studies have been made, as described above, and these can be extended, for example, speci®cally to study gluconeogenesis in diabetics and tumor-bearing patients using 13C or 31 P MRS combined with an alanine stress test. In addition, the interpretation that metabolite changes in end-stage liver disease can provide an index of hepatocyte regeneration, rather than re¯ecting production of ®brous tissue or in¯ammatory cell activity, suggests that hepatic 31P MRS may provide a speci®c marker which could be used to guide management decisions and as a prognostic assessment in patients with liver disease. Further studies correlating functional grades and etiology to absolute metabolite levels are appropriate. The clinical management of a range of pathologies relies on the interpretation of histological results from liver biopsy. Since liver biopsy has a de®ned risk, a noninvasive technique of providing the same information would be of value, particularly when serial liver biopsies are required, for example in patients undergoing liver transplantation. In such patients hepatic MRS may be used to provide a guide to optimal timing of transplantation, to indicate good or poor liver metabolic status for noninvasive assessment of donor livers prior to transplan-

IN VIVO HEPATIC MRS OF HUMANS

tation68 and also to provide a useful early marker of rejection after transplantation. For patients with hepatic malignancy, 31P MRS may provide an early marker of response to treatment, and 19F MRS can be used to measure the pharmacokinetics of speci®c anticancer drugs and their metabolites. Other areas of study include the metabolism of the liver in patients with tumor at distant sites. Hepatic MRS may provide insights into the mechanisms responsible for cachexia. Many of the clinical and biochemical applications of hepatic MRS to date have relied on the interpretation of changes in metabolite ratios, rather than absolute metabolite concentrations. However, in a number of pathologies, metabolite ratios alone do not adequately re¯ect the underlying metabolic changes, particularly if there is an alteration in ATP levels. Recourse to absolute metabolite concentrations is therefore required. This may be dif®cult to achieve within the con®nes of a clinical examination because the low S/N of hepatic spectra means it is time consuming to measure the NMR characteristics of individual resonances. Improvements in sensitivity and/or spectral resolution may be achieved using proton decoupling techniques, higher ®eld strengths, and more ef®cient rf coil arrangements. However, detailed interpretation of clinical data will continue to rely on results from animal models. A wide range of metabolite resonances can be detected using 1H MRS,1 and, if the technical problems in acquiring hepatic 1H MRS spectra can be overcome, hepatic 1H MRS will provide additional and new information. In conclusion, hepatic MRS can provide a noninvasive measure of hepatic function, which has a range of clinical and biochemical applications in diagnosis, prognosis, and assessment of treatment ef®cacy in liver disease.

9 RELATED ARTICLES Animal Methods in MRS; Applications of 19F-NMR to Oncology; Cells and Cell Systems MRS; Chemical Shift Imaging; High-Field Whole Body Systems; pH Measurement In Vivo in Whole Body Systems; Proton Decoupling During In Vivo Whole Body Phosphorus MRS; Quantitation in In Vivo MRS; Single Voxel Whole Body Phosphorus MRS; Spatial Localization Techniques for Human MRS; Spectroscopic Studies of Animal Tumor Models; Susceptibility Effects in Whole Body Experiments; Tissue and Cell Extracts MRS; Tissue Behavior Measurements Using Phosphorus-31 NMR; Whole Body Studies: Impact of MRS. 10

REFERENCES

1. J. D. Bell, I. J. Cox, J. Sargentoni, C. J. Peden, D. K. Menon, C. S. Foster, P. Watanapa, R. A. Iles, and J. Urenjak, Biochim. Biophys. Acta, 1993, 1225, 71. 2. R. A. Iles, A. N. Stevens, and J. R. Grif®ths, Prog. NMR Spectrosc., 1982, 15, 49. 3. R. D. Cohen, R. M. Henderson, R. A. Iles, J. P. Monson, and J. A. Smith, `Metabolic Acidosis', Ciba Foundation Symposium 84, Pitman, London, 1982, p. 20. 4. R. D. Oberhaensli, G. J. Galloway, D. J. Tayor, P. J. Bore, and G. K. Radda, Br. J. Radiol., 1986, 59, 695.

11

5. P. R. Luyten, G. Bruntink, F. M. Sloff, J. W. A. H. Vermeulen, J. I. van der Heijden, J. A. den Hollander, and A. Heerschap, NMR Biomed., 1989, 4, 177. 6. D. J. Meyerhoff, G. S. Karczmar, G. B. Matson, M. D. Boska, and M. W. Weiner, NMR Biomed., 1990, 3, 17. 7. I. J. Cox, D. K. Menon, J. Sargentoni, D. J. Bryant, A. G. Collins, G. A. Coutts, R. A. Iles, J. D. Bell, I. S. Benjamin, S. Gilbey, H. J. F. Hodgson, and M. Y. Morgan, J. Hepatol., 1992, 14, 265. 8. E. J. Murphy, B. Rajagopalan, K. M. Brindle, and G. K. Radda, Magn. Reson. Med., 1989, 12, 282. 9. T. E. Bates, S. R. Williams, and D. G. Gadian, Magn. Reson. Med., 1989, 12, 145. 10. R. K. Gupta and R. D. Moore, J. Biol. Chem., 1980, 255, 3987. 11. S. M. Cohen, J. Biol. Chem., 1983, 258, 14 294. 12. R. A. Iles, I. J. Cox, J. D. Bell, L. M. Dubowitz, F. Cowan, and D. J. Bryant, NMR Biomed., 1990, 3, 90. 13. M. Barany, D. G. Spigos, E. Mok, P. N. Venkatasubramanian, A. C. Wilbur, and B. G. Langer, Magn. Reson. Imaging, 1987, 5, 393. 14. P. Canioni, J. Alger, and R. G. Shulman, Biochemistry, 1983, 22, 4974. 15. T. Jue, J. A. B. Lohman, R. J. Ordidge, and R. G. Shulman, Magn. Reson. Med., 1987, 5, 377. 16. T. Jue, D. L. Rothman, B. A. Tavitian, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1989, 86, 1439. 17. D. L. Rothman, I. Magnusson, L. D. Katz, R. G. Shulman, and G. I. Shulman, Science, 1991, 254, 573. 18. M. Ishihara, H. Ikehira, S. Nishikawa, T. Hashimoto, K. Yamada, J. Shishido, T. Ogino, K. Cho, S. Kobayashi, M. Kawana, T. Matumoto, T. A. Iinuma, N. Arimizu, and Y. Tateno, Am. J. Physiol. Imaging, 1992, 7, 32. 19. N. Beckmann, R. Fried, I. Turkalj, J. Seelig, U. Keller, and G. Stalder, Magn. Reson. Med., 1993, 29, 583. 20. J. Alger, K. Behar, D. L. Rothman, and R. G. Shulman, J. Magn. Reson., 1984, 56, 334. 21. A. Heerschap, P. R. Luyten, J. I. van der Heyden, L. J. M. P. Oosterwaal, and J. A. den Hollander, NMR Biomed., 1989, 2, 124. 22. M. Saner, G. McKinnon, and P. Boesiger, Magn. Reson. Med., 1992, 28, 65. 23. J. L. Evelhoch, Invest. New Drugs, 1989, 7, 5. 24. M. D. Boska, B. Hubesch, D. J. Meyerhoff, D. B. Tweig, G. S. Karczmar, G. B. Matson, and M. W. Weiner, Magn. Reson. Med., 1990, 13, 228. 25. H. Bomsdorf, T. Helzel, D. Kunz, P. Roschmann, O. Tschendel, and J. Wieland, NMR Biomed., 1988, 1, 151. 26. H. Barfuss, H. Fischer, D. Hentschel, R. Ladebeck, A. Oppelt, R. Wittig, W. Duerr, and R. Oppelt, NMR Biomed., 1990, 3, 31. 27. W. P. Aue, Rev. Magn. Reson. Med., 1986, 1, 21. 28. R. Oberhaensli, B. Rajagopalan, G. J. Galloway, D. J. Taylor, and G. K. Radda, Gut, 1990, 31, 463. 29. V. Rajanayagam, R. R. Lee, Z. Ackerman, W. G. Bradley, and B. D. Ross, J. Magn. Reson. Imaging, 1992, 2, 183. 30. I. R. Young, I. J. Cox, G. A. Coutts, and G. M. Bydder, NMR Biomed., 1989, 2, 329. 31. I. J. Cox, D. J. Bryant, B. D. Ross, I. R. Young, D. G. Gadian, G. M. Bydder, S. R. Williams, A. L. Busza, and T. E. Bates, Magn. Reson. Med., 1987, 5, 186. 32. I. J. Cox, G. A. Coutts, D. G. Gadian, P. Ghosh, J. Sargentoni, and I. R. Young, Magn. Reson. Med., 1991, 17, 53. 33. M. J. Blackledge, R. D. Oberhaensli, P. Styles, and G. K. Radda, J. Magn. Reson., 1987, 71, 331. 34. S. D. Buchhal, W. J. Thoma, J. S. Taylor, S. J. Nelson, and T. R. Brown, NMR Biomed., 1989, 2, 298. 35. P. C. Dagnelie, D. K. Menon, I. J. Cox, J. D. Bell, J. Sargentoni, G. A. Coutts, J. Urenjak, and R. A. Iles, Clin. Sci., 1992, 83, 183.

12 IN VIVO HEPATIC MRS OF HUMANS 36. F. Terrier, P. Vock, J. Cotting, R. Ladebeck, J. Reichen, and D. Hentschel, Radiology, 1989, 171, 557. 37. C. Segebarth, A. R. Grivegnee, R. Longo, P. R. Luyten, and J. A. den Hollander, Biochemie, 1991, 73, 105. 38. I. Magnusson, D. L. Rothman, L. D. Katz, R. G. Shulman, and G. I. Shulman, J. Clin. Invest., 1992, 90, 1323. 39. R. D. Oberhaensli, B. Rajagopalan, D. J. Taylor, G. K. Radda, J. E. Collins, J. V. Leonard, H. Schwarz, and N. Herschkowitz, Lancet, 1987, ii, 931. 40. J. E. Seegmiller, R. M. Dixon, G. J. Kemp, P. W. Angus, T. E. McAlindon, P. Dieppe, B. Rajagopalan, and G. K. Radda, Proc. Natl. Acad. Sci. USA, 1990, 87, 8326. 41. D. J. Meyerhoff, M. D. Boska, A. M. Thomas, and M. W. Weiner, Radiology, 1989, 173, 393. 42. P. W. Angus, R. M. Dixon, B. Rajagopalan, N. G. Ryley, K. J. Simpson, T. J. Peters, D. P. Jewell, and G. K. Radda, Clin. Sci., 1990, 78, 33. 43. T. Munakata, R. D. Grif®ths, P. A. Martin, S. A. Jenkins, R. Shields, and R. H. T. Edwards, NMR Biomed., 1993, 6, 168. 44. J.-F. Dufour, C. Stoupis, F. Lazeyras, P. Vock, F. Terrier, and J. Reichen, Hepatology, 1992, 15, 835. 45. D. K. Menon, J. Sargentoni, S. D. Taylor-Robinson, J. D. Bell, I. J. Cox, D. J. Bryant, G. A. Coutts, K. Rolles, A. K. Burroughs, and M. Y. Morgan, Hepatology, 1995, 21, 417. 46. H. Sakuma, K. Itabashi, K. Takeda, T. Hirano, Y. Kinosada, T. Nakagawa, M. Yamada, and T. Nakano, J. Magn. Reson. Imaging, 1991, 1, 701. 47. N. Beckmann, J. Seelig, and H. Wick, Magn. Reson. Med., 1990, 16, 150. 48. R. D. Oberhaensli, B. Rajagopalan, D. J. Taylor, G. K. Radda, J. E. Collins, and J. V. Leonard, Pediatr. Res., 1988, 23, 375. 49. B. Kalderon, R. M. Dixon, B. Rajagopalan, P. W. Angus, R. D. Oberhaensli, J. E. Collins, J. V. Leonard, and G. K. Radda, Pediatr. Res., 1992, 32, 39. 50. K. D. Hagspiel, C. von Weymarn, G. McKinnon, R. Haldemann, B. Marincek, and G. K. von Schulthess, J. Magn. Reson. Imaging, 1992, 2, 527. 51. W. Negendank, NMR Biomed., 1992, 5, 303. 52. D. J. Meyerhoff, G. S. Karczmar, F. Valone, A. Venook, G. B. Matson, and M. W. Weiner, Invest. Radiol., 1992, 27, 456. 53. I. R. Francis, T. L. Chenevert, B. Gubin, L. Collomb, W. Ensminger, S. Walker-Andrews, and G. M. Glazer, Radiology, 1991, 180, 341. 54. G. M. Glazer, S. R. Smith, T. L. Chenevert, P. A. Martin, A. N. Stevens, and R. H. Edwards, NMR Biomed., 1989, 1, 184.

55. A. Schilling, B. Gewiese, G. Berger, J. Boese-Landgraf, F. Fobbe, D. Stiller, U. Gallkowski, and K. J. Wolf, Radiology, 1992, 182, 887. 56. I. J. Cox, J. D. Bell, C. J. Peden, R. A. Iles, C. S. Foster, P. Watanapa, and R. C. N. Williamson, NMR Biomed., 1992, 5, 114. 57. R. M. Dixon, P. W. Angus, B. Rajagopalan, and G. K. Radda, Br. J. Cancer, 1991, 63, 953. 58. G. Brinkmann, and U. H. Melchert, Magn. Reson. Imaging, 1992, 10, 949. 59. R. D. Oberhaensli, D. Hilton-Jones, P. J. Bore, L. J. Hands, R. P. Rampling, and G. K. Radda, Science, 1986, 2, 8. 60. J. M. Maris, A. E. Evans, A. C. Mclaughlin, G. J. D'Angio, L. Bolinger, H. Manos, and B. Chance, New Engl. J. Med., 1985, 312, 1500. 61. D. K. Menon, M. Harris, J. Sargentoni, S. Taylor-Robinson, I. J. Cox, and M. Y. Morgan, Gastroenterology, 1995, 108, 776. 62. R. M. Dixon, P. W. Angus, B. Rajagopalan, and G. K. Radda, Hepatology, 1992, 16, 943. 63. W. Wolf, M. J. Albright, M. S. Silver, H. Weber, U. Reichardt, and R. Sauer, Magn. Reson. Imaging, 1987, 5, 165. 64. C. A. Presant, W. Wolf, M. J. Albright, K. L. Servis, R. Ring, D. Atkinson, R. L. Ong, C. Wiseman, M. King, D. Blayney, P. Kennedy, A. El-Tahtawy, M. Singh, and J. Shani, J. Clin Oncol, 1990, 8, 1868. 65. W. Semmler, P. Bachert-Baumann, F. Guckel, F. Ermark, P. Schlag, W. J. Lorenz, and G. van Kaick, Radiology, 1990, 174, 141. 66. M. P. Findlay, M. O. Leach, D. Cunningham, D. J. Collins, G. S. Payne, J. Glaholm, J. L. Mansi, and V. R. McCready, Ann. Oncol., 1993, 4, 597. 67. P. Jynge, T. Skjetne, I. Gribbestad, C. H. Kleinbloesem, H. F. Hoogkamer, O. Antonsen, J., Krane, O. E. Bakoy, K. M. Furuheim, and O. G. Nilsen, Clin. Parmacol. Ther., 1990, 48, 481. 68. R. F. Wolf, R. L. Kamman, E. L. Mooyaart, E. B. Haagsma, R. P. Bleichrodt, and M. J. Slooff, Transplantation, 1993, 55, 949.

Biographical Sketch Isobel Jane Cox. b 1959. B.A. (Nat. Sci.), 1981, University of Cambridge, UK; M.Sc., 1982, Ph.D., 1984 (supervisors Peter S. Belton and Robin K. Harris), University of East Anglia, UK. Introduced to clinical MRS on joining I.R. Young's team at GEC Hirst Research Centre. Lecturer in Diagnostic Radiology, RPMS, 1986±present. Approx. 50 publications. Current research speciality: development and applications of liver and brain MRS.

KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS

Kidney, Prostate, Testicle, and Uterus of Subjects Studied by MRS

TMA (% of t = 0)

E

Michael W. Weiner

Uosm (mOsm kg–1)

University of California, San Francisco, CA, USA

1 KIDNEY

Shah et al.1 reported a technique for using the stimulated echo amplitude mode (STEAM) sequence for 1H MRS of the human kidney. The results demonstrated the presence of trimethylamines (TMAs). A prominent peak observed at 5.8 ppm was from urea. Avision et al.2 used volume localized 1H MRS to detect and measure changes in medullary trimethylamines in the human kidney (Figure 1). Proton magnetic resonance spectra were obtained from the human renal medulla using a stimulated echo localization sequence. In addition to residual water and lipid, TMAs were identi®ed at 3.25 ppm. In normal volunteers, overnight dehydration led to a signi®cant increase of urine osmolality, and an increase in medullary TMAs (Figure 2). Water loading caused a water

6.0

5.0

4.0

3.0

2.0

1.0

0.0

ppm

Figure 1 Water suppressed, volume localized 1H spectrum from kidney. Water suppression consisted of a Silver±Hoult phase swept adiabatic fast passage pulse for selective water inversion followed by an inversion±recovery time of 0.8 second to allow the water Z magnetization to null. This sequence was delivered at the start of the volume localized stimulated echo sequence. TE = 68 ms, TM = 44 ms, TR = 3 s. Number of scans = 128. Filtering was 5 Hz of exponential line broadening. The resonances are water (4.75 ppm), lipids (0.9±1.4 ppm), and TMA (3.25 ppm)Ðwhich is shown expanded (8) above the main spectrum. (Reproduced by permission of the National Academy of Sciences from Avison et al.2)

E 120 100 80 60 40 20 0

1000

1000

800

800

600

600

400

400

200

200

0

0 0

2 4 Time (h)

6

25

Figure 2 Time-course of changes in medullary TMA levels and Uosm in the four volunteers studied. TMA levels are expressed as the TMA area (%) at the point of maximal dehydration (i.e. t = 0). Error bars are ‹ SEM. *p < 0.05 versus t = 0 value for the TMA time-course. **p < 0.05 versus t = ÿ15 for the Uosm time-course. E, euvolemic; D, dehydration; W, water load. (Reproduced by permission of the National Academy of Sciences from Avison et al.2)

diuresis and a signi®cant reduction in medullary TMAs within 4 hours. These results are consistent with the view that TMAs may play an osmoregulatory role in the medulla of the normal human kidney.

1.2

¥8

W

120 100 80 60 40 20 0

–15

1.1 Proton MRS of Kidney

D

1

Phosphorus-31 MRS of Kidney

Jue et al.3 demonstrated that 31P MRS signals can be obtained from the normal human kidney. Matson et al.4 also demonstrated the feasibility of obtaining 31P magnetic resonance spectra from human kidneys using the image selected in vivo spectroscopy (ISIS) localization technique. Boska et al.5 obtained spatially localized 31P magnetic resonance spectra from healthy normal human kidneys and from well functioning renal allographs (Figure 3). Little or no phosphocreatine (PCr) in all spectra veri®ed the absence of muscle contamination and was consistent with proper volume localization. The PME/ATP ratio (PME, phosphomonoester) was slightly elevated in transplanted kidneys (1.1) compared with normal healthy kidneys (0.8). Despite the practical problems produced by organ depth, respiratory movement, and tissue heterogeneity, these results demonstrate the feasibility of obtaining 31P magnetic resonance spectra from human kidneys. Bretan et al.6,7 reported their clinical experience with pretransplant assessment of renal viability using 31P MRS to study 40 renal transplant recipient patients (Figure 4). The purpose of their study was to develop and investigate the use of MRS in the clinical transplant setting by correlation of pretransplant MRS parameters with subsequent renal function. Kidneys were studied during simple hypothermic storage within their sterile containers using an external 31P MRS surface coil. Mean storage times were about 38 hours. Cold storage times did not correlate with subsequent clinical renal function. However,

2 KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS Pi (a)

PDE

PME

(b) γ α

β

PCr

15.00

10.00

5.00

0.00

–5.00 –10.00 –15.00 –20.00 –25.00 ppm

15.00

10.00

5.00

0.00

–5.00 –10.00 –15.00 –20.00 –25.00 ppm

Figure 3 Phosphorus-31 ISIS spectra of (a) the healthy normal kidney and (b) of the well functioning kidney transplant. Acquisition parameters: TR = 2.0 seconds, acquisition time = 1 hour, 90 pulse set for the region of interest, distance of the center of the volume of interest (VOI) from the 14 cm surface coil = 70 mm for normal kidney and 42 mm for transplanted kidney, size of the VOI is 25  45  55 mm = 62 ml for normal kidney and 25  50  54 mm = 68 ml for transplanted kidney. (Reproduced by permission of Springer-Verlag from Boska et al.5)

selected 31P MRS data did. ATP was present in 11 kidneys and was associated with the best subsequent renal function. Only 36% of these patients required dialysis, compared with 71 patients without detectable ATP who required a posttransplant dialysis. Kidneys with ATP had the highest PME/Pi (Pi, inorganic phosphate) ratios; in general, the higher the PME/Pi ratio, the better the renal function posttransplant. The intracellular pH did not correlate with subsequent renal function. The authors suggested that MRS provided a better correlation with renal function after transplantation than existing methods. Grist et al.8 used 31P MRS to investigate the effects of rejection on renal transplants (Figures 5 and 6). The PDE/PME (PDE, phosphodiester) and Pi/ATP ratios in the transplants with rejection differed signi®cantly from the corresponding metabolite ratios of patients without rejection. A PDE/PME ratio exceeding 0.8 had a sensitivity of 100% and a speci®city of 86% for predicting rejection. A Pi/ATP ratio greater that 0.6 had a sensitivity of 72% and a speci®city of 86% for predicting rejection. The authors concluded that 31P MRS may be useful as a noninvasive method for evaluating renal metabolism during episodes of transplant rejection. 2 PROSTATE 2.1 Proton MRS of Prostate Normal prostate has a very high concentration of citrate, a unique feature of this tissue related to the function of prostate cells to secrete citrate into the semen. Most of the MRS of prostate has focused on investigating changes of citrate associated with benign prostatic hypertrophy (BPH) and prostatic carcinoma. Schick et al.9 investigated the signal characteristics of citrate at low ®eld strengths at 1.5 T using spatially selected spectroscopy and theoretical methods. In vivo localized spectroscopy of small volume elements (2 ml) using a double spin echo method within the prostate gland provided citrate signals.

Volume selected proton spectra with different echo times were recorded. Thomas et al.10 performed 1H MRS of normal and malignant human prostates in vivo (Figures 7 and 8). The results demonstrated that water-suppressed 1H MRS spectra could be obtained from the prostate. Normal healthy prostate showed a large resonance from citrate. The spectrum from a malignant human prostate showed a much lower level of citrate. The results also suggested that the low concentration of citrate might be useful for the identi®cation of prostate cancer. Schnall et al.11 performed localized 1H MRS of the human prostate in vivo using an endorectal surface coil. High levels of citrate were observed in all regions of normal prostate and benign prostatic hypertrophy. The citrate levels in regions containing tumor were variable. The presence of high citrate levels in one case of prostate cancer was con®rmed from extracts. Schiebler et al.12 reported high-resolution 1H MRS of human prostate perchloric acid extracts (Figure 9). The citrate peak area was higher in benign prostatic hyperplasia than in adenocarcinoma. However, citrate peak areas from the normal peripheral zones were not signi®cantly different from those found in adenocarcinomas. A sharp peak at 2.05 ppm that was seen in four out of thirteen adenocarcinoma samples and only one out of thirteen in the BPH samples was assigned to N-acetylneuraminic acid. Fowler et al.13 also obtained 1H NMR spectra from perchloric extracts of tissue samples from human prostate. Statistically signi®cant differences between the normals, the benign prostratic hypertrophy, and the cancer groups occurred for metabolite ratios of creatine, citrate, and phosphorylcholine. None of the ratios correlated with the Gleason grade of the cancer samples. Different sections of large tumors often yielded substantially different ratios. Yacoe et al.14 reported in vitro 1H MRS of normal and abnormal prostate cells (Figure 10). Proton MRS was used to determine if cell strains derived from prostatic cancers could be distinguished from normal prostate. Prostatic cancer cells had lower concentrations of

KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS

3

Pi (a)

1 d PME

ATP 2

Viable cold stored 9 month old human cadaveric kidney 24

16

8

0

–8

–16

a

3

b c –24

–32 ppm

Buffer Pi

f

g

4 10

(b)

e

0

–10

–20

–30 ppm

Figure 5 Contiguous 31P magnetic resonance spectra obtained from slices of kidney. Note the presence of signi®cant PCr at the surface, consistent with muscle tissue. The deepest slice shows a large PME peak and little PCr, consistent with renal tissue. Peaks from left to right correspond to PMEs (a), Pi (b), PDEs (c), PCr (d) and - (e), - (f), and - (g) phosphates of ATP. (Reproduced by permission of Williams & Wilkins from Grist et al.8)

Pi

In the past several years Kurhanewicz, Vigneron and their colleagues have published a series of reports using 1H MRSI of the prostate in a clinical setting. They have used an endorectal coil16 and a high spatial resolution technique.17 This

PME

PDE

(c)

30

20

10

0

–10

10

0

–10

–20

–30 ppm

10

0

–10

–20

–30 ppm

10

0

–10

–20

–30 ppm

–20 ppm

Figure 4 Ex situ in vivo magnetic resonance spectrum of (a) a viable pediatric and (b) an adult kidney. PME, Pi, PDS (phosphodiester), and ATP (adenosine triphosphate) peaks are de®ned. (Reproduced by permission of Williams & Wilkins from Bretan et al.6)

citrate compared with normal prostate epithelium, but the differences were small and not statistically signi®cant. However, both the cancer and normal prostate cells were washed, which may have removed diffusable citrate. Kurhanewicz et al.15 performed 1H MRS and enzymatic assays of human prostate adenocarcinomas and prostate DU145 xenographs grown in nude mice. The results showed that citrate concentrations in primary human adenocarcinomas were signi®cantly lower than those observed for normal benign hyperplastic prostatic tissue. There was a 10-fold reduction of citrate associated with DU145 xenographs compared with primary prostate cancer. These ®ndings support the hypothesis that citrate concentrations are low in prostate cancer.

(b)

(a)

Figure 6 Phosphorus-31 magnetic resonance spectra from a patient with (a) a normally functioning renal allograft, (b) a patient with cyclosporine nephrotoxicity, and (c) a patient with moderate cellular rejection. Note the increase in Pi and PDE in the patient with rejection compared with the normal control subject or the patient with cyclosporine toxicity. PCr is a contaminant from extrarenal tissue. (Reproduced by permission of Williams & Wilkins from Grist et al.8)

4 KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS 2

2

1 7 6

7 3

3 5

6 1

4

5 4

12.00

10.00

8.00

6.00

4.00

2.00

0.00

–2.00

12.00

10.00

8.00

6.00

4.00

2.00

0.00

–2.00

ppm

ppm

Figure 7 Proton magnetic resonance spectrum prostate in a healthy 26-year-old subject with TE = 40 ms. The total acquisition time was 1.5 minutes. Resonance assignments: 1, 2, water and MDPA from the external standard, situated in the plane of the surface coil; 3, residual water; 4, spermine/creatine/PCr; 5, 6, the methylene protons of citrate; 7, methylene protons of spermine and lipids. (Reproduced by permission of the Society for Magnetic Resonance Imaging from Thomas et al.10)

Figure 8 Proton magnetic resonance spectrum of a malignant human prostate with binomial water suppression. Number of scans = 16; repetition time, 5 seconds; TE = 80 ms; total acquisition time = 80 seconds. The residual water and the resonances from the external standard were identi®ed at 4.8 and 7.8 ppm, respectively. A 16-step phase cycling was used. Resonance assignments: 1, 2, external standard; 3, residual water; 4, 5, C-2 and C-5 protons of the citrate molecule; 6, 7, methylene protons of spermine and lipids. (Reproduced by permission of the Society for Magnetic Resonance Imaging from Thomas et al.10)

approach has been used to study the effects of hormone ablation18 and cryosurgery,19 and for the detection of local recurrence.20,21 2.2 Phosphorus-31 MRS of Prostate Kurhanewicz et al.22 reported the use of a 31P MRS transrectal probe for studies of the human prostate (Figures 11±13). The preliminary results indicated that transrectal 31P MRS may characterize 31P metabolites in normal prostates, benign prostatic hyperplastia, and malignant prostates. The preliminary results suggested that malignant prostates are characterized by signi®cantly decreased levels of PCr and increased levels of PME compared to healthy prostates. Thomas et al. evaluated some of the problems encountered with transrectal 31P MRS of human prostate to determine the optimal conditions for these studies.23 The authors investigated the reproducibility of 31P MRS, regional differences of 31P metabolites, the T1 relaxation times, and metabolic alterations associated with disease. The PME/ATP ratio was highest in the upper region and lowest in the lower region. Similarly, the PDE/ATP ratio was highest in the upper region and lowest in the lower region. In contrast, the PCr/ATP was lowest in the upper region but was increased in the lower region. The PME/ATP ratio of normal subjects (0.09 ‹ 0.1) was increased in the patients with BPH (1.5 ‹ 0.1) and signi®cantly increased in patients with cancer (1.7 ‹ 0.2). The PCr/ATP ratio in normal subjects (1.5 ‹ 0.2) was not signi®cantly

reduced in BPH, but it was reduced to 0.9 in prostatic cancer. The PME/PCr ratio in normal subjects was 0.7, was not signi®cantly increased in BPH to 1.4 but was signi®cantly increased in prostate cancer to 2.2. Hering and Muller reported 31P MRS and 1H MRI of the human prostate with a transrectal probe.24 Fourteen patients were evaluated with 1H MRI and seven patients with 31P MRS. The PME/ATP ratio was higher in patients with prostatic cancer. Narayan et al. investigated the ability of 31P MRS to characterize normal human prostate as well as prostate with benign and prosthetic hypertrophy and malignant neoplasms (Figure 14).25 Normal prostate had PCr/ATP, PME/ATP, and PME/PCr ratios of 1.2, 1.1, and 0.9, respectively. Malignant prostates had PCr/ATP ratios that were lower than normal prostates. Malignant prostates had PME/ATP ratios that were higher than normal prostates. Using the PME/PCr ratio it was possible to differentiate metabolically malignant prostates from normal prostates with no overlap of individual ratios. 2.3

Carbon-13 MRS of Prostate

Halliday et al. obtained high-quality, high-resolution proton decoupled natural abundance 13C NMR spectra from various human tumors, including prostate (Figure 15).26

KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS C

5

PCr L

(a) PME ATP Pi PDE

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5 ppm

L (b)

10

3.5

3.0

2.5

0

–5 ppm

–10

–15

–20

Figure 11 Phosphorus-31 magnetic resonance spectrum of a normal prostate from a normal subject (26 years old), TR = 20 seconds; NS = 100. (Reproduced by permission of the Society for Magnetic Resonance Imaging from Thomas et al.23)

C

4.0

5

2.0

1.5

1.0

0.5 ppm

Figure 9 (a) In vitro 1H NMR spectra at 360 MHz, demonstrating a normal pattern with a large citrate peak seen in a sample of benign prostatic hyperplasia with a glandular predominance (C, citrate; L, lactate). Standard not shown. (b) In vitro 1H NMR spectra at 360 MHz for adenocarcinoma showing the expected low citrate peak. Standard not shown. (Reproduced by permission of Williams & Wilkins from Schiebler et al.12)

When prostatic adenocarcinoma was compared with adjacent hyperplastic tissue, the tumors were found to contain larger amounts of triacylglycerols, smaller amounts of citrate, and acid mucins. The citrate-to-lipid ratio appeared to differentiate malignant from nonmalignant prostates.26 Halliday et al. obtained 13C NMR spectra from prostate tumor cell lines.27 The results showed that the amount of taurine was increased, and tyrosine was decreased in androgen sensitive rat prostatic tumors in comparison to androgen responsive malignant or normal tissue. The authors concluded that the amount of these

C PME

–80

PCr

–60

ATP –40

Pi

(b)

PDE

0

Hz

–20

20 40 (a) 60

C

80 4.5

4.0

3.5

3.0

2.5 ppm

2.0

1.5

1.0

0.5

Figure 10 Two-dimensional J resolved spectrum of an HClO4 extract of a normal peripheral zone epithelial cell strain, obtained at 500 MHz. The chemical shift is resolved in the horizontal axis and the J coupling constant of complex peaks is resolved in the vertical axis. The complex peaks assigned to citrate (C) are pointed out. A one-dimensional projection in the chemical shift dimension is plotted at the top. (Reproduced by permission of Williams & Wilkins from Yacoe et al.14)

10.00

5.00

0.00

–5.00 –10.00 ppm

–15.00

–20.00

Figure 12 Phosphorus-31 magnetic resonance spectra from human prostates in patients with (a) BPH and (b) prostatic cancer. (Reproduced by permission of the Society for Magnetic Resonance Imaging from Thomas et al.23)

6 KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS 11

PCr (b) 31

ATP PME (a)

Pi

PDE

46

30

32 33

41 45

34

40 49

12

22

28 27 18 25 19 23

13 14 15 16

39 42

43

5 4 3 1 2

17

44

10 9 8 7 6

36 35

37

(b) 13 11 15 14

28

(a)

25 46

(c)

31

45

10.00

5.00

0.00

–5.00 –10.00 ppm

–15.00

32

41

44 49

16

40 43 ba

42

21

18

24

34

10 6 7 5 4 3

17

2

39

1 37

Std.

–20.00 180

Figure 13 Phosphorus-31 magnetic resonance spectra from (a) upper, (b) middle, and (c) lower regions of the prostate. (Reproduced by permission of the Society for Magnetic Resonance Imaging from Thomas et al.23)

160

140

120

100 80 d (ppm)

60

40

20

0

Figure 15 Natural abundance proton decoupled 100.6 MHz 13C NMR spectrum of (a) human prostate with benign prostatic hypertrophy. This proton decoupled spectrum was taken from 2.24 g of tissue, at a temperature of 310 K, with a total of 17 271 scans. (b) A spectrum of poorly differentiated adenocarcinoma of human prostate from the same individual as (a), taken from 3.51 g of tissue with the same parameters, except for the use of 11 241 scans in total. It should be noted that peak 11 is truncated. Assignments for the numbered resonances are given elsewhere.26 These spectra are representative of seven hyperplastic prostates, three of which contained adenocarcinoma, plus a fourth sample of adenocarcinoma, with the following exception: lactate was increased relative to the control tissue only in the depicted case and resonances from acidic mucins were not present in all of the tumors. (Reproduced by permission of Williams & Wilkins from Halliday et al.26)

amino acids discriminates androgen-sensitive from insensitive rat prostatic tissues. Sillerud et al. followed this up with an in vivo 13C NMR study of human prostate (Figure 16).28 High levels of citrate were measured in the human prostate in vivo as well as the tissue samples of human and rat prostate in vitro.

3 TESTICLE abnormal testicle; furthermore, the PME/PDE ratio was also reduced in patients with primary testicular failure. In patients with azoospermia, there were signi®cant differences in the same peak area ratios between patients with primary testicular failure and those with chronic tubular

3.1 Phosphorus-31 MRS of Testicle Chew et al. investigated the clinical feasibility of 31P MRS to assess the metabolic integrity of the human testicle (Figures 17 and 18).29 The PME/ATP ratio was greatly reduced in the

(a)

(b)

PCr

(c)

PCr

PME

g-ATP a-ATP

PME

PCr

b-ATP Pi

PME a-ATP Pi g-ATP b-ATP PDE

20

10

0

–10

–20 ppm

Pi

PDE

a-ATP b-ATP g-ATP

PDE

20

10

0

–10

–20 ppm

20

10

0

–10

–20 ppm

Figure 14 In vivo 31P magnetic resonance prostate spectra. (a) Normal volunteer, (b) patient with BPH, and (c) patient with prostatic cancer. (Reproduced by permission of Williams & Wilkins from Narayan et al.25)

KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS

7

Citrate C-2,5

[13C]NaCN Citrate C-4

Citrate C-1, 6

HC

CH

C-3

200.0

180.0

160.0

140.0

120.0

100.0 ppm

80.0

60.0

40.0

20.0

0.0

Figure 16 Proton decoupled, natural abundance 100.614 MHz 13C NMR in vitro spectrum from 3.88 g of human benign hypertrophic prostate tissue sample. This spectrum is the average of 13 600 scans at a temperature of 283 K. The signal at 165 ppm is from 15 l of a 0.15 M [13C]sodium cyanide standard. (Reproduced by permission of Williams & Wilkins from Sillerud et al.28)

after vasectomy signi®cantly differed from the control GPC/ total phosphate ratio, which appropriately re¯ected complete vasal occlusion. The results suggested that a signi®cant portion of seminal GPC is derived from epidymal secretion and that 31 P MRS is useful for monitoring GPC/total phosphate levels when assessing epidymal function in male infertility.

PME/b-ATP = 1.92 PME ATP

Pi PDE

4

10

0

–10

–20

ppm

Figure 17 Characteristic 31P magnetic resonance spectrum from the in vivo normal human testicle. Present are the three peaks due to ATP, small signals from PDE and Pi, and a large contribution from the PME peak. This spectrum was acquired with 400 signals averaged in 13.5 minutes. (Reproduced by permission of the Radiological Society of North America from Chew et al.29)

obstruction. Bretan et al. compared 31P MRS of human testicle with conventional semen analysis.30 The glycerophosphorylcholine (GPC)/total phosphate ratio in azoospermic men

UTERUS

High-resolution 1H MRS was used as an adjunct to conventional and histological diagnosis of cervical neoplasia.31 Cervical biopsy specimens were examined with 1H MRS and the results compared with histology. A high-resolution lipid spectrum was observed in 39 of 40 invasive carcinomas whereas 119 preinvasive samples showed little or no lipids but were characterized by a strong unresolved peak between 3.8 and 4.2 ppm. Peak ratios of the methylene/methyl and the unresolved/methylene resonances allowed accurate distinction between invasive and preinvasive malignancy.

5

RELATED ARTICLES

In Vivo Hepatic MRS of Humans; NMR Spectroscopy of the Human Heart; Proton Decoupling During In Vivo Whole Body Phosphorus MRS; Proton Decoupling in Whole Body Carbon-13 MRS; Quantitation in In Vivo MRS; Whole Body Studies: Impact of MRS.

8 KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS (a)

(b)

PME ATP

Pi PDE

10

0

–10

–20 ppm

20

0

–20

ppm

Figure 18 (a) Phosphorus-31 magnetic resonance spectrum from an azoospermic testicle due to primary testicular failure. It is characterized by the same peaks found in the normal spectrum, but the PME/ -ATP peak area ratio is substantially lower (1.32 in this case). (b) Phosphorus-31 magnetic resonance spectrum from an azoospermic testicle due to chronic ductal obstruction. It is characterized by the same peaks found in the normal spectrum, but the PME/ -ATP peak area ratio is lower than normal (1.51 in this case). (Reproduced by permission of the Radiological Society of North America from Chew et al.29)

6 REFERENCES 1. N. J. Shah, T. A. Carpenter, I. D. Wilkinson, L. D. Hall, A. K. Dixon, C. E. L. Freer, K. Prosser, and D. B. Evans, Magn. Reson. Med., 1991, 20, 292. 2. M. J. Avison, D. L. Rothman, T. W. Nixon, W. S. Long, and N. J. Siegel, Proc. Natl. Acad. Sci. USA, 1991, 88, 6053. 3. T. Jue, D. L. Rothman, J. A. B. Lohman, E. W. Hughes, C. C. Hanstock, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1988, 85, 971. 4. G. B. Matson, D. B. Twieg, G. S. Karczmar, T. J. Lawry, J. R. Gober, M. Valenza, M. D. Boska, and M. W. Weiner, Radiology, 1988, 169, 541. 5. M. D. Boska, D. J. Meyerhoff, D. B. Twieg, G. S. Karczmar, G. B. Matson, and M. W. Weiner, Kidney Int., 1990, 38, 294. 6. P. N. Bretan, N. Baldwin, A. C. Novick, A. Majors, K. Easley, T. Ng, N. Stowe, P. Rehm, S. B. Streem, and D. R. Steinmuller, Transplantation, 1989, 48, 48. 7. P. N. Bretan, N. Baldwin, A. C. Novick, A. Majors, K. Easley, T. C. Ng, N. Stowe, S. Streem, D. Steinmuller, P. Rehm, and R. Go, Transplant. Proc., 1989, 21, 1266. 8. T. M. Grist, H. C. Charles, and H. D. Sostman, Am. J. Roentgenol., 1991, 156, 105. 9. F. Schick, H. Bongers, S. Kurz, W. I. Jung, M. Pfeffer, and O. Lutz, Magn. Reson. Med., 1993, 29, 38. 10. M. A. Thomas, P. Narayan, J. Kurhanewicz, P. Jajodia, and M. W. Weiner, J. Magn. Reson., 1990, 87, 610. 11. M. D. Schnall, R. Lenkinski, B. Milestone, and H. Y. Kressel, Proc. 9th Ann Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 288. 12. M. L. Schiebler, K. K. Miyamoto, M. White, S. J. Maygarden, and J. L. Mohler, Magn. Reson. Med., 1993, 29, 285. 13. A. H. Fowler, A. A. Pappas, J. C. Holder, A. E. Finkbeiner, G. V. Dalrymple, M. S. Mullins, J. R. Sprigg, and R. A. Komoroski, Magn. Reson. Med., 1992, 25, 140. 14. M. E. Yacoe, G. Sommer, and D. Peehl, Magn. Reson. Med., 1991, 19, 429. 15. J. Kurhanewicz, R. Dahiya, J. M. Macdonald, L. H. Chang, T. L. James, and P. Narayan, Magn. Reson. Med., 1993, 29, 149.

16. H. Hricak, S. White, D. Vigneron, J. Kurhanewicz, A. Kosco, D. Levin, J, Weiss, P. Narayan, and P. Carroll, Radiology, 1994, 193, 703. 17. J. Kurhanewicz, D. Vigneron, H. Hricak, P. Carroll, P. Narayan, and S. Nelson, Radiology, 1996, 198, 795. 18. H. Chen, H. Hricak, C. L. Kalbhen, J. Kurhanewicz, D. Vigneron, J. Weiss, and P. Carroll, Am. J. Roentgenol., 166, 1157. 19. J. Kurhanewicz, H. Hricak, D. B. Vigneron, S. Nelson, F. Parivar, K. Shinohara, and P. R. Carroll, Radiology, 1996, 200, 489. 20. F. Parivar, H. Hricak, J. Kurhanewicz, K. Shinohara, D. B. Vigneron, S. J. Nelson, and P. R. Carroll, Urology, 1996, 48, 594. 21. F. Parivar and J. Kurhanewicz, Curr. Opin. Urol., 1998, 8, 83. 22. J. Kurhanewicz, A. Thomas, P. Jajodia, M. W. Weiner, T. L. James, D. B. Vigneron, and P. Narayan, Magn. Reson. Med., 1991, 22, 404. 23. M. A. Thomas, P. Narayan, J. Kurhanewicz, P. Jajodia, T. L. James, and M. W. Weiner, J. Magn. Reson., 1992, 99, 377. 24. F. Hering and S. Muller, Urolog. Res., 1991, 19, 349. 25. P. Narayan, P. Jajodia, J. Kurhanewicz, A. Thomas, J. MacDonald, B. Hubesch, M. Hedgcock, C. M. Anderson, T. L. James, E. A. Tanagho, and M. Weiner, J. Urol., 1991, 146, 66. 26. K. R. Halliday, C. Fenoglio-Preiser, and L. O. Sillerud, Magn. Reson. Med., 1988, 7, 384. 27. K. R. Halliday, L. O. Sillerud, and D. Mickey, Proc. 11th Ann Mtg. Soc. Magn. Reson. Med., Berlin, 1992, p. 492. 28. L. O. Sillerud, K. R. Halliday, R. H. Griffey, C. Fenoglio-Preiser, and S. Sheppard, Magn. Reson. Med., 1988, 8, 224. 29. W. M. Chew, H. Hricak, R. D. McClure, and M. F. Wendland, Radiology, 1990, 177, 743. 30. P. N. Bretan, D. B. Vigneron, R. D. McClure, H. Hricak, R. A. Tom, M. Moseley, E. A. Tanagho, and T. L. James, Urology, 1989, 33, 116. 31. E. J. Delikatny, P. Russell, J. C. Hunter, R. Hancock, K. H. Atkinson, C. van Haaften-Day, and C. E. Mountford, Radiology, 1993, 188, 791.

Acknowledgements Supported by NIH grant R01AG10897 and the DVA Medical Research Service.

KIDNEY, PROSTATE, TESTICLE, AND UTERUS OF SUBJECTS STUDIED BY MRS

Biographical Sketch Michael W. Weiner. b 1940. B.S., 1961, Johns Hopkins. M.D., 1965, SUNY Upstate Medical Center. Intern and Resident, Mount Sinai Hospital 1965±67. Resident and Fellow in Metabolism, Yale University, 1967±70. Fellow in Biochemistry Institute for Enzyme Research, University of Wisconsin, 1970±72. Faculty at University of Wisconsin,

9

1971±74, Stanford University, 1974±80, University of California San Francisco, 1980±present. Director, Magnetic Resonance Department, Veterans Affairs Medical Center. Professor of Medicine and Radiology, University of California San Francisco. Research interests include application of MRI to investigation of human metabolism and diagnosis of disease.

LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI

1

Liver, Pancreas, Spleen, and Kidney MRI David Stark and Ashley Davidoff University of Massachusetts, MA, USA

1 INTRODUCTION The upper abdomen presents unique diagnostic challenges. Numerous organs with diverse physiological functions are in close proximity with each other, abut the cardiopulmonary system through a thin diaphragm, and have open contact with the dependent organs of the pelvis. Thus, tumors, in¯ammatory disease, and other pathology of the upper abdomen may present clinically with symptoms attributable to an organ involved only secondarily, or they may appear to arise from the thorax or pelvis, leading the clinician away from the primary problem. Imaging of the abdomen is confounded by its large anatomic area, the mobility of the organs and their ¯uids, and the presence of tissues of varying imaging characteristics. This article brie¯y introduces the advantages and limitations of magnetic resonance imaging (MRI) methods as applied to the evaluation of abdominal cancer. After the technical introduction, which provides an understanding of the philosophy of probing the abdomen for known or unsuspected disease, the remainder of the discussion reviews the MRI characteristics of the major upper abdominal neoplasms. Virtually all diagnostic medicine, and certainly diagnostic imaging, is guided by three hierarchical objectives: 1. detection of disease; 2. characterization of disease; and 3. staging of disease. Detection itself is the most important challenge, as it begins with the patients themselves deciding whether a symptom is suf®ciently abnormal to warrant a visit to the doctor. After several clinical steps, imaging is most commonly used to determine whether an anatomic area (head, neck, chest, abdomen, pelvis, or musculoskeletal system) or physiologic organ system (e.g. hepatobiliary system) is normal or abnormal. This simple, binary decision is critical in selecting patients for more intensive investigation at considerable expense and no small risk of iatrogenic complications. Characterization of disease commences once an abnormality is found. In the abdomen, nearly all patients develop benign masses if they live long enough (e.g. renal cyst or nodular hyperplasia of the prostate). Where imaging for other reasons results in detection of these benign masses, imaging must also solve the diagnostic dilemma it created and, at low cost, identify these lesions as benign and of no clinical signi®cance. Staging is essential to determining prognosis and therapy once treatable disease has been characterized. In many ways, staging is a subset of detection as it is the process by which multiple lesions are discriminated from a single abnormality, and the extent of disease within an organ or spread to adjacent

Figure 1 Normal abdomen. T1-weighted spin echo image at the level of the pancreatic head shows a normal common bile duct, gall bladder, hepatic artery, and con¯uence of the splenic and portal veins. The absence of a motion artifact and the low signal intensity of kidneys relative to that of the liver are clues that this is a short TE sequence

areas is measured. Since staging often involves monitoring of treatment, it presents unique problems because the background normal structures may be altered by surgery, radiation, or progression of the disease itself. As society raises ethical and economic questions concerning the merit of treating various conditions, imaging becomes more important as a noninvasive low-cost method of determining which treatments are appropriate and effective. 2

IMAGING TECHNIQUES

MRI is similar to computerized tomography (CT) in providing a full ®eld of view of the abdomen, and a permanent record of the entire examination, to a comparable level of anatomic detail (spatial resolution). MRI has the further advantage of allowing selection of sagittal, coronal, or oblique planes of section. However, in the abdomen virtually all imaging is performed in the transverse plane (see Figure 1). MRI in the 1990s, like CT in the 1970s, is getting faster, with the expectation that speed will solve the problem of motion. Debate rages concerning the optimal methods for obtaining T1 or T2 contrast and the best signal-to-noise ratio (S/N) per unit time. These technical details are addressed elsewhere in this volume. 3

IMAGE QUALITY

Speci®c magnetic resonance (MR) pulse sequences and their user-selectable timing parameters are of utmost importance for lesion±liver contrast and image quality. MR image quality can be quanti®ed by calculating the lesion±liver contrast-to-noise ratio (CNR) and the liver S/N.1 Motion artifacts contribute

2 LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI most to (systematic) noise in abdominal imaging. Vascular ghost artifacts arising from the aorta and inferior vena cava may obscure focal hepatic abnormalities, especially in the left lobe. Changing the phase-encoding direction results in artifacts projected outside the region of interest.2,3 Signal averaging or implementation of presaturation pulses are highly effective methods of motion artifact reduction in the upper abdomen. The CNR is an effective parameter for quantitating pulse sequence performance with respect to lesion detection. A useful rule is that spleen±liver CNR will, on average, match cancer± liver CNR. This rule holds for all pulse sequences, on all machines, and at all ®eld strengths, since, on average, cancer and spleen have the same proton density, T1 and T2. Thus, the spleen can serve as a reference signal intensity, and the spleen±liver CNR will be predictive of pulse sequence performance for liver cancer detection.

4 PULSE SEQUENCE PERFORMANCE Theoretical calculations suggest that the optimal pulse sequence for hepatic tumor detection varies with ®eld strength. However, clinical results have not con®rmed any signi®cant ®eld strength differences. Performance variations due to gradients, other hardware, and software appear to be more substantial. The dominant factor appears to be the level of commitment of the practicing radiologist who implements and maintains quality imaging protocols tailored to the speci®c machine and its generation (upgrade level). In an area such as the abdomen, of which relatively few MRI examinations are conducted, the necessary expertise and attention to detail is too often lacking. Properly performed T1- or T2-weighted sequences yield comparable CNR values and hence ef®cacy at liver cancer detection. Cancer or lymphoma of the spleen is poorly detected by any method, since cancer and spleen have the same proton density, T1, and T2, on average. T1-weighted sequences generally offer greater S/N per unit time than do T2-weighted sequences. This fact, combined with excellent fat±tissue T1 contrast, leads to the choice of short TE T1-weighted sequences for imaging the pancreas, kidneys, adrenal glands, and the remainder of the retroperitoneum and mesentery. While T1-weighted images offer superior anatomic resolution and are generally superior for detection of abdominal pathology, it is well established that T2-weighted sequences are preferred for tissue characterization, especially in the liver and adrenal glands. Less is known about the kidney, where both benign (solid or cystic) and malignant lesions are also common. 4.1 Contrast Media Parenchymal tissue contrast is enhanced by intravenous or arterial administration of magnetopharmaceuticals. Commonly gadolinium diethylenetriaminetetraacetic acid, Gd-DTPA, (Magnevist) or the newer nonionic contrast agents, such as GdDTPA-BMA (Omniscan) or GdDO3A (ProHance) are given by the intravenous route. These agents are designed for rapid excretion and do not signi®cantly interact with body physiology. Binding to blood proteins or cells is negligible. Excretion is by passive glomerular ®ltration. Thus, these agents, introduced to

the blood stream, will penetrate capillaries in the body and have an extracellular distribution, although in the brain they are limited to the vascular space by the blood±brain barrier. It is the molecular weight of these agents (less than 1000 Da) that determines their diffusion into the extracellular space of the abdominal organs. Intravenous contrast agents are concentrated in the renal tubules, distinguishing functioning renal tissue from all but the most vascular kidney tumors. Unfortunately, in the liver, spleen, and pancreas, the blood supply, capillary permeability, and capacity of the extracellular space of tumors is quite variable. The kinetics and degree of enhancement of the tumors is often described in general terms such as `hypovascular' or `hypervascular'. Unfortunately, enormous biological variation exists, and tumors of a common cell type (e.g. colon adenocarcinoma) show wide variations in enhancement characteristics. Efforts to enhance selectively liver parenchyma and improve the contrast between normal liver and cancer include rapid (`dynamic-bolus') intravenous infusion to exploit the dual (portal venous and hepatic arterial) vascular supply of normal liver as opposed to cancers, which are typically supplied by arteries and may have fewer vessels than normal liver. Dissatisfaction with intravenous techniques has stimulated attempts to administer contrast agents at angiography via selective cannulation of the hepatic artery or via the superior mesenteric artery, looking at the portal venous phase of contrast circulation. This latter approach may improve tumor±liver image contrast, and some authors have claimed improved sensitivity for detecting liver lesions. However, due to variations in the portal venous perfusion of the liver, nonneoplastic perfusion inhomogeneities create numerous false-positive diagnoses of cancer, lowering speci®city and limiting the value of this technique for planning therapy. For more than a decade, investigators have attempted to adapt cholegraphic agents, taken up by hepatocytes and excreted into the biliary system, for use in the detection of liver cancer.4 Similarly, colloidal or particulate magnetopharmaceuticals have been used to target the normal hepatic Kupffer cells in an attempt to contrast liver against cancer, which does not show phagocytic uptake of magnetic particles. Unfortunately, attempts to target particulate agents in bulk to normal liver have been complicated by toxic side effects. Newer superparamagnetic iron oxide agents under development may show improved safety pro®les.5

5 5.1

THE LIVER Metastases

The liver is the most common organ in the body to which cancer spreads from primary disease elsewhere. Gastrointestinal neoplasms, including those in the pancreas, nearly always involve the liver before metastasizing to other organs. In addition, breast, lung, renal, and ovarian and other common malignancies spread hematogenously, lymphatically or transperitoneally to the liver. Indeed, liver failure is one of the most common causes of death in cancer patients, and liver metastases serve as a basis for determining prognosis and monitoring therapy. Serologic studies of liver function, including enzymes leaked into the blood stream when the liver is injured [serum

LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI

glutamic oxaloacetate transaminase, (SGOT), serum glutamic pyruvic transaminase (SGPT), and alkaline phosphatase], have such poor sensitivity and speci®city that they cannot be justi®ed for use as an independent diagnostic test. When metastatic cancer to the liver is suspected or a reasonable risk exists, imaging must be performed. Unfortunately, imaging also has its limitations and pitfalls. Metastatic spread of cancer is thought to begin as a single cell or a small cluster of cells, well below the resolution limit of imaging. Indeed, imaging cannot reliably detect liver lesions smaller than 5 mm in size. As a metastasis grows exponentially it is therefore submillimeter in size for the vast majority of its `life cycle' before it enlarges to the point of causing symptoms and ultimately killing the patient. Therefore, it is obvious that the majority of metastases present in an individual, or speaking epidemiologically, the majority of metastases present in a population, are below the threshold size for detection by any means. 5.2 Imaging: Detection Liver cancer detection does offer clinically useful information because most patients with metastatic cancer have multiple lesions, some of which are suf®ciently large (>1 cm diameter) to be detected reliably. The accuracy of imaging for detecting liver metastases is the subject of endless disputes and revision of the radiological literature. The major problem with the literature is the lack of suitable `gold standards'. It is simply impossible to collect a large series of patients for whom autopsy and histologic con®rmation of imaging ®ndings are available. Thus, radiologists frequently resort to comparative studies where one modality is compared to another, and a third technique (such as surgery) or a third imaging modality is used as the arbiter of `truth'. Such comparative studies consistently overestimate the accuracy of detection of liver lesions.6±9 It is now generally accepted that MRI is the best test, followed by CT, ultrasonography, and then angiography. CT angioportography (CTAP) techniques are equal to or slightly better than MRI; however, the angiographic CT methods are generally not available and are nearly 10 times more expensive. Therefore, when the clinician is able to focus the diagnostic question of the detection of liver metastases, MRI is now the preferred technique. Unfortunately, clinicians often have mixed objectives and request inspection of the adrenal glands, retroperitoneum, and other abdominal organs. As a practical matter, in such unfocused clinical situations, ultrasonography or CT are preferred, despite their inaccuracy, as widely available, cheap, and, arguably, acceptable (i.e. `cost-effective') compared to MRI. In summary, for patients at risk of metastatic cancer it is an economic and public policy issue whether the best test (MRI) is made available. Given the cost of misdirected cancer therapy in patients with undiagnosed metastases (in societies where cancer treatment is available), it is likely that diagnostic imaging pays for itself. Unfortunately, neither MRI nor CT is available to many patients. For a variety of nonscienti®c reasons, doctors often accept ultrasonography as the primary screening test, notwithstanding the hidden costs of false-negative and false-positive examinations. The dominant worldwide use of ultrasonography for liver imaging may in fact be cost-effective, as this rapid and inexpensive method does reliably identify the majority of patients

3

Table 1 Relaxation Times for Liver, Hemangioma, Metastases, and Hepatocellular Carcinoma at 0.6 Ta Tissue Normal liver Hemangioma Metastases Hepatocellular carcinoma

T1

T2

499 ‹ 140 1010 ‹ 497 691 ‹ 100 569 ‹ 133

48 ‹ 11 143 ‹ 51 71 ‹ 21 87 ‹ 17

a

Values are the mean ‹ SD calculated from four or more spin echo measurements.

having metastatic cancer. Since the treatments available have such a poor outcome, is it any wonder that misdiagnosis and inappropriate therapy combined do not noticeably alter this miserable outcome? Despite government pressure to sacri®ce health and life that is not measurable or statistically cost-effective, medical ethics dictates that in those cases having a solitary or questionable liver lesion at ultrasonography, CT should be made available. Setting aside economic issues, stateof-the-art oncologic practice requires use of either CT or MRI to screen for liver metastases in the abdomen. 5.3

Lesion Characterization

MRI is more likely than CT to demonstrate internal nodular structure, rings and hemorrhage. In particular, T2-weighted MR images are useful as they show edema at the border of active lesions which may be manifested as rings or as geographic zones of increased signal intensity. T1-weighted images are less sensitive to edema; as a result, metastases are virtually always the same size or larger on T2-weighted images. 5.4

Staging

Staging of hepatic neoplasms principally involves identifying the lobe and segment of disease. Patients with larger tumor burdens and receiving chemotherapy can be monitored by scan-to-scan comparison of tumor diameters or estimates of tumor volume. Some idea about the extent of differentiation that is possible is given by Table 1. Surgical approaches to hepatic metastases depend primarily on con®dence that all of the tumors have been identi®ed. If there is reason to believe that a solitary metastasis exists or that a cohort of a few similar sized lesions represent the only disease in the liver, then the question of resection depends upon the functional hepatic reserve and the technical ability to remove the lesions. Lesions invading an adjacent structure, such as the diaphragm or inferior vena cava, or lesions in dif®cult areas, such as the porta hepatis, make surgery challenging and risky. Undoubtedly, the greatest risk to resective therapy is undiagnosed residual disease. 6

HEPATOCELLULAR CARCINOMA

Also known as `hepatoma', primary malignancy of hepatocytes [i.e. hepatocellular carcinoma (HCC)] comprises 1% of all cancers in the USA. Worldwide, variation in nutritional fac-

4 LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI tors and the incidence of viral hepatitis10,11 account for major differences in the incidence and nature of HCC. It is far more common in Japan, the rest of Asia, and sub-Saharan Africa in association with chronic viral hepatitis. Articles can be found in the literature citing a wide range of sensitivity, speci®city, and overall accuracy for ultrasonography, CT and MRI.13,14,15 It is evident that HCC is more dif®cult to detect with any of the imaging methods than is metastatic cancer. The principal reason for this dif®culty is that HCC most often occurs against a background of chronic hepatitis, fatty liver and cirrhosis.11±14 Fatty change within malignant hepatocytes serves to decrease tumor±liver contrast on conventional MR images, although the presence of fat can be exploited to increase contrast if chemical shift selective techniques are used. Ongoing regeneration and areas of nodular hyperplasia serve to mask or mimic hepatoma. Bands of scar tissue in the liver have increased water content, and the long T1/T2 signal intensity characteristics are very similar to hepatoma. 6.1 Lesion Characterization Characterization of hepatoma and discrimination from metastatic cancer or benign hepatic masses is rarely possible with ultrasonography, CT, or scintigraphy. MRI, however, has the ability to detect three features characteristic of hepatoma. Each of these features is present in approximately 30% of hepatomas, and therefore one or more features can be found in a majority of cases. First, a capsule of compressed liver or scar tissue may create a sharp boundary with adjacent liver tissue. The capsule has long T1/long T2 signal intensity characteristics, and is usually best seen on T1-weighted images. The capsule must be distinguished from low signal intensity blood vessels on T1-weighted spin echo images. Hepatic adenoma may also show peripheral capsules indistinguishable from those of hepatoma. However, hepatic adenoma is an unusual tumor with different demographics. Tumor and capsules can be distinguished from the rings of metastases, as the latter are better seen on T2-weighted images and are not well seen at all on T1-weighted images. Images from a hepatic adenoma study are shown in Figure 2. Second, hepatomas accumulate intracellular triglyceride. Fatty accumulation can be detected using a variety of chemical shift methods such as phase contrast frequency-selective imaging. Although fatty accumulation may occur in injured hepatocytes in other conditions, including adenoma, focal nodular hyperplasia, hepatitis, and dietary disturbances, it is usually possible to distinguish focal fatty in®ltration from fat within a mass, as the latter displaces normal blood vessels, while focal fat in®ltration usually does not. Third, hepatoma has a propensity to grow into hepatic and portal veins, and MRI is able to demonstrate this morphology to advantage. Certainly ultrasonography also has this capability; however, the ability of ultrasonography to distinguish the solid tumor from adjacent liver makes it more dif®cult to follow the tumor into vessels.17

7 BENIGN DISEASE: CAVERNOUS HEMANGIOMA Hemangioma is a vascular malformation characterized by a cavernous collection of blood spaces (Figure 3). Blood ¯ow is

Figure 2 Hepatic adenoma. (a) A CT scan enhanced with bolus infusion of iodinated contrast media shows a large enhancing lesion in the left hepatic lobe. This ®nding is nonspeci®c and resembles cancer or any other solid neoplasm. (b) A catheter angiogram suggests a benign solid tumor, most likely hepatic adenoma. (c) MRI (SE 1500/ 100) shows a heterogenous mass that is isointense with the spleen, consistent with any solid neoplasm; surgery con®rmed the diagnosis

very slow and virtually undetectable, except for the occasional peripheral site of venous entry. Hemangioma is exceedingly common in both sexes and all races around the world. Approximately 15±20% of the adult population has such a lesion.8

LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI

5

Figure 3 Cavernous hemangioma of the liver. (a) A CT scan shows a 3 cm lesion in the posterior segment, right hepatic lobe, with a hypodense appearance identical to cancer. (b) MRI (SE 300/15) shows the lesion to have a low intensity, similar to cancer (very slightly darker than the spleen). Note a second, 1 cm lesion in the anterior segment not seen on the CT scan. (c) MRI (SE 2400/180) shows the lesions to be hyperintense relative to spleen, and nearly isointense with cerebrospinal ¯uid; this is consistent with blood in cavernous hemangiomas. (d) Lesions as small as 3 mm diameter can be characterized by MRI

7.1 Lesion Characterization MRI has been remarkably successful in identifying the ¯uid content of cavernous hemangiomas by its long T2 relaxation time (Table 1).15,16 Long TE spin echo images, which reduce the signal intensity of solid tissue relative to ¯uid, show cavernous hemangiomas, cerebrospinal ¯uid, bile, and gastric contents as extremely bright structures. Properly performed, with TE in excess of 120 ms and successful suppression of motion artifact, cavernous hemangiomas are homogeneous and sharply circumscribed on the MR image. Tumors, on the other hand, are ill-de®ned, heterogeneous, and lower in signal intensity due to their gelatinous (solid) make-up. Contrast-enhanced blood-pool studies provide an alternative method of identifying hemangiomas and other vascular lesions. CT uses iodine based diagnostic pharmaceuticals; scintigraphy uses technetium-labeled red blood cells, and MRI uses the paramagnetic agent Gd-DTPA. In addition, iron oxide particles and other magnetic contrast media have

been shown to be effective blood-pool markers. The concept behind these studies depends principally on the kinetics and distribution of the vascular agents. Cavernous hemangiomas tend to enhance uniformly over 5±30 min and may retain contrast longer than the circulating blood pool. Solid tumors show more rapid wash-in and wash-out of the contrast agent and usually show less peak enhancement than do cavernous hemangiomas. As these ®ndings depend upon the cardiac output, circulation time, timing of contrast administration, and the speci®cs of the perfusion pattern of the vascular malformation, contrast-enhanced studies are less predictable than plain T2-weighted MRI. Nevertheless, MRI can be used in both methods. Controlled studies have shown that MRI can discriminate cavernous hemangioma from solid neoplasm with an overall accuracy of 90%. If, in addition to this T2-weighted technique, a contrast study is done, the accuracy increases further. With MRI one can perform both an unenhanced and enhanced scan within a single 1-h examination.

6 LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI 8 THE PANCREAS Adenocarcinoma is a malignant mucinous neoplasm of the pancreas that arises from the ductal epithelium. It is the fourth most common cause of death from cancer in the USA and accounts for 3% of all cancers. Adenocarcinoma represents 75% of nonendocrine pancreatic malignancies. The disease usually affects people over the age of 70 years, with a male/ female ratio of 1.5: 1.0. Proposed etiologic factors include cigarette smoking and coffee consumption, alcohol consumption and dietary intake of fat, protein and a high number of calories.7,18,19 Disease states of the pancreas that are reportedly associated with the development of pancreatic carcinoma include diabetes mellitus and familial pancreatitis. The etiologic association with chronic pancreatitis is not clear. Because the pancreas is not encased by a capsule, early spread into the surrounding retroperitoneal fat is common. Local structures then become involved, including the portal vein, superior mesenteric vein, splenic vein, gastroduodenal artery, bile duct and duodenum. Local lymph-node involvement and hematogenous spread to the liver via the portal vein are also seen. 8.1 Imaging: Detection Lesions in the head of the pancreas are generally visualized by both ultrasonography and CT. The former is limited by surrounding bowel gas or obesity in about 30% of patients. Maneuvers such as altering respiration, ®lling the stomach with water, placing the patient in the upright position and repeat scanning are helpful. Lesions in the tail of the pancreas are dif®cult to detect by ultrasonography, but visualization can be improved by using the spleen as a window. With CT, tail and body lesions are sometimes obscured by a partial volume artifact from a bowel not ®lled with contrast. Intraoperative ultrasound scanning is very sensitive to small and impalpable tumors, particularly for small, functioning tumors such as insulinomas. MRI has little or no role at the present time due to inferior anatomic resolution and no better contrast than ultrasonography or CT. 8.2 Characterization Ultrasonography, CT, and MRI are equally nonspeci®c in distinguishing pancreatic cancer from benign masses. Pancreatitis is the most important differential diagnosis. In most cases, radiologically guided ®ne-needle aspiration biopsy (FNAB) is necessary for diagnosis.

9 THE SPLEEN The spleen rarely causes clincial symptoms and is most often imaged as a bystander to a study directed at another abdominal organ. Trauma and suspected rupture is the most frequent indication for imaging the spleen itself. While MRI would be preferred, CT is usually selected because of its availability and resistance to motion artifacts.

Oncologic disease within the spleen is infrequently identi®ed, even though more than half of patients with widespread malignancy may have splenic involvement at autopsy. Melanoma probably accounts for most of the metastases found in the spleen, although various studies have described the origin of metastases, in decreasing order of frequency, as lung, prostate, colon, stomach, melanoma, ovary, and pancreas. 9.1

Lymphoma

Imaging of lymphoma has traditionally been insensitive to splenic involvement. Splenic size is unreliable, as one-third of patients with splenomegaly do not have lymphomatous involvement, and one-third of patients without splenomegaly have such an involvement. In patients with aggressive disease, such as the poorly differentiated nodular type, involvement becomes probable, but not necessarily visible.20 Ultrasonography, CT, and scintigraphy have unsatisfactory accuracy rates, ranging between 54% and 75%. MRI is beginning to show some promise. When the water content of lymphoma is high, such as in the large lesions of histiocytic lymphoma, they become detectable on T2-weighted sequences. MRI using superparamagnetic iron oxide is showing promise in demonstrating splenic involvement, but these research ®ndings need further clinical veri®cation.21

10

THE KIDNEY

Renal disease, like diseases of the liver and pancreas, can be divided between diffuse and focal processes. Most diffuse disease is in¯ammatory or metabolic in nature. Solid renal masses are cancers until proven otherwise, usually at the expense of nephrectomy given the habit of urologists to avoid referral of patients for FNAB, a less expensive alternative practice by radiologists. Since the aging population at risk of cancer has a high prevalence of benign simple cysts, it has been a challenge to identify benign lesions noninvasively, in order to save good kidneys from the knife. Unfortunately, features demonstrable by ultrasonography, CT, and MRI add little to the low pretest probability of malignancy. The few cancers picked up as cystic structures with nodular irregularities or septations by ultrasonography, calci®cations or density changes on CT, or hemorrhage on MRI are outnumbered by similar ®ndings in nonmalignant cysts falsely diagnosed as positive for malignancy. While the identi®cation of renal cysts by MRI should be as straightforward as identi®cation of hepatic hemangioma, comparable research has not been performed to establish quantitative or qualitative criteria. Furthermore, renal cell carcinoma is more dif®cult than other cancers to distinguish from cysts because of its hypervascularity, tendency for necrosis, and the resultant long T1/T2, approaching that of cysts. MRI studies in the kidney have largely consisted of anatomic descriptions recapitulating the ®ndings known from CT, albeit less clearly. The sole application of MRI to the diagnosis and management of renal disease is inspection of the draining veins for extension of tumor. For all other questions, ultrasonography and CT are preferred. As a result, very little renal MRI is practiced.

LIVER, PANCREAS, SPLEEN, AND KIDNEY MRI

11

SUMMARY

Abdominal imaging is historically anatomic, based on the delineation of high contrast fatty tissue boundaries. Ultrasonography and CT are widely available, familiar to clinicians, competitive in diagnostic quality, and cheap. MRI of the abdomen is best established in the liver for detection and differential diagnosis of focal masses.

12

RELATED ARTICLES

In Vivo Hepatic MRS of Humans; Kidney, Prostate, Testicle, and Uterus of Subjects Studied by MRS; Lung and Mediastinum MRI; Male Pelvis Studies Using MRI; MRI of the Female Pelvis.

13

REFERENCES

1. R. E. Hendrick, T. R. Nelson, and W. R. Hendee, Magn. Reson. Imag., 1984, 2, 193. 2. G. M. Bydder, J. M. Pennock, R. E. Steiner, S. Khenia, J. A. Payne, and I. R. Young, Magn. Reson. Imag., 1985, 3, 251. 3. D. R. Bailes, D. J. Gilderdale, G. M. Bydder, A. G. Collins, and D. M. Firmin, J. Comput. Assist. Tomogr., 1985, 9: 835. 4. Y. M. Tsang, M. Chen, and G. Elizondo, in `Sixth Annual Meeting, Society for Magnetic Resonance Imaging, Boston, MA, 1988', 6 (S1), 124. 5. D. D. Stark, R. Weissleder, G. Elizondo, P. F. Hahn, S. Sains, L. E. Todel, G. J. Wittenberg, and J. T. Ferrucci, Radiology, 1988, 168, 297. 6. D. D. Stark, J. Wittenberg, R. J. Butch, and J. T. Ferrucci, Radiology 1987, 165, 399. 7. D. G. Mitchell and D. D. Stark, `Hepatobiliary MRI', Mosby, St Louis, 1992. 8. K. Okuda, K. G. Ishak, `Neoplasms of the Liver', Springer, Berlin, 1987.

7

9. J. P. Heiken, P. J. Weyman, J. K. Lee, D. M. Balfe, D. Picus, E. M. Brunt, and M. W. Flyte, Radiology, 1989, 171, 47. 10. P. F. Hahn, D. D. Stark, S. Saini, E. Rummeny, G. Elizonilo, R. Weissleder, J. Wittenberg, and J. T. Ferrucci, Am. J. Roentgenol., 1990, 154, 287. 11. M. Ebara, M. Ohto, Y. Watanabe, K. Kimura, H. Saisho, Y. Tsuchiya, K. Ohuda, N. Arimuzu, F. Konilo, and H. Ikcheva, Radiology, 1986, 159, 371. 12. P. R. Ros, B. J. Murphy, J. E. Buck, G. Olmedilla, and Z. Goodman, Gastrointest. Radiol., 1990, 15, 233. 13. H. Yoshida, Y. Itai, K. Ohtomo, T. Kohubo, M. Minami, and N. Yashiro, Radiology, 1989, 171, 339. 14. E. Rummeny, R. Weissleder, D. D. Stark, S. Saini, C. C. Compton, W. Bennett, P. F. Hohn, J. Wittenburg, R. A. Malt, and J. T. Ferruci, Am. J. Roentgenol., 1989, 152, 63. 15. Y. Itai, K. Ohtomo, S. Furui, T. Yamauchi, M. Minomi, and N. Yashiro, Am. J. Roentgenol., 1985, 145, 1195. 16. D. D. Stark, R. C. Felder, J. Wittenberg, S. Saini, R. J. Butch, M. E. White, R. R. Edelman, P. R. Mueller, J. F. Simeone, and A. M. Cohen, Am. J. Roentgenol., 1985, 145, 213. 17. D. D. Stark, P. F. Hahn, C. Trey, M. E. Clouse, and J. T. Ferrucci, Am. J. Roentgenol., 1986, 146, 1141. 18. A. L. Cubilla and P. J. Fitzgerald, Clin. Bull., 1978, 8, 143. 19. R. E. Schultz and N. J. Finkler, Mt. Sinai J. Med., 1980, 47, 622. 20. M. Federle and A. A. Moss, CRC Crit. Rev. Diagn. Imaging, 1983, 10, 1. 21. R. Weissleder, G. Elizondo, D. D. Stark, P. F. Hahn, J. Margel, J. F. Gonzalez, F. Saini, W. E. Todel, and J. T. Farruci, Am. J. Roentgenol., 1989, 152, 175.

Biographical Sketch David D. Stark. b 1952. B.A., 1974, M.D., 1978, Harvard University, USA. Postdoctoral work, University of California, San Francisco. Successively, instructor, assistant professor and associate professor of radiology, Harvard Medical School, Massachusetts General Hospital. Currently Professor of Radiology, University of Massachusetts Medical Center, USA. Approx. 200 publications. Current research specialties: contrast media, clinical applications of MRI, quantitation of tissue iron.

MALE PELVIS STUDIES USING MRI

1

Male Pelvis Studies Using MRI Hedvig Hricak Memorial Sloan-Kettering Cancer Center, New York, NY, USA

and William T. Okuno University of California, San Francisco, CA, USA

1 INTRODUCTION Superb soft tissue contrast resolution and multiplanar imaging capability have placed MR imaging in an eminent position in the evaluation of pelvic anatomy and pathology. In the male pelvis, much work has been performed in studying the prostate gland, seminal vesicles, testes, penis, and urethra. While the results of MR imaging in these areas were promising from the very beginning, these applications are still wrapped in controversy, and the indications are not generally accepted by many clinicians. As MR techniques keep evolving and overall image resolution is improved, MRI is gaining recognition and converting many skeptics into ardent advocates.

2 PROSTATE GLAND 2.1 Anatomy The mature prostate is composed of both glandular and nonglandular tissue. The glandular portion is differentiated into three major zones: peripheral (70%); central (25%); and transition (5%) zones.1 A small amount of glandular tissue is also present in the periurethral glands. Zonal differentiation is clinically signi®cant because most (68%) prostate carcinomas arise in the peripheral zone whereas benign prostatic hyperplasia (BPH) usually originates in the transition zone.2 Nonglandular tissue within the prostate includes the urethra and anterior ®bromuscular stroma. The normal prostate gland demonstrates a homogeneous intermediate signal intensity on T1-weighted images, regardless of patient age or magnetic ®eld strength; T2-weighted images are necessary to delineate the zonal anatomy of the prostate (Figure 1).3 The peripheral zone is of high signal intensity, equal to or greater than that of the adjacent periprostatic fat. In contrast, the central zone has a lower signal intensity than the surrounding peripheral zone.3 These differences in signal characteristics are thought to be related to the presence of striated muscle and fewer glandular elements in the central zone.3 The transition zone also has a lower signal intensity than the peripheral zone. The signal intensities of the central and transition zones are similar at all imaging parameters, and the two can be differentiated only by knowledge of their respective anatomic locations.3 On Gd chelate-enhanced T1-

Figure 1 Prostate gland, normal anatomy: T2-weighted image, endorectal coil, coronal plane of section. The peripheral zone (P) of the prostate gland demonstrates homogeneous high signal intensity. Seminal vesicles (SV) are seen cephalad to the prostate gland. They are convoluted in Nature and have a grape-like appearance

weighted images, the zonal anatomy of the gland is depicted but not as consistently as on T2-weighted images.4 2.2 2.2.1

Pathology Congenital Anomalies

Agenesis and hypoplasia of the prostate are frequently encountered with other congenital anomalies of the genitourinary tract. While MRI can establish the diagnosis, this frequently is an associated ®nding and not the clinical reason for the study. Congenital cysts of the prostate are the most common anomalies of this gland. Prostatic utricle and MuÈllerian duct cysts are midline structures that arise at the level of the verumontanum and extend cephalad (Figure 2). Their signal intensity is nonspeci®c and depends on the nature of their content, ranging from serous to proteinaceous and hemorrhagic ¯uid.5 2.2.2

In¯ammatory Disease

MR imaging does not play a role in the diagnosis of prostatic in¯ammatory disease. When discovered incidentally, acute prostatitis will demonstrate diffuse glandular enlargement and the peripheral zone sometimes exhibits a low signal intensity on the T2-weighted image.6 Thus far, there is only limited experience in MR imaging of prostate abscess. If this diagnosis is clinically suspected, transrectal ultrasound (TRUS) or CT should be the modalities of choice. 2.2.3

Benign Prostatic Hyperplasia

Benign prostatic hyperplasia (BPH) is a glandular and/or stromal proliferation in the transition zone and periurethral glands. The prostate gland is enlarged, and often the hyperplastic tissue is separated from the compressed peripheral zone

2 MALE PELVIS STUDIES USING MRI

Figure 3 Benign nodular hyperplasia: T2-weighted image, coronal plane of section. The prostate gland is enlarged. The heterogeneous signal intensity is secondary to benign nodular hyperplasia. A large hyperplastic nodule (*) protrudes into the urinary bladder (B)

Figure 2 Prostatic utricle: T2-weighted image, coronal plane of section. The utricle (U) is seen as a midline cyst of high signal intensity extending superior to the verumontanum. In the right peripheral zone, the low signal intensity region is a focus of prostate cancer (C). E, right ejaculatory duct; CZ, central gland

tissue by the surgical pseudocapsule (Figure 3). MR imaging of BPH includes a spectrum of ®ndings, based on the varying histologic composition and depending on the imaging sequence utilized.7 Although MRI provides excellent morphologic de®nition of BPH, it is not possible to differentiate benign from malignant disease of the prostate. The multiplanar imaging capability of MRI allows accurate volumetric measurement of the hyperplastic prostate gland and is especially useful in glands greater than 100 g, where ultrasound is limited in accuracy.8 Because of its accuracy in volumetric measurement and depiction of zonal anatomy, MR imaging has been used in the follow-up of patients on androgen deprivation therapy in order to monitor the changes in gland size and in differential volume reduction in zones of the gland.9 2.2.4

Prostate Carcinoma

Cancer of the prostate is the most common cancer in men and ranks as the second most common cause of cancer death

in American men.2,10 MR imaging is not applicable for the detection of prostate carcinoma. The strength of MRI is in the staging of biopsy-proven prostate cancer, although this application continues to be controversial and is a subject of numerous ongoing investigations. On T2-weighted MR images, prostatic carcinoma (PCa) is most commonly shown as a decreased signal intensity area within the high-signal-intensity normal peripheral zone. The detection of PCa on MRI is applicable only to the tumors located in the peripheral zone. The normal low or heterogeneous signal intensity of the transition and central zones precludes tumor detection.11 Even in the peripheral zone, tumor detection may be hampered by postbiopsy changes (Figure 4).11 Depending on the time interval between biopsy and MRI scan, the biopsy changes may cause either under- or overstaging of tumor presence and extent. Recent studies have demonstrated that MRI should not be performed for at least 2 weeks after biopsy.12 The ®nding of low signal intensity within the peripheral zone is not speci®c for prostate cancer.13 A number of other etiologies such as chronic prostatitis, dystrophic changes, scar, previous trauma, postbiopsy hemorrhage, radiation, and hormonal changes can all cause low signal intensity within the peripheral zone.13 Recently, in vivo MR spectroscopic imaging of the prostate gland has been developed to increase the speci®city of cancer detection in the peripheral zone. This three-dimensional proton spectroscopic technique allows the simultaneous acquisition of multiple spectra from the prostate gland with a tissue voxel resolution down to 0.24 ml.14,15 Since the metabolic changes of prostate cancer cause an increase in choline and a decrease in citrate levels, the ratio of the area under the choline spectroscopic peak to that under the citrate peak is increased in areas of cancer. The increased choline/citrate ratio in areas of prostate cancer enables the differentiation of cancer from the other causes of low T2 signal in the peripheral zone listed above (Figure 5).16 In addition, the ability to

MALE PELVIS STUDIES USING MRI

Figure 4 Postbiopsy hemorrhage: T1-weighted image, transaxial plane of section. In this patient, hemorrhage after a prostate biopsy is seen as increased signal intensity diffusely in the peripheral zone (PZ) and more focally in the right central gland (CZ). Normally on T1weighted images, the entire prostate shows homogeneous intermediate signal intensity

3

estimate the spatial extent of prostate cancer has important implications for assessing the ef®cacy of various cancer treatments.17 Prostate cancer, particularly the mucinous type, can demonstrate high signal intensity indistinguishable from the surrounding peripheral zone.18 While prostate cancer detection rates as high as 92% have been reported,11 the results of large multicenter studies are disappointingly low, with only 60% of lesions greater than 5 mm in any one dimension being detected on MRI scans.19 Attempts have also been made to measure tumor volume by MRI. As with ultrasound, the results are inaccurate, and smaller lesions are overestimated in 44% of patients and larger lesions underestimated in 55% of patients.19 Prostate cancer staging can follow either the TNM or Jewitt classi®cation.20 The TNM stage 1 or Jewitt stage A are tumors not applicable to MRI detection. Most of those cancers are within the transition zone, the area where MRI can neither depict nor stage the disease. TNM stage 2±Jewitt B disease in-

Figure 5 MR spectroscopic imaging of the prostate gland: transaxial T2-weighted image and MR spectra. The spectrum from the normal peripheral zone (a) shows the normal relationship between the choline and citrate peaks. The spectrum from the prostate cancer (b) shows increased choline and reduced citrate levels

4 MALE PELVIS STUDIES USING MRI

Figure 6 Prostate carcinoma with extracapsular extension on the left: T2-weighted image, transaxial plane of section. Carcinoma (*) demonstrates low signal intensity. Prostate capsule (white solid arrows). There is a breach of the capsule (open white arrows) with direct cancer invasion to the neurovascular bundle. Normal neurovascular bundle on the right side (black and white arrowhead)

dicates tumor con®ned to the prostate gland. The low-signalintensity tumor is seen within the peripheral zone, and while the lateral margin can bulge, the bulge is smooth in contour.21,22 With the endorectal coil, direct visualization of the prostate capsule increases the con®dence level in the evaluation of TNM stage 2. In TNM stage 3 and Jewitt C disease, the ®ndings of importance are extracapsular extension and seminal vesicle invasion. MRI ®ndings of extracapsular extension on endorectal coil include (a) bulge of the prostate gland with an irregular margin, (b) contour deformity with step-off or angulated margin, (c) breech of the capsule with direct tumor extension, (d) obliteration of the fat in the rectoprostatic angle, and (e) asymmetry, of the neurovascular bundles (Figure 6).12,21,22 Seminal vesicle invasion is diagnosed when there is (a) demonstration of contiguous low-signal-intensity tumor extension into and around the seminal vesicles, and/or (b) tumor extension along the ejaculatory duct resulting in nonvisualization of the ejaculatory duct, decreased signal intensity of seminal vesicles, and loss of seminal vesicle wall on T2-weighted images (Figure 7). While transaxial planes of section are essential in the evaluation of extracapsular extension, the invasion of the seminal vesicles is facilitated by the evaluation of transaxial and coronal planes of section.12,23±25 In recently reported studies using the Jewitt classi®cation and endorectal coil, the accuracy for extracapsular extension was 82% and accuracy for seminal vesicle invasion 97%.12 These reports are in accordance with previously published results by Schnall et al.,24 although they are higher than the latest publications by Chelsky et al.25 In the evaluation of lymph node metastases, reports in the literature testify to the superiority of MRI over CT. However, none of the studies has suf®ciently high numbers of positive nodes for

Figure 7 Prostate carcinoma with extracapsular and seminal vesicle extension: T2-weighted image, coronal plane of section. Prostate cancer (*) is seen at the base of the prostate gland on the right. Irregular outer margin indicates extracapsular invasion. There is also evidence of direct tumor invasion (white arrows) into the right seminal vesicle (SV)

statistically meaningful analysis. The accuracy of staging of cancer of the prostate depends on the type of coil used and appears to be most accurate when the combination of endorectal and surface multicoil system is used.11,19±26 Contrastenhanced images do not contribute to either tumor detection or staging except in rare instances when they help in the detection of seminal vesicle invasion.4

3 3.1

SEMINAL VESICLES Anatomy

The seminal vesicles are paired, androgen-dependent accessory sex glands. On T1-weighted images, normal seminal vesicles demonstrate homogeneous medium signal intensity similar to that of the adjacent pelvic muscle. On T2-weighted images, seminal vesicles demonstrate a grape-like con®guration with the high-signal-intensity ¯uid differentiated from the low signal intensity of the inner convolutions and outer wall (Figure 1).27 The size and the signal intensity of the seminal vesicles depends on patient's hormonal status and will decrease with patient's age, hormonal replacement therapy, radiation therapy, or severe alcoholism.27 On contrast-enhanced T1weighted images, the internal architecture of the seminal vesicles is depicted with the convolutions and wall demonstrating enhancement while the vesicular ¯uid remains of low signal intensity.4,6 Although great variation in the size of normal seminal vesicles can occur in adult men of the same age, there is a tendency for a decrease in size with advancing years. Seminal vesicles are usually symmetric in size, but asymmetry has been reported in up to 10% of patients.6

MALE PELVIS STUDIES USING MRI

3.2 Pathology 3.2.1

Congenital Anomalies

Congenital anomalies of the seminal vesicles include the absence of the seminal vesicles and more commonly seminal vesicle cysts. In diagnosing the absence of seminal vesicles, T1-weighted transaxial images are most useful.27 In the evaluation of seminal vesicle cysts, MRI provides precise anatomical localization but the signal intensity depends on its ¯uid composition.5 Blood is often present within the cyst with its signal intensity depending on the age of the hemorrhage. 3.2.2

In¯ammatory Disease

In¯ammation of the seminal vesicles is usually associated with epidydimitis and the associated ascending spread of infection through the prostate gland. The MRI appearance varies with the stage of in¯ammation. Seminal vesicles are often enlarged in the acute stage and small in the chronic phase. Hemorrhage within the seminal vesicles is common in subacute infection.27 Chronic in¯ammation results in small atrophic seminal vesicles often of lower than normal signal intensity on T2-weighted images.6 3.2.3

Tumors

The benign tumors of seminal vesicles (leiomyomas predominate) are more common than primary malignant neoplasms which are usually adenocarcinomas. The vast majority of malignant tumors of the seminal vesicles are secondary usually from prostate carcinoma, but extension from cancer of the bladder or rectum can be seen as well. The differentiation between benign and malignant tumors is usually based on the morphologic ®ndings.28 Benign tumors appear as smooth, wellmarginated masses while malignant invasion of the seminal

5

vesicles usually results in masses of low signal intensity and are irregular in con®guration.28 Loss of normal adjacent tissue planes also suggests secondary involvement of the seminal vesicles (Figure 7). Tumor extension into the seminal vesicles is best seen on T2-weighted images, in the sagittal and coronal planes.6,28

4 4.1

PENIS/URETHRA Anatomy

The penis is composed of three erectile bodies: the two lateral corpora cavernosa and the ventromedial corpus spongiosum (Figure 8).29 Each of the three erectile bodies is enveloped by a ®brous sheathÐthe tunica albuginea. Buck's fasciaÐa common ®brous sheathÐdivides the penis into its dorsal compartment by enclosing the two corpora cavernosa and the ventral compartment by enclosing the corpus spongiosum. The male urethra extends from the bladder neck to the fossa navicularis in the glans penis. It is divided into the prostatic, membranous, and penile portions of the urethra. On T1weighted images, the three erectile bodies demonstrate medium signal intensity, and they all increase in signal intensity on T2weighted images (Figure 8). While the corpus spongiosum demonstrates homogeneous high signal intensity, the normal corpora cavernosa can be either homogeneous or heterogeneous in signal intensity depending on the blood volume within the erectile tissue.29 The tunica albuginea and Buck's fascia appear as a low-signal-intensity stripe surrounding the corporeal bodies on T2-weighted images and provide excellent contrast with the high-signal-intensity corpora.29 The use of contrast enhancement allows demonstration of the penile anatomy on T1-weighted images.6

Figure 8 Normal anatomy of the penis. (a) Transaxial and (b) coronal plane of section: T2-weighted images. Corpora cavernosa (cc), corpus spongiosum (cs), tunica albuginea (short arrows), and Buck's fascia (long black arrow)

6 MALE PELVIS STUDIES USING MRI 4.2 Pathology 4.2.1

Congenital Anomalies

Epispadia, hypospadia, or duplication of the penis (penis diphallus) are the three groups of most commonly encountered congenital anomalies of the penis. The ability of MRI to delineate each of the corpora allows precise de®nition of the type of anomaly and provides depiction of associated anomalies involving the perineum and the remaining genitourinary tract.29 The clinical indications, however, are rare, the most common being evaluation of penis diphallus. 4.2.2

In¯ammatory Disease

Peyronie's disease is an in¯ammatory condition of unknown etiology characterized by the development of ®brous plaque involving the tunica albuginea and extending into the corpora cavernosa. On T2-weighted MR images, the lower-signal-intensity plaque can be depicted in contrast to the higher-signalintensity corpora cavernosa.30,31 MRI is not the primary method for diagnosing this condition. MRI, however, can provide information about the location and size of the plaque as well as the degree of cavernosal involvement. Furthermore, MRI can differentiate between the acute and chronic form of Peyronie's disease, especially when contrast media are used. 4.2.3

Trauma

Penile trauma usually results from direct blunt injury to the erect penis causing fracture of the penis or rupture of the corpora cavernosa. On T2-weighted MR images, the diagnosis of penile fracture is based on the ®nding of interruption of the normal low-signal-intensity tunica albuginea. Peripenile hematoma is often detected.29,32 MRI is valuable in the presurgical evaluation of urethral trauma (grade 3) associated with complex pelvic injury. Separation between the prostatic apex, membranous, and bulbous urethra can occur in superior anteroposterior or lateral direction. Misalignment of greater than 2 cm in the superior direction and greater than 1 cm in the lateral direction necessitates a different surgical approach (perineal versus suprapubic with removal of the symphysis pubis).33,34 MRI can greatly assist in planning the surgical approach but thin-section T2weighted images in all three orthogonal planes are essential for the assessment of this complicated problem.34 4.2.4

Tumors

Penile carcinoma accounts for only about 1% of all male cancers, and carcinoma of the male urethra is even more rare.35 When cancers are located in the glans penis or in the penile shaft, the locoregional tumor extent can usually be determined by physical examination. However, when the tumor involves the root of the penis or bulbomembranous urethra, the clinical evaluation is limited and MRI has a role in local staging of disease.29 On MRI, penile and urethral carcinomas are usually not distinguishable as most lesions requiring radiologic assessment are advanced and have spread to involve adjacent structures. The use of T2-weighted images demonstrates the tumor of different signal intensity (usually lower) as compared with the adjacent normal high-signal-intensity erectile bodies (Figure 9). MRI, however, is not speci®c for cancer detection and cannot be distinguished from in¯ammatory disease.6,29 This is es-

pecially true for smaller lesions of the urethra. However, MRI can be of great help in locoregional tumor staging, and in therapy planningÐsurgery versus radiation therapy.29 Metastasis to the penis can also be well evaluated by MRI. Metastasis demonstrates diffuse low signal intensity in®ltrating the erectile bodies on T2-weighted images. Differentiation between primary and metastatic lesions is not possible.29

5

TESTIS

5.1

Anatomy

On T1-weighted images, the testis demonstrates intermediate signal intensity, similar to that of the corpora cavernosa or corpus spongiosum and lower than that of subcutaneous fat. On T2-weighted images, the testis demonstrates high signal intensity in contrast to the lower-signal-intensity tunica albuginea (a dense ®brous connective tissue capsule) covering the testis (Figure 10).36 The mediastinum testis has signal characteristics similar to tunica albuginea and is seen as a low-signal-intensity stripe invaginating into the high-signal-intensity testicular parenchyma (Figure 10).36 The parietal and visceral layers of the tunica vaginalis are often separated by a small amount of normal serous ¯uid. As in a hydrocele, this normal ¯uid surrounds the testis completely except for the posteromedial bare area. The signal intensity of the epididymis is similar to the testis on T1-weighted images and much lower than the testis on T2weighted images.36 5.2 5.2.1

Pathology Undescended Testis

An undescended testis is de®ned as a testis located outside the scrotum. MR diagnosis of undescended testes relies on the ®nding of an elliptical mass demonstrated along the expected path of testicular descent.37 However, it has recently been reported that dynamic gadolinium-enhanced MR angiography improves the detection of atrophic undescended testes by showing enhancement of the pampinform plexus.38 When the undescended testis is in the inguinal canal, it is usually oval in shape while the intraabdominal testis assumes a more rounded con®guration. The large ®eld of view and multiplanar image capability of MRI allows precise identi®cation of the undescended testis as being either high scrotal, intracanalicular, or intraabdominal in location. When the intraabdominal testis is located close to the internal inguinal ring, it is easily depicted by MRI. However, high abdominal testis is dif®cult to demonstrate with MRI and CT is the procedure of choice. MR imaging allows differentiation between the undescended testis, gubernaculum, and lymph nodes.37 Lymph nodes lie outside the expected descent of the testis. The differentiation between the gubernaculum and the testis relies on their respective differences in signal intensity. The gubernaculum characteristically demonstrates very low signal intensity on both T1- and T2weighted images as its predominant histologic composition is ®brous tissue. The signal intensity of the testis varies depending on the degree of atrophy but is never as low as a ®brous cord.

MALE PELVIS STUDIES USING MRI

7

Figure 9 Cancer of the penis. (a) Transaxial and (b) coronal plane of section: T2-weighted images. Tumor (T) is seen invading both corpora cavernosa (cc), spectum corpora cavernosa as well as corpus spongiosum (cs). In the coronal plane of section, the zonal anatomy of the penis is not visualized, and there is demonstration of tumor extension through the tunica albuginea (small black arrows)

5.2.2

Testicular Tumors

Testicular tumor is the most common malignancy among men of 15 to 34 years of age although it accounts for only 1% of all cancers in males.35 The majority (90±95%) of testicular tumors are malignant germ cell tumors. The homogeneous high signal intensity of the normal testicular tissue serves as an excellent background for the depiction of intratesticular pathology.36,39,40 On T2-weighted images, testicular tumors usually

demonstrate a lower signal intensity than the adjacent normal testicular tissue (Figure 11). Although MR imaging has a high sensitivity rate for the detection of testicular tumors, its ®ndings are variable and are not speci®c for tumor histology. While on T2-weighted images testicular tumors are mostly hypointense to the testis, they may demonstrate a signal intensity similar or even higher than testicular tissue. Tumors can be homogeneous or heterogeneous in signal pattern. While seminomas are typically homogeneous and hypointense, the

8 MALE PELVIS STUDIES USING MRI as sharper marginated lesions with ¯uid signal intensity characteristics.36 5.2.3

Figure 10 Normal testis, transaxial plane of section: T2-weighted image. Testis (T) demonstrates homogeneous high signal intensity. Tunica albuginea (solid arrow) is seen as a low-signal-intensity stripe. Mediastinum testis (open arrows)

nonseminomatous tumors are often heterogeneous and exhibit various signal intensities.40 However, exceptions to the rule can be encountered.36 Furthermore, differentiation between primary and secondary testicular tumors is not possible, and the MR appearance of benign tumors is similar to testicular cancer.41 Testicular cysts can usually be differentiated from testicular tumors especially when contrast media are used. They appear

Acute epididymitis is the most common in¯ammatory lesion in the scrotum. In acute epididymitis, the epididymis is enlarged and demonstrates heterogeneous and often higher signal intensity than normal. Hemorrhage may complicate acute epididymitis producing signal intensity commensurate with its age. Reactive hydrocele is often present. In chronic epididymitis, the signal intensity of the epididymis is reduced, a ®nding best appreciated on T2-weighted images.39 Associated orchitis, when present, appears as a homogeneous or heterogeneous hypointense lesion within the normal sized or enlarged testis. The signal intensity change is most commonly present in the mediastinum region. Fibrotic thickening of the tunica albuginea can occur following epididymitis, and the thickening may sometimes be dif®cult to differentiate from small testicular cancer, thus the term `pseudotumor of the tunica albuginea' has been introduced.42 5.2.4

Spermatic Cord Torsion

Torsion is considered a surgical emergency since delaying intervention may result in irreversible damage to the testis. MRI plays no role and is not a primary imaging modality in diagnosing testicular torsion. Reports on MRI ®ndings in acute torsion in humans are sporadic and not well documented. The testis may be normal or enlarged and of heterogeneous signal intensity. With testicular torsion, the spermatic cord can enlarge, exhibiting a high signal intensity due to edema. A twisted cord can be seen as multiple low-intensity curvilinear structures rotating in a `whirlpool' pattern which is best seen in a plane perpendicular to its axis.43 Torsion of the testicular or epididymal appendices can be identi®ed with MRI by their typical location and the signal characteristics of hemorrhage.43 5.2.5

Figure 11 Seminoma of the left testis. Testicular tumor (*) demonstrates low signal intensity as compared with the adjacent remaining higher signal intensity testicular tissue. Normal right testis (T)

In¯ammatory Disease

Fluid Collection and Benign Scrotal Masses

Fluid collections whether hydrocele, hematocele, or pyocele all indicate abnormal ¯uid located between the parietal and visceral layers of the tunica vaginalis. The signal of hydrocele is typical for ¯uid (low signal intensity on T1-weighted image and high intensity on T2-weighted image).36,39 The signal intensity of hematocele varies with the age of the hemorrhagic ¯uid, and the proteinaceous component of pyocele usually makes the ¯uid signal intensity on T1-weighted images higher than that of hydrocele. Spermatocele is a retention cyst of small tubules which connect the rete testis to the head of epididymis. On MRI scan, the diagnosis of spermatocele can be made by its typical location (usually in the head of the epididymis) and its cystic nature.6 In scrotal pathology, MRI serves as a problem-solving modality when ultrasound ®ndings are equivocal, technically inadequate, or there is a discrepancy between the physical examination and ultrasound ®ndings. Recent studies have shown that when ultrasound and physical examination of the scrotum are inconclusive, MRI can improve patient management and reduce treatment costs.37±46

MALE PELVIS STUDIES USING MRI

6 RELATED ARTICLES Coils for Insertion into the Human Body; MRI of the Female Pelvis. 7 REFERENCES 1. J. E. McNeal, Monogr. Urol., 1983, 4, 5. 2. T. A. Stamey, J. E. McNeal, F. S. Freiha, and E. A. Redwine, J. Urol., 1988, 139, 1235. 3. H. Hricak, G. C. Dooms, J. E. McNeal, A. S. Monk, M. Marotti, A. Avalloro, M. Pelzer, E. C. Proctor, and E. A. Tanagho, Am. J. Roentgenol., 1987, 148, 51. 4. S. A. Mirowitz, J. J. Brown, and J. P. Heiken, Radiology, 1993, 186, 153. 5. S. Thurnher, H. Hricak, P. R. Carroll, R. S. Pobiel, and R. A. Frilly, Radiology, 1988, 167, 631. 6. H. Hricak, in `MRI of the Pelvis. A Text Atlas', eds. H. Hricak and B. M. Carrington, Martin Dunitz, London, 1991, p. 313. 7. W. G. Way, Jr., J. J. Brown, J. K. Lee, E. Gutierrez, and G. L. Andriole, Magn. Reson. Imag., 1992, 10, 341. 8. A. Rahmouni, A. Yang, C. M. Tempany, T. Frenkel, J. Epstein, P. Walsh, P. K. Leichner, C. Ricci, and E. Zerhouni, J. Comput. Assist. Tomogr., 1992, 16, 935. 9. C. M. Tempany, A. W. Partin, E. A. Zerhouni, S. J. Zinreich, and P. C. Walsh, Prostate, 1993, 22, 39. 10. S. H. Landis, T. Murray, S. Bolden, and P. A. Wingo, CA Cancer J. Clin., 1999, 49, 8. 11. M. D. Schnall, R. E. Lenkinski, H. M. Pollack, Y. Imai, and H. Y. Kressel, Radiology, 1989, 172, 570. 12. S. F. Quinn, D. A. Franzini, T. A. Demlow, D. R. Rosencrantz, J. Kim, R. M. Hanna, and J. Szumowski, Radiology, 1994, 190, 323. 13. K. Lovett, M. D. Rifkin, P. A. McCue, and H. Choi, JMRI, 1992, 2, 35. 14. J. Kurhanewicz, D. B. Vigneron, S. J. Nelson, H. Hricak, J. M. MacDonald, B. Konety, and P. Narayan, Urology, 1995, 45, 459. 15. J. Kurhanewicz, D. B. Vigneron, H. Hricak, P.Narayan, P. Carroll, and S. J. Nelson, Radiology, 1996, 198, 795. 16. Y. Kaji, J. Kurhanewicz, H. Hricak, D. L. Sokolov, L. R. Huang, S. J. Nelson, and D. B. Vigneron, Radiology, 1998, 206, 785. 17. J. Kurhanewicz, D. B. Vigneron, H. Hricak, F. Parivar, S. J. Nelson, K. Shinohara, and P. Carroll, Radiology, 1996, 200, 489. 18. E. Outwater, M. L. Schiebler, J. E. Tomaszewski, M. D. Schnall, and H. Y. Kressel, JMRI, 1992, 2, 597. 19. M. D. Rifkin, E. A. Zerhouni, C. A. Gatsonis, L. E. Quint, D. M. Paushter, J. I. Epstein, W. Hamper, P. C. Walsh, and B. J. McNeil, N. Engl. J. Med., 1990, 323, 621. 20. M. Graf, P. Hermanek, R. V. P. Hutter, L. H. Sobin, G. Wagner, and C. Wittekind (eds.), `TNM Atlas: Illustrated Guide to the TNM/pTNM-Classi®cation of Malignant Tumors, 4th edn', Springer-Verlag, Berlin, 1997. 21. E. K. Outwater, R. O. Petersen, E. S. Siegelman, L. G. Gomella, C. E. Chernesky, and D. G. Mitchell, Radiology, 1994, 193, 333. 22. K. K. Yu, H. Hricak, R. Alagappan, D. M. Chernoff, P. Bacchetti, and C. J. Zaloudek, Radiology, 1997, 202, 697. 23. H. Hricak, G. C. Dooms, R. B. Jeffrey, A. Arallone, D. Jacobs, W. K. Benton, P. Narayan, and E. A. Tanagho, Radiology, 1987, 162, 331. 24. M. D. Schnall, Y. Imai, J. Tomaszewski, H. M. Pollack, R. Lenkinski, and H. Y. Kressel, Radiology, 1991, 178, 797. 25. M. D. Schnall, R. E. Lenkinski, H. M. Pollack, Y. Imai, and H. Y. Kressel, Radiology, 1989, 172, 570.

9

26. M. J. Chelsky, M. D. Schnall, E. J. Seidmon, and H. M. Pollack, J. Urol., 1993, 150, 391. 27. E. Secaf, R. N. Nuruddin, H. Hricak, R. D. McClure, and B. Demas, Am. J. Roentgenol., 1991, 156, 989. 28. R. D. McClure and H. Hricak, Urology, 1986, 27, 91. 29. H. Hricak, M. Marotti, T. J. Gilbert, T. F. Lue, L. H. Wetzel, J. W. McAninch, and E. A. Tanagho, Radiology, 1988, 169, 683. 30. G. Helweg, W. Judmaier, W. Buchberger, K. Wicke, H. Oberhauser, R. Knapp, O. Ennemoser, and D. Zur Nedden, Am. J. Roentgenol., 1992, 158, 1261. 31. R. Vosshenrich, I. Schroeder-Printzen, W. Weidner, U, Fischer, M. Funke, and R. H. Ringert, J. Urol., 1995, 153, 1122. 32. M. Fedel, S. Venz, R. Andreessen, F. Sudhoff, and S. A. Loening, J. Urol., 1996, 155, 1924. 33. C. M. Dixon, H. Hricak, and J. W. McAninch, J. Urol., 1992, 148, 1162. 34. Y. Narumi, H. Hricak, N. A. Armenakas, C. M. Dixon, and J. W. McAninch, Radiology, 1993, 188, 439. 35. D. M. Parkin, P. Pisani, and J. Ferley, CA Cancer J. Clin., 1999, 49, 33. 36. S. Thurnher, H. Hricak, P. R. Carroll, R. Pobiel, and R. A. Frilly, Radiology, 1988, 167, 631. 37. P. J. Fritzsche, H. Hricak, B. A. Kogan, M. L. Winkler, and E. A. Tanagho, Radiology, 1987, 164, 169. 38. W. W. Lam, P. K. Tam, V. H. Ai, K. L. Chan, W. Cheng, F. L. Chan, and L. Leong, J. Pediatr. Surg., 1998, 33, 123. 39. L. L. Baker, P. C. Hajek, T. K. Burkhard, L. Dicepua, H. M. Landa, G. R. Leopold, J. R. Hesselink, and R. F. Mattrey, Radiology, 1987, 163, 93. 40. J. O. Johnson, R. F. Mattrey, and J. Phillipson, Am. J. Roentgenol., 1990, 154, 539. 41. F. V. Coakley, H. Hricak, and J. C. Presti Jr, Urol. Clin. North Am. 1998, 25, 375. 42. R. B. Poster, B. A. Spirt, A. Tamsen, and B. V. Surya, Radiology, 1989, 173, 561. 43. M. A. Trambert, R. F. Mattrey, D. Levine, and D. P. Merthoty, Radiology, 1990, 175, 53. 44. H. Derouet, H. U. Braedel, G. Brill, K. Hinkeldey, J. Steffens, and M. Ziegler, Urol. Ausg., 1993, 32, 327. 45. B. M. Cramer, E. A. Schlegel, and J. W. Thueroff, Radiographics, 1991, 11, 9. 46. A. D. Serra, H. Hricak, F. V. Coakley, B. Kim, A. Dudley, A. Morey, B. Tschumper, and P. R. Carroll, Urology, 1998, 51, 1018.

Biographical Sketches Hedvig Hricak. b 1946. M.D., 1970, University of Zagreb, Croatia; Ph.D., 1992, Karolinska Institute, Stockholm, Sweden. University of California, San Francisco, Faculty in Radiology, 1982±99; Faculty in Urology, 1986±present; Faculty in Radiation Oncology, 1991±99. Department of Radiology, Sloan-Kettering Cancer Center, New York, 2000±present. Involved in clinical MR imaging of gynecologic and urologic diseases since the introduction of this technique. Over 150 papers on MR imaging in peer-reviewed journals and co-author of four major textbooks on the subject. Member of numerous editorial boards of prestigious journals in radiology and radiation oncology and serves on national and international committees of prestigious professional organizations. William T. Okuno. b 1967. M.D., 1993, University of Illinois, USA. Radiology Residency, Massachusetts General Hospital, 1994±1998; Abdominal Imaging Fellowship, University of California, San Francisco, 1998±1999; Lutheran General Hospital, Park Ridge, Illinois, 1999±present. Special interests: abdominal imaging.

MRI OF THE FEMALE PELVIS

MRI of the Female Pelvis Robert C. Smith, Michael J. Varanelli, Leslie M. Scoutt, and Shirley McCarthy Yale University School of Medicine, New Haven, CT, USA

1 INTRODUCTION The article will describe the clinical applications and technical considerations of MRI of the female pelvis. Due to its improved tissue contrast capability (compared with computerized tomography and ultrasound), MRI is the technique of choice for depicting normal uterine and cervical anatomy as well as a variety of benign and malignant conditions affecting these structures. MRI is also capable of delineating a number of nongynecological abnormalities of the pelvis. Recent technical advances have dramatically improved the resolution that can be obtained when imaging the female pelvis. At the same time, these technical improvements have greatly reduced the imaging time. Both of these factors should improve the accuracy of MRI not only for detecting abnormalities but in the staging of gynecological malignancies. Cost± bene®t analyses have shown MRI of the female pelvis to be cost-effective. 2 MRI TECHNIQUE Imaging of the female pelvis requires both T1- and T2weighted images.1 T1-weighted images are useful for lymph node detection and to characterize regions which contain blood products and/or fat. T2-weighted images are necessary to depict the internal anatomy of the gynecological organs, characterize masses and other abnormalities, and determine the origin of a mass as uterine or ovarian. T1-weighted images are typically acquired in the axial plane from the level of the aortic bifurcation through the pubic symphysis. The TR (repetition time) used is between 500 and 600 ms, and the minimum possible TE (echo time) is used (usually 10±15 ms). Superior and inferior saturation pulses are routinely used to diminish the intravascular signal and therefore help decrease pulsation artifacts and distinguish lymph nodes from vascular structures. In some cases, use of a longer TE (up to 20 ms) will help diminish intravascular signal. Respiratory compensation is always utilized to diminish ghost artifacts from the high-signal subcutaneous fat.2 Other typical parameters would include a ®eld of view (FOV) of 24±30 cm, section thickness of 5 mm, intersection spacing of 2.0±2.5 mm, a frequency matrix size of 256, a phase matrix size of 128 or 192, and one or two signal averages. If there is a large amount of bowel within the pelvis, glucagon can be administered to help diminish artifacts from bowel peristalsis. In the majority of cases, glucagon is not necessary. T2-weighted images should be acquired through the uterus and ovaries in at least two orthogonal planes. The main purpose of sagittal images is to depict the uterine and cervical

1

zonal anatomy. Depending upon the orientation of the uterus, the axial or coronal plane can be chosen to image the uterus in a plane perpendicular to its long axis. Both the axial and coronal planes will usually depict the ovaries to good advantage. The T2-weighted images should be acquired using fast spin echo (FSE) or turbo spin echo (TSE) pulse sequences.3,4 Using these, high-resolution T2-weighted images can be acquired through the entire pelvis in less than 5 min.

3

FSE IMAGING

When FSE T2-weighted images are acquired in a body coil, typical imaging parameters would include a TR of 4500± 6000 ms, a TE of 100±130 ms, a 5 mm section thickness, a 2.0±2.5 mm intersection spacing, a minimum phase matrix size of 192, a FOV of 24±28 cm, and two to four signal averages. If a small FOV is used with a phase matrix size of 256, this usually necessitates use of four signal averages to maintain the signal-to-noise ratio (S/N) at an acceptable level. In addition to the above parameters, the echo train length (ETL) and the echo spacing (ESP) must also be speci®ed when using FSE sequences. With conventional spin echo sequences, a single phase-encoding step is acquired during each TR interval. Even if multiple echoes are measured (as with proton density and T2-weighted images), each is acquired with the same phase-encoding step. The multiple echoes are used to generate multiple images (of different TE) at each imaging location. In order to reconstruct an image, the number of phaseencoding steps must equal the size of the phase matrix. It is this fact which makes the imaging time of conventional spin echo sequences so long. FSE sequences acquire multiple phase-encoding steps during each TR interval. This is achieved by applying multiple 180 pulses after each 90 pulse and separately phase encoding the resulting spin echoes. The ETL is equal to the number of separately phase-encoded echoes acquired following each 90 pulse. The maximum value of the ETL is usually 16 or 32. Recent development of `single-shot' techniques using a halfFourier reconstruction allows an ETL of 128 or 192 and acquisition of each image in about 1 s. Once all of the separately phase-encoded echoes have been acquired in one section, data are then acquired in the next section. The spacing between the separate phase-encoded echoes is the ESP. The minimum value of ESP is usually 6±8 ms and depends on the bandwidth. In almost all circumstances the minimum value of ESP should be used. Regardless of the imaging technique, each phase-encoding step uses a different strength phase-encoding gradient. Echoes acquired with the weak phase-encoding gradients provide most of the signal (and hence contrast) of an image, and echoes acquired with the strong phase-encoding gradients provide most of the spatial resolution of an image. With the fast spin echo technique, multiple separately phase-encoded echoes are used to generate a single image at each location. The overall contrast of such images is determined by the echo times at which the weak phase-encoding steps are acquired.5±7 If the weak phase-encoding steps are used for the early echoes, the image will appear as either proton density or T1 weighted (depending on the TR). If the weak phase-encoding steps are used for the late echoes, the image will appear as T2 weighted.

2 MRI OF THE FEMALE PELVIS For this reason, the echo time of an FSE image is usually referred to as an effective TE. When the user selects an effective TE, the phase-encoding steps are assigned to the appropriate echoes of each multiecho train following each 90 pulse. Severe artifacts can result from this technique. Late echoes have diminished signal due to T2 decay. If the strong phase-encoding gradients are applied to the late echoes (as with proton density- and T1-weighted FSE images), there may be no useful signal left. Since the strong phase-encoding gradients provide the spatial resolution of an image, without these phase-encoding steps there will be a loss of spatial resolution. This is manifest as image blurring, which can be severe. Image blurring is minimized by using a shorter ETL (usually 4 or 6) when acquiring proton density- or T1-weighted FSE images. T2-weighted FSE images never suffer from this blurring effect since the strong phase-encoding gradients are always applied to the early echoes. Early echoes suffer minimal signal loss due to T2 decay. Therefore, T2-weighted FSE images can usually be acquired with the maximum possible value of ETL. The imaging time of a conventional spin echo sequence is given by TR  (phase matrix size)  (number of signal averages). The imaging time of an FSE sequence is this same product divided by ETL. However, when using FSE sequences, the number of sections that can be acquired for a given TR will be diminished. This results from the fact that many more echoes must be measured in each section before moving on to the next section. Depending on the choice of ETL, this will usually necessitate the use of a longer TR with FSE sequences. The shorter the ETL, the more sections that can be acquired per TR interval. For example, using a TR of 3000 and an ETL of 8 will give approximately the same number of sections as using a TR of 6000 and an ETL of 16. The imaging times will be the same. The advantage of using a longer TR is that it gives a true T2-weighted image. The trade-off is that use of a shorter ETL will improve the S/N since fewer late echoes are acquired. One must always keep in mind that the maximum value of TE that can be achieved is equal to ETL  ESP. Therefore, use of a shorter ETL will limit the maximal value of TE. When imaging the pelvis in a body coil with FSE sequences, one can use either an ETL of 8 (with a TR of 3000±4000 ms) or an ETL of 16 (with a TR of 6000±8000 ms). It is recommended that a 192 or 256 phase matrix is used. There is usually a signi®cant relative image degradation when a 128 phase matrix size is used with FSE images.

4 MULTICOIL IMAGING When imaging with a single surface coil there is a marked improvement in the S/N, but the FOV that can be achieved is severely limited (roughly of the order of the diameter of the coil). In addition, the relative phase of the signal will depend upon the orientation of the surface coil within the magnet. If multiple surface coils are used together and their signals are combined into a circuit, there can be signal loss due to phase differences between the two signals. This can be minimized by careful orientation of the coils. A multicoil (also called a phased-array coil) consists of multiple surface coils which act independently in a receive-only

mode. Each separate coil of the multicoil inputs its signal into separate receiver channels. The signals from the individual coils are then used to reconstruct an individual image for each coil. These separate images are then recombined into a single composite image. In this case the relative phases of the individual signals is irrelevant, as the signals are only combined after magnitude reconstruction. By using a multicoil one gets the improved S/N of a surface coil, and the FOV that can be achieved is comparable to a body coil.8±10 The only disadvantage of a multicoil is that four separate receiver channels must be used, which is expensive. All pelvic imaging should now be performed using a multicoil and FSE pulse sequences. When using a multicoil, the signal from subcutaneous fat immediately adjacent to the coils will be markedly increased and can result in signi®cant ghost artifacts. This can be eliminated by the placement of saturation pulses (within the ®eld of view) through the subcutaneous fat both anteriorly and posteriorly.

5

NORMAL ANATOMY

On T1-weighted images the uterus has a homogeneous low signal intensity. The appearance is similar to that of the uterus on computed tomography images, as no internal architecture is visible. On T2-weighted images, a zonal architecture of the uterus is readily identi®ed. The appearance of the zonal anatomy depends upon whether the patient is pre- or postmenopausal and the phase of the menstrual cycle.11±13 In virtually all premenopausal patients, three distinct zones of signal intensity can be seen in the uterine corpus (Figure 1).12 An inner zone of high signal corresponds to the endometrium. An adjacent zone of low signal intensity corresponds to the so-called junctional zone. Histological studies have shown this zone to represent an inner layer of myometrium which has an increased nuclear area.14 The remainder of the myometrium has intermediate signal intensity but can be quite variable, depending on the hormonal milieu of the patient. Previous studies have shown that the inner bright zone progressively increases in thickness during the menstrual cycle with a peak thickness occurring late in the secretory phase. The thickness may vary from 4 mm early in the proliferative phase to 13 mm late in the secretory phase. The junctional zone shows no signi®cant change in thickness during the menstrual cycle. In patients taking oral contraceptives (combined estrogen and progestin), endometrial thickness is markedly reduced after several months. In addition, the outer myometrium of these patients shows relatively increased signal intensity compared with those not using oral contraceptives. Ultrasound studies have been performed in an attempt to determine an approximate upper limit of normal for endometrial stripe thickness in postmenopausal patients.15,16 Regardless of hormonal replacement therapy, an upper limit of 8 mm has been suggested. No similar large series have as yet been performed with MRI. However, a comparative study of ultrasound and MRI has shown that the MRI measurement of endometrial thickness is almost always smaller than the corresponding ultrasound measurement.17 The cervix also has a unique zonal anatomy on T2-weighted MRI (Figure 1). Four distinct zones of signal intensity have been described:4,18 a thin inner bright zone which corresponds

MRI OF THE FEMALE PELVIS

6

Figure 1 Sagittal FSE multicoil image demonstrating the zonal anatomy of the uterus. The endometrium is the central bright stripe (curved arrow). Note the surrounding low-signal junctional zone and intermediate-signal outer myometrium. Within the cervix, the central high signal represents the canal (straight arrow) and contained mucus. The next layer is the intermediate-intensity signal of the cervical mucosa which is surrounded by the low-intensity signal of the ®brous stroma. Finally, the outer cervical stroma has an intermediate signal intensity and is continuous with the myometrium

to the endocervical canal; an adjacent thin zone of intermediate signal intensity corresponding to the cervical mucosa; a thicker zone of very low signal intensity thought to correspond to the predominantly ®brous portion of the wall of the cervix; and an outer zone which is continuous with the myometrium and is isointense with the myometrial signal. Previous studies have shown little variation in the appearance of the cervix during the course of the menstrual cycle.13 In addition, there is no apparent difference in the appearance of cervical zonal anatomy between premenopausal and postmenopausal women, or between those using and not using oral contraceptives. The normal MRI appearance of the vagina is less complex. The vaginal wall typically shows low signal intensity, and the vaginal canal typically appears as a bright stripe of high signal intensity corresponding to vaginal secretions. The vagina is surrounded by the high signal intensity perivaginal venous plexus. The ovaries appear homogeneously hypointense on T1weighted images. Follicles appear as small, round, homogeneous areas of high signal intensity on T2-weighted images. The ovarian stroma is of low to intermediate signal intensity on T2-weighted images. With high-resolution imaging, it is almost always possible to identify the ovaries in premenopausal patients, and most of the time in postmenopausal patients.

3

BENIGN DISEASES OF THE UTERUS AND CERVIX

The most common benign mass of the uterus is the leiomyoma, also commonly referred to as a ®broid. These masses are sharply circumscribed, usually spherical in shape, and typically have low signal intensity on all pulse sequences. When ®broids become large, they can undergo degeneration, which is usually manifest as central increased signal intensity on T2weighted images. MRI is uniquely able to determine the intrauterine location of these masses. Fibroids are usually described as being submucosal, intramural, or subserosal in location. Submucosal ®broids are commonly associated with abnormal menstrual bleeding (Figure 2). Intramural ®broids are not usually associated with abnormal menstrual bleeding but are a common cause of uterine enlargement (Figure 3). Subserosal ®broids can be on a stalk and therefore can undergo torsion. Subserosal ®broids can also be confused with an adnexal mass on physical examination and on other imaging modalities. When ®broids cause signi®cant clinical symptoms and surgery is contemplated, the surgical approach is dependent upon the precise location of the ®broids. Some submucosal ®broids can be removed hysteroscopically. Intramural and subserosal ®broids can only be removed using a transabdominal approach. In addition to their location, another important surgical consideration is the size and increased vascularity of these lesions. Hormonal therapy is sometimes used to decrease their size and vascularity prior to possible surgery. MRI can be used to follow precisely the size and vascularity during such therapy. Adenomyosis is de®ned as the presence of endometrial tissue within the myometrium. This tissue is usually not functional. The appearance of adenomyosis on MRI scans is rather characteristic.19 It appears as focal or diffuse ill-de®ned thickening of the junctional zone (Figure 4). It has low signal

Figure 2 Sagittal FSE multicoil image demonstrating a submucosal ®broid (arrow)

4 MRI OF THE FEMALE PELVIS

Figure 3 Coronal FSE multicoil image demonstrating a large intramural ®broid (arrow) displacing the endometrial stripe

intensity on T2-weighted images, being isointense with the junctional zone. In some cases, adenomyosis can appear as small foci of high signal intensity within the myometrium on T2-weighted images. Adenomyosis can result in uterine enlargement and abnormal menstrual bleeding. It is therefore important to differentiate this condition from ®broids. MRI is particularly useful in making this distinction.

Figure 4 Sagittal FSE multicoil image demonstrating adenomyosis (arrow). Note the diffuse irregular thickening of the junctional zone with tiny areas of hyperintensity characteristic of the disease

Figure 5 Axial FSE multicoil image demonstrating diffuse thickening of the endometrium (arrow) consistent with endometrial hyperplasia

Endometrial hyperplasia is thought to represent a physiological response of the endometrium to unopposed estrogenic stimulation. Some forms of endometrial hyperplasia are known to be precursors of endometrial carcinoma. On MRI, endometrial hyperplasia appears as thickening of the high-signal endometrium on T2-weighted images (Figure 5).1 However, MRI cannot distinguish between hyperplasia and early carcinoma. Endometrial polyps can be sessile or pedunculated masses projecting into the endometrial cavity. They can be associated with abnormal uterine bleeding. However, endometrial carcinoma can sometimes have a polypoid con®guration and cannot be reliably distinguished from simple polyps. On T2-weighted images, polyps may appear as intermediate-signal masses (with respect to the low-signal junctional zone and the high-signal endometrium) but can also be isointense with the endometrium. Gadolinium-enhanced T1-weighted images may be useful in detecting endometrial polyps. The enhancing polyp can be outlined by nonenhancing ¯uid within the endometrial cavity. Pedunculated submucosal ®broids can also appear as polypoid ®lling defects within the endometrial cavity on MRI. Their signal intensity is usually signi®cantly lower than that of endometrial polyps, which helps in their distinction. The two most common uterine anomalies that need to be differentiated are the septate uterus and the bicornuate uterus.20 In the septate uterus, the external contour of the uterus is normal (Figure 6). In the bicornuate uterus (Figure 7), there are two separate horns of the endometrial cavity, and the external contour of the uterus is signi®cantly indented at the fundus (as there is no uterine tissue in the space between the two horns). By imaging through the fundus along the long axis of the uterus, these two entities can be reliably distinguished. The signal characteristics of the septum may re¯ect myometrial or ®brous tissue, and hence signal behavior is not useful in classi-

MRI OF THE FEMALE PELVIS

7

Figure 6 Axial FSE multicoil image of a septate uterus. The contour of the uterine fundus is normal, and a large myometrial septum (arrow) is present

5

MALIGNANT DISEASE OF THE UTERUS

Endometrial carcinoma is the most common invasive malignancy of the female genital tract. Unlike cervical carcinoma, there is not a well documented progression from precursor lesions to invasive carcinoma. Known risk factors for the development of endometrial carcinoma include obesity, nulliparity, late menopause, diabetes mellitus, hypertension, polycystic ovarian syndrome, estrogen-producing tumors, and unopposed exogenous estrogen supplementation. MRI has been shown to be useful in evaluating patients with known endometrial carcinoma. The vast majority of patients with endometrial carcinoma (up to 75%) will have stage I disease at the time of diagnosis. Tumor grade and depth of myometrial invasion are the two important prognostic factors which can affect therapy in these patients.21±23 Tumor grade will be known based upon histological ®ndings. However, depth of myometrial invasion can only be determined preoperatively with the use of imaging studies. On T1-weighted images, most endometrial carcinomas are isointense with the uterus unless they contain hemorrhagic areas. On T2-weighted images, most endometrial carcinomas (large enough to be detected as distinct masses) have a signal intensity intermediate between normal endometrium (higher signal intensity) and normal myometrium (lower signal intensity). Multiple studies have been performed evaluating the ability of MRI to depict accurately the depth of myometrial invasion.24±31 Using T2-weighted images and/or gadoliniumenhanced T1-weighted images, these studies have shown an accuracy of 75±95% in distinguishing super®cial from deep myometrial invasion (Figure 8). More importantly, these studies have shown that an intact junctional zone has a 100%

Figure 7 Axial FSE multicoil image of a bicornuate uterus. The contour of the uterine fundus is abnormal, with a large indentation separating the two endometrial cavities. There is a large low-signal intensity leiomyoma in the right horn

fying anomalies. The septate uterus is more commonly associated with infertility, and can be treated surgically through a hysteroscopic approach. There are only a few benign lesions of the cervix commonly seen on MRI. Nabothian cysts represent dilated cervical glands usually seen following in¯ammation. These are very commonly visualized on MRI as homogeneous, round, sharply circumscribed areas of very high signal intensity on T2-weighted images. Rarely, ®broids can be seen originating from the cervix. These are otherwise identical to those originating within the uterus.

Figure 8 Axial multicoil image acquired following gadolinium administration demonstrates a large tumor extending beyond the uterus (arrow)

6 MRI OF THE FEMALE PELVIS negative predictive value (NPV) in excluding myometrial invasion, and that segmental disruption of an otherwise intact junctional zone has a 100% positive predictive value (PPV) in detecting at least super®cial myometrial invasion. Unfortunately, most patients with endometrial carcinoma are postmenopausal, and uterine zonal anatomy may not be as conspicuous as that seen in premenopausal patients. Thus, these data are based on a small number of patients. In addition, patients with large intraluminal polypoid tumors can have signi®cant expansion of the endometrial cavity with resultant distortion of zonal anatomy. Accurate assessment of myometrial invasion can be impossible in such cases.32 In patients without visible zonal anatomy, the presence of myometrial invasion must be presumed, based upon the appearance of the endometrial/myometrial interface. If this interface is irregular, invasion is presumed to be present, and if this interface is smooth, invasion is presumed to be absent. However, these ®ndings do not have a high PPV or NPV, and this limits their usefulness. All patients with clinically suspected stage I endometrial carcinoma must have evaluation of the cervix for possible tumor involvement. Endocervical curettage can be unreliable in this assessment. A few studies indicate an NPV of nearly 100% for MRI detection of cervical involvement.25,29 The positive data regarding MRI evaluation of cervical involvement in patients with endometrial carcinoma are more limited. This is because only a small number of patients will have cervical involvement, since the vast majority have stage I disease at the time of presentation. In addition, some patients may have distention of the cervical canal by clot or debris, which can give a false-positive diagnosis of cervical involvement. Uterine sarcomas are relatively rare tumors accounting for only 3±5% of all uterine cancers.33 The three most common histological variants are malignant mullerian mixed tumor (MMMT), leiomyosarcoma (LMS), and endometrial stromal sarcoma (ESS). MMMT and LMS each account for about 40% of all uterine sarcomas, while ESS accounts for 10±15%. The staging system for uterine sarcomas is the same as that used for endometrial carcinoma. The incidence of sarcomatous change in preexisting uterine leiomyomas is reported to be between 0.1 and 0.8%.33 A small percentage (4%) of these patients will have a history of prior pelvic radiotherapy. The presenting clinical symptoms include vaginal bleeding, pelvic pain, and pelvic mass. The diagnosis should be suspected if rapid uterine growth occurs, especially in a postmenopausal patient. MMMTs are histologically composed of a mixture of sarcoma and carcinoma. Almost all of these tumors occur after menopause. A history of prior pelvic radiotherapy can be elicited in up to 35% of cases.33 Postmenopausal bleeding is the most common presentation. The tumor usually grows as a large polypoid mass with areas of necrosis and hemorrhage. It spreads in a manner identical to endometrial carcinoma but tends to be more aggressive with signi®cant myometrial invasion in almost all cases. Endometrial stromal tumors are rare tumors composed of cells resembling normal endometrial stroma. They can be divided into three types based upon mitotic activity, vascular invasion, and prognosis.33 The endometrial stromal nodule is a benign lesion con®ned to the uterus. Endolymphatic stromal myosis in®ltrates the myometrium, may extend beyond the

uterus, and can metastasize. ESS, the third type of stromal tumor, is differentiated from stromal myosis mainly on the basis of mitotic activity and its much more aggressive course. Stromal tumors usually occur in perimenopausal patients. The MRI appearance of LMS and endometrial stromal tumors is not well known. The MRI ®ndings in a series of seven patients with surgically proven MMMT showed this tumor to have an appearance similar to endometrial carcinoma.34 The only feature that might suggest the diagnosis is that these tumors are usually very large and show deep myometrial invasion. However, when detected early, these tumors have an appearance identical to early endometrial carcinoma. Trophoblastic tissue can give rise to a variety of tumors.35 Complete hydatidiform mole, partial hydatidiform mole, invasive mole, and choriocarcinoma arise from villous trophoblastic tissue, while placental site trophoblastic tumor arises from nonvillous trophoblastic tissue. Complete hydatidiform mole has been shown to be the result of a purely paternal conceptus. This genetic abnormality results in trophoblastic differentiation and proliferation without development of an embryo. Partial hydatidiform mole is seen almost exclusively in triploid conceptuses, with two paternal chromosomal complements and one maternal chromosomal complement. Often a malformed fetus is found in association with a partial mole. The partial mole itself usually arises from only a portion (or part) of the placenta. Clinical presentation of these entities can be abnormal uterine enlargement, vaginal bleeding, elevated human chorionic gonadotropin levels, or the absence of fetal heart sounds. Invasive moles are thought to develop in previously existing complete moles. They are most commonly seen in the early months following evacuation of a complete hydatidiform mole. The hallmark of this entity is myometrial invasion. Choriocarcinoma is a malignant neoplasm which most commonly occurs after a molar pregnancy (sometimes remotely) but can be seen after a normal pregnancy, abortion, ectopic pregnancy, and possibly de novo. Placental site trophoblastic tumor arises from the nonvillous trophoblast which in®ltrates the placental site in normal pregnancy. It is considered an atypical form of choriocarcinoma. The importance of recognizing this tumor type pathologically lies in its less aggressive behavior compared with choriocarcinoma. This behavior makes this tumor amenable to surgery but often resistant to chemotherapy. There are few data regarding the role of MRI in the evaluation and management of patients with trophoblastic tumors.36,37 MRI is usually performed in patients with a known diagnosis of persistent mole following therapy or patients developing invasive mole or choriocarcinoma. A few studies have shown that prior to therapy most tumors show heterogeneous signal intensity on T2-weighted images and distort or obliterate the normal zonal anatomy. These tumors only rarely appear as endometrial masses. Many tumors show markedly increased vascularity as evidenced by visualization of tortuous, dilated vessels within the tumor and/or the adjacent myometrium. Previous studies have also shown that in patients responding to chemotherapy there was a progressive decrease in uterine size, tumor size, tumor vascularity, and a progressive improvement in visualization of normal zonal anatomy. All patients completely responding to therapy will show normal uterine

MRI OF THE FEMALE PELVIS

7

size, normal zonal anatomy and no evidence of tumor following completion of chemotherapy.

8 MALIGNANT DISEASE OF THE CERVIX Invasive cervical cancer is thought to develop over time from noninvasive precursor lesions.38 These precursor lesions are referred to as cervical intraepithelial neoplasia (CIN). CIN is divided pathologically into three grades: CIN 1 (minor dysplasia), CIN 2 (moderate dysplasia), and CIN 3 (severe dysplasia). CIN 3 is synonymous with carcinoma-in-situ. Available evidence indicates that up to 40% of CIN 3 lesions and a lesser proportion of CIN 1 and CIN 2 lesions would progress to invasive cancer if untreated. Imaging studies play no role in the detection of cervical carcinoma or its precursor lesions. On T1-weighted images, cervical carcinoma is of intermediate signal intensity and usually isointense with the uterine corpus and cervix. On T2-weighted images, cervical carcinoma is usually of intermediate signal intensity relative to the lowsignal ®brous cervical stroma and high-signal cervical and endometrial canals. It is the sharp contrast with the ®brous stroma that makes T2-weighted imaging so crucial in depicting the tumor and its depth of invasion. Most patients with CIN or microinvasion will have a normal appearance on MRI.39 However, some patients with early invasive disease can also have a normal appearance on MRI. Prior studies have shown that the detection of a macroscopic lesion on MRI had a 100% PPV in determining the presence of at least invasive disease. A normal appearance on MRI has a less than 100% NPV in excluding the presence of invasive disease. Thus, a normal MRI requires further investigation to exclude early invasive disease. It is important to note that there are few studies in the literature evaluating this point, and few studies have been performed with high-resolution imaging. The therapy of cervical carcinoma is dependent upon the stage of the disease at the time of diagnosis. The most crucial determination is the presence or absence of tumor invasion into the parametrium. Patients without parametrial invasion will usually be treated surgically. The presence of parametrial invasion precludes surgical therapy. A number of MRI studies have shown that the ®nding of a completely intact ring of low-signal intensity cervical stroma has a 100% NPV in excluding parametrial invasion (Figure 9).39±42 Unfortunately, focal areas of disruption of the stromal ring or full-thickness involvement of the ®brous stroma by tumor have a signi®cantly lower than 100% PPV in determining parametrial invasion The ability of MR imaging to accurately determine vaginal, pelvic sidewall, bladder, and rectal involvement in patients with cervical carcinoma has been evaluated in only a limited number of studies. The NPV of MR in excluding involvement of these structures is probably close to 100%. The PPV is dif®cult to assess because of the small number of positive cases. There is little if any role for gadolinium-enhanced MR imaging in the staging of cervical carcinoma.24,27,43 Several studies have indicated that gadolinium-enhanced images consistently overestimate the depth of cervical invasion. In addition, there is overestimation of involvement of the parametrium, bladder, and vagina with gadolinium-enhanced images.

Figure 9 Coronal FSE multicoil image demonstrating a large tumor in the right cervix. The thin rim of intact ®brous stroma (curved arrows) indicates that parametrial invasion is absent

9

BENIGN DISEASE OF THE OVARIES

In premenopausal patients the normal ovaries almost always contain multiple small follicles (Figure 10). These appear as round, unilocular, sharply circumscribed high-signal areas on T2-weighted images. They are usually less than 1.0 cm across but can attain a size of 2.5 cm and still be considered normal follicles. A corpus luteum cyst or follicular cyst can have an identical appearance to normal follicles other than their slightly larger size. They can sometimes contain hemorrhage, which will be seen as a high signal on T1-weighted images. Theca lutein cysts can be seen in association with excessive levels of hCG, generally associated with trophoblastic disease. These cysts are usually multilocular, bilateral, and very large. It is well known that even postmenopausal patients may have simple cysts within their ovaries.44 Studies in the ultrasound literature have shown that if a cyst is simple, less than 3.0 cm in size, and shows no abnormal ¯ow on Doppler evaluation, it is most probably benign. There are no similar studies in the MRI literature. An infrequently seen benign disorder of the ovaries is polycystic ovarian disease. This is thought to result from unopposed estrogenic stimulation and chronic anovulation. This disorder is commonly associated with obesity, hirsutism, and oligomenorrhea. On MRI, the ovaries in these patients usually appear mildly enlarged, and contain multiple small cysts in the periphery with an abundant central stroma. A common cystic lesion of the adnexal region is the paratubal cyst. These cysts develop from wolf®an duct remnants within the broad ligament, separate from the ovary. These are almost always benign in nature. Their appearance is indistinguishable from simple ovarian cysts, other than their

8 MRI OF THE FEMALE PELVIS

Figure 10 Axial FSE multicoil image demonstrating multiple follicles within the ovaries. The appearance suggests polycystic ovarian disease

extraovarian origin. It is this latter point that is key to their diagnosis. Although there is a paucity of published data, paratubal cysts usually displace an otherwise normal appearing ovary and therefore appear as a separate structure.45 Another cystic adnexal lesion is the Gartner's duct cyst.45 This develops in a remnant of the muÈllerian duct. Their signal

characteristics are identical to other cystic lesions. They can be diagnosed by their paravaginal location (separate from the ovaries) and usually tubular con®guration. They are usually well demonstrated on high-resolution MRI. These patients commonly present with dyspareunia. Benign epithelial tumors of the ovary include serous and mucinous cystadenomas (Figure 11). The appearance of these tumors at MR imaging is variable, and can be indistinguishable from their malignant counterparts. Benign cystadenomas can appear as simple cystic intraovarian masses. The presence of internal septations, solid components, or papillary projections makes malignancy more likely. Intravenous gadolinium can sometimes be helpful in detecting solid components, papillary projections or even septations.46 Ovarian ®bromas, ®brothecomas, and Brenner tumors are solid ovarian masses that can contain ®brous tissue, and, therefore, may show diffuse or focal areas of low signal intensity on all pulse sequences. Fibromas are typically extremely low signal intensity, while ®brothecomas typically contain multiple high-signal foci. One must be careful to distinguish these masses from pedunculated subserosal ®broids. The only way to do so de®nitively is to identify them as intraovarian. Two ovarian lesions which can occasionally be dif®cult to distinguish are the ovarian dermoid and an ovarian endometrioma.47 The dermoid tumor is the most common form of benign teratoma, and can be de®nitively characterized by its fat content. Endometriosis is a disease characterized by rests of normal functional endometrium in abnormal locations. The most common location of endometriosis is the ovary. This can occur as a large mass, referred to as an endometrioma, which can be characterized by its blood content. On MRI, a dermoid and an endometrioma can have identical signal characteristics on both T1- and T2-weighted images. The most de®nitive way to distinguish these lesions is by using sequences which selectively suppress the signal from fat- or water-containing structures.47 The fat in dermoids is of high signal on T1-weighted images obtained with and without water suppression, and is of low signal on T1-weighted images obtained with fat suppression (Figure 12). Endometriomas appear as a high signal on T1-weighted images obtained with and without fat suppression, and appear as a low signal on T1-weighted images obtained with water suppression. Unfortunately, endometriomas are indistinguishable from other hemorrhagic cysts. Multiple such cysts make the diagnosis of endometriosis more likely (Figure 13). Sometimes hemosiderin can be visualized within and on the surface of the ovaries. Hemosiderin appears as tiny foci of low signal, most conspicuous on T2-weighted images. MRI is insensitive in detecting small implants of endometriosis involving bowel, the bladder, or other peritoneal surfaces.

10

Figure 11 Sagittal conventional spin echo image of a mucinous cystadenoma (curved arrow). Note the thin septations and lack of solid elements. The signal behavior of the mass is identical to urine in the bladder (straight arrow) on all sequences

MALIGNANT DISEASE OF THE OVARIES

The most common malignant neoplasms of the ovary are of epithelial origin, particularly serous and mucinous cystadenocarcinomas (Figure 14). These tumors often appear primarily cystic in nature. On MRI, the cystic portions will appear as very high-signal areas on T2-weighted images. They also, however, commonly contain internal septations and solid elements.

MRI OF THE FEMALE PELVIS

9

Figure 12 Complex dermoid containing both ¯uid and fat elements. Note that the fat (arrow) is isointense with pelvic fat on the (a) T1-weighted and (b) T2-weighted images. On the (c) water suppression and (d) fat suppression images, the fat (arrows) suppresses on the latter sequence

The solid components are of lower signal on the T2-weighted images, and often show enhancement following intravenous administration of gadolinium.46 However, these tumors can sometimes appear indistinguishable from simple cysts. In general, even when purely cystic in appearance, these masses will be larger in size than other simple benign cysts. A variety of other malignant neoplasms that are primarily solid or mixed can involve the ovaries. Ovarian malignancies usually spread from the surface of the ovaries into the peritoneal cavity. Most commonly, metastases involve the omentum, mesentery, and peritoneal surface of solid viscera. Due to artifacts from the bowel, MRI is not as

sensitive as computerized tomography in detecting these metastases. Intravenous gadolinium is essential if one attempts to detect these metastases on MRI.

11

RELATED ARTICLES

Male Pelvis Studies Using MRI; Multi Echo Acquisition Techniques Using Inverting Radiofrequency Pulses in MRI; Whole Body Machines: NMR Phased Array Coil Systems.

10 MRI OF THE FEMALE PELVIS

Figure 13 Axial conventional spin echo images demonstrating endometriosis. Note on (a) the intermediate weighted sequence, multiple hyperintense lesions (arrows) are seen. On (b) the T2-weighted image, some of these lesions become hypointense whereas others remain hyperintense consistent with blood of various ages. Also note tethering of the rectum (arrow) secondary to adhesions, an intramural ®broid (arrowhead) and the bladder (b) pushed anteriorly

Figure 14 Axial FSE multicoil image of a clear cell adenocarcinoma of the ovary (arrows). Note both solid and cystic elements indicative of a malignancy

12

REFERENCES

1. R. C. Smith and S. M. McCarthy, Radiol. Clin. N. Am., 1994, 32, 109. 2. M. L. Wood and R. M. Henkelman, Med. Phys., 1986, 13, 794. 3. R. C. Smith, C. Reinhold, R. C. Lange, T. R. McCauley, R. Kier, and S. McCarthy, Radiology, 1992, 184, 665.

4. R. C. Smith, C. Reinhold, T. R. McCauley, R. C. Lange, R. T. Constable, R. Kier, and S. M. McCarthy, Radiology, 1992, 184, 671. 5. J. Hennig, A. Nauerth, and H. Friedburg, Magn. Reson. Med., 1986, 3, 823. 6. J. Hennig and H. Friedburg, Magn. Reson. Imag., 1988, 6, 391. 7. R. V. Mulkern, P. S. Melki, P. Jakab, N. Higuchi, and F. A. Jolesz, Med. Phys., 1991, 18, 1032. 8. C. E. Hayes, and P. B. Roemer, Magn. Reson. Med., 1990, 16, 181. 9. C. E. Hayes, N. Hattes, and P. B. Roemer, Magn. Reson. Med., 1991, 18, 309. 10. P. B. Roemer, W. A. Edelstein, C. E. Hayes, S. P. Souza, and O. M. Mueller, Magn. Reson. Med., 1990, 16, 192. 11. C. L. Janus, H. P. Wiczyk, and N. Laufer, Magn. Reson. Imag., 1988, 6, 669. 12. A. R. Lupetin, in `Magnetic Resonance Imaging', ed. D. D. Stark and W. G. Bradley, C. V. Mosby, St Louis, 1988, p. 1270. 13. S. McCarthy, C. Tauber, and J. Gore, Radiology, 1986, 160, 119. 14. L. M. Scoutt, S. D. Flynn, D. J. Luthringer, T. R. McCauley, and S. M. McCarthy, Radiology, 1991, 179, 403. 15. M. C. Lin, B. B. Gosink, S. I. Wolf, M. R. Feldesman, C. A. Stuenkel, P. S. Braly, and D. H. Pretorius, Radiology, 1991, 180, 427. 16. T. J. Dubinsky, H. R. Parvey and N. Maklad, Am. J. Roentgenol., 1997, 169, 145. 17. D. G. Mitchell, L. Schonholz, P. L. Hilpert, R. G. Pennell, L. Blum, and M. D. Rifkin, Radiology, 1990, 174, 827. 18. T. R. McCauley, L. M. Scoutt, and S. B. Flynn, JMRI, 1991, 1, 319. 19. D. G. Mitchell, Radiol. Clin. N. Am., 1992, 30(4), 777. 20. J. S. Pellerito, S. M. McCarthy, M. B. Doyle, M. G. Glickman, and A. H. Decherney, Radiology, 1992, 183, 795. 21. W. T. Creasman, C. P. Morrow, B. N. Bundy, M. D. Homesley, J. E. Graham, and P. B. Heller, Cancer, 1987, 60, 2035. 22. W. T. Creasman and J. C. Weed in `Gynecologic Oncology', 2nd edn, ed. M. Coppleson, Churchill Livingstone, London, 1992, p. 780. 23. M. H. Lutz, P. B. Underwood, A. Kreutner, and M. C. Miller, Gynecol. Oncol., 1978, 6, 83.

MRI OF THE FEMALE PELVIS 24. Y. Hirano, K. Kubo, Y. Hirai, S. Okada, K. Yamada, S. Sawano, T. Yamashita, and Y. Hiramatsu, RadioGraphics, 1992, 12, 243. 25. H. Hricak, J. L. Stern, M. R. Fisher, L. G. Shapeero, M. L. Winkler, and C. G. Lacey, Radiology, 1987, 162, 297. 26. H. Hricak, L. V. Rubinstein, G. M. Gherman, and N. Karstaedt, Radiology, 1991, 179, 829. 27. H. Hricak, B. Hamm, R. C. Semelka, C. E. Cann, T. Nauert, E. Secaf, J. L. Stern, and K. J. Wolf, Radiology, 1991, 181, 95. 28. H. H. Lien, V. Blomlie, C. Trope, J. Kaern, and V. M. Abeler, Am. J. Roentgenol., 1991, 157, 1221. 29. H. V. Posniak, M. C. Olson, C. M. Dudiak, M. J. Castelli, J. Dolan, R. A. Wisniewski, J. H. Isaacs, S. K. Sharma, and V. Bychkov, RadioGraphics, 1990, 10, 15. 30. S. Sironi, G. Taccagni, P. Garancini, C. Belloni, and A. DelMaschio, Am. J. Roentgenol., 1992, 158, 565. 31. S. Sironi, E. Colombo, G. Villa, G. Taccagni, C. Belloni, P. Garancini, and A. DelMaschio, Radiology, 1992, 185, 207. 32. L. M. Scoutt, S. M. McCarthy, S. D. Flynn, R. C. Lange, F. Long, R. C. Smith, S. K. Chambers, E. I. Kohorn, P. Schwartz and J. T. Chambers, Radiology, 1995, 194, 567. 33. J. R. Lurain and M. S. Piver in `Gynecologic Oncology', 2nd edn, ed. M. Coppleson, Churchill Livingstone, London, 1992, p. 827. 34. L. G. Shapeero and H. Hricak, Am. J. Roentgenol., 1989, 153, 317. 35. F. J. Paradinas, in `Gynecologic Oncology', 2nd edn, ed. M. Coppleson, Churchill Livingstone, London, 1992, p. 1013. 36. J. W. Barton, S. M. McCarthy, E. I. Kohorn, L. M. Scoutt, and R. C. Lange, Radiology, 1993, 186, 163. 37. H. Hricak, B. E. Demas, C. A. Braga, M. R. Fisher, and M. L. Winkler, Radiology, 1986, 161, 11. 38. M. Coppleson, K. H. Atkinson, and F. C. Dalrymple, in `Gynecologic Oncology', 2nd edn, ed. M. Coppleson, Churchill Livingstone, London, 1992, p. 571. 39. K. Togashi, K. Nishimura, T. Sagoh, S. Minami, S. Noma, I. Fujisawa, Y. Nakano, J. Konishi, H. Ozasa, I. Konishi, and T. Mori, Radiology, 1989, 171, 245. 40. H. Hricak, C. G. Lacey, L. G. Sandles, Y. C. F. Chang, M. L. Winkler, and J. L. Stern, Radiology, 1988, 166, 623. 41. S. H. Kim, B. I. Choi, H. P. Lee, S. B. Kang, Y. M. Choi, M. C. Han, and C. W. Kim, Radiology, 1990, 175, 45.

11

42. S. Sironi, C. Belloni, G. L. Taccagni, and A. DelMaschio, Am. J. Roentgenol., 1991, 156, 753. 43. Y. Yamashita, M. Takahashi, T. Sawada, K. Miyazaki, and H. Okamura, Radiology, 1992, 182, 643. 44. D. Levine, B. B. Gosink, S. I. Wolf, M. R. Feldesman, and D. H. Pretorius, Radiology, 1992, 184, 653. 45. R. Kier, Am. J. Roentgenol., 1992, 158, 1265. 46. S. K. Stevens, H. Hricak, and J. L. Stern, Radiology, 1991, 181, 481. 47. R. Kier, R. C. Smith, and S. M. McCarthy, Am. J. Roentgenol., 1992, 158, 321.

Biographical Sketches Robert Smith. b 1960. B.A. (Mathematics), 1981, Johns Hopkins University, M.D., 1985, Yale University. Radiology Resident and MR Fellow, Yale University, 1986±91. Assistant Professor, Yale University, 1991±96. Associate Professor, and Director of MRI, Yale University, 1996±97. Research interests: fast spin echo MRI, clinical applications of multicoil arrays, basic MRI physics. Michael J. Varanelli. b 1970. B.Sc. (Biology), Bucknell University, 1992. M.D., University of Pittsburgh, 1996. Radiology Resident, Yale University, 1997±present. Leslie M. Scoutt, b 1952. B.A. (Biology), Wesleyan University, 1974, M.D., University of Rochester, 1978. Radiology Resident, Beth Israel Hospital, Boston, MA, 1982±85. Cross-Sectional Imaging Fellow, Yale University, 1985±87. Assistant Professor, Yale University, 1987± 94. Associate Professor, Yale University, 1994±present. Research interests: MRI and ultrasound of the female pelvis, breast and vascular ultrasound. Shirley McCarthy. b 1949. B.A. (Biological Sciences), 1971, State University of New York, Albany Ph.D. (Mammalian Physiology), 1975, Cornell University, M.D., 1975, Yale University School of Medicine. Associate Professor, Yale University, 1989±present. Chief of MRI, Yale University, 1987±present. Research interests: gynecological imaging, cost-effective analyses of MRI use.

PEDIATRIC BODY MRI

Pediatric Body MRI Rosalind B. Dietrich and Gerald M. Roth University of California, Irvine, Orange, CA, USA

1 INTRODUCTION Although initially slow to become established, magnetic resonance imaging (MRI) now plays a vital role in the diagnostic imaging evaluation of many pediatric diseases and disorders. The ability to produce multiplanar images in a noninvasive manner without the use of ionizing radiation and with only minimal patient preparation has made MRI a useful complement to other cross-sectional modalities such as ultrasound and computerized tomography (CT). The most frequent indications for an MRI study in the pediatric population include the evaluation of congenital anomalies and the characterization of tumors and other mass lesions.

2 CONGENITAL ANOMALIES MRI plays an important role in evaluating both complex congenital anomalies, where ultrasound is inadequate, and simple congenital anomalies, where sonography is not possible, is incomplete or is suboptimal. Congenital anomalies are well visualized with MRI due to its provision of excellent anatomical detail and its multiplanar capabilities. The majority of lesions are adequately assessed using multiplanar T1- and T2weighted spin echo imaging. Occasionally, additional sequences are required, especially if the presence of fat or ¯owing blood is to be determined. The ability of MRI to differentiate vessels containing ¯owing blood from other mediastinal structures without the use of bolus injection of contrast material makes it an ideal choice for evaluating anomalies of the great vessels.1,2 Preoperative mapping may alleviate the need for angiography in selected patients. In addition, in such lesions as the double aortic arch and the right aortic arch with an aberrant left subclavian artery and a left ligamentum arteriosum, MRI can evaluate for the presence of esophageal and/or tracheal compression that may be associated with these lesions. In the entity known as the `pulmonary sling', where the left pulmonary artery arises from the right pulmonary artery and passes between the trachea and the esophagus, the presence of tracheal compression may be similarly assessed. Other, more rare, symptomatic abnormalities, as well as the asymptomatic great vessel anomalies can also be evaluated. In patients with a right aortic arch, MRI can determine the presence or absence of mirror image branching and can demonstrate any associated cardiac anomalies. Cardiac gated T1-weighted images in the axial and coronal planes can often provide suf®cient information for these diagnoses to be made, although sometimes images parallel to the course of the aorta are very useful. The most common congenital pediatric abnormalities involving the lung include congenital lobar emphysema, cystic

1

adenomatoid malformation, and sequestration. The ®rst two entities can be evaluated using MRI, although plain ®lm radiography and/or CT examination are usually suf®cient. MRI, however, can be very useful in cases of sequestration, as it may be able to demonstrate the anomalous feeding vessels and draining veins.3 The corresponding lung tissue may appear solid or aerated on MRI, depending on its connection, if any, with the bronchial tree. Abnormalities of the diaphragm, such as diaphragmatic hernias and eventrations, can also be evaluated by MRI. Coronal and/or sagittal T1-weighted images are useful in determining the presence or absence of a portion of the diaphragm, as well as identifying which, if any, of the normal abdominal components are in the chest.4 In patients with anterior abdominal wall defects, such as gastroschisis and omphalocele, MRI can be used to determine which abdominal organs have protruded through the defect and de®ne their position (Figure 1). In the abdomen, congenital anomalies of the hepatobiliary system such as choledochal cysts, biliary atresia, and polycystic liver disease can all be evaluated using MRI, but are more commonly diagnosed using ultrasound. MRI plays a more important role in the visualization of abnormal vascular anatomy in such anomalies as the Budd±Chiari malformation and discontinuity of the inferior vena cava. In the Budd±Chiari malformation, MRI may identify obliteration of the hepatic veins and may help determine if a congenital web or a thrombus is responsible for the condition. In the retroperitoneum, abnormalities in renal position and development are readily seen with MRI. Renal ectopia, whether high (intrathoracic) or low (pelvic) can be differentiated from renal agenesis more easily with MRI than with ultrasound, since visualization of the abdominopelvic cavity by MRI is not limited by the presence of bowel gas or bone, as it is with ultrasound. Fusion anomalies, such as crossed fused ectopia and horseshoe kidney are also well demonstrated by MRI.5±7 Crossed fused ectopia, like most congenital renal anomalies, is best imaged using coronal T1-weighted images, and is diagnosed when an abnormally positioned fused kidney is visualized on one side and there is no identi®able renal tissue on the contralateral side. Horseshoe kidneys, on the other hand, are easier to diagnose on axial T1-weighted images, as the fusion of the lower poles is often more apparent in this plane. In children with renal agenesis, the ipsilateral adrenal gland develops a discoid rather than a chevron-shaped appearance, and can be seen as an elongated linear structure on coronal magnetic resonance (MR) images.4 Although all of the renal cystic diseases can be demonstrated using MRI, ultrasound remains the primary imaging modality for this class of diseases. MRI is most useful in clarifying confusing cases and in demonstrating complications such as hemorrhage. Simple renal cysts appear as homogeneous, well-de®ned masses on the MR image and are of low signal intensity on T1-weighted sequences and of very high signal intensity on T2-weighted sequences. The cyst wall and the ¯uid in the cyst may be indistinguishable. If hemorrhage into the cyst occurs, or if the cyst has a high protein concentration for another reason, inhomogeneous high signal intensity can be demonstrated on both T1- and T2-weighted sequences. Despite these guidelines, there is a wide range of variability in the appearance of hemorrhagic cysts, and in some cases it is not

2 PEDIATRIC BODY MRI

Figure 1 Omphalocele. (a) Sagittal view (SE 350/11): multiple loops of bowel and a large volume of the liver have herniated into the omphalocele sac. (b) Axial view (FSE 4000/102Ef): liver (L), gall bladder (GB), bowel (B), spleen (Sp) and stomach (St) are identi®ed in the omphalocele sac. (c) Coronal view (SE 350/12): both kidneys are abnormally superior in position, lying just below the diaphragm

always possible to distinguish a hemorrhagic cyst from an infected cyst or even from a neoplasm. MRI is an excellent modality for the evaluation of congenital anomalies of the genital tract. In disorders of sexual differentiation, MRI can clearly show the absence of, or abnormal location of, pelvic organs in children with ambiguous genitalia or genetic abnormalities.7±9 Imaging in both the axial and sagittal planes, and with both T1- and T2-weighted sequences is often necessary in these patients with Turner syndrome, testicular feminization, hermaphrodism, or one of the various forms of pseudohermaphrodism. MRI can also evaluate the spectrum of MuÈllerian duct anomalies, ranging from uterus didelphys (two uteri, two cervices, and two vaginas) through bicornuate uterus (two uteri, single vagina, and cervix) to uterus septus (single uterus, cervix, and vagina with a uterine septum).10±12 The distinction between the last two entities is crucial if surgical intervention is planned. MRI can often make this distinction in postpubertal girls. On T2-weighted images in a plane axial to the body of the uterus, the bicornuate uterus will demonstrate a medium signal intensity strip of myometrium separating the two low signal intensity junctional zones that the septate uterus will not. MRI can also evaluate the urinary anomalies, such as renal agenesis or renal malposition, which are often associated with MuÈllerian duct anomalies. Renal anomalies can also occur in patients with the Mayer± Rokitansky±Kuster±Hauser syndrome.13 In this syndrome, which is characterized by aplasia of both the uterus and upper vagina, the spectrum of renal anomalies includes unilateral agenesis, unilateral or bilateral ectopia, horseshoe kidney, malrotation, and collecting system abnormalities. Vertebral anomalies have also been reported. Because MRI can be used to evaluate the vagina noninvasively, it is the modality of choice for patients with this disorder, as well as for patients with isolated vaginal atresia.

In patients with hematometrocolpos, T1- and T2-weighted images can demonstrate a high signal intensity collection within a markedly dilated upper vagina and uterine cavity14±16 (Figure 2). These dilated structures may compress the adjacent bladder or rectum. The ¯uid collection may extend into the fallopian tube, leading to a dilated tortuous fallopian tube visualized extending into the abdomen. MRI has also proved useful in identifying undescended testes in male children prior to surgery. Ultrasound is performed initially and, if it is unsuccessful, MRI is then performed.17,18 The undescended testis is usually in the inguinal canal, but if it is in the abdomen or pelvis, the testis can be adjacent to the lateral bladder wall, the psoas muscle, the iliac vessels or in the retroperitoneum or the super®cial inguinal pouch. On T1-weighted images, the testis is of medium signal intensity and can frequently be identi®ed if it is surrounded by fat, which is of high signal intensity. On T2weighted images, the central portion of the testis is of high signal intensity with a surrounding rim of medium signal intensity, and therefore can be easily distinguished from muscles and lymphadenopathy. Anorectal anomalies, such as imperforate anus and ectopic anus, are also well visualized using MRI. Both these disorders are due to failure of descent of the hindgut. In imperforate anus, the rectum terminates in a blind pouch, whereas in ectopic anus, the more common of the two entities, there is a ®stulous connection between the pouch and another structure, such as the perineum, vestibule, vagina, urethra, bladder, or cloaca. Plain ®lm radiography, contrast examination, and ultrasonography have all been helpful in the preoperative evaluation of these patients. MRI, however, can noninvasively give a multiplanar view of the hindgut, puborectalis sling and adjacent structures, de®ning the anatomy to better advantage.19±22 In this regard, axial and coronal T1-weighted images can demonstrate the anatomy of the puborectalis sling, determine whether

PEDIATRIC BODY MRI

3

Figure 2 Hematometros involving one side of a uterus didelphys. (a) Coronal view (SE 600/18): high signal intensity blood products are identi®ed in a markedly distended uterine cavity; a second uterine cavity (arrows) with an endometrial stripe is identi®ed compressed on the right. (b) Axial view (FSE 4000/102Ef): the distended left uterine cavity occupies most of the volume of the pelvis, displacing the left ovary (Ov) anteriorly

Figure 3 Mediastinal teratoma. (a) Coronal view (SE 625/11): a mediastinal mass is identi®ed with locules of differing signal intensity; a few small foci of bright signal intensity are noted within the mass. (b) Coronal view (SE 625/11, chemical lipid presaturation): using a lipid-selective presaturation pulse, the formerly bright areas in the lesion become dark, verifying that they represent fat; the presence of fat in the lesion histologically characterizes it as a teratoma

or not it is hypoplastic, and de®ne its relationship to the hindgut. 3 TUMORS AND OTHER MASS LESIONS Due to its multiplanar capability, its ability to distinguish vessels with ¯owing blood from other structures, without the need for bolus injection of contrast material, and its superior contrast resolution compared with that of CT, MRI is an extremely useful tool in the evaluation of mass lesions.4 Once a lesion has been discovered on a chest radiograph or on an ab-

dominal or pelvic ultrasound, MRI can help to characterize the lesion, identify the organ of origin, de®ne its extent, clarify its relationship to adjacent vessels, and evaluate for distant metastasis. MRI can classify a lesion as either cystic, solid, or mixed. Although some solid lesions such as teratomas, lipomas, and hemorrhagic lesions may demonstrate characteristic MR appearances, the majority of solid lesions are impossible to differentiate using signal intensity alone. Most such solid lesions are of medium signal intensity on T1-weighted images and of high signal intensity on T2-weighted images. However, when the signal characteristics of a lesion are combined with information about its organ of origin, its relationship to adja-

4 PEDIATRIC BODY MRI cent vessels and/or its sites of metastasis, a de®nitive diagnosis can often be made. MRI can be used before treatment, to plan a surgical approach or a radiation port, as well as after treatment, to assess for residual or recurrent tumor after surgery, chemotherapy and/or radiation therapy. Gadolinium chelate contrast agents may be useful in characterizing vascular masses and in evaluating renal and perirenal lesions.23 In these instances, T1-weighted images obtained dynamically during, and/or immediately after, intravenous contrast administration may more clearly de®ne the borders of the lesion, characterize its internal architecture, accentuate surrounding adenopathy, and/or uniquely de®ne the histology of the lesion. In the thorax, MRI is particularly useful in the evaluation of masses arising in the posterior mediastinum, as it may noninvasively demonstrate the presence or absence of intraspinal extension.24,25 These posterior mediastinal masses are most frequently of neurogenic origin and include such entities as neuroblastoma, ganglioneuroma, and neuro®broma. They are all of medium signal intensity on T1-weighted images and of high signal intensity on T2-weighted images. Coronal and axial images are useful in demonstrating intraspinal extension, if any, and in clarifying the relationship of the lesion to adjacent vasculature. MRI can also demonstrate lesions in the anterior and middle compartments of the mediastinum, and may even have a slight edge over CT in selected cases.26 For example, coronal and/or sagittal images may help de®ne the extent of neck masses that secondarily involve the mediastinum or lung apices. These lesions include cystic hygroma, hemangioma, and ®bromatosis. Fat selective sequences may also be helpful in distinguishing mediastinal teratomas from other mass lesions (Figure 3). Additionally, as with posterior compartment masses, MRI may help clarify the relationship of the lesion to adjacent vessels and other structures, due to its multiplanar capability, its ability to distinguish vessels with ¯owing blood from other structures, without the need for bolus injection of contrast material, and its superior contrast resolution. In the abdomen and pelvis, MRI is an excellent modality with which to differentiate masses arising from the liver, kidney, adrenal gland, and paraspinal regions. Some of the more common lesions arising from these areas warrant discussion.

In children with hemangioendotheliomas, MRI can often demonstrate the mass in the liver and may be able to map the feeding and draining vessels. An abrupt caliber decrease in the abdominal aorta distal to the origin of the celiac axis may also be observed.4 Serial imaging after gadolinium chelate administration can often uniquely identify the hepatic lesion as a hemangioendothelioma due to its distinctive temporal pattern of peripheral to central enhancement. The most common malignant tumors of the liver, hepatoblastomas and hepatocellular carcinomas, can also be evaluated using MRI. Both lesions demonstrate variable signal intensity on both T1- and T2-weighted images (Figure 4). Because MRI can usually identify the portal and hepatic veins, it can assess for vascular invasion by the tumor and can also de®ne the extent of the tumor with respect to the segmental anatomy of the liver, thus aiding in determining resectability of the lesion. Wilms' tumor is the most common solid abdominal mass and the most common primary renal neoplasm in children, with a peak incidence between the ages of 1 and 3 years. An increased incidence of Wilms' tumor is noted in children with aniridia, hemihypertrophy, neuro®bromatosis, genitourinary abnormalities, and the Beckwith±Wiedemann syndrome. The tumor presents most often as a unilateral abdominal mass, although the lesion is bilateral in 5±10% of cases. The initial diagnosis is usually made by ultrasound, which demonstrates a solid renal mass. MRI, however, may be superior to both ultrasound and CT in diagnosing and de®ning the extent of Wilms' tumor.27,28 In this area of the body, coronal T1weighted images can often de®ne the organ of origin of a lesion extremely well, separating renal lesions from those arising in the adjacent liver or adrenal gland. With a combination of coronal and axial planes, and T1- and T2-weighting, a study can de®ne the full extent of the lesion, its possible invasion of adjacent organs or vessels, as well as assess for the presence of associated lymphadenopathy. In selected cases, intravenous administration of a gadolinium chelate and/or use of MR angiographic techniques may be necessary.23 Neuroblastoma is the most common extracranial solid malignant tumor in children. Neuroblastoma and its more differentiated forms, ganglioneuroblastoma and ganglioneuroma, arise from primitive sympathetic neuroblasts and therefore may

Figure 4 Hepatoblastoma. (a) Coronal and (b) sagittal (SE 300/20) views: a large mass, hetereogeneous in signal intensity, projects inferiorly from the liver and invades the portal vein; the speckled high signal intensity areas probably represent hemorrhage. (c, d) Axial views (SE 2500/30,80): although markedly heterogeneous, the lesion is predominantly low in signal intensity

PEDIATRIC BODY MRI

5

Figure 5 Neuroblastoma. (a) Coronal view (SE 300/16): an enlarged right adrenal gland (arrows) displaces the right kidney inferiorly. (b) Axial view (FSE 2500/17Ef): the liver is also enlarged and ®lled with metastases, leading to a salt and pepper heterogeneity to its signal intensity

Figure 6 Neuroblastoma. (a) Coronal view (SE 616/15): a large multilobulated medium signal intensity mass surrounds the aorta, displacing it and the inferior vena cava to the right and anteriorly. (b) Axial view (SE 2128/15): a portion of the mass is seen extending into the bony spinal canal (arrow); the left psoas muscle (P) is elevated and displaced to the left by this high signal intensity tumor

arise from the adrenal gland (most commonly) or from anywhere else along the sympathetic chain from the nasopharynx to the presacral region. Children with neuroblastoma may present with a palpable abdominal mass or with symptoms referable to metastases, such as bone pain. The staging of neuroblastoma is based on both local extent as well as on the presence or absence of metastases. Clinically, however, it is more important to determine resectability of the lesion, as surgery remains the treatment of choice. On MRI, neuroblastoma is usually of medium signal intensity on T1-weighted images and of high signal intensity on T2-weighted images. It is less well de®ned than Wilms' tumor and if it arises from the adrenal gland it can displace the kidney inferiorly and/or laterally (Figure 5). As with Wilms' tumor, MRI can play a role in de®ning the full extent of a lesion, identifying its organ of origin, assessing for possible invasion of adjacent organs or

vessels, and demonstrating the presence or absence of metastases.29±31 Evidence of tumor extending into the spinal canal or encasing the retroperitoneal vessels can classify the neuroblastoma as unresectable (Figure 6). Neonatal adrenal hemorrhage is another entity that may present as an asymptomatic abdominal mass. Differentiation of neonatal adrenal hemorrhage from neuroblastoma is crucial, and can usually be done using ultrasound or MRI. On ultrasonography, neonatal adrenal hemorrhage is usually anechoic and avascular, in contradistinction to neuroblastoma, which is usually echogenic and vascular. In some instances, adrenal hemorrhage does not appear cystic on ultrasonography and MRI may be extremely useful in helping to make this distinction. On MR images neonatal adrenal hemorrhage behaves similarly to hematomas elsewhere in the body, with a changing MR appearance as the hematoma evolves (Figure 7). More

6 PEDIATRIC BODY MRI

Figure 7 Neonatal adrenal hemorrhage. (a, b) Coronal views (SE 450/11): an enlarged right adrenal gland (arrows) is identi®ed (a) that does not demonstrate enhancement after gadolinium administration (b). (c) Coronal view (FSE 4000/102Ef) and (d) axial view (FSE 3500/102Ef): blood products in varying stages of oxidation are identi®ed in the enlarged right adrenal gland (arrows)

Figure 8 Presacral teratoma. (a) Sagittal view (SE 800/23): a presacral mass is identi®ed with components that are hypointense, isointense, and hyperintense to muscle. (b) Sagittal view (SE 800/16, chemical lipid presaturation): using a lipid-selective presaturation pulse, the formerly hyperintense component is suppressed to the same degree as the adjacent subcutaneous fat; the presence of fat in the lesion histologically characterizes it as a teratoma. (c) Sagittal view (FSE 3000/102Ef): the presence of chemical shift spatial misregistration can also be used to verify the presence of fat in this lesion (arrows); when this artifact is present, lipid-selective presaturation pulses are not necessary to characterize the lesion

PEDIATRIC BODY MRI

speci®cally, on T1-weighted images acute hematomas are isointense to muscle, whereas subacute hematomas are hyperintense to muscle. Neuroblastoma usually demonstrates homogeneous medium signal intensity on T1-weighted images. On T2weighted images, neonatal adrenal hemorrhage may be of high signal intensity or of low signal intensity, depending on the chemical state of the blood products. In rare cases of cystic neuroblastoma with hemorrhage, however, the distinction may be dif®cult. In these cases, evaluation of the liver for the presence of metastasis (low signal intensity on T1-weighted images and high signal intensity on T2-weighted images) may help make the diagnosis of neuroblastoma. Rhabdomyosarcoma is the most common pediatric soft tissue sarcoma and can occur in almost any primary site except the brain. In children, it is most frequently found in the pelvis and genitourinary tract or in the head and neck. In the genitourinary tract, the tumor most frequently arises in the bladder; other common sites are the urethra, prostate, and vagina. Bladder lesions most commonly arise from the submucosa of the trigone or the bladder base and then in®ltrate the bladder wall and adjacent structures including the urethra, prostate, vagina, and uterus. Because of the propensity for local invasion, it can be dif®cult to determine if the primary site was the bladder or the prostate in males or the bladder or the vagina in females. The multiplanar capabilities of MRI make it well suited for demonstrating bladder wall thickening, demarcating the inferior, lateral and posterior extent of the tumor, and detecting distant metastases.6,24 Sacrococcygeal teratomas are the most common tumors of the caudal region in children. These lesions are derived from all three germinal cell layers and have a characteristic MR appearance. Speci®cally, on T1- and T2-weighted images a large presacral mass is identi®ed that contains rounded, wellde®ned areas of different signal intensity. Often one of these locules will contain fat, which can be demonstrated using a lipid-selective presaturation pulse or by simply looking for chemical shift misregistration. (Figure 8). Since these lesions displace the rectum anteriorly, as do all lesions in the presacral space, they can be distinguished from lesions arising in the pelvic cavity. The differential diagnosis for presacral masses in infants also includes anterior meningocele, rectal duplication, neuroblastoma, lymphoma, and lipoma.

4 SUMMARY During the relatively short time that MRI has been applied to the evaluation of pediatric diseases and disorders, it has proven to be a useful tool in the diagnostic evaluation of a variety of entities, many of which we have touched upon in this article. More detailed information about some of these diseases and disorders can be found in the references given at the end of this article, and it is the authors' hope that the reader will learn from these references and, perhaps in time, add to them.

5 RELATED ARTICLES Abdominal MRA; Brain MRS of Infants and Children; Lung and Mediastinum MRI; Male Pelvis Studies Using MRI;

7

MRI of the Female Pelvis; Liver, Pancreas, Spleen, and Kidney MRI.

6

REFERENCES

1. B. D. Fletcher and M. D. Jacobstein, Am. J. Roentgenol., 1986, 146, 941. 2. G. S. Bisset III, J. L. Strife, D. R. Kirks, and W. W. Bailey, Am. J. Roentgenol., 1987, 149, 251. 3. M. L. Pessar, R. L. Soulen, J. S. Kan, S. Kadir, and E. A. Zerhouni, Pediatr. Radiol., 1988, 18, 229. 4. R. B. Dietrich, in `Magnetic Resonance Imaging', 2nd edn, ed. D. D. Stark and W. G. Bradley, Jr, Mosby-Year Book, St Louis, 1992, Vol. 2, Chap. 59. 5. R. B. Dietrich and H. Kangarloo, Radiology, 1986, 159, 215. 6. R. B. Dietrich, in `Magnetic Resonance Imaging of Children', ed. M. D. Cohen and M. K. Edwards, B. C. Decker, Philadelphia, 1990, Chap. 21. 7. A. Daneman and D. J. Alton, Radiol. Clin. North Am., 1991, 29, 351. 8. J. Gambino, B. Caldwell, R. B. Dietrich, I. Walot, and H. Kangarloo, Am. J. Roentgenol., 1992, 158, 383. 9. E. Secaf, H. Hricak, and C. A. Gooding, Pediatr. Radiol., 1994, 24, 291. 10. H. Hricak and M. J. Popovich, in `Magnetic Resonance Imaging of the Body', 2nd edn, ed. C. B. Higgins, H. Hricak, and C. A. Helms, Raven, New York, 1992, Chap. 31. 11. M. C. Mintz, D. I. Thickman, D. Gussman, and H. Y. Kressel, Am. J. Roentgenol., 1987, 148, 287. 12. B. M. Carrington, H. Hricak, R. N. Nuruddin, E. Secaf, R. K. Laros Jr, and E. C. Hill, Radiology, 1990, 176, 715. 13. H. Hricak, Y. C. F. Chang, and S. Thurnher, Radiology, 1988, 169, 169. 14. K. Togashi, K. Nishimura, K. Itoh, I. Fujisawa, Y. Nakano, K. Torizuka, H. Ozasa, and M. Oshima, Radiology, 1987, 162, 675. 15. C. Hugosson, H. Jorulf, and Y. Bakri, Pediatr. Radiol., 1991, 21, 281. 16. R. B. Dietrich and H. Kangarloo, Radiology, 1987, 163, 367. 17. P. J. Fritzsche, H. Hricak, B. A. Kogan, M. L. Winkler, and E. A. Tanagho, Radiology, 1987, 164, 169. 18. A. H. Troughton, J. Waring, and A. Longstaff, Clin. Radiol., 1990, 41, 178. 19. S. J. Pomeranz, N. Altman, J. J. Sheldon, T. A. Tobias, K. P. Soila, L. J. Jakus, and M. Viamonte, Magn. Reson. Imag., 1986, 4, 69. 20. Y. Sato, K. C. Pringle, R. A. Bergman, W. T. C. Yuh, W. L. Smitt, R. T. Soper, and E. A. Franken, Radiology, 1988, 168, 157. 21. K. McHugh, N. E. Dudley, and P. Tarr, Pediatr. Radiol., 1998, 25, 33. 22. A. Vade, H. Reyes, and A. Wilbur, Pediatr. Radiol., 1989, 19, 179. 23. D. D. Kidney, R. B. Dietrich, and A. K. Goyal, Pediatr. Radiol., 1998, 28, 322. 24. R. B. Dietrich and H. Kangarloo, Am. J. Roentgenol., 1986, 146, 251. 25. M. J. Siegel, G. A. Jamroz, H. S. Glazer, and C. L. Abramson, J. Comput. Assist. Tomogr., 1986, 10, 593. 26. M. J. Siegel and G. D. Luker, MRI Clin. North Am., 1996, 4, 599. 27. T. G. Belt, M. D. Cohen, J. A. Smith, D. A. Cory, S. McKenna, and R. Weetman, Am. J. Roentgenol., 1986, 146, 955. 28. H. Kangarloo, R. B. Dietrich, R. M. Ehrlich, M. I. Boechat, and S. A. Feig, Urology, 1986, 28, 203. 29. B. D. Fletcher, S. Y. Kopiwoda, S. E. Strandjord, A. D. Nelson, and S. P. Pickering, Radiology, 1985, 155, 699.

8 PEDIATRIC BODY MRI 30. M. D. Cohen, R. M. Weetman, A. J. Provisor, W. McGuire, S. McKenna, B. Case, A. Siddiqui, D. Mirksh, and I. Seo, Am. J. Roentgenol., 1984, 143, 1241. 31. B. D. Fletcher and S. C. Kaste, Urol. Radiol., 1992, 14, 263.

Biographical Sketches Rosalind B. Dietrich. b 1953. M.B., Ch.B, 1976, University of Manchester School of Medicine, UK. Internship and radiology residency, Cedars-Sinai Medical Center, Los Angeles, CA, USA, 1979±84. Successively, fellow and Assistant Professor of Pediatric Radiology, University of California, Los Angeles, CA, USA, 1984±1990. Professor of Radiology and Director of MRI, University of California, Irvine,

Orange, CA, USA, 1990±1995. Director of Research 1996±present. Approx. 65 publications. Research specialties: applications of MRI in the evaluation of the pediatric brain and body; MRI of brain maturation and white matter diseases. Gerald M. Roth. b 1962. A.B. (biochemical sciences), 1984, Harvard University, USA; M.D., 1988, Columbia University College of Physicians and Surgeons, USA. Internship, Cedars-Sinai Medical Center, Los Angeles, CA, USA. Radiology residency, Hospital of the University of Pennsylvania, Philadelphia, PE, USA, 1989±1993. Successively, MRI Fellow, Faculty, Department of Radiological Sciences, University of California at Irvine, 1993±present. Approx. 5 publications. Research interests: applications of MRI in the chest, abdomen and pelvis; optimization of MR scan protocols.

TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR

Tissue Behavior Measurements Using Phosphorus-31 NMR

1

Pathological processes involving hypoxia and ischemia are particularly amenable to 31P MRS assessment using absolute or relative quantitation of the PCr and Pi peaks. The PCr/Pi ratio has been the most commonly used marker of tissue energy reserve under these conditions. Phosphorus-31 MRS may also provide an indication of the viability of isolated donor organs prior to transplantation. In this article, we discuss the role of in vivo 31P MRS in the examination of tissue bioenergetics in human studies.

Simon D. Taylor-Robinson Hammersmith Hospital, London, UK

and

2

Claude D. Marcus HoÃpital Robert DebreÂ, Reims, France

1 INTRODUCTION Phosphorus-31 MRS provides a noninvasive method of assessment of mobile phosphorus-containing compounds. A typical in vivo 31P MR spectrum contains seven resonances (Figure 1). Phospholipid cell membrane precursors, adenosine monophosphate (AMP) and glycolytic intermediates (sugar phosphates) contribute to the phosphomonoester (PME) peak. Phospholipid cell membrane degradation products and endoplasmic reticulum contribute to the phosphodiester (PDE) peak. Information on tissue bioenergetics can be obtained from inorganic phosphate (Pi), phosphocreatine (PCr) and the three nucleoside triphosphate resonances. A measurement of intracellular pH (pHi) can be calculated from the chemical shift of the Pi peak.

PCr

PDE

31

P MRS AND CELLULAR ENERGY STATUS

The energy metabolism of each cell is dependent on the synthesis and utilization of compounds which contain highenergy phosphate bonds such as ATP and PCr.1 ATP is present in all cells, but PCr is limited to those tissues containing creatine and the enzyme creatine kinase (CK), such as skeletal muscle and brain. ATP has a pivotal role in cellular bioenergetics. The demand for ATP is usually reasonably constant under normal resting conditions. The hydrolysis of ATP to ADP (adenosine 5'-diphosphate) and Pi releases potential energy from high-energy phosphate bonds for all activities involved in maintaining intracellular homeostasis and the specialized functions which may be unique to each cell type. Phosphocreatine acts as an energy reservoir in tissues such as muscle and brain. The enzyme CK splits PCr to provide an energy source for ATP resynthesis (Figure 2). Oxidative phosphorylation, which involves an electron transport chain in the mitochondrial membrane, is the process which provides most of the ATP for each cell under conditions of adequate oxygen supply. In a range of situations the requirements for ATP cannot be met by oxidative phosphorylation in the mitochondria. For example, oxidative phosphorylation is impaired in hypoxia and may be inadequate in normal exercising muscle. The shortfall in ATP production is met by glycolysis in the cytoplasm. Lactic acid may accumulate as a consequence of this process. Phosphocreatine is utilized under these conditions and as PCr falls, Pi increases, but any reduction in ATP is minimized because of the buffering effect of CK. Only when the PCr pool is completely consumed do the tissue ATP levels fall appreciably, leading to a rise in both ADP and Pi.

Oxidative phosphorylation and/or glycolysis

Pi g

PME

a Cr

bATP

ATP

CK

+20

+10

0

–10

–20

ppm

Figure 1 An unlocalized 31P MR spectrum from the head of a healthy adult volunteer. There are seven resonances. PME, phosphomonoester; Pi, inorganic phosphate; PDE, phosphodiester; PCr, phosphocreatine; ATP, adenosine triphosphate

PCr (+H+)

ADP

Pi

Figure 2 The interrelationship between ATP and PCr. Cr, creatine; CK, creatine kinase; ADP, adenosine diphosphate

2 TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR Mitochondrial oxidative metabolism may also be affected by changes in pHi,2 but the rate of ATP biosynthesis remains constant under normal conditions. The concentration of ADP is a major factor in the rate of ATP production.2,3 The majority of ADP is not directly detectable using NMR methods because of tissue binding. The ability of a particular tissue to synthesize ATP may be calculated from the phosphorylation potential, given by equation (1) ‰ATPŠ=‰ADPŠ‰PiŠ

…1†

A further indication of cellular energy status is given by equation (2): ‰PCrŠ=‰CrŠ‰PiŠ

…2†

Absolute or relative concentrations of tissue PCr, Pi, and ATP can be obtained using 31P MRS. The rate of ATP biosynthesis, the phosphorylation potential, ADP concentrations, and the kinetics of CK may be calculated from 31P MRS measurements.2,3 The PCr/Pi and PCr/ATP ratios are commonly used MR indices of cellular energy status or `bioenergetic reserve', as they re¯ect equations (1) and (2). The oxidative phosphorylation pathway can be assessed using the PCr/Pi ratio. Glycolytic intermediates contribute to the PME peak and therefore an indirect assessment of glycolysis may be obtained from quantitation of this resonance. This is of particular importance in assessment of glycolytic disorders in muscle and in dynamic studies of liver metabolism. 3

31

P MRS AND INTRACELLULAR pH

Phosphorus-31 MRS can be used to measure intracellular pH from the chemical shift of the Pi peak with reference to the PCr resonance in tissues such as brain and muscle where PCr is present. In tissues such as liver where PCr is absent the reference used is ATP. The MR signal from Pi is thought not to represent the total intracellular levels of Pi. It is unclear why the remainder is not detected, but it may be bound in the mitochondria. It is not known whether there are pH differences between the cytoplasm and the mitochondria. Split Pi resonances representing intracellular compartmentation have not been observed in human 31P MRS in vivo. Different body tissues are more susceptible to ischemia and hypoxia than others. In normal exercising muscle, anaerobic glycolysis may take place with the accumulation of lactic acid as a normal sequence of events. The large PCr reservoir in muscle ensures an energy source for these anaerobic reactions. The accumulation of lactate in the brain is more likely to be of pathological consequence because cerebral function is particularly sensitive to hypoxic insults. The measurement of intracellular acidosis may be used as a marker of hypoxia± ischemia and cellular dysfunction in conditions such as stroke or birth asphyxia. Under these circumstances, the PCr/Pi ratio is reduced. Lactate accumulation leads to a reduction in pHi and a change in the chemical shift of Pi. Intracellular pH may change with time: for example, an intracellular alkalosis can develop in ischemic brain tissue of stroke patients over an extended time period.4 The underlying mechanisms behind such pH changes are not known. However, the MR measurement of pHi may be used to discriminate between diseased and

healthy tissue in combination with indices of bioenergetic reserve such as PCr/Pi. 4

31

P MRS AND PHOSPHOLIPID METABOLISM

The PME and PDE peaks are multicomponent. Phosphoethanolamine (PE) and phosphocholine are cell membrane precursors and contribute to the PME peak with signal from glycolytic intermediates and AMP. Glycerophosphorylethanolamine (GPE) and glycerophosphorylcholine (GPC), which are cell membrane degradation products, contribute to the PDE resonance. Phosphoenolpyruvate and endoplasmic reticulum form other contributory factors. The relative contributions of these compounds to the PME and PDE peaks may change with disease. In situations where there is rapid cell turnover, the PME resonance may be elevated due to an increase in PE and phosphocholine. Similarly, under conditions of rapid cell death, after tumor embolization or chemotherapy, for example, an increased contribution of GPE and GPC to the PDE peak may be expected. Phosphorus±proton decoupling can be used to resolve further these resonances in vivo. 5

31

P MRS AND TRANSPLANT ORGAN VIABILITY

Organ transplantation is a steadily expanding surgical ®eld. Increased patient survival rates have been achieved because of improved operative techniques, anti-rejection chemotherapy, and methods for harvesting and preserving donor organs. The success of any transplant procedure is dependent on the quality of the donor graft and this is a re¯ection of organ storage methods. The use of more physiological preservation ¯uids has led to increased storage times, allowing donor organs to be transported between transplant centers. Despite these advances, tissue damage, caused by cold preservation, is an important factor in patient morbidity and mortality. Phosphorus MRS provides a noninvasive assessment of the viability of the isolated donor organ prior to transplantion. The standard indices of bioenergy reserve such as PCr/ATP and PCr/Pi ratios may be appropriate markers in heart transplantation, but the 31 P MR spectra from healthy liver and kidney contain no appreciable PCr. Therefore, other indices such as the PME/Pi ratio have to be employed. Adenosine 5'-triphosphate begins to degenerate to ADP and Pi immediately each organ has been harvested. With time, ADP further degenerates to AMP which contributes to the PME peak. The PME/Pi ratio may re¯ect the ability of the isolated organ to rephosphorylate AMP to ATP on transplantation. Speci®c transplant studies are considered later in this chapter. 6

CLINICAL APPLICATIONS

The development of whole body magnets has allowed clinical 31P MRS studies to be undertaken in patients with a variety of pathological processes, facilitating comparisons with healthy volunteers and offering the possibility of disease monitoring in response to treatment. MR measurements of bioenergetic reserve such as the PCr/Pi ratio have been proposed as predictors of outcome in hypoxic and/or

TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR

ischemic conditions such as birth asphyxia. A review of the role of 31P MRS in some of the major disease processes follows. 6.1 6.1.1

31

P MRS and the Adult Brain Stroke

Stroke is the most common adult neurological condition and is a prominent cause of mortality in developed countries. Early diagnosis of ischemia facilitates more appropriate treatment, and delay may result in an irreversible loss of neuronal function. Ischemia and the consequent tissue hypoxia result in a depletion in PCr, ATP levels being maintained initially. The acute stages of stroke are characterized by a decreased PCr/Pi ratio in the 31P MR spectrum and an intracellular acidosis.5 These changes may be detectable before MRI changes become evident.5 A combination of 31P MRS and MRI may aid early diagnosis, and help to monitor the brain's response to treatment. Persistent cerebral ischemia results in irreversible cell damage and neuronal death. A reduction in phosphorus signal has been seen in patients with chronic stroke, consistent with a reduction in viable cells in the area of infarction.4 The pHi has been noted to change with time, resulting in a rebound intracellular alkalosis which may persist.4 The reasons underlying this change remain unclear. 6.1.2

Transient Ischemia

A transient ischemic attack is de®ned as a reversible neurological de®cit that lasts for 24 h or less. The diagnosis is therefore retrospective and treatment is aimed at preventing recurrence. One study of two patients suggested that the total phosphorus signal was reduced in the absence of MRI changes.4 This may be due to ischemia, but remains dif®cult to explain. 6.1.3

Epilepsy

Temporal lobe epilepsy may be unresponsive to standard antiepileptic therapy and a small number of patients require surgical resection of the epileptogenic focus. The de®nition of the pathological area needs careful preoperative planning. MRI studies have been used to obtain hippocampal volumes6 but the role of 31P MRS has been limited. A reduced PME, probably as a result of underlying hippocampal sclerosis, has been noted in some studies. An increased Pi and an unexplained intracellular alkalosis have been noted in most studies,4,7 all of which have involved relatively small numbers of patients. 6.1.4

Alzheimer's Disease

The results of 31P MRS studies have been disappointing. Bottomley and colleagues8 found no changes in either metabolite ratios or absolute concentrations of metabolites. 6.1.5

Multiple Sclerosis

This condition is common in temperate climates and is characterized by episodes of focal neurological de®cit which relapse and remit over a period of many years. The classic histological lesions are plaques of demyelination. MRI has

3

revolutionized the diagnosis of this condition, but the results of P MRS studies are less clear-cut. In one study9 there was a decrease in the PCr/ATP ratio in some patients with active disease, whereas in another study10 there was an increase in the PCr/ATP ratio with disease activity. 31

6.1.6

Other Cerebral Conditions

Decreased PCr/Pi has been observed in the cerebral 31P MR spectra from patients with migraine and mitochondrial cytopathies. This suggests an alteration in bioenergetic reserve. Chronic hepatic encephalopathy is de®ned as the neuropsychiatric impairment observed in patients with cirrhosis of the liver. The results of some 31P MRS studies have suggested altered brain energy metabolism in these patients (see also Systemically Induced Encephalopathies: Newer Clinical Applications of MRS). 6.2

31

P MRS and Pediatric Brain Studies

The normal 31P MR spectrum from a healthy neonatal brain is signi®cantly different from adult spectra. The PME signal is much larger in the neonate and this varies with gestational age. Azzopardi and colleagues11 found that the PME is smaller and the PDE larger in the healthy full-term infant than the healthy preterm infant, related most probably to the changes in membrane lipids with myelin formation. The PCr/Pi ratio increases with gestational age, indicating an increased phosphorylation potential with brain development. The measured intracellular pH appears not to vary with the gestational age of healthy newborn infants. 6.2.1

Hypoxic±Ischemic Encephalopathy

Birth asphyxia of the newborn infant has been extensively investigated using 31P MRS. This condition is almost unique because there are well-de®ned MR indices of prognosis. Reduced PCr/Pi ratio has been correlated with outcome.12 Spectra obtained in the initial 24 h after birth may be normal, but a fall in PCr and a rise in Pi (reduced PCr/Pi ratio) may develop over the ensuing hours and days (Figure 3). This re¯ects defective oxidative phosphorylation as a result of birth trauma. Intracellular pH tends to rise in a delayed response to the hypoxic±ischemic insult and there may also be a reduction in ATP levels. The metabolite ratios tend to return to normal within 2 weeks in neonates who recover. The reduction in PCr/ Pi ratio is proportional to the degree of subsequent neurodevelopmental impairment and to reduced cranial growth in the ®rst year of life.12 In the severest cases, where the neonates subsequently die, the PCr and ATP may be almost undetectable. The Pi often rises out of proportion to the reduction in PCr. This delayed or secondary response to ischemic injury is poorly understood, but may be partly due to the toxic effects of neurotransmitters such as glutamate, which may induce mitochondrial membrane disruption through the generation of free radicals.13 Phosphorus-31 MR spectroscopy may be utilized to monitor the effectiveness of treatment designed to prevent this secondary energy failure. The development of suitable therapeutic regimens is an area of current and future research. However, the PCr/Pi ratio is already being used as a predictor of patient outcome and may give the pediatrician insight into planning future management decisions.

4 TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR

PME

cate this matter. The pHi of normal human myocardium is pH 7.15 ‹ 0.03.18

Pi

6.3.2

PDE

PCr

gATP aATP

bATP

Spectral abnormalities in human myocardium mainly re¯ect ischemic conditions.19 The PCr/ATP ratio decreases signi®cantly during hand grip exercise20 in patients with a high degree of coronary artery stenosis. This ratio is also reduced in advanced stages of heart failure, in left ventricular hypertrophy, or dilated cardiomyopathy.21 An increase in normal PCr/ATP ratio has been shown during treatment for heart failure.17 Apart from blood contamination, an increased PDE/ATP ratio might re¯ect an accumulation of cell membrane degradation products in patients with a decreased left ventricular ejection fraction.22 6.3.3

+10

0

–10

–20

ppm

Figure 3 An unlocalized 31P MR spectrum from the head of a birth asphyxiated neonate. The PCr peak is reduced and the Pi peak is elevated

6.3 Myocardial Metabolism Heart muscle is well supplied with oxygen and mitochondria. It relies on glycolysis to a much smaller extent than skeletal muscle, which has a higher concentration of glycolytic enzyme and fewer mitochondria. The regulation of cardiac metabolism and its relation to mechanical function has been widely studied in animals and humans14,15 (see also Cardiovascular NMR to Study Function and NMR Spectroscopy of the Human Heart). 6.3.1

Measurements in Normal Myocardium

Some of the major problems encountered in cardiac studies are related to signal contamination of the myocardium from the chest wall and from blood circulating through the cardiac chambers. The chest wall contains about four times more PCr than the normal myocardium. Signal from blood usually includes appreciable amounts of ATP and PDE, and an intense 2,3-diphosphoglycerate (2,3-DPG) resonance, which may obscure the Pi and PME resonances. Under such circumstances the pHi and the PCr/Pi ratio cannot be measured.16 Inorganic phosphate and pHi measurements still remain dif®cult in humans. Localization techniques may be used to minimize signal contamination from chest wall or blood. Correction for residual blood contamination and saturation effects may also be made. The PCr/ATP ratio is the most frequently used index of bioenergy reserve in cardiac studies. Results from human and animal studies are comparable.17 This ratio remains relatively constant during the cardiac cycle and in exercise. It is also highly reproducible. The PDE/ATP ratio is dif®cult to measure accurately because of interference from the 2,3-DPG signal. The available data show considerable biological variation in human studies. Different examination techniques may compli-

Measurements in Myocardial Diseases

Cardiac Transplantation

Advances in operative technique and in antirejection therapy have led to cardiac transplantation being used as a viable treatment for patients with end-stage heart disease. Endomyocardial biopsies with histological grading are the gold standard for detection of rejection. No reliable noninvasive alternative is available at present. Phosphorus-31 MRS has been studied in this context. Animal models of cardiac transplantation have demonstrated an early decrease in PCr/ATP and PCr/Pi ratios.23 A parallel rise in PDE/ATP was found in two studies which preceded the onset of histologically detectable rejection by 1 or 2 days.23,24 The increase in the PDE peak measurements may be due to immunological or reperfusion injury. The abnormalities of highenergy phosphate during acute, severe rejection are reversible with antirejection therapy.25 In human studies no correlation was found between the decrease in the PCr/ATP ratio measurements and the biopsy grading. Severe rejection may therefore not be distinguishable from mild or moderate rejection by 31P MRS alone.26 The discrepancy between animal and human studies may be explained by the different time courses for MRS examinations and the rather more heterogeneous human population and study conditions. 6.4

31

P MRS and the Kidney

Spectral localization is required because there are considerable regional differences in renal function and metabolism. The cortex is dependent on oxidative phosphorylation and the inner medulla is more reliant on glycolysis for energy requirements. Localization has proved dif®cult owing to the anatomical position and the displacement on respiratory excursion. No large-scale studies have been undertaken in the renal failure patients, but transplanted kidneys are positioned in a relatively super®cial, static position in the anterior pelvis and are therefore much more amenable to MRS examination. The normal 31P MR spectrum does not contain a large PCr signal arising from the kidney. There is a relatively large PME peak compared with liver and muscle. Urinary Pi may contribute to the PDE resonance, but this only becomes signi®cant when the collecting ducts are distended through obstructive disease.27

TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR

6.4.1

The Isolated Donor Kidney

Phosphorus-31 MRS can be used to assess the viability of isolated donor organs, kept in physiological preservation ¯uid, prior to transplantation. An alternative index of tissue energy reserve has to be used because the renal PCr pool is small. Adenosine 5'-triphosphate degrades to ADP and Pi fairly rapidly, but provided there is suf®cient cellular AMP, rephosphorylation to ATP should be possible. Adenosine monophosphate resonates in the PME region of the spectrum and therefore the PME/Pi ratio has been used as an indicator of viability. Human studies have positively correlated the PME/Pi ratio with subsequent postoperative renal function.28 The best renal function was observed in kidneys where ATP was seen in the donor organ spectra. 6.4.2

The Transplanted Kidney

Phosphorus-31 MRS may be used to investigate renal failure, graft rejection, or organ viability postoperatively. The PME/ATP ratio may be slightly higher in the transplanted kidney than in healthy volunteers. Animal studies have shown an elevated Pi/ATP ratio with renal failure or ischemia due to poor graft viability. Organ rejection or renal dysfunction due to cyclosporin toxicity may produce a similar picture.27 Such changes are nonspeci®c, but perhaps may be used to monitor the effectiveness of antirejection chemotherapy. 6.5

31

P MRS and Liver

The position of the liver renders it much more amenable to MRS investigation. The MR spectrum contains no PCr signal arising from the liver itself. Most studies have concentrated on changes in PME/ATP and PDE/ATP ratios with disease, re¯ecting changes in hepatic phospholipid and carbohydrate metabolism. The Pi/ATP ratio remains relatively constant. Clinical interest has focused on spectral changes under conditions such as alcoholic liver disease, cirrhosis, primary and secondary liver tumors, and the dynamic changes in metabolites following infusions of fructose, alanine, and alcohol. This subject is discussed in detail in another article (see In Vivo Hepatic MRS of Humans). Tissue behavior measurements have been used to measure the effectiveness of chemoembolization therapy for hepatic cancers. The resulting ischemia can be assessed using PME/Pi or PME/ATP ratios, both of which fall after successful treatment.29 The PME/Pi ratio has been used to assess the viability of isolated donor organs prior to transplantation in a similar fashion to the renal studies already mentioned.30 6.6

31

P MRS in Tumors

The dif®culty with reporting metabolic changes in tumors, as re¯ected by 31P MRS, is that the results are variable, depending on the size, location, and precise histology. In human tumors, the typical metabolite characteristics include lower PCr and higher PME and PDE levels.31 Variations in the PME and PDE levels may re¯ect different rates of membrane synthesis, catabolism, or metabolic turnover. (See also Spectroscopic Studies of Animal Tumor Models; In Vivo Hepatic MRS of Humans; and Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy).

5

The PCr/Pi ratio is reduced in most high-grade tumors of the brain, re¯ecting an increased demand for ATP in rapid growth. The measured pHi is often found to be alkaline under these circumstances.32 The reasons for this intracellular alkalinization remain unclear. Treatment such as radiotherapy or chemotherapy may induce hypoxia, ischemia, or necrosis in tumors. These changes may lead to an increase in Pi33 and acidosis.34 The early decrease in PME seems to be a sensitive indicator of changes in the phospholipid metabolism of the cell membrane.31 However, because of the wide interindividual variability, changes should be related to initial measurements performed in patients before the start of therapy. 31

P MRS and Muscle

6.7

Phosphorus-31 MRS allows investigation of muscle bioenergetics at rest, during exercise, and during the recovery period. Normal muscle has a low metabolic rate and a high energy capacity at rest, re¯ected in a high PCr/Pi ratio. During exercise the ATP levels are maintained at the expense of the PCr pool. Anaerobic glycolysis results in lactate accumulation and a reduced pHi. In the recovery phase PCr regenerates and the PCr/Pi ratio rises to preexercise levels. The observed pHi also returns to normal, because oxidative phosphorylation then provides the bulk of the energy requirements. Phosphorus-31 MRS may be utilized as a screening test in patients with exercise intolerance. The various muscle pathologies are discussed in a separate article (see Peripheral Muscle Metabolism Studied by MRS). We brie¯y consider some of the major conditions where tissue behavior measurements are important. 6.7.1

Mitochondrial Myopathies

Mitochondrial myopathies, where there is defective oxidative phosphorylation, are characterized by a reduced resting energy state and a decreased PCr/Pi ratio, which falls rapidly to very low levels on exercise.35 The capacity for exercise is reduced and the resynthesis of PCr after exercise is impaired, because this is dependent on mitochondrial function. Therefore there is a prolonged recovery phase before the PCr/Pi ratio reaches resting levels. Intracellular acidosis is often observed, which may become marked on exercise. 6.7.2

Glycolytic Disorders

Speci®c enzyme de®ciencies in the glycolytic pathway block the production of lactic acid and therefore the pHi does not fall in anaerobic exercise as it does in normal muscle. This failure of intracellular acidi®cation is characteristic.35 During exercise an accumulation of glycolytic intermediates can be measured indirectly from the 31P MR spectrum, because sugar phosphates contribute to the PME resonance.

7

CONCLUSIONS

In vivo 31P MRS provides a noninvasive assessment of tissue bioenergetics and phospholipid metabolism. Comparisons may be made between healthy and diseased tissue. Measurements of the PCr/Pi ratio and pHi may provide insights into

6 TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR pathogenic mechanisms and in the future, the ef®cacy of therapeutic intervention in cardiac and cerebral hypoxic±ischemic states. Energy reserves at rest and during exercise may be monitored in muscle disease, and the PME/Pi ratio is a noninvasive indicator of viability in isolated donor organs prior to transplantation. In vivo 31P MRS remains predominantly a research tool, but it has proved to be clinically useful in birthasphyxiated babies and in studies of muscle disease.

8 RELATED ARTICLES Brain MRS of Human Subjects; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Cardiovascular NMR to Study Function; In Vivo Hepatic MRS of Humans; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; NMR Spectroscopy of the Human Heart; Peripheral Muscle Metabolism Studied by MRS; Quantitation in In Vivo MRS; Spectroscopic Studies of Animal Tumor Models; Systemically Induced Encephalopathies: Newer Clinical Applications of MRS; Whole Body Studies: Impact of MRS.

9 REFERENCES 1. E. E. Conn, P. K. Stumpf, G. Bruening, and R. H. Doi, `Outlines of Biochemistry', 5th edn., Wiley, New York, 1987. 2. S. Nioka, B. Chance, M. Hilberman, H. V. Subramanian, J. S. Leigh, Jr., R. L. Veech, and R. E. Forster, J. Appl. Physiol., 1987, 62, 2094. 3. B. Chance, J. S. Leigh, Jr., J. Kent, K. McCully, S. Nioka, B. J. Clark, J. M. Maris, and T. Graham, Proc. Natl. Acad. Sci., U.S.A., 1986, 83, 9458. 4. J. W. Hugg, G. B. Matson, D. B. Twieg, A. A. Maudsley, D. Sappey-Marinier, and M. W. Weiner, Magn. Reson. Imaging, 1992, 10, 227. 5. J. Kucharczyk, M. Moseley, J. Kurhanewicz, and D. Norman. Invest. Radiol., 1989, 24, 951. 6. C. R. Jack, F. W. Sharbrough, C. K. Twomey, G. D. Cascino, K. A. Hirschorn, W. R. Marsh, A. R. Zinsmeister, and B. Scheithauer. Radiology, 1990, 175, 423. 7. J. W. Hugg, K. D. Laxxer, G. B. Matson, A. A. Maudsley, C. A. Husted, and M. W. Weiner. Neurology, 1992, 42, 2011. 8. D. G. M. Murphy, P. A. Bottomley, J. A. Salerno, C. DeCarli, M. J. Mentis, C. L. Grady, D. Teichberg, K. R. Giacometti, J. M. Rosenberg, C. J. Hardy, M. B. Schapiro, S. I. Rapoport, J. R. Alger, and B. Horwitz. Arch. Gen. Psychiatry, 1993, 50, 341. 9. T. A. D. Cadoux-Hudson, A. Kermode, B. Rajagopalan, D. Taylor, A. J. Thompson, I. E. C. Ormerod, W. I. McDonald, and G. K. Radda. J. Neurol. Neurosurg. Psychiatry, 1991, 54, 1004. 10. J. M. Minderhoud, E. L. Mooyaart, R. L. Kamman, A. W. Teelken, M. C. Hoogstraten, L. M. Vencken, E. J's. Gravenmade, and W. van den Burg. Arch. Neurol., 1992, 49, 161. 11. D. Azzopardi, J. S. Wyatt, P. A. Hamilton, E. B. Cady, D. T. Delpy, P. L. Hope, and E. O. R. Reynolds. Pediatr. Res., 1989, 25, 440. 12. D. Azzopardi, J. S. Wyatt, E. B. Cady, D. T. Delpy, J. Baudin, A. L. Stewart, P. L. Hope, P. A. Hamilton, and E. O. R. Reynolds. Pediatr. Res., 1989, 25, 445. 13. J. S. Wyatt, A. D. Edwards, D. Azzopardi, and E. O. R. Reynolds. Arch. Dis. Child., 1989, 64, 953.

14. C. B. Higgins, M. Saeed, M. Wendland, and W. M. Chew. Invest. Radiol., 1989, 24, 962. 15. S. Schaefer. Am. J. Cardiol., 1990, 66, 45F. 16. A. de Roos, J. Doornbos, S. Rebergen, P. van Rugge, P. Pattynama, and E. E. van der Wall. J. Radiol., 1992, 14, 97. 17. S. Neubauer, T. Krahe, R. Schindler, M. Horn, H. Hillenbrand, C. Entzeroth, H. Mader, E. P. Kromer, G. A. J. Riegger, K. Lackner, and G. Ertl. Circulation, 1992, 86, 1810. 18. A. de Roos, J. Doornbos, P. R. Luyten, L. J. M. P. Oosterwaal, E. E. van der Wall and J. A. den Hollander. J. Magn. Reson. Imaging, 1992, 2, 711. 19. S. Schaefer, G. G. Schwartz, J. R. Gober, B. Massie, and M. W. Weiner. Invest. Radiol., 1989, 24, 969. 20. R. G. Weiss, P. A. Bottomley, C. J. Hardy, and G. Gerstenblith. N. Engl. J. Med., 1990, 323, 1593. 21. H. Sakuma, K. Takeda, T. Tagami, T. Nakagawa, S. Okamoto, T. Konishi, and T. Nakano. Am. Heart J., 1993, 125, 1323. 22. W. Auffermann, W. M. Chew, C. L. Wolfe, N. J. Tavares, W. W. Parmley, R. C. Semelka, T. Donnelly, K. Chatterjee, and C. B. Higgins. Radiology, 1991, 179, 253. 23. R. C. Canby, W. T. Evanochko, L. V. Barrett, J. K. Kirklin, D. C. McGif®n, T. T. Sakai, M. E. Brown, R. E. Foster, R. C. Reeves, and G. M. Pohost. J. Am. Coll. Cardiol., 1987, 9, 1067. 24. C. D. Fraser, Jr., V. P. Chacko, W. E. Jacobus, R. L. Soulen, G. M. Hutchins, B. A. Reitz, and W. A. Baumgartner. Transplantation, 1988, 46, 346. 25. P. McNally, N. Mistry, J. Idle, J. Walls, and J. Freehally. Transplantation, 1989, 48, 1068. 26. P. A. Bottomley, R. G. Weiss, C. J. Hardy, and W. A. Baumgartner. Radiology, 1991, 181, 67. 27. M. D. Boska, D. J. Meyerhoff, D. B. Twieg, G. S. Karczmar, G. B. Matson, and M. W. Weiner. Kidney Int., 1990, 38, 294. 28. P. N. Bretan, Jr., N. Baldwin, A. C. Novick, A. Majors, K. Easley, T. Ng, N. Stowe, P. Rehm, S. B. Streem, and D. R. Steinmuller. Transplantation, 1989, 48, 48. 29. D. J. Meyerhoff, G. S. Karczmar, F. Valone, A. Venook, G. B. Matson and M. W. Weiner. Invest. Radiol., 1992, 27, 456. 30. R. F. E. Wolf, R. L. Kamman, E. L. Mooyaart, E. B. Haagsma, R. P. Bleichrodt, and M. J. H. Slooff. Transplantation, 1993, 55, 949. 31. W. Negendank. NMR Biomed., 1992, 5, 303. 32. W.-D. Weiss, W. Heindel, K. Herholz, J. Rudolf, J. Burke, J. Jeske, and G. Friedman. J. Nucl. Med., 1990, 31, 302. 33. A. Schilling, B. Gewiese, G. Berger, J. Boese-Landgraf, F. Fobbe, D. Stiller, V. Gallkowski, and K. J. Wolf. Radiology, 1992, 182, 887. 34. M. W. Dewhirst, H. D. Sostman, K. A. Leopold, H. C. Charles, D. Moore, R. A. Burn, J. A. Tucker, J. M. Harrelson, and J. R. Oleson. Radiology, 1990, 174, 847. 35. Z. Argov and W. J. Bank. Ann. Neurol., 1991, 30, 90.

Biographical Sketches Simon. D. Taylor-Robinson. b 1960. M.B., B.S., 1984, London; M.R.C.P., 1989, (UK) M.D., 1996, London. Introduced to NMR by G. M. Bydder. Honorary lecturer, medicine, Royal Free Hospital and School of Medicine, London, 1992±94. Research fellow and Honorary Senior Registrar at NMR Unit, Royal Postgraduate Medical School, Hammersmith Hospital, London, 1992±94. Senior Registrar, Division of Gastroenterology, Department of Medicine, Royal Postgraduate Medical School, Hammersmith Hospital, London, 1994±1996. Senior Lecturer and Honorary Consultant Physician, Department of Medicine, Imperial College Medical School, Hammersmith Hospital, London 1996±present. Approx. 60 publications. Research interests include clinical application of NMR to liver disease and transplant organ viability.

TISSUE BEHAVIOR MEASUREMENTS USING PHOSPHORUS-31 NMR Claude D. Marcus. b 1957. M.D., 1986, Reims, France. Chef de clinique assistant des hopitaux, 1986±1990; praticien hospitalier temps plein centre hospitalier universitaire de Reims, France, 1990±present. Introduced to NMR by Professor D. Doyon (CHU Kremlin-Biceˆtre,

7

France). Titulaire du Diplome d'Universite d'imagerie par resonance magneÂtique, 1988. Research interests include clinical applications of NMR to brain and breast diseases.

IMAGING AND SPECTROSCOPY OF MUSCLE

Imaging and Spectroscopy of Muscle Chris Boesch and Roland Kreis University of Bern, Switzerland

1 INTRODUCTION Examinations of extremities and skeletal muscle have been among the ®rst applications of NMR in the human body. At the advent of in vivo NMR, magnets were available with suf®ciently high ®eld strength for magnetic resonance spectroscopy (MRS) but their limited bore size allowed only small animals or human extremities to be examined. The development of surface coils (see Spatial Localization Techniques for Human MRS) allowed localized acquisition of NMR signalsÐa technique that was mainly used for 31P MRS and to a lesser extent for 13C MRS. For a while, classical 31P MRS experiments (see Peripheral Muscle Metabolism Studied by MRS, Tissue Behavior Measurements Using Phosphorus-31 NMR) were the major application of MRS and led to a substantial contribution to knowledge in muscle physiology. In the following years, the development of whole-body magnets introduced NMR applications to clinical medicine (see Whole Body Studies: Impact of MRS). MRI of extremities, muscle tissue, and joints became a routine diagnostic tool (see ESR Probes as Field Detectors in MRI, MRI of Musculoskeletal Neoplasms, Skeletal Muscle Evaluated by MRI). The combination of MRI and MRS in whole-body magnets led not only to improved anatomical localization de®ned by gradients but also to better description of the sensitive volume by MRI in the same magnet. The wider bore of these systems allowed for examinations of larger muscle groups, for example those of the thigh. Gradient-based volume selection and improved localization sequences particularly promoted the development of 1H MRS methods. The number of NMR applications in perfused muscles and animals is huge but will not be covered in this chapter, which focuses on human skeletal muscle. At present, publications on 31 P MRS in small-bore and whole-body systems still represent a considerable portion of the physiological applications of MR (see also Peripheral Muscle Metabolism Studied by MRS). An increasing number of new applications, however, combine and use other MR techniques and other nuclei. Some of these applications will be described in this chapter.

2 STRUCTURAL ORDER AND COMPARTMENTATION IN MUSCLE TISSUE Skeletal muscle is a biological structure that is highly organized at different spatial levels. (a) Skeletal muscles are composed of various types of muscle ®ber with inherently different composition and metabolism. (b) Bulk fat along fasciae around muscles forms macroscopic plates that run almost parallel with the axes of extremities. (c) Muscle ®bers may

1

extend parallel to the muscle (`fusiform' arrangement) or with a certain `pennation' angle between muscle and ®bers (`unipennate' with one major direction or `bipennate' if two different types of ®ber orientation exist that are attached to one tendon). (d) Myo®brils form the well-known striated structures (sarcomeres) including actin and myosin molecules, which are spatially organized on the microscopic level. (e) Membranes and cell organelles such as mitochondria separate a speci®c metabolite from the same metabolite in the cytoplasm, leading to different pools and transport processes. Structural order and compartmentation in biological tissue can in¯uence MR signals such that measurable parameters or artifacts are generated, depending on the way the experiment and observer are using or neglecting that information. Several effects, which will be discussed below, can be attributed to structure or compartmentation. 1. the relaxation times of muscle, tendons, and cartilage, depend on ®ber orientation and type;1±3 2. the muscle- or subject-speci®c metabolite content, energy kinetics, and recruitment pattern based on distributions of ®ber types;3±7 3. the isotropic and ordered compartments, as distinguished by 23Na double quantum ®ltered (DQF) MR spectra;8,9 4. susceptibility effects, which are different for extramyocellular (bulk) (EMCL) fat and for intramyocellular lipids (spherical droplets) (IMCL);10,11 5. magnetization transfer (MT) of the invisible ®xed pool of molecules to the moving and visible pool;1 6. restricted diffusion measurable by specialized MRI and MRS sequences;12±16 7. the dipolar coupling effects of creatine (Cr) and/or phosphocreatine (PCr) resonances owing to anisotropic motional averaging;17,18 8. differences in the NMR visibility of Cr/PCr and other metabolites as a result of compartmentation and/or limited motional freedom.19±21 MR is a unique tool with which to investigate order and compartmentation in biological tissues. Even if these are not the major target of a speci®c examination, their effects should always be kept in mind when interpreting images and spectra.

3

3.1

USE OF MRI FOR HUMAN MUSCLE PHYSIOLOGY MRI for Volume Localization

The combination of MRI and MRS improves the anatomical localization of spectra considerably. MRS examinations in small-bore systems often have to rely on palpation and approximate positioning of the surface coil. With appropriate imaging sequences, it was possible to show that palpation incorrectly identi®ed ¯exor muscle margins by more than 15 mm in 50% of attempts.22 In principle, MRI makes it possible to place the region of interest with an accuracy of about 1 mm. Even if images are acquired prior to MRS, it may not be feasible to place the sensitive volume of a surface coil with the same precision. However, the subsequent use of gradient-based localization schemes and a proper ®xation of the examined extremity allow for a volume selection within adequate

2 IMAGING AND SPECTROSCOPY OF MUSCLE accuracy. Since the ®ber orientation, composition, and activation are muscle speci®c and can vary even within one muscle,23 combined MRI/MRS systems are, therefore, necessary to obtain MR parameters from one speci®c muscle type, as shown in several examples below.

3.2 Activity-dependent MRI The amount and distribution of water in skeletal muscle is changed by muscular activity.2,23±27 Subsequent variations in relaxation times make MR signal intensity highly sensitive to changes in the water distribution, oxygenation, pH, and other factors. Fleckenstein et al.28 suggested imaging parameters with which it was possible to detect and document muscular activity (see also Skeletal Muscle Evaluated by MRI). This technique can be used to study intersubject variations of normal anatomy and muscular activation. Ergometers to be used in an MR system have to be redesigned to ®t into the restricted space of the patient bore. The technique described can help to document the activity imposed by such an ergometer. While it is clear that activity-dependent MRI is a valuable and practical tool to investigate muscular activity, it is not so clear which changes in MR parameters lead to this effect. It seems that water in¯ow and pH changes contribute only partially to the observed increase in T2 relaxation.25,27

3.3 Muscle Volume Determination using MRI Muscle mass is a variable parameter and is dependent on training, ageing, temporary immobilization, disease, and other factors. Knowing muscle mass is, therefore, crucial for an estimation of training effects and also for calculations of total body content of metabolites and components of the muscular cell. Many localized observations of physiological parameters, such as biopsies, need additional data on the absolute volume of the muscle, since changes in concentration could be a result of increased muscle mass (i.e. simple dilution with unchanged total content) or an effective increase in total content. It is obvious that MRI with its excellent soft tissue contrast is predestined to be used for in vivo morphometric measurements.4,29±33 However, all approaches have speci®c advantages and disadvantages, for example three-dimensional versus multislice techniques, nonlinearity of the spatial representation, different ways of image analysis, and partial volume effects. The image acquisition can be done either by three-dimensional sequences or using sequential two-dimensional data (i.e. multislice techniques) (Figure 1). As long as discrete slices are evaluated, the so-called Cavalieri principle needs to be respected.31 It states that the starting point of a series of slices should be arbitrarily set such that every position has equal probability. This avoids systematic over- or underestimation of a volume through speci®c sampling of the structure. In principle, images can be measured in one acquisition as long as the muscle does not exceed the homogeneous volume of the magnet, which is typically about 50 cm in diameter. However, the linearity of the gradients within the homogeneous volume and subsequently the accuracy of the mapping is signi®cantly reduced in the outer portions of the volume because of the nonlinearity of the gradients. Therefore, imaging of

larger muscles may be more appropriate in multiple steps. The imperfect spatial accuracy of MR systems is usually improved by retrospective image-correction algorithms, which use the known spatial characteristics of the gradients. This procedure is acceptable for diagnostic images where the relative positions are much more important than the absolute dimensions. For quantitative measurements, however, a continuous monitoring of these effects is necessary, especially if variations over longer time periods are investigated, such as training effects on muscular volume. The cross-sectional area and actual volume can be determined by several methods:29,31±33 simple threshold, voxel counting within operator-de®ned borders, elaborated tissue segmentation algorithms, or point counting. Simple threshold methods need a homogeneous rf sensitivity of the coil since substantial inhomogeneities in signal intensity will lead to wrong assignment of voxels. Computed tissue segmentation, often based on different MRI scans, uses several steps to distinguish between tissues, in many cases with human interaction. Image intensity threshold can be followed by boundary tracing, edge detection ®lters, morphological erosion, seed-growing algorithms,29,30 etc. Point counting31 is an interactive method very popular in morphometry, for example to analyze electron micrographs. Grids are placed over the region of interest (Figure 1) and crosses that hit the structure of interest are counted. The probability for a hit is proportional to the area covered by the structure. If large enough numbers are counted (typically over 100 hits per structure), variations resulting from the counting procedure are negligible and the number of hits represents the desired area almost perfectly. Point counting is a very robust method but is very time consuming and, to a certain extent, operator dependent. All digitization methods are somewhat susceptible to partial volume effects: the misinterpretation of voxels that contain more than one type of tissue. If a T1-weighted MR sequence is used that leads to a strong signal of fat, voxels that are only half ®lled with adipose tissue will produce suf®cient signal to be counted as fat. Since this happens in all voxels at the border of fat and muscle tissue, fat content will be systematically overestimated. Optimized imaging sequences and appropriate windowing of the signal intensity may help to reduce this problem.

3.4

Use of MRI for Determination of Muscle Fiber Orientation and Composition

In addition to assessment of gross morphology, histological characterization of muscle tissue by means of MR is desirable. So far, ®ber orientations in muscle and the `pennation angle' between muscle ®bers and axis of the muscle have been measured by ultrasound. It has been proposed that MR images would show suf®cient morphological detail to see striations generated by fat, which runs parallel and between the muscle fascicles, and that this would allow for an assessment of the fascicle pennation.4,33 One interesting ®nding of such a threedimensional analysis was a considerable variation of pennation angles within one muscle;4 for example a range of 5 to 50 was observed in the vastus medialis. This noninvasively obtained information helps to evaluate muscle mechanics. Further suggestions have been made to use the effect of

IMAGING AND SPECTROSCOPY OF MUSCLE

3

Figure 1 Selected axial slices of the lower leg with an illustration of morphometry based on the point counting method. A grid is placed arbitrarily over the anatomical structure to be measuredÐhere the tibialis anterior muscleÐand points lying within the structure of interest are counted. The number of counted points is proportional to the volume of the structure when a series of slices is evaluated. The error introduced by the placement of the grid and the counting becomes negligible if more than 100 points are counted overall

anisotropic diffusion1,14,16 within heart or skeletal muscle tissue to visualize ®ber orientation in MR images.13,15 So far, ®ber composition has mainly been assessed by 31P MRS using differences in PCr content and pH (see Peripheral Muscle Metabolism Studied by MRS). The method requires relatively large sensitive volumes and, therefore, it is inherently susceptible to partial volume effects: a mixture of signals from different muscles that are partially activated and resting. Activity-dependent MR images and gradient-based volume selection may improve these observations in future. The determination of ®ber composition by MRI is based on differences in relaxation times of water.3 Using the effect of gadolinium diethylenetriamine pentaacetic acid (DTPA) on MR

images of rabbit muscle, larger extracellular space has been observed in slow-twitch (red) muscle (see Skeletal Muscle Evaluated by MRI).7 4

4.1

USE OF MRS FOR HUMAN MUSCLE PHYSIOLOGY The Use of Different Nuclei to Observe Muscular Metabolism

The high-energy phosphates visible in the 31P MR spectrum and the pH titration of the inorganic phosphate (Pi) relative to

4 IMAGING AND SPECTROSCOPY OF MUSCLE PCr made 31P MRS the favorite tool for biochemists and physiologists interested in muscular metabolism (see Peripheral Muscle Metabolism Studied by MRS). The application of 13C MRS for examinations of muscular metabolism attracted the attention of physiologists because glycogen is 100% visible in the 13C MR spectrum despite the very large molecular weight of this molecule. However, the technical requirements for 1H-decoupled 13C MRS are considerably higher than those usually provided in a standard clinical scanner. Until recently, there had only been some exploratory studies using 1H MRS. This is particularly remarkable since 1H is the most frequently used nucleus in brain examinations. Originally, the reasons for this obvious disinterest were of a technical nature (water suppression, eddy currents for short echo times), but later the sparse attention to 1H MRS of muscle may have resulted from a general feeling that its information content would be rather low. It will be shown below that this impression was wrong. Other nuclei such as 23Na have only rarely been used to study muscle metabolism.8,9 However, we expect that studies of order on the molecular level will bene®t from the potential of less popular nuclei in the future. Some very elegant examinations of muscular metabolism use combinations of different nuclei, for example 13C with 31P MRS:34±37 the 13C studies follow the depletion and recovery of glycogen while 31P is used to monitor phosphorylation. Combinations of different nuclei and the fusion of MRS and MRI in the same session will be vital approaches in the future development of MR of human muscle metabolism and will prove the enormous versatility of MR. 4.2 Classical

31

P MRS of Human Muscle

Many of the classical studies of human muscle using 31P MRS are covered in other monographs in these volumes and these provide a wide range of references to work in this area (see Peripheral Muscle Metabolism Studied by MRS, Phosphorus-31 Magnetization Transfer Studies In Vivo, Proton Decoupling During In Vivo Whole Body Phosphorus MRS, Single Voxel Whole Body Phosphorus MRS, Tissue Behavior Measurements Using Phosphorus-31 NMR, Whole Body Studies: Impact of MRS, pH Measurement In Vivo in Whole Body Systems). The ®rst report on a muscular disease documented by 31P MRSÐMcArdle diseaseÐpromoted the application of MRS in vivo. However, widespread distribution of 31P MRS in clinical routine and the use of this modality in all-day diagnostics did not occur as expected and it remained an excellent, but somewhat specialized, tool for research in physiology and pathology. Beside the well-known observation of high-energy phosphates by 31P MRS and the determination of intracellular pH, MT experiments have also examined the creatine kinase reaction (see also Phosphorus-31 Magnetization Transfer Studies In Vivo). The very recent development of genetically manipulated mice (e.g. creatine kinase knock-out mice) may help in the understanding of many aspects of 31P MRS. Proton decoupling during acquisition of phosphorus spectra (see Proton Decoupling During In Vivo Whole Body Phosphorus MRS) allows signals such as those from glycerophosphorylcholine and glycerophosphorylethanolamine

to be distinguished, which may help to improve understanding of membrane metabolism. Diffusion-weighted 31P MRS of PCr and PiÐtogether with MT techniquesÐmay provide more information on the creatine kinase reaction and the role and nature of Cr/PCr as energy shuttle, reservoir, or buffer in the myocyte. 4.3

Application of

13

C MRS in Human Muscle

The main application of 13C MRS in human muscle has been the observation of glycogen depletion and repletion, with and without labeling of blood glucose at the C1 position by 13 C, and in some experiments controlled by clamp techniques.38±40 Speci®c labeling of other compounds and subsequent isotopomer analysis has been very successful in whole animals and isolated organs but has hardly been used in humans. Despite the fact that 13C MRS is extremely useful and elegant, only a few groups worldwide are using this method in studies of human metabolism. This can be explained by the technical requirements (broadband acquisition, 1H decoupling) and by the low signal-to-noise ratio (SNR), which would bene®t from ®eld strengths higher than those available in routine scanners. Studies with substrates labeled with 13C are expensive and clamp studies require additional experience. The increasing number of MR systems at ®eld strengths of 3±4.7 T will hopefully overcome the technical limitations and prove the value of 13C MRS studies. Four main factors characterize 13C MRS: (a) the low gyromagnetic ratio, which is about 25% of that of 1H; (b) the low natural abundance of 13C (most carbons are NMR invisible in the form of 12C); (c) the large chemical shift dispersion of about 250 ppm compared with about 10 ppm for 1H; and (d) the direct chemical bonds to 1H of most 13C atoms. All these factors have methodological consequences. The low gyromagnetic ratio leads not only to a lower resonance frequency but also to an inherently low sensitivity of 13C, which is about 0.016 compared with 1H. The natural abundance of 1.1% for 13 C further reduces signal intensity. However, the low natural abundance does not only have negative consequences; it also allows for the use of 13C-labeled substrates as tracers, which is obviously not feasible for 1H or 31P. The large chemical shift dispersion per se is advantageous since the separation of different resonances increases. This makes an unspoiled observation of a speci®c metabolite much easier than in the much more crowded 1H MR spectrum. However, the large chemical shift dispersion leads to a spatial misregistration that may be intolerable when popular gradient-based volume localization methods are used directly. The fact that gradients have not been widely used for localization of 13C MR spectra is also a consequence of the very short T2 of glycogen; the signals would have decayed at echo times that are typically used in point-resolved spectroscopy (PRESS) or stimulated echo acquisition (STEAM) sequences. (For additional information on localization sequences see Spatial Localization Techniques for Human MRS). The use of image-selected in vivo spectroscopy (ISIS) overcame the problem of fast transverse relaxation since the magnetization is kept in the longitudinal direction during the localization procedure.41 Localization of 13C nuclei can also be accomplished indirectly via the coupled 1H nuclei and subsequent polarization transfer.42±46 Because of very short 1H T2, these methods are not useful for localization of glycogen. Most

IMAGING AND SPECTROSCOPY OF MUSCLE glycogen C1

creatine

160

140

120

100

ppm

Figure 2 Carbon-13 MR spectrum of the thigh at 1.5 T (GE SIGNA, Milwaukee, WI), without (lower trace) and with (upper trace) continuous wave decoupling of the protons (decoupler S.M.I.S., UK). The decoupling and NOE build up of the doublet of glycogen C1 leads to an improved signal-to-noise ratio that allows a reduction of measurement time (lower trace 10 000 scans, upper trace 4000 scans)

applications, therefore, apply surface coils for localized transmission and reception of the signals, using pulse-and-acquire sequences. Nevertheless, volume localization of 13C MR spectra is often a critical issue for several reasons: (a) adjacent muscle groups may have differences in metabolite levels but usually contribute to an overall signal; (b) huge signals from subcutaneous fat may mask the tiny resonances from metabolites (e.g., signal contributions from muscular glycogen in the abdominal wall may not be separable from liver glycogen); (c) sensitivity pro®les of surface coils are complicated and make absolute quanti®cation much more dif®cult than methods with a rectangular shape of the selected region. Some methods use bottles ®lled with known solutions that replace the human body to calibrate the metabolite levels;47 others use Cr as an internal standard.48 The chemical bonds between carbon atoms and adjacent 1H lead to substantial heteronuclear spin±spin coupling and to splitting of the resonance lines, which further reduce the amplitudes. With 1H decoupling by irradiation on a second rf channel, it is possible to cancel the coupling effect, restoring the full resonance amplitude (Figure 2). In addition, irradiation on the 1H frequency prior to acquisition leads to the so-called nuclear Overhauser enhancement (NOE), which results in an additional 50±90% in signal intensity49 under in vivo conditions (Figure 2). Glycogen is a large molecule with a molecular weight of 107 to 109. Since macromolecules are usually not visible by MRS because of their very short T2 relaxation, it was surprising that glycogen turned out to be 100% visible.50,51 It seems that the internal molecular mobility leads to effective decoupling from fast relaxation processes. The observation of glycogen in human muscle (and liver) is meanwhile one of the `workhorse' applications of in vivo MRS. NOE effect and short repetition times guarantee a reasonable SNR within approximately 10 min even at 1.5 T. Higher ®eld strengths, however, are desirable to increase SNR and/or temporal resolution.

5

Studies of noninsulin-dependent diabetes mellitus (NIDDM) patients36 showed a slower glycogen repletion compared with healthy volunteers. Combination with 31P MRS was then used to identify the limiting step in glycogen synthesis more closely (see below). Patients suffering from glycogen storage diseases48,52 were found to have higher levels of glycogen in muscle and liver. Studies on normal subjects37,43,47,53,54 revealed important physiological information on glycogen as an energy reserve for the human muscle, for example changing with different diets, with different exercise intensity, and as a function of the insulin-dependent and independent control mechanisms. Studies of lipid metabolism using 13C MRS would be very attractive since the degree of saturation and chain length could be evaluated.55±57 However, the dif®culties in separating signals from subcutaneous adipose tissue and muscular lipids (IMCL, see below) restrict this method to studies of bulk fat so far. Since 13C has a very low natural abundance, substrates can be enriched at speci®c positions in the molecule. This can either be used to distinguish between `old' and `new' fractions of a metabolite pool, e.g., to identify glycogen that is newly built in the muscle, or it can be used to distinguish between several potential biochemical pathways. One of the most popular applications is labeling of glucose at the C1 position and the follow-up by 13C MRS during its incorporation into muscular glycogen in clamp studies.36,50 Most 13C labeling experiments with subsequent isotopomer analysis (i.e., analysis of the coupling patterns of mixtures of the same compound with different degrees of labeling), have been done in animals, perfused organs, or plasma samples.58±63 An example in skeletal muscle is the observation of [1-13C]glucose incorporation into intramuscular [1-13C]-glycogen, [3-13C]-lactate, and [3-13C]-alanine.59 Carbon-13 MRS requires much higher enrichment of the 13C isotope than does mass spectrometry; it does, however, have the advantage of organ selectivity (e.g., compared with analysis of breathing air) and chemical speci®city. 4.4 4.4.1

Application of 1H MRS in Human Muscle General Features

Proton MR spectra of skeletal muscle were believed to be swamped by the large signal from fat, which would render all other metabolites invisible. Recent work has shown that the spectrum is indeed very complex, but it is rich in information and re¯ects several aspects of the physiology of exercise. A typical 1H MR spectrum of human muscle tissue acquired by a PRESS sequence (see Spatial Localization Techniques for Human MRS) is shown in Figure 3. Based on high-resolution spectra of muscle extracts and in vivo rat and frog muscle, the peaks visible in spectra of human muscle in vivo have been assigned early on5,64±68 to the following compounds: lipids (methyl protons 0.9 ppm, aliphatic chain methylene protons 1.3 ppm, aliphatic protons near double bonds and carboxyl group at 1.7±2.5 ppm and protons of unsaturated carbons at 5.4 ppm); Cr/PCr (methyl at 3.03 ppm (Cr3), methylene at 3.93 ppm (Cr2)); trimethylammonium group-containing metabolites (TMA), such as choline, phosphocholine, glycerophosphocholine, and carnitine at 3.2 ppm; a line that has been tentatively assigned to taurine at approximately 3.4 ppm, depending on the orientation of the muscle; and two his-

6 IMAGING AND SPECTROSCOPY OF MUSCLE IMCL (-CH2-)n

EMCL (-CH2-)n TMA

Cr3

Cr2 EMCL (-CH3)

tidine protons of the dipeptide carnosine at 7.0 and 8.0 ppm (anserine in some species). A peak at the unusual position of 78 ppm (Figure 4), which is invisible under the experimental conditions used for Figure 3, could be detected and assigned to deoxymyoglobin.69,70 Lactate at 1.3 ppm can also be observed only under speci®c experimental conditions.6,21,68,71±73 As seen above, skeletal muscle is highly organized at different spatial levels and many aspects of order in¯uence 1H MR spectra. These aspects are summarized in a previous paragraph and for the different metabolites separately below.

IMCL IMCL (-CH3)

EMCL X3

4.0

2.0

3. 0

1.0

ppm

Figure 3 PRESS-localized 1H MR spectrum of the human tibialis anterior muscle. The orientation of the ®bers in this muscle parallel to the magnetic ®eld leads to the separation of the signals from intra- (IMCL) and extramyocellular lipids (EMCL), and the splitting of the creatine (methylene Cr2, methyl Cr3) resonances. Trimethylammonium (TMA) covers one part of the Cr3 triplet and the coupling partner of resonance X3, which can be tentatively assigned to taurine. Acetylcarnitine at 2.13 ppm is not visible at rest; while lactate at 1.3 ppm is only visible with editing techniques and during workload/ischemia (a)

Deoxymyoglobin

4.4.2

Lactate

The signal of the methyl group of lactate is overlaid by the much larger signals from lipids in human muscle. However, using the hetero- or homonuclear spin±spin interaction and editing techniques by inversion and decoupling,68 zero- or double-quantum ®lters6,21,71±73 are able to observe the lactate signal without the huge overlapping resonances. Promoted by the physiological importance of this metabolite, observations of lactate have been among the ®rst applications of 1H MRS in human muscle. 4.4.3

pH

A second important parameter for the description of muscular metabolism is the intracellular pH, usually determined by

(b) Deoxymyoglobin (au)

Proximal

1/k = 10 s

Cuff

Deoxymyoglobin (au)

Distal

100

80

ppm

1/k = 36 s

0

4

8

Time (min)

Figure 4 (a) Spectra showing the resonance of deoxymyoglobin from the lower leg in a patient with peripheral arterial disease. (b) Signal development during ischemia and recovery proximal and distal to the lesion (au, arbitary units). Arrows indicate time points when the spectra shown on the left were acquired. The time resolution in the graphs is 3.5 s, one spectrum represents a 16 s acquisition (30 Hz apodization). The time constant for recovery is 10 s proximal and 36 s distal from the lesion (unpublished data; clinical support by Dr I. Baumgartner is greatly appreciated)

IMAGING AND SPECTROSCOPY OF MUSCLE

the frequency difference between Pi and PCr in the 31P MR spectrum. Proton MRS provides the same information, since the aromatic signals of histidine in the carnosine molecule are titrating and can, therefore, be used to determine intracellular pH. A comparison between titration observed in 31P and 1H MR spectra68 showed the accuracy of 1H MRS. Because of the fairly low concentrations of carnosine, the 1H MRS measurement may be less sensitive than 31P MRS during exercise. However, 1H MRS has the advantage of identical sensitivity independent of exercise scheme, while the undershoot of Pi after exercise and the low Pi content in resting muscle can make pH measurements with 31P MRS impossible in some instances. 4.4.4

Myoglobin

Proton 1H MRS can also be used during exercise or ischemia to estimate the degree of intracellular oxygenation by measuring deoxymyoglobin resonances. The signals from oxymyoglobin are in the normal frequency range of an 1H MR spectrum and, therefore, are overlapped by many other resonances at 1.5 T. Only at 7 T, the resonance of the -methyl group of Val-E11 at ÿ2.8 ppm can be observed separately under oxygenated conditions.74 However, while this resonance disappears with deoxygenation, the F8 proximal histidyl N- proton shifts to 78 ppm,69,70,75 and becomes visible also at moderate ®eld strength. This histidine resonance has dramatically shortened T2 relaxation times and is, therefore, very broad, but the chemical shift separation from the large resonances between 0 and 10 ppm is suf®cient to detect it reliably. Figure 4 shows the increase of deoxymyoglobin during ischemia in human skeletal muscle, observed at 1.5 T. 4.4.5

ene group of Cr/PCr appears as a prominent dipolar doublet and the methyl group as a less well-de®ned triplet (see Figure 5). This ®nding was unexpected since in vivo MRS had been seen as spectroscopy of metabolite solutions where dipolar coupling is averaged out and dipolar effects are restricted to relaxation processes. Using more elaborate methods such as one-dimensional zero- and double-quantum ®ltering, twodimensional J-resolved spectroscopy, two-dimensional constant time COSY, and longitudinal order separation spectroscopy,18 it was possible to prove unequivocally that peaks of Cr methylene result from a pair of dipolar coupled protons.78 Most other resonance peaks, including the methyl group of Cr, were also found to beÐat least partiallyÐdependent on the angle between muscle ®bers and magnetic ®eld. The form of the orientation dependence and the approximate size of the coupling have been con®rmed at a higher ®eld in rat muscle.19 One can now speculate about the mechanisms behind the observed effects. Irrespective of the chemical nature of the metabolites involved, there are three main explanations to account for incomplete temporal averaging of dipolar couplings. First of all, the observed molecules may be hindered from isotropic tumbling by being constrained into small elongated spaces between the actin/myosin chains. Second, they might be temporarily bound to macromolecules that themselves are strictly ordered within the muscle cells. Finally, dipolar-coupled peaks might originate from large molecules that are permanently bound to ordered structures in muscle but have enough sidechain mobility to be observable and partially average dipolar couplings. The most likely explanation of the three is that PCr and/or free Cr together with their hydration spheres are large enough to be hindered from isotropic tumbling in the elongated

Acetylcarnitine

After heavy workload, a peak at 2.13 ppm can be detected in localized spectra of human muscle,76 while spectra from resting muscle do not feature a sharp singlet at this position in general. Based on its chemical shift, its singlet nature, its approximate concentration, and physiological behavior, the peak was tentatively assigned to the acetyl group of acetylcarnitine.76 Carnitine has long been known for its vital role in the transportation of fatty acids into mitochondria for -oxidation. More recently, it was recognized that carnitine is equally important as a buffering system for a potential surplus of acetyl groups.77 The export of acetyl groups in the form of acetylcarnitine out of the mitochondria helps to keep the acetyl CoA/ coenzyme A ratio balanced such that the nonacetylated form of the co-factor can play its roles in the TCA cycle and in pyruvate dehydrogenation.

7

Cr3

TMA Cr2 X3

Cr3

Cr2 TMA X3

4.4.6

Creatine

Considering the concentration and chemical shift differences of Cr and PCr in muscle tissue, one could expect that the resulting signals would be well above the noise level and would represent the sum of both, i.e., total Cr, since these species would not be separable at 1.5 T. A detailed analysis of Cr in 1H MR spectra, however, showed several unexpected features17±20,78 such as an orientation-dependent dipolar splitting for several compounds that were not assigned at the very beginning.17 In a subsequent Cr-loading study,78 one of these metabolites has been identi®ed as Cr and/or PCr. The methyl-

4.0

3.0

ppm

Figure 5 Spectra from the same voxel in the tibialis anterior muscle show the effect of orientation upon creatine Cr2 (methylene) and Cr3 (methyl). The lower spectrum has been acquired with the leg parallel to the magnetic ®eld, the upper spectrum at the magic angle (54 between muscle and ®eld). At the magic angle, splitting owing to dipolar coupling vanishes. TMA, trimethylammonium; X3 tentatively assigned to taurine

8 IMAGING AND SPECTROSCOPY OF MUSCLE spaces between actin/myosin chains. Because of charge distributions on these molecules, speci®c or unspeci®c coupling to the actin/myosin complexes may exacerbate the ordering. It has been postulated previously that some of this space is inaccessible to PCr. Since Cr and PCr are involved in the creatine kinase equilibrium and may serve not only as a short-term energy storage but also for some forms of energy transportation, a motional restriction as observed in these experiments would be of crucial importance since the overall reaction could be limited in vivo by transport processes. Additional experiments lead to further questions about the exact nature of the Cr resonances: (a) in a postmortem study of rat muscle, the doublet of Cr methylene was shown to disappear from the MR spectrum on a time scale similar to the disappearance of PCr;19 and (b) in human skeletal muscle, Cr/ PCr during and after heavy exercise showed decrease and recovery similar to PCr in the 31P MR spectrum.20 It is surprising that depletion and recovery appear to be related to the PCr content of human muscle, since the Cr signals in 1H MRS have hitherto always been associated with total Cr, not PCr. This ®nding is, however, in agreement with the signal behavior observed in rat muscle postmortem.19 A straightforward but still questionable explanation for that observation would be that only the protons from PCr are detected by 1H MRS while free Cr is invisible. The reduced visibility of Cr may or may not have similar roots to the incomplete dipolar averaging, MT effects, restricted diffusion, anisotropy of relaxation times, and the questioned visibility of lactate. The calculation of ADP concentrations based on ATP, free Cr, pH, and the creatine kinase reaction constant would be questionable if reduced visibility of Cr was an indication for reduced availability of this substrate. Therefore, ®ndings on reduced visibility of creatine would have a serious effect on the modeled phosphorylation kinetics derived from 31P MRS. Diffusion-weighted spectroscopy may be another tool to identify restricted motion of speci®c metabolites. An excellent overview of diffusion effects in human muscle can be found in Nicolay et al.12 This study shows restricted motion of PCr compared with water in rat hindleg muscle. 4.4.7

Intramyocellular Lipids

IMCL are stored in droplets in the cytoplasm of muscle cells and are a form of stored energy readily accessed during long-term exercise. When Schick et al.11 compared the lipid resonances in calf muscle and fat tissue, they observed two compartments of triglycerides with a resonance frequency shift of 0.2 ppm. They assigned the resonance at 1.5 ppm to lipids in fat cells and hypothesized that the resonance at 1.3 ppm could be attributed to lipids located inside muscle cells (i.e., in their cytoplasm) experiencing different bulk susceptibility (see Figure 3). This assignment was veri®ed by Boesch et al.10 who demonstrated that IMCL signals scale linearly with voxel size, water and Cr signals, while EMCL signals do not. Furthermore the resonance frequency shift of EMCL and the orientation dependence of their spectral pattern could be attributed to the spatial arrangement of these lipids (plate-like structures for EMCL versus spheroid droplets for IMCL). Experimental data agree very well with theoretical estimations of susceptibility effects. Inter- and intraindividual reproducibility studies indicate that the error of the method is about 6% and that IMCL levels differ signi®cantly between identical muscles in different

[au]

6

IMCL

workload

4

2 creatine 0 0

20 40 60 [hours after workload]

80

100

Figure 6 Depletion and repletion of intramyocellular lipids (IMCL) in tibialis anterior muscle after strenuous exercise (3 h bicycle training). In this example, the recovery phase of IMCL can be characterized by an exponential ®t with a time constant of about 40 h. Further experiments have shown that recovery is strongly dependent on the diet. Creatine levels stay constant within the experimental accuracy (au, arbitary units; graph adapted from Boesch et al.10)

subjects, different muscles in the same subject, as well as intraindividually in the same muscle when measured at 1 week intervals. The accuracy of IMCL determination by 1H MRS is suf®cient to follow IMCL depletion and recovery after exercise in single individuals (Figure 6). Dietary modulation of IMCL levels has been investigated by biopsy studies, with the wellknown advantages and disadvantages of this invasive method. A consequence of the invasiveness is the scarcity of results and follow-up measurements. With 1H MRS it is now feasible to follow IMCL depletion and repletion more or less continuously and in different muscle groups. A correlation of increased IMCL levels and insulin sensitivity has been shown by 1H MRS in diabetic patients.79,80 Since it is not trivial to separate EMCL and IMCL, the comparison of measurements in obese patients with normal weight volunteers could be prone to systematic errors. Comparing cohorts of patients with volunteers, weight and muscular fat in®ltration should be matched between the groups to avoid systematic differences between the resulting EMCL contaminations of the spectra.

5

CONCLUSIONS

MRI and MRS of muscular physiology has occurred in distinct separate centers with different equipment for a considerable time. A combination of the two complementary modalities is now developing while classical pulse and acquire 31 P MRS has become less prominent. Multinuclear MRS in whole-body systems using combinations of MRI and MRS with elaborate MR techniques such as diffusion weighting, editing, and MT seem to lead the way to future MR studies of muscle physiology. Less popular nuclei such as 23Na have the potential to elucidate order in biological tissues further. Decoupling of 31P and 13C spectra, and the use of whole-body magnets at higher ®eld, have already promoted the application

IMAGING AND SPECTROSCOPY OF MUSCLE

of MRS in humans. This use will increase as soon as a larger number of versatile, high-®eld MR systems is installed. Recent results obtained with localized 1H MRS in skeletal muscle have shown that it is a very attractive tool to study muscle physiology both at rest and during exercise. It now appears that with 1H MRS alone one can, in principle, determine pH, oxygenation, lactate levels, substrate use (IMCL, potentially also glycogen in addition to 13C MRS), acetyl group buffering, and possibly PCr levels. It has also become thoroughly evident that the molecular basics of muscle tissue leads to most interesting NMR ®ndings that can by no means be modeled by a single isotropic compartment consisting of an aqueous solution. Incompletely averaged dipolar couplings, reduced visibility of Cr, MT effects on Cr, the postulated compartmentation of lactate, and the susceptibility induced separation between IMCL and EMCL are all re¯ections of the complex and partly oriented architecture of muscle tissues, where (partial) compartmentation, chemical exchange, varying susceptibility, structural anisotropy, and protein interactions are essential ingredients.

6 RELATED ARTICLES ESR Probes as Field Detectors in MRI; MRI of Musculoskeletal Neoplasms; Peripheral Muscle Metabolism Studied by MRS; pH Measurement In Vivo in Whole Body Systems; Phosphorus-31 Magnetization Transfer Studies In Vivo; Proton Decoupling During In Vivo Whole Body Phosphorus MRS; Proton Decoupling in Whole Body Carbon-13 MRS; Quantitation in In Vivo MRS; Single Voxel Localized Proton NMR Spectroscopy of Human Brain In Vivo; Single Voxel Whole Body Phosphorus MRS; Skeletal Muscle Evaluated by MRI; Spatial Localization Techniques for Human MRS; Tissue Behavior Measurements Using Phosphorus-31 NMR; Whole Body Studies: Impact of MRS.

7 REFERENCES 1. R. M. Henkelman, G. J. Stanisz, J. K. Kim, and M. J. Bronskill, Magn. Reson. Med., 1994, 32, 592. 2. W. C. Cole, A. D. LeBlanc, and S. G. Jhingran, Magn. Reson. Med., 1993, 29, 19. 3. J. M. Bonny, M. Zanca, O. Boesp¯ug-Tanguy, V. Dedieu, S. Joandel, and J. P. Renou, Magn. Reson. Imag., 1998, 16, 167. 4. S. H. Scott, C. M. Engstrom, and G. E. Loeb, J. Anat., 1993, 182, 249. 5. J. Hu, M. R. Willcott, and G. J. Moore, J. Magn. Reson., 1997, 126, 187. 6. J. A. Kmiecik, C. D. Gregory, Z. P. Liang, D. E. Hrad, P. C. Lauterbur, and M. J. Dawson, Magn. Reson. Med., 1997, 37, 840. 7. I. K. Adzamli, F. A. Jolesz, A. R. Bleiler, R. V. Mulkern, and T. Sandor, Magn. Reson. Med., 1989, 11, 172. 8. R. Reddy, L. Bolinger, M. Shinnar, E. Noyszewski, and J. S. Leigh, Magn. Reson. Med., 1995, 33, 134. 9. T. Kushnir, T. Knubovets, Y. Itzchak, U. Eliav, M. Sadeh, L. Rapoport, E. Kott, and G. Navon, Magn. Reson. Med., 1997, 37, 192. 10. C. Boesch, H. Slotboom, H. Hoppeler, and R. Kreis, Magn. Reson. Med., 1997, 37, 484. 11. F. Schick, B. Eismann, W. I. Jung, H. Bongers, M. Bunse, and O. Lutz, Magn. Reson. Med., 1993, 29, 158.

9

12. K. Nicolay, A van der Toorn, and R. M. Dijkhuizen, NMR Biomed., 1995, 8, 365. 13. A. van Doorn, P. H. Bovendeerd, K. Nicolay, M. R. Drost, and J. D. Janssen, Eur. J. Morphol., 1996, 34, 5. 14. P. J. Basser, NMR Biomed., 1995, 8, 333. 15. T. G. Reese, R. M. Weisskoff, R. N. Smith, B. R. Rosen, R. E. Dinsmore, and V. J. Wedeen, Magn. Reson. Med., 1995, 34, 786. 16. E. W. Hsu and S. Mori, Magn. Reson. Med., 1995, 34, 194. 17. R. Kreis and C. Boesch, J. Magn. Reson. Series B, 1994, 104, 189. 18. R. Kreis and C. Boesch, J. Magn. Reson. Series B, 1996, 113, 103. 19. V. Ntziachristos, R. Kreis, C. Boesch, and B. Quistorff, Magn. Reson. Med., 1997, 38, 33. 20. R. Kreis, B. Jung, J. Slotboom, J. Felblinger, and C. Boesch, J. Magn. Reson., 1999, 137, 350. 21. L. Jouvensal, P. G. Carlier, and G. Bloch, Magn. Reson. Med., 1997, 38, 706. 22. J. L. Fleckenstein, L. A. Bertocci, R. L. Nunnally, R. W. Parkey, and R. M. Peshock, Am. J. Roentgenol., 1989, 153, 693. 23. E. R. Weidman, H. C. Charles, R. Negro-Vilar, M. J. Sullivan, and J. R. MacFall, Invest. Radiol., 1991, 26, 309. 24. T. Ogino, H. Ikehira, N. Arimizu, H. Moriya, K. Wakimoto, S. Nishikawa, H. Shiratsuchi, H. Kato, F. Shishido, and Y. Tateno, Ann. Nucl. Med., 1994, 8, 219. 25. L. L. Ploutz-Snyder, S. Nyren, T. G. Cooper, E. J. Potchen, and R. A. Meyer, Magn. Reson. Med., 1997, 37, 676. 26. G. Yue, A. L. Alexander, D. H. Laidlaw, A. F. Gmitro, E. C. Unger, and R. M. Enoka, J. Appl. Physiol., 1994, 77, 84. 27. H. A. Cheng, R. A. Robergs, J. P. Letellier, A. Caprihan, M. Icenogle, and L. J. Haseler, J. Appl. Physiol., 1995, 79, 1370. 28. J. L. Fleckenstein, R. C. Canby, R. W. Parkey, and R. M. Peshock, Am. J. Roentgenol., 1988, 151, 231. 29. M. A. Elliott, G. A. Walter, H. Gulish, A. S. Sadi, D. D. Lawson, W. Jaffe, E. K. Insko, J. S. Leigh, and K. Vandenborne, MAGMA, 1997, 5, 93. 30. N. Mitsiopoulos, R. N. Baumgartner, S. B. Heyms®eld, W. Lyons, D. Gallagher, and R. Ross, J. Appl. Physiol., 1998, 85, 115. 31. N. Roberts, L. M. Cruz-Orive, N. M. K. Reid, D. A. Brodie, M. Bourne, and R. H. T. Edwards, J. Microsc., 1993, 171, 239. 32. T. Fukunaga, R. R. Roy, F. G. Shellock, J. A. Hodgson, M. K. Day, P. L. Lee, H. Kwong-Fu, and V. R. Edgerton, J. Orthop. Res., 1992, 10, 926. 33. M. V. Narici, L. Landoni, and A. E. Minetti, Eur. J. Appl. Physiol., 1992, 65, 438. 34. K. F. Petersen, R. Hendler, T. Price, G. Perseghin, D. L. Rothman, N. Held, J. M. Amatruda, and G. I. Shulman, Diabetes, 1998, 47, 381. 35. D. L. Rothman, R. G. Shulman, and G. I. Shulman, J. Clin. Invest., 1992, 89, 1069. 36. G. I. Shulman, D. L. Rothman, T. Jue, P. Stein, R. A. DeFronzo, and R. G. Shulman, N. Engl. J. Med., 1990, 332, 223. 37. R. Roussel, P. G. Carlier, J. J. Robert, G. Velho, and G. Bloch, Proc. Natl. Acad. Sci. USA, 1998, 95, 1313. 38. N. Beckman, in `Carbon-13 NMR spectroscopy of biological systems', ed. N. Beckmann, Academic Press, New York, 1995, p. 269. 39. J. Seelig and A. P. Burlina, Clin. Chim. Acta, 1992, 206, 125. 40. R. Badar-Goffer and H. Bachelard, Essays Biochem., 1991, 26, 105. 41. R. Gruetter, E. J. Novotny, S. D. Boulware, D. L. Rothman, G. F. Mason, G. I. Shulman, R. G. Shulman, and W. V. Tamborlane, Proc. Natl. Acad. Sci. USA, 1992, 89, 1109. 42. R. Gruetter, G. Adriany, H. Merkle, and P. M. Andersen, Magn. Reson. Med., 1996, 36, 659. 43. M. Saner, G. McKinnon, and P. Boesiger, Magn. Reson. Med., 1992, 28, 65.

10 IMAGING AND SPECTROSCOPY OF MUSCLE 44. R. Kreis, J. Slotboom, J. Felblinger, and C. Boesch, Proc. Vth Ann Mtg. (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, p. 1438. 45. H. Watanabe, Y. Ishihara, K. Okamoto, K. Oshio, T. Kanamatsu, and Y. Tsukada, J. Magn. Reson., 1998, 134, 214. 46. A. J. van den Bergh, H. J. van den Boogert, and A. Heerschap, J. Magn. Reson., 1998, 135, 93. 47. R. Taylor, T. B. Price, L. D. Katz, R. G. Shulman, and G. I. Shulman, Am. J. Physiol., 1993, 265, E224. 48. P. Jehenson, D. Duboc, G. Bloch, M. Fardeau, and A. Syrota, Neuromuscul. Disord., 1991, 1, 99. 49. G. Ende and P. Bachert, Magn. Reson. Med., 1993, 30, 415. 50. R. Roussel, P. G. Carlier, C. Wary, G. Velho, and G. Bloch, Magn. Reson. Med., 1997, 37, 821. 51. R. Taylor, T. B. Price, D. L. Rothman, R. G. Shulman, and G. I. Shulman, Magn. Reson. Med., 1992, 27, 13. 52. N. Beckman, J. Seelig, and H. Wick, Magn. Reson. Med., 1990, 16, 150. 53. A. van den Bergh, S. Houtman, A. Heerschap, N. J. Rehrer, H. J. van den Boogert, B. Oeseburg, and M. T. E. Hopman, J. Appl. Physiol., 1996, 81, 1495. 54. T. B. Price D. L. Rothman, R. Taylor M. J. Avison, G. I. Shulman, and R. G. Shulman, J. Appl. Physiol., 1994, 76, 104. 55. C. Wary, G. Bloch, P. Jehenson, and P. G. Carlier, Anticancer Res., 1996, 16, 1479. 56. C. T. Moonen, R. J. Dimand, and K. L. Cox, Magn. Reson. Med., 1988, 6, 140. 57. E. L. Thomas, S. C. Cunnane, and J. D. Bell, NMR Biomed., 1998, 11, 290. 58. J. G. Jones, M. A Solomon, A. D. Sherry, F. M. H. Jeffrey, and C. R. Malloy, Am. J. Physiol., 1998, 275, E843. 59. B. M. Jucker, A. J. Rennings, G. W. Cline, K. F. Petersen, and G. I. Shulman, Am. J. Physiol., 1997, 273, E139. 60. E. D. Lewandowski, C. Doumen, L. T. White, K. F. LaNoue, L. A. Damico, and X. Yu, Magn. Reson. Med., 1996, 35, 149. 61. A. D. Sherry, P. Zhao, A. Wiethoff, and C. R. Malloy, Magn. Reson. Med., 1994, 31, 374. 62. L. A. Bertocci, J. G. Jones, C. R. Malloy, R. G. Victor, and G. D. Thomas, J. Appl. Physiol., 1997, 83, 32. 63. H. Bachelard, Dev. Neurosci., 1998, 20, 277. 64. S. R. Williams, D. G. Gadian, E. Proctor, D. B. Sprague, D. F. Talbot, I. R. Young, and F. F. Brown, J. Magn. Reson., 1985, 63, 406. 65. P. A. Narayana, J. D. Hazle, E. F. Jackson, L. K. Fotedar, and M. V. Kulkarni, Magn. Reson. Imag, 1988, 6, 481. 66. M. Barany and P. N. Venkatasubramanian, NMR Biomed., 1989, 2, 7. 67. H. Bruhn, J. Frahm, M. L. Gyngell, K. D. Merboldt, W. Haenicke, and R. Sauter, Magn. Reson. Med., 1991, 17, 82. 68. J. W. Pan, J. R. Hamm, H. P. Hetherington, D. L. Rothman, and R. G. Shulman, Magn. Reson. Med., 1991, 20, 57.

69. Z. Wang, E. A. Noyszewski, and J. S. Leigh, Magn. Reson. Med., 1990, 14, 562. 70. C. Brillault-Salvat, E. Giacomini, L. Jouvensal, C. Wary, G. Bloch, and P. G. Carlier, NMR Biomed., 1997, 10, 315. 71. J. E. van Dijk, D. K. Bosman, R. A. F. M. Chamuleau, and W. M. M. J. Bovee, Magn. Reson. Med., 1991, 22, 493. 72. R. E. Hurd and D. Freeman, NMR Biomed., 1991, 4, 73. 73. G. Bloch, L. Jouvensal, and P. G. Carlier, Magn. Reson. Med., 1995, 34, 353. 74. U. Kreutzer, D. S. Wang, and T. Jue, Proc. Natl. Acad. Sci. USA, 1992, 89, 4731. 75. E. A. Noyszewski, E. L. Chen, R. Reddy, Z. Wang, and J. S. Leigh, Magn. Reson. Med., 1997, 38, 788. 76. R. Kreis, B. Jung, S. Rotman, J. Slotboom, J. Felblinger, and C. Boesch, Proc. Vth Ann Mtg. (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, p. 162. 77. E. P. Brass and W. R. Hiatt, Life Sci., 1994, 54, 1383. 78. R. Kreis, M. Koster, M. Kamber, H. Hoppeler, and C. Boesch, Magn. Reson. Med., 1997, 37, 159. 79. J. Machann, F. Schick, S. Jacob, O. Lutz, H. U. HaÈring, and C. D. Claussen, MAGMA, 1998, p. 220. 80. D. T. Stein, L. S. Szczepaniak, R. L. Dobbins, P. Snell, and J. D. McGarry, Proc. VIth Ann Mtg. Soc. (Int.) Magn. Reson. Med., Sydney, 1998, p. 388.

Biographical Sketch Chris Boesch. b 1951. M.S. (Physics diploma) ETH Zurich (Swiss Federal Institute of Technology) Switzerland, 1976; Ph.D. ETH Zurich, 1979 (High-resolution NMR studies of polypeptide conformation, supervisor K. WuÈthrich); Studies in Medicine, 1977±86; M.D. University Zurich 1986. Research assistant at the University Children's Hospital in Zurich, Switzerland: Installation of a 2.35T/40 cm MR system in a clinical setting, patient-monitoring systems for pediatric and intensive care patients, MRI and MRS studies of brain development, 1985± 90; Professor and Director of MR Spectroscopy and Methodology at the University of Bern, Switzerland, 1991±present. Current research specialties: in vivo spectroscopy (1H, 31P, and 13C) of brain, muscle, and liver; methodology of in vivo NMR: patient monitoring. Roland Kreis. b 1958. M.S. (diploma in chemistry) ETH (Federal Institute of Technology) Zurich, Switzerland, 1983; Ph.D. ETH Zurich, 1989 (Zero-®eld NMR, supervisor R. R. Ernst). Boswell Fellow at Caltech and Huntington Medical Research Institutes, Pasadena, CA, USA, 1989±91; Assistant Professor at Department for Clinical Research (MR Spectroscopy and Methodology) University of Bern, Switzerland, 1991±present. Research interests: quantitative clinical spectroscopy using methods for optimized data acquisition and proBreakcessing while studying cerebral, musculoskeletal and cardiac (patho) physiology.

1

IMAGING OF TRABECULAR BONE

Imaging of Trabecular Bone Felix W. Wehrli University of Pennsylvania Medical School, Philadelphia, PA, USA

1 INTRODUCTION Bone is a composite material consisting of an inorganic phaseÐcalcium apatite, Ca10(PO4)6(OH)2, corresponding to about 65% of total volumeÐand an organic phaseÐessentially collagenÐaccounting for most of the remaining 35%. From an architectural point of view, bone can be subdivided into cortical and trabecular, the latter providing most of the strength of the axial skeleton (e.g., the vertebral column) and the portions of the appendicular skeleton near the joints. Trabecular bone is made up of a three-dimensional network of struts and plates, the trabeculae, which are on the order of 100±150 m in width and spaced 300±1000 m apart. Like engineering materials, trabecular bone derives its mechanical strength from its inherent elastic properties, its volume density, and its structural arrangement. Bone is constantly renewed through a process called `bone remodeling', a term referring to a dynamic equilibrium between bone formation and bone resorption, controlled by two essential types of cells: the osteoblastsÐbone-forming cellsÐand the osteoclastsÐbone-resorbing cells. During bone formation, osteoblasts eventually become imbedded in bone, turning into osteocytes, which presumably act as piezoelectric sensors transmitting signals to the osteoblasts to induce bone formation. Since the seminal work of Wolff,1 it has been known that bone grows in response to the forces to which it is subjected (see, for example, Roesler2). Therefore, weightlessness and physical inactivity are well-known factors inducing bone loss. The most common pathologic process leading to bone loss is osteoporosis.3 Among the various etiologies, postmenopausal osteoporosis, which results from increased osteoclast activity, is the most frequent form of the disease, af¯icting a substantial fraction of the elderly female population and, increasingly, the male population. The most common clinical manifestations are fractures of the hip and vertebrae. If detected early, calcium supplements and estrogen replacement are effective forms of therapeutic intervention. Further, the development of drugs inhibiting osteoclastic activity is in progress. Bone mineral density is the most widely invoked criterion for fracture risk assessment, typically measured by dual energy X-ray absorptiometry (DEXA), which is based on the measurement of the attenuation coef®cient in a quantitative radiographic procedure, or by quantitative computed tomography (QCT). Whereas both methods measure bone mineral density with suf®cient precision, neither provides information on the properties or structural arrangement of the bone. NMR, however, has the potential to probe structure as well as chemical composition of bone, both relevant to biomechanical competence.

2

DIRECT DETECTION OF BONE MINERAL BY NMR

31

P

The dif®culties of detecting phosphorus in the solid state in vivo are considerable but do not seem insurmountable. The problems are symptomatic of high-resolution NMR in the solid state in general: a combination of long T1 (on the order of minutes) and short T2 (on the order of 100 s), as well as additional line broadening by anisotropic chemical shift and dipolar coupling. Brown et al. ®rst demonstrated the feasibility of quantitative analysis by solid state 31P spectroscopy in human limbs as a means of measuring bone mineral density noninvasively.4 Imaging adds an additional level of complexity, since the short lifetime of the signal demands short gradient duration, a requirement that can only be reconciled with gradients of high amplitude and large slew rates, which are both dif®cult to achieve in large sample volumes. Ackerman et al. produced one-dimensional spin echo images at 7.4 T with echo times on the order of 1 ms and ¯ip-back 180 rf pulses as a means to restore the longitudinal magnetization, inverted by the phase reversal pulse.5 The same group of workers reported twodimensional images of chicken bone at 6 T by means of a combination of back-projection and 1H±31P cross polarization for sensitivity enhancement, with echo times as short as 200 s.5 In the cross-polarization technique, the 31P magnetization is derived from that of dipolar-coupled protons, which have shorter T1 and thus permit shorter pulse sequence recycling times. Making use of the dependence of the cross-polarization rate on the proton±phosphorus dipolar coupling, the same workers showed that different phosphate species can be distinguished, demonstrating the presence of a minor HPO42ÿ species in immature chicken bone.7 Very recently, Wu et al. have measured three-dimensional mineral density of hydroxyapatite phantoms and specimens of bone ex vivo.8 While this work is at an early stage, it clearly has unique potential for nondestructive assaying of the chemical composition and its age-related changes of bone parameters that might in part explain the increased fragility of bone in older individuals.

3

IMPLICATIONS OF BONE DIAMAGNETISM ON NMR LINE BROADENING

Another approach toward assessing the properties of trabecular boneÐspeci®cally, as its architectural arrangement is concernedÐexploits the diamagnetic properties of bone mineral. By virtue of the higher atomic number of its elemental composition (i.e., calcium and phosphorus), mineralized bone is more diamagnetic than marrow constituents in the trabecular marrow cavities, which consist mainly of water and lipids (i.e., oxygen, carbon, and hydrogen). Note that in this article, the following de®nitions and notation will be used for the contribution from magnetic ®eld inhomogeneity to the total dephasing rate R*2 : 1/T*2, irrespective of the notation in the original literature: 1/T2' : R2' & * 1 2
Hz = 1/T2 ÿ 1/T2, with
Hz representing the full width at half maximum of the magnetic ®eld histogram in the sampling volume such as the imaging voxel and R2' the effective transverse relaxation rate.

2 IMAGING OF TRABECULAR BONE It is well known that near the interface of two materials of different magnetic susceptibility, and depending on the geometry of the interface, the magnetic ®eld is inhomogeneous. Among the ®rst to investigate these effects systematically were Glasel and Lee, who studied deuteron relaxation of beads of different size and susceptibility, suspended in 2H2O.9 Speci®cally, they showed that the linewidth 1/T*2 scaled with
, the difference in volume susceptibility between the beads and deuterium oxide. The transverse relaxation rate 1/T2 was found to increase linearly with reciprocal bead size, an observation that could be reconciled with diffusion in the induced magnetic ®eld gradients. Similar phenomena were reported by Davis et al., who measured proton NMR linewidths at 5.9 T in powdered bone suspended in various solvents and found T*2 to decrease with decreasing grain size and thus increased surfaceto-volume ratio.10 Rosenthal et al. measured R2' in specimens from human vertebrae following marrow removal and immersion in water at 0.6 T, reporting a value of 6.1 sÿ1, more than an order of magnitude smaller than that found for powdered bone at 5.9 T.11 Wehrli et al. ®rst reported a gradient-echobased method for measuring the line width in vivo in humans in vertebral trabecular bone marrow and found R*2 in the vertebrae to be increased by a factor of two to three relative to those in the intervertebral discs.12 One of the earliest studied magnetically heterogeneous systems in biology is lung tissue, where the local magnetic ®eld distortions are caused by air in the alveoli, and some of the concepts described here have parallels in imaging pulmonary parenchyma.13 Transverse relaxation enhancement from diffusion in intrinsic microscopic gradients has received increased attention in conjunction with the blood oxygenation leveldependent (BOLD) contrast phenomenon, resulting from physiologic variations of deoxyhemoglobin in capillary vessels during functional activation.14,15 However, since trabeculae are considerably larger than the venules of the capillary bed (100± 200 m versus 10±20 m), and diffusion of the protons in the marrow spaces is small (on the order of 10ÿ5 cm2 sÿ1), diffusion-induced shortening of T2 is expected to be negligible. Consequently, the effect of the susceptibility-induced inhomogeneous ®eld is essentially line broadening. Suppose there is a distribution
B(x,y,z) across the sample volume, such as an imaging voxel of dimension
x
y
z; then the transverse magnetization Mxy(t) can be written as Mx;y …t† ˆ

M0 eÿt=T2
x
y
z Z Z Z  ei
B…x;y;z†t dx dy dz
x


y


z

…1†

For a Lorentzian ®eld distribution, the integral in Equation (1) can be described as an additional damping term, yielding 0

Mx;y …t† ˆ M0 eÿt=T2 eÿt=T2

…2†

T2' is, therefore, the time constant for inhomogeneity-induced spin dephasing. While the line broadening, in general, of course, is not Lorentzian, it will be seen subsequently that the assumption of a single exponential time constant is often a valid approximation.

3.1

Susceptibility of Bone

Although there is plenty of evidence for the susceptibility hypothesis, a quantitative determination of the volume magnetic susceptibility of bone has not been reported until recently. In preliminary experiments the author determined the susceptibility of bone using a susceptibility matching technique.16 For this purpose, powdered bone from bovine femoral head was suspended in a cylindrical sample tube with a coaxial capillary containing water and serving as a reference, both aligned along the axis of a superconducting magnet. Potassium ferricyanide, K4Fe(CN)6, which is highly diamagnetic, was then added incrementally to the suspension. This operation resulted in a decrease in linewidth and an increase in the bulk magnetic susceptibility (BMS) shift. For concentric cylinders, the BMS shift is given as17
 ˆ

B0 jd ÿ w j 3 2

…3†

with w and d being the volume susceptibilities of water and ferricyanide solution, respectively. Whereas the critical concentration at which the solution matched the susceptibility of bone could not be attained owing to limited solubility of K4Fe(CN)6, the matching concentration was determined by extrapolation of the line broadening±concentration curve, from which a BMS shift of ÿ0.95 ‹ 0.13 ppm was obtained (corresponding to d ÿ w = ÿ2.85 ppm). Hence, bone is considerably less diamagnetic than calcium hydroxyapatite. Based on these earlier experimental approaches, Hopkins and Wehrli conducted a more rigorous study to determine the absolute susceptibility of bone, with potassium chloride as the diamagnetic additive.18 Speci®cally, they showed that the susceptibility of the suspension, susp, is related to the line broadening ÿÿÿ0 [difference between the line width in the presence (subscript p) and absence (subscript 0) of bone powder], as follows: susp ˆ p ÿ

1 ÿ fp …ÿ ÿ ÿ0 † fp kH0

…4†

where fp represents the volume fraction of the bone powder and k is an empirical proportionality constant, which is a function of particle size, shape, orientation, diffusion, and so forth. Therefore, extrapolation of the straight line de®ned by equation (4) to ÿÿÿ0 = 0 directly provides the susceptibility of the powder. The concentration-dependent decrease in line width and increase in BMS shift is shown in Figure 1. The susceptibility of bovine rib bone was found to be ÿ11.3 (‹0.25)  10ÿ6 (S.I.), indicating bone to be about 2.3 ppm more diamagnetic than water, which is somewhat less than the earlier experiments suggested. 3.2

Theoretical Considerations and Computer Modeling

Consider two adjoining materials of susceptibilities m and b. The induced magnetic surface charge density  at some location on the interface between the two materials is then given as  ˆ
H 0
n

…5†

IMAGING OF TRABECULAR BONE

3

Figure 1 Spectra of 1H at 400 MHz from suspensions of bovine rib bone in CaCl2 solutions of increasing volume fraction of salt, ranging from 5 to 25% (bottom to top). The dotted line spectrum offset from the top spectrum is the best ®t Lorentzian line used to estimate the spectral parameters. Small symmetric resonances about the DOH reference line are spinning sidebands. (With permission from Hopkins and Wehrli18)

where n is the unit vector normal to the interface and
 = m ÿ b. The additional ®eld Hi(r) resulting from the magnetic charges at the phase boundary can be estimated from the Coulomb integral19 Z H i …r† ˆ

…r0 †

r ÿ r0 jr ÿ r0 j3

dS0

…6†

where the integration is over the surface S of the interface, with r' and r representing the locations of the source and ®eld points, respectively. The induced ®eld is thus proportional to the difference in susceptibility of the two adjoining materials, the strength of the applied ®eld, and the inverse square of the distance between source and ®eld location. For an array of trabeculae, the ®eld should be highly inhomogeneous, and a relationship is expected to exist between the magnetic ®eld distribution within the volume of interest and the number density, thickness, and orientation of trabeculae. Ford et al.20 developed a three-dimensional model that resembles strut-like trabecular bone as found in the vertebrae21 to predict the line broadening behavior of protons in the marrow spaces, as a means to investigate the structural dependence of R2'. The model consists of a tetragonal lattice of interconnected parallelepipeds (`struts') of square cross-section, differing in susceptibility from the medium by
. When oriented so that the two parallel faces of each transverse strut are normal to the direction of the applied ®eld H0 then, according to Equation (5), these two faces will have uniform charge density  = ‹
 H0. The other faces of the transverse (and longitudinal) struts will have  = 0, and so do not contribute to Hi(r). If the ®eld is oriented arbitrarily relative to the lattice, all faces will be uniformly polarized, with a charge density  =
 H0 cos  where  is the angle between H0 and the unit vector normal to any given polarized surface. An analytical expression exists for the induced ®eld given by the integral of Equation (6) for a

rectangular lamina; consequently, the total ®eld at any one point in space can be calculated as a sum of contributions from all charge-bearing faces. In this manner, a histogram of the ®eld for the unit cell of this lattice was obtained by randomly placing ®eld points within the unit cell, and R2' was calculated by ®tting the Fourier transform of the ®eld histogram to a decaying exponential. The model predicts nearly exponential decay within the experimentally practical range of echo times (about 10±50 ms). Further, R2' is predicted to increase with both the number density of transverse struts and their thickness. Since the latter two quantities scale with material density, this ®nding appears unremarkable, since it would imply that R2' merely measures bone mineral density. However, if both strut thickness and number density are varied in an opposite manner (so as to keep the material density constant), the model indicates that R2' will increase as strut thickness decreases and number density increases. These predictions, which have been con®rmed in analogous physical models,22 underscore the importance of the distribution of the material, and suggest that different etiologies of bone loss (e.g., trabecular thinning as opposed to loss of trabecular elements) might be distinguishable. Extending the numerical approaches described above, Hwang and Wehrli computed the magnetic ®eld distribution in trabecular bone of human and bovine origin on the basis of a surface model derived from isotropic high-resolution threedimensional NMR images.23 Surfaces were modeled with triangular elements of constant magnetic surface charge density, which allows the induced ®eld to be computed from the charged surfaces. The method was applied to computing histograms of the induced ®elds in specimens of trabecular bone. The width of the induced ®eld distributions was found to be narrowest when the polarizing ®eld was parallel to the preferred orientation of the trabeculae, con®rming previous experimental ®ndings,24 which provides further support for the

4 IMAGING OF TRABECULAR BONE

y

(a)

H0, perpendicular

bution of mutually orthogonal cylindrical columns and struts.25 They showed that beyond a critical time within which signal decay is Gaussian, signal evolution can be described by a single exponential time constant:     1 0 2 …7† R2 /
H0 &h ‡ &v ÿ &h sin # 2 where
 is the susceptibility difference between bone and marrow, & h and & v are the densities of the struts and columns, and W is the angle between the magnetic ®eld B0 and the columns. Yablonskiy et al. demonstrated the angular dependence of R2' in a simple trabecular bone phantom composed of parallel polyethylene ®laments and found their experimental ®ndings to agree well with their theory.26 Similar ®ndings were reported by Selby et al. in microphantoms consisting of cylindrical Pyrex rods.27 Previously, Chung et al. had demonstrated the anisotropic behavior of R*2, in vivo in the distal radius where trabeculae are highly ordered, following the anatomic axis.24 This observation was con®rmed by Yablonskiy et al., who found R2' in the radius to be twice as large with the axis of the wrist perpendicular compared with parallel to the direction of the ®eld.26

x

z

H0, parallel (b)

Fraction of field points

0.02

Parallel

4

Perpendicular

0.01

0.00 –0.4

–0.2

0.0

0.2

0.4

0.6

0.8

Figure 2 Induced magnetic ®eld in trabecular bone from bovine tibia. (a) Shaded three-dimensional surface display derived from threedimensional NMR micrograms. (b) Histogram of the induced magnetic ®eld resulting from the bone's diamagnetism with respect to the marrow constituents

anisotropic nature of the effect. Figure 2 shows a histogram of the induced magnetic ®eld, derived from three-dimensional MR micrograms of bovine tibia with the magnetic ®eld oriented along two orthogonal directions, together with a projection image of the specimen. The dependence of the induced ®eld, expressed in terms of the reversible contribution to R2', was also investigated by Yablonskiy and Haacke, who derived an analytical expression for R2' in a model of trabecular bone consisting of a distri-

RELATIONSHIP BETWEEN TRABECULAR ARCHITECTURE, LINE BROADENING, AND MECHANICAL COMPETENCE

Trabecular bone is well known to be anisotropic, with the orientation of the trabeculae following the major stress lines. In the vertebrae, for example, the preferred orientation of the trabeculae is along the body axis, in response to the compressive forces acting in this direction. The role of the horizontal trabeculae is to act as cross ties preventing failure by buckling. It has also been shown that, during aging, horizontal trabeculae are lost preferentially.28 If the static magnetic ®eld is applied parallel to the inferior± superior axis of the vertebrae, the horizontal trabeculae (i.e., those orthogonal to the ®eld) are polarized and, therefore, are expected to be the principal cause of the susceptibilityinduced line broadening. Chung et al. measured the mean spacing of horizontal trabeculae (i.e., the reciprocal of horizontal trabecular number density) in cadaver specimens of human lumbar trabecular bone after bone marrow removal and suspension of the bone in water, using NMR microscopy and digital image processing.29 They found a positive correlation between water proton R2', measured at 1.5 T, and mean number density of the horizontal trabeculae (r = 0.74, p < 0.0001). The critical role of the horizontal trabeculae in the vertebrae in conferring compressive strength is illustrated with the correlation between R2' measured with the polarizing ®eld parallel to the body axis and Young's modulus of elasticity (stiffness) for compressive loading. Figure 3(a) shows the relationship between anatomic axis, trabecular orientation, the orientation of the applied magnetic ®eld, and the direction of compressive loading. A strong association between stiffness and R2' exists over a wide range of values (r = 0.90), corresponding to trabecular bone of very different morphologic composition

,,  


 ,, ,

IMAGING OF TRABECULAR BONE

Horizontal trabeculae

5

Vertical trabeculae

Stress

5.1

Stress

Marrow spaces (water)

B0

(a)

600

Young’s modulus (MPa)

500 400 300 200 100 y = 53.3 + 19.9x, r = 0.91 0 0

10

20

30

R¢2 (s–1) (b)

Figure 3 (a) Cross-section through a cylindrical trabecular bone specimen (schematic) used for R2', structural, and stress analysis. The cylinder axis is parallel to the anatomic inferior±superior axis, aligned with the external ®eld polarizing predominantly horizontal trabeculae, which cause line broadening of the proton resonances in the marrow spaces. Compressive loading is applied along the cylinder axis. (b) Young's modulus of elasticity obtained from compression tests in 22 cylindrical specimens from the lumbar vertebral bodies of 16 human subjects aged 24±86 years, plotted as a function of R2' for the water protons in the intertrabecular spaces (r = 0.91, p < 0.0001). (Modi®ed from Chung et al.29)

Figure 3(b). From these data, it is inferred that a global measurement of R2' in trabecular bone is able to predict the compressive strength of this highly complicated structure. Subsequently, Jergas et al. evaluated the ability of R*2 to predict the elastic modulus for uniaxial loading in specimens of the human proximal tibia.30 They found correlations ranging from 0.87 to 0.95 in specimens in which the marrow was removed, but much weaker associations in another set of specimens where the measurements were conducted with the bone marrow intact.

5

IN VIVO QUANTITATIVE NMR OF TRABECULAR BONE Measurement and Data Analysis

Bone marrow has cellular (hematopoietic) and fatty components, with the relative fractions varying widely, depending on anatomical site and age. The major chemical constituents of the two types of marrow are water and fatty acid triglycerides. This chemical heterogeneity of bone marrow complicates in vivo measurement of T*2. A linewidth measurement by means of image-guided localized spectroscopy has the advantage of providing T*2 for each spectral component.31,32 Schick, in an excellent review on bone marrow NMR in vivo, described the potential of localized spectroscopy as a means to probe osteoporosis in the calcaneus, showing dramatic reductions in the linewidth of the CH2 lipid resonance in response to trabecular bone loss.32 Image-based (nonspectrally resolved) techniques, typically conducted by means of gradient echo33 or asymmetric spin echo techniques,34 have the advantage of providing information at multiple skeletal locations rapidly, allowing generation of maps of R*235 or R2'.36 By collecting an array of images with incrementally stepped time for inhomogeneity dephasing (gradient echo delay or echo offset), the pixel amplitudes can be ®tted to some model for signal decay. These methods are less sensitive to magnetic ®eld inhomogeneity arising from effects unrelated to susceptibility-induced gradients, since the ®eld across an imaging voxel of a few cubic millimeters is, in general, quite homogeneous. The presence of multiple chemically shifted constituents causes an amplitude modulation that has the characteristics of an interferogram. The latter can be expressed as the modulus of the vector sum of the individual phase-modulated spectral components:33,37 n  n n 1=2 XX X   I…t† ˆ I i ˆ I0i eÿt=T2i I0j eÿt=T2j cos…
!ij t† iˆ1

…8†

iˆ1 jˆ1

where I0i is the initial amplitude of the ith chemically shifted constituent,
!ij is the chemical shift difference in rad sÿ1 between nuclei i and j, t is the dephasing time (e.g., the echo time TE in a gradient echo), and the summation is over all spectral components n. Typically, the most abundant spectral components are those of the CH2 protons of fatty acids and of water, which are separated by chemical shift () = 3.35 ppm. It has been shown that I(t) can be ®tted to a two-component interferogram with T*2 and fat and water signal amplitudes as adjustable parameters and assuming T*2, fat & T*2, water.33 Multiparameter curve ®ts are hampered by the dif®culty of locating the global minimum, and require a relatively large number of images. One alternative is to suppress one spectral component,35 which sacri®ces some of the signal-to-noise ratio and is relatively sensitive to global magnetic ®eld inhomogeneity, demanding that the ®eld across the sample volume vary less than the chemical shift. Another approach to suppress the modulation is to sample the interferogram at the modulation frequency, ideally in such a manner that the two components are in phase with one another.38 This condition is satis®ed for sampling at multiples of the modulation period T = 2/ H0, which is 4.65 ms at 1.5 T ®eld strength.

6 IMAGING OF TRABECULAR BONE 0

ln (S/S0)

–1

–2

–3

–4 10

20

30

40

50

TE (ms) (b)

Figure 4 (a) First of a series of eight coronal gradient echo images for simultaneous measurement of T*2 in the hip and lumbar spine, obtained by collecting 128 data samples every 4.65 ms, from a single gradient echo train so as effectively to demodulate the signal (see text for details). (b) Plot of signal measured in the trochanter [see region indicated in (a)] versus echo time, obtained from ®ve successive scans. Solid lines are linear leastsquare ®ts, affording T*2 = 12.55 ‹ 0.38 ms

Another approach consists of deconvolving the amplitudemodulated signal with a reference signal.39 If the marrow composition is known (e.g., all fatty, such as in most of the appendicular skeleton), the reference signal could be derived from subcutaneous fat or marrow in the diaphysis, locations where no line broadening occurs. In the time domain, deconvolution can be achieved by simple division of the marrow signal at the location of interest (trabecular bone marrow) by the reference signal, resulting in a demodulated signal that ®ts to a decaying exponential of rate constant R2'. Dif®culties of this approach lie in the uncertainties of the marrow composition and large-scale static ®eld inhomogeneities, which may differ at the region of interest and the reference region. Rather than computing T*2 by ®tting the mean signal from a two-dimensional array of pixels (often called the `region of interest'), it is desirable to perform this calculation pixel by pixel for the generation of T*2 maps. This, however, requires that misregistration be minimal between acquisitions, which may, for example, be achieved with a multiple echo pulse sequence that collects gradient echoes of the same polarity at multiples of the chemical shift modulation period. The precision achievable in vivo with this technique is illustrated with the data obtained from ®ve separate scans in the same test subject in Figure 4. The data also show that chemical shift modulation is completely suppressed and that the decay is exponential to a high degree (Figure 4b). Excellent precision was also reported by Funke et al., who used a 16-echo gradient echo pulse sequence to give a coef®cient of variation of 2.5% from 11 successive scans, which included repositioning following each scan.40

Since variations in R2 (for example from variations in marrow composition) could mask effects resulting from changes in trabecular architecture or density, a direct measurement of R2' would be preferable. The asymmetric echo technique34,35 measures R2' but is inherently inef®cient as it requires separate image acquisitions for each time increment. Recently, Ma and Wehrli reported a new multislice pulse sequence capable of measuring both R2' and R2 in a single scan.36 The method, termed GESFIDE (gradient echo sampling of FID and echo), is based on sampling the descending and ascending portion of a Hahn echo with a train of gradient echoes. The transient signal before and after the phase-reversal rf pulse decays with rate constants R*2 = R2+R2' and Rÿ 2 = R2ÿR2', respectively. If, further, the time interval between successive gradient echoes is set equal to the fat±water chemical shift modulation period (see above), R*2 and Rÿ 2 can be determined by curve ®tting and R2 and R2' determined algebraically. Salient features of the method are its insensitivity to rf pulse imperfections, as well as its high precision and ef®ciency. Figure 5 shows the evolution of the GESFIDE signal in a region where R2' > R2, along with a computed R2' parameter image. 5.2

Dependence of R*2 and R2' on Image Voxel Size and Field Strength

For a perturber that is smaller than the imaging voxel but much larger than the molecular scale (also referred to as `mesoscopic' scale41) the susceptibility-induced line broadening should be independent of image voxel size. However, if the

IMAGING OF TRABECULAR BONE

(a)

7

(b)

6

'

Log (S/S0)

5

180… pulse

4

3

2 0

20

40

60

80

TE (ms)

Figure 5 The GESFIDE method. (a) Evolution of the gradient echo signal from the region of interest in the greater trochanter before and after the phase-reversal rf pulse (S, signal; points obtained from ®ve successive scans). Prior to the phase-reversal pulse, the signal evolves with a time constant R*2 = R2+R2', subsequently with Rÿ 2 = R2+R2'. The change in the sign of the slope indicates that R2' >>R2. (b) Computed R2' map. Note the high intensity for structures pertaining to trabecular bone, consistent with enhanced R2' as a result of susceptibility-induced line broadening. (Modi®ed from Ma and Wehrli36)

voxel size decreases below the typical range of gradients induced by the ®eld pertubing trabeculae, then the likelihood of this voxel falling in a region suf®ciently removed from the ®eld gradients induced by trabeculae increases. As a consequence, the mean T2' is expected to increase with decreasing pixel size. Majumdar and Genant studied this effect in trabecular bone of various densities.35 They found that the T2' histogram becomes wider and more asymmetric with decreasing pixel size. However, if the voxel size is large relative to the range of the gradients, the value of T2' becomes independent of voxel size, which is the case for pixels on the order of 1.5±2 mm. The distribution of T2' as a function of image resolution is shown in Figure 6. The induced magnetic ®eld is proportional to the polarizing static ®eld. In the absence of diffusion (static dephasing regime) one would, therefore, expect a linear relationship between trabecular bone marrow R2' and ®eld strength. Parizel et al. reported the ®eld strength dependence of R*2 from measurement of the gradient echo signal decay at 1.5, 1.0, and 0.2 T in vertebral bone marrow in a single volunteer42 and found R*2 to increase with ®eld strength. More recently, Song et al. performed a detailed study of the ®eld strength dependence of R2 and R2' in the calcaneus at 1.5 and 4 T by means of the GESFIDE technique.43 They found R2' to scale nearly proportionally with ®eld strength while R2 was almost ®eld-strength invariant. The data, which show that at 4 T R2' accounts for nearly 90% of the total relaxation rate (R*2 = R2+R2'), are summarized in Table 1.

5.3

Clinical Studies in Patients with Osteoporosis

The most likely association involving R*2 is with bone marrow density (BMD). In fact, the theory developed by

Figure 6 Smoothed T2' histograms obtained from T2' maps computed from axial images in the epiphysis of the distal femur, a site of dense trabeculation. As image resolution increases (smaller pixel size), the histogram broadens and becomes skewed. Means are found to vary between 41.7 ms at 0.32 mm pixels size and 25.7 ms at 1.25 mm pixel size. (Modi®ed, with permission, from Majumdar and Genant35)

8 IMAGING OF TRABECULAR BONE Table 1 Field dependence of cancellous bone marrow relaxation rates Field (T) 1.5 4.0

Relaxation rate ‹SD (sÿ1)a R2 R2' 16.4 ‹ 1.6 19.2 ‹ 3.4

65.4 ‹ 11.2 178.4 ‹ 37.6

a

SD, standard deviation from values for ®ve subjects. R2' (at 4 T)/R2' (at 1.5 T) = 2.73. Source: Song et al.43

Yablonskiy and Haacke predicts a linear relationship between R2' and the volume fraction of the perturber.25 A possible relationship between these parameters was suggested by the observed T*2 shortening in the distal femur from the diaphysis (lowest trabeculation) toward the metaphysis and epiphysis of the bone (highest trabeculation).31 Majumdar et al. determined R2' in intact specimens of human vertebral trabecular bone after bone marrow removal and suspending the bone in saline.44 The measurements afforded a linear correlation between R2' and BMD (in mg cmÿ3), obtained from QCT, with a slope of 0.20 ‹ 0.02 sÿ1 mgÿ1 cm3 (r = 0.92). The same group extended these studies in vivo in normal volunteers to the distal radius and proximal tibia.35 Con®rming earlier work, they found both BMD and R2' to depend on the anatomical site of measurement. In excellent agreement with in vitro data, they measured 0.20 ‹ 0.01 sÿ1 mgÿ1 cm3 (r = 0.88) for the combined data from both anatomic sites. Since inception of the T*2 method, several groups have evaluated its potential for assessing osteoporosis and have compared it with other modalities, notably DEXA and QCT.33,40,45±47 An early pilot study on a small group of patients with clinically established osteoporosis of the spine (n = 12) and an equal number of control subjects showed the former to have signi®cantly prolonged bone marrow T*2 in lumbar vertebra L5.33 A larger follow-up study was designed to explore whether image-based measurements of T*2 could provide an index of the integrity of trabecular bone as a possible criterion for predicting fracture risk. The value of R*2 was measured in 146 non-black women at 1.5 T ®eld strength by means of ®ve successive gradient echoes, spaced 4.6 ms apart to minimize fat±water chemical shift modulation. Data were ®tted to an exponential model and data sets with p > 0.05 for the Pearson correlation rejected. The control population (n = 77, mean age 46.6 ‹ 14.9 years) consisted of women with mean spinal BMD >0.9 g cmÿ2 (DEXA) or >90 mg cmÿ3 (QCT), and no vertebral deformities. The patient population (n = 59; mean age 59.7 ‹ 10.2 years) was made up of women with osteoporosis of the spine, exhibiting at least one radiographic deformity of the thoracic vertebrae and/or BMD below the cutoff for controls. The extent of deformity was determined as a mean deformity index, DIav. The value of R*2 was signi®cantly lower in osteoporosis for all L vertebrae (p < 0.001), except for L1. The best discriminator was the average of L3±L5 (R*2 av) for which means and standard errors obtained were 64.8 ‹ 1.2 sÿ1 and 53.4 ‹ 1.2 sÿ1 in controls and osteoporotics, respectively (p < 0.0001). Both R*2 and BMD correlated with DIav, the correlation with R*2 being slightly stronger (r = 0.40; p < 0.0005, versus r = 0.36; p < 0.001). Finally, R*2 av was signi®cantly cor-

related with mean BMD (r = 0.54; p < 0.0001, slope = 31.4 sÿ1 g cmÿ2, p < 0.0001). Overall, these ®ndings corroborate the results of the prior study in that subjects with osteoporosis have lower R*2 values of their vertebral marrow and show that MR may have the potential to distinguish patients with fractures from those without this condition. Nevertheless, the results fell short of demonstrating MRI's superiority over bone densitometry. A more recent study from the same laboratory, undertaken with more advanced methodology, involving measurements of R2' (rather than R*2) at multiple skeletal sites, indicates the complementary nature of this parameter to BMD.48 In this work, the rate constants R2' and R2 and the marrow fat fraction were measured in the lumbar vertebrae and proximal femur by the GESFIDE method.36 Sixteen gradient echoes were collected, eight each before and after the phase-reversal pulse, at TE 4.6, 6.9, 9.2, 11.5, 13.8, 18.4, 23.0, and 27.6 ms at 57.0, 61.6, 66.2, 70.8, 73.1, 75.4, 77.7, and 80 ms, the latter being an rf echo. Fat and water signal amplitudes were computed from echoes 1±3 by three-point Dixon processing,49 and from these, the volume fractions of the two constituents. The value of R2' was moderately (positively) correlated with BMD at all sites (r = 0.46±0.69; p < 0.0001), albeit with different slopes, indicative of the different trabecular orientation relative to the static ®eld at the various anatomical locations. For example, in the femoral neck, the slope was twice that found for the average R2' from lumbar vertebrae 2±4 (38.0 sÿ1 g cmÿ2 versus 19.2 sÿ1 g cmÿ2). The value of R2' classi®ed the subjects well at all sites, but the strength of the discrimination was greater at the femoral sites, a ®nding that agrees with a small patient study by Machann et al.,50 and suggests that skeletal sites rich in yellow marrow (such as the calcaneus) are better predictors of osteoporosis than measurements in the vertebrae. This observation can be understood when the different susceptibilities of fat (i.e., yellow marrow) and water (i.e., hematopoietic marrow) are considered. It is well known that fat is less diamagnetic than water (see, for example, Hopkins and Wehrli18) thus resulting in a greater absolute susceptibility difference
 between bone and bone marrow, which increases the sensitivity of R2' to variations in bone volume fraction [Equation (7)]. In fact, when the spine R2' values are normalized by correcting for varying bone marrow composition, achieved by computing R2' as it would be observed for an all yellow marrow composition, the slope of the R2' versus BMD correlation increased from 19.2 to 28.4 sÿ1 g cmÿ2. Finally, the value of R2' in this study was found to be predictive of vertebral fracture status. The latter was de®ned by measuring vertebral deformities in the thoracic and lumbar spine in terms of standard deviations from normalcy and by selecting a threshold beyond which a fracture was considered present. When only a single parameter was included in the logistic regression, DEXA BMD was found to predict fracture status better than R2', with Ward's triangle exhibiting the strongest correlation (r2 = 0.48). Further, the femoral sites were more predictive than the lumbar spine, a ®nding that applies to both R2' and BMD. However, combining MR and BMD signi®cantly improved prediction. The strongest association was found for the combination of R2' measured in the greater trochanter and BMD at the Ward's triangle, affording a 30% increase in the strength of the correlation relative to BMD alone (r2 = 0.62). By contrast, the combination of multiple

IMAGING OF TRABECULAR BONE

Figure 7 Images at microscopic dimension showing trabecular structure: (a)±(c) orthogonal views obtained from a 643 array of 3D spin echo array of images acquired at 9.4 T on a specimen of trabecular bone from bovine tibia (114 m  114 m  139 m voxel size); (a) transverse and (b), (c) longitudinal sections. Note the preferential orientation of the trabeculae along the inferior±superior direction. (d) Shaded surface display of the same array of data, resolution-enhanced by means of a subvoxel tissue classi®cation technique using Bayesian segmentation60

BMD sites did not yield stronger associations. These data are promising in that they underscore the complementary nature of R2' and the potential role of structure in affecting the trabecular bone's mechanical competence.

6 HIGH-RESOLUTION IMAGING OF TRABECULAR BONE STRUCTURE 6.1 In Vitro Cancellous Bone NMR Microscopy Whereas the measurement of the induced magnetic ®eld inhomogeneity provides structural information indirectly, highresolution MRI at microscopic dimensions has the potential for nondestructive mapping of three-dimensional trabecular morphology, as an alternative to conventional microscopy from sections51 or tomographic X-ray microscopy.52±54 Bone is well suited for imaging by NMR since it appears with background intensity and, therefore, provides excellent contrast with marrow, which has high signal intensity. Ideally, the image voxel size is smaller than a typical structural element (i.e., a trabecula), in which case partial volume blurring is minimal and the resulting histogram is bimodal, allowing segmentation (generation of a binary image) by setting the intensity threshold midway between the marrow peak and background. Clearly, the voxel size needed to satisfy this requirement depends on the trabecular width, which is species dependent. Bovine trabeculae are 200±300 m thick, those in humans between 100 and 200 m, whereas the trabecular thickness in rats is only 50±70 m, consequently demanding considerably higher resolution. Most of the work reported so far has been conducted on small-bore microimaging systems at 300 or 400 MHz proton

9

resonance in human cadaver or biopsy specimens55±58 or animals, at resolutions ranging from 25 to 120 m. Ex vivo, the quality of the images is optimized by substituting the marrow with gadolinium-doped water as a means to optimize the signal-to-noise ratio. The background gradients from the susceptibility-induced ®elds can cause artifacts in the form of signal loss from intravoxel phase dispersion [Equation (1)]. Since these phase losses are recoverableÐassuming diffusion to be negligibleÐspin echo detection is advantageous. Alternatively, the dephasing time should be minimal; this can, for example, be achieved with projection±reconstruction techniques, as applied in microimaging of lung parenchyma.59 At marginal resolution the accuracy of the 3D structures derived from NMR images can be improved with such techniques as subvoxel tissue classi®cation.60 Based on a statistical model for the noise and partial volume averaging, the number of subvoxels containing marrow can be determined and the most likely spatial arrangement ascertained using probabilistic arguments for the interaction between adjacent subvoxels. Chung et al. developed algorithms for the measurement of the structural parameters of interest (bone volume fraction, mean trabecular thickness, and mean trabecular plate separation55 based on two-dimensional images. These techniques, however, are merely the digital imaging adaptations of stereology, the methodology practiced by the histomorphometrist who performs measurements on anatomic sections which are then extrapolated to the third dimension.61 However, trabecular bone is inherently three-dimensional and anisotropic. Figure 7 shows three orthogonal images from bovine trabecular bone illustrating the bone's anisotropic structure, along with a surface projection image enhanced by subvoxel classi®cation.60 6.2

Relationship between Architecture and Strength

Knowledge of the three-dimensional architecture of cancellous bone allows one of the key questions to be addressed, i.e., whether structure is predictive of the bone's strength. While still controversial, it is known that apparent density (essentially bone volume fraction, i.e., the amount of bone present per unit volume) predicts anywhere from 40±80% of Young's modulus (and thus ultimate strength); the remainder is generally attributed to architecture.62 To investigate whether architecture is predictive of the modulus of elasticity for compressive loading, Hwang et al. acquired three-dimensional images at 78 m isotropic resolution of cadaver specimens cored from the ultradistal radius, after the cores were tested nondestructively in compression along the bone's anatomic axis. Rather than segmenting the images, bone volume fraction (BVF) images were generated by ®tting the histogram to a two-peaked noiseless histogram convolved with Rician noise, using maximum likelihood methods. In this manner the true BVF can be found for each pixel as the probability of ®nding bone at that location. This idea was then extended to two-point probabilities, for example the probability of ®nding bone at two neighboring locations xi and xi+n along a row of voxels, P(xi, xi+n) = BVF(xI)BVF(xi+n). Averaging the two-point probabilities over all locations yields the spatial autocorrelation function (AFC). Because of the quasiregular nature of the trabecular lattice, the AFC has a maximum at the average spacing between trabeculae along that direction. This concept can be extended and other parameters characterizing

10 IMAGING OF TRABECULAR BONE Male 76 years

Male 80 years

Male 53 years

Bone volume fraction

0.16

0.14

0.12

Young’s modulus (MPa)

316

676

543

Tranverse

Longitudinal

Three-dimensional

Figure 8 NMR images of three samples of trabecular bone illustrating its structural variability. Note that the sample from the 76-year-old man has the lowest modulus in spite of having the highest bone volume fraction. However, the bone of the 80- and 50-year-old donors is considerably more tubular, which explains their greater Young's modulus.

the trabecular network de®ned. One useful parameter is tubularity, which quanti®es how `tubular' the bone is along its anatomic axis, which is the direction along the radius (z). Tubularity is given as AFCz(1)/AFC(0), where AFCz(1) is the average of the product of BVF in corresponding voxels of successive slices. The signi®cance of this parameter in determining cancellous bone strength is illustrated with Figure 8 showing cross-sectional and projection images of three radius specimens of varying BVF. It is noteworthy that the specimen with the highest BVF is only half as strong as the two other specimens that are both more tubular. A detailed analysis on specimens from 23 cadavers of widely varying BVF, tubularity, and longitudinal spacing (spacing between trabeculae orthogonal to the direction of loading) jointly best predicted Young's modulus, accounting for over 90% of its variance (as opposed to BVF, which explained only 50%). Clearly, a perfectly tubular structure could not fail by loading in the direction of these trabeculae. However, as the deviation from such an ideal geometry decreases, the transverse trabeculae, acting as cross-ties, become increasingly important. 6.3 In Vivo Microimaging In vivo imaging of the microarchitecture of cancellous bone at voxel sizes suf®cient to resolve individual trabeculae is considerably more challenging than measurements on small

specimens in vitro. The main dif®culty is to achieve suf®cient signal to noise ratio in scan times tolerated by the patient, typically on the order of 10 min. A second problem is subject movement, which, even on a minute scale on the order of a pixel (100±200 m), can cause blurring and artifacts, precluding derivation of accurate structural parameters. Restraining the portion of the body being scanned is usually not suf®cient to prevent motional blurring. However, the incorporation of navigator echoes into the pulse sequence proves to be effective.63 Additional echoes, which are not phase-encoded, generate projections of the object from which the displacement can be measured and a commensurate phase correction applied to the k-space data. The effect of alternate navigator echoes incorporated into a three-dimensional spin echo sequence was found to eliminate motion degradation effectively (Figure 9).43 The third problem is the considerably lower resolution achievable in vivo, which, while suf®cient to visualize the trabecular architecture, demands more sophisticated methods for image restoration and analysis. In spite of these dif®culties, a growing body of literature has accumulated during the past few years, highlights the potential of in vivo micro-MRI to probe bone structure.64±68 The image voxel sizes reported for cancellous bone imaging in humans range from 210ÿ5 mm3 in the distal phalanx of the middle ®nger69 to 210ÿ2 mm3 in the calcaneus.68 At these resolutions, the histogram is monomodal (instead of

IMAGING OF TRABECULAR BONE

11

Figure 9 (a) FLASE (fast large-angle spin echo) sequence with navigator echoes alternating between frequency and phase-encoding axes (137 m3137 m3350 m3 voxel size). The zeroeth moment of all three gradients are nulled before the navigators are acquired. Also, to maintain steady state, the navigator gradients are refocused after collecting motion data. (b) Image of the distal radius from a 56-year-old female at this voxel size, before motion correction. (c) Image after motion correction. (With permission from Song et al.43)

bimodal as it is at higher resolution) and, therefore, does not provide simple criteria for segmenting the images into bone and bone marrow. The limitations imposed by marginal resolution to segment the images and derive structural parameters by image processing has been addressed in various ways. Majumdar et al. set the threshold at half the peak maximum (observed for cortical bone) in the grayscale-inverted histogram.65 In recent work on the calcaneus, Link et al. compared structural measures in postmenopausal subjects who had suffered hip fractures with those in age-matched controls.68 They found that both apparent trabecular bone volume and apparent trabecular separation were stronger predictors of hip fracture than BMD of the proximal femur. Gordon et al. applied an adaptive threshold to eliminate errors from intensity variations caused by the inhomogeneous reception pro®le of the surface coil, by thresholding the image against the low-pass ®ltered version of itself, followed by a second threshold at 50% of the maximum.66 From the threshold and subsequent skeleton images they determined hole size and a connectivity index and

found these parameters to be correlated with subject age (positively and negatively). Whereas the above methods cannot retrieve the true BVF, they, nevertheless, can provide parameters that can effectively characterize the trabecular network and provide clinically relevant information. True BVF images have been obtained in vivo in the distal radius by means of a histogram deconvolution technique.70 In brief, a model histogram is convolved with Rician noise (the type of noise encountered in magnitude NMR images71) and the result compared with the observed histogram. The resulting error then serves as input to improve the noiseless histogram, and the process is repeated until the error falls below a predetermined level. Finally, the actual noiseless image is computed using both intensity and connectivity arguments. Wehrli et al. obtained BVF images in this manner in the distal radius of a small cohort of patients with and without vertebral fractures.67 The parameters that proved to be successful in predicting the elastic modulus in vitro were also evaluated to explore whether architecture is associated with the

12 IMAGING OF TRABECULAR BONE degree of vertebral deformities, measured in terms of a fracture index, Dfract. Whereas none of the individual structural measures was found to correlate with the Dfract, a highly signi®cant relationship (R = 0.78; p = 0.0016) was found between Dfract and a function of tubularity and longitudinal spacing. These data further emphasize the role of architecture in determining trabecular bone's resistance to fracture, and highlight the prospects of in vitro micro-MRI as a noninvasive modality to assess trabecular microstructure.

7 RELATED ARTICLES Lung and Mediastinum: A Discussion of the Relevant NMR Physics; Susceptibility and Diffusion Effects in NMR Microscopy; Susceptibility Effects in Whole Body Experiments.

8 REFERENCES 1. J. Wolff, `Das Gesetz der Transformation der Knochen', A. Hirschwald, Berlin, 1892. 2. H. Roesler, J. Biomech., 1987, 20, 1025. 3. B. Riggs and L. Melton, `Osteoporosis', Raven Press, New York, 1988. 4. C. E. Brown, J. R. Allaway, K. L. Brown, and H. Battocletti, Clin. Chem., 1987, 33, 227. 5. J. L. Ackerman, D. P. Raleigh, and M. J. Glimcher, Magn. Reson. Med., 1992, 25, 1. 6. J. Moore, L. Garrido, and J. Ackerman, Magn. Reson. Med., 1995, 33, 293. 7. Y. Wu, M. Glimcher, C. Rey, and J. Ackerman, J. Mol. Biol., 1994, 244, 423. 8. Y. Wu, J. L. Ackerman, D. A. Chesler, J. Li, R. M. Neer, J. Wang, and M. J. Glimcher, Calcif. Tissue Int., 1998, 62, 512. 9. J. Glasel and K. Lee, J. Am. Chem. Soc., 1974, 96, 970. 10. C. A. Davis, H. K. Genant, and J. S. Dunham, Invest. Radiol., 1986, 21, 472. 11. H. Rosenthal, K. R. Thulborn, D. I. Rosenthal, S. H. Kim, and B. R. Rosen, Invest. Radiol., 1990, 25, 173. 12. F. W. Wehrli, J. C. Ford, D. A. Gusnard, and J. Listerud, Proc. VIIIth Annu. Mtg Soc. Magn. Reson. Med., Amsterdam, 1989, 1, 217. 13. D. C. Ailion, T. A. Case, D. Blatter, A. H. Morris, A. Cutillo, C. H. Durney, and S. A. Johnson, Bull. Magn. Reson., 1984, 6, 130. 14. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. Goldberg, R. Weisskoff, B. Poncelet, D. N. Kennedy, B. Hoppel, M. Cohen, R. Turner, H. Cheng, T. Brady, and B. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 15. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellerman, S. G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1992, 89, 5951. 16. P. C. Lauterbur, B. V. Kaufman, and M. K. Crawford, in `Biomolecular Structure and Function', ed. P. F. Agris. Academic Press, New York, 1978, p. 343. 17. S. Chu, Y. Xu, J. Balschi, and C. Springer, Magn. Reson. Med., 1990, 13, 239. 18. J. A. Hopkins and F. W. Wehrli, Magn. Reson. Med., 1997, 37, 494. 19. A. Morrish, `The Physical Principles of Magnetism', Robert E. Krieger, Malabar, FL, 1983. 20. J. C. Ford, F. W. Wehrli, and H. W. Chung, Magn. Reson. Med., 1993, 30, 373. 21. L. J. Gibson, J. Biomech., 1985, 18, 317.

22. K. Engelke, S. Majumdar, and H. K. Genant, Magn. Reson. Med., 1994, 31, 380. 23. S. Hwang and F. Wehrli, J. Magn. Reson. B, 1995, 109, 126. 24. H. Chung and F. W. Wehrli, Proc. XIIth Annu. Mtg Soc. Magn. Reson. Med., New York, 1993, 1, 138. 25. D. A. Yablonskiy and E. M. Haacke, Magn. Reson. Med., 1994, 32, 749. 26. D. A. Yablonskiy, W. R. Reinus, E. M. Haacke, and H. Stark, Magn. Reson. Med., 1996, 37, 214. 27. K. Selby, S. Majumdar, D. C. Newitt, and H. K. Genant, J. Magn. Reson. Imaging, 1996, 6, 549. 28. L. Mosekilde, Bone, 1990, 11, 67. 29. H. Chung, F. W. Wehrli, J. L. Williams, and S. D. Kugelmass, Proc. Natl. Acad. Sci. USA, 1993, 90, 10 250. 30. M. Jergas, S. Majumdar, J. Keyak, I. Lee, D. Newitt, S. Grampp, H. Skinner, and H. Genant, J. Comput. Assist. Tomogr., 1995, 19, 472. 31. J. C. Ford and F. W. Wehrli, Magn. Reson. Med., 1991, 17, 543. 32. F. Schick, Prog. Nucl. Magn. Reson. Spect., 1996, 29, 169. 33. F. W. Wehrli, J. C. Ford, M. Attie, H. Y. Kressel, and F. S. Kaplan, Radiology, 1991, 179, 615. 34. G. L. Wismer, R. B. Buxton, B. R. Rosen, C. Fisel, R. Oot, T. Brady, and K. Davis, J. Comput. Assist. Tomogr., 1988, 12, 259. 35. S. Majumdar and H. K. Genant, JMRI, 1992, 2, 209. 36. J. Ma and F. W. Wehrli, J. Magn. Reson. B, 1996, 111, 61. 37. F. W. Wehrli, T. G. Perkins, A. Shimakawa, and F. Roberts, Magn. Reson. Imag., 1987, 5, 157. 38. J. C. Ford and F. W. Wehrli, JMRI, 1992, 2(P), 103. 39. F. W. Wehrli, J. Ma, J. A. Hopkins, and H. K. Song, J. Magn. Reson., 1998, 131, 61. 40. M. Funke, H. Bruhn, R. Vosshenrich, O. Rudolph, and E. Grabbe, Fortschr. Geb. Rontgen. (Neuen Bildgeb Verfahr), 1994, 161, 58. 41. D. A. Yablonskiy, Magn. Reson. Med., 1997, 39, 417. 42. P. M. Parizel, B. van Riet, B. A. van Hasselt, J. W. Van Goethem, L. van den Hauwe, H. A. Dijkstra, P. J. van Wiechem, and A. M. De Schepper, J. Comput. Assist. Tomogr., 1995, 19, 465. 43. H. K. Song, F. W. Wehrli, and J. Ma, JMRI, 1997, 7, 382. 44. S. Majumdar, D. Thomasson, A. Shimakawa, and H. K. Genant, Magn. Reson. Med., 1991, 22, 111. 45. H. Sugimoto, T. Kimura, and T. Ohsawa, Invest. Radiol., 1993, 28, 208. 46. F. W. Wehrli, J. C. Ford, and J. G. Haddad, Radiology, 1995, 196, 631. 47. S. Grampp, S. Majumdar, M. Jergas, D. Newitt, P. Lang, and H. K. Genant, Radiology, 1996, 198, 213. 48. F. W. Wehrli, J. A. Hopkins, H. K. Song, S. N. Hwang, J. D. Haddad, and P. J. Snyder, Proc. VIth Annu. Mtg (Int.) Soc. Magn. Reson. Med., Sydney, 1998, p. 456. 49. G. H. Glover and E. Schneider, Magn. Reson. Med., 1991, 18, 371. 50. J. Machann, F. Schick, D. Seitz, et al, Proc. IVth Annu. Mtg (Int.) Soc. Magn. Reson. Med., New York, 1996, 2, 1093. 51. A. Odgaard, K. Andersen, F. Melsen, and H. J. Gundersen, J. Microsc., 1990, 159, 335. 52. L. A. Feldkamp. S. A. Goldstein, A. M. Par®tt, G. Jesion, and M. Kleerekoper, J. Bone Miner. Res., 1989, 4, 3. 53. J. H. Kinney, N. E. Lane, and D. L. Haupt, J. Bone Miner. Res., 1995, 10, 264. 54. P. RuÈegsegger, B. Koller, and R. Muller, Calcif. Tissue Int., 1996, 58, 24. 55. H. W. Chung, F. W. Wehrli, J. L. Williams, S. D. Kugelmass, and S. L. Wehrli, Proc. Natl. Acad. Sci. USA, 1995, 10, 803. 56. H. W. Chung, F. W. Wehrli, J. L. Williams, and S. L. Wehrli, J. Bone Miner. Res., 1995, 10, 1452.

IMAGING OF TRABECULAR BONE 57. S. N. Hwang, F. W. Wehrli, and J. L. Williams, Med. Phys., 1997, 24, 1255. 58. M. Wessels, R. P. Mason, P. P. Antich, J. E. Zerwekh, and C. Y. Pak, Med. Phys., 1997, 24, 1409. 59. S. L. Gewalt, G. H. Glover, L. W. Hedlund, G. P. Cofer, J. R. MacFall, and G. A. Johnson, Magn. Reson. Med., 1993, 29, 99. 60. Z. Wu, H. Chung, and F. W. Wehrli, Magn. Reson. Med., 1994, 31, 302. 61. A. M. Par®tt, in `Bone Histomorphometry: Techniques and Interpretation', ed. R. R. Recker, CRC Press, Boca Raton, FL, 1981, p. 53. 62. M. J. Ciarelli, S. A. Goldstein, J. L. Kuhn, D. D. Cody, and M. B. Brown, J. Orthop. Res., 1991, 9, 674. 63. R. L. Ehman and J. P. Felmlee, Radiology, 1989, 173, 255. 64. S. Majumdar, D. Newitt, M. Jergas, A. Gies, E. Chiu, D. Osman, J. Keltner, J. Keyak, and H. Genant, Bone, 1995, 17, 417. 65. S. Majumdar, H. K. Genant, S. Grampp, D. C. Newitt, V.-H. Truong, J. C. Lin, and A. Mathur, J. Bone Miner. Res., 1997, 12, 111. 66. C. L. Gordon, C. E. Webber, N. Christoforou, and C. Nahmias, Med. Phys., 1997, 24, 585. 67. F. W. Wehrli, S. N. Hwang, J. Ma, H. K. Song, J. C. Ford, and J. G. Haddad, Radiology, 1998, 207, 833; erratum in Radiology, 1998, 206, 347.

13

68. T. M. Link, S. Majumdar, P. Augat, J. C. Lin, D. Newitt, Y. Lu, N. E. Lane, and H. K. Genant, J. Bone Miner. Res., 1998, 13, 1175. 69. H. Jara, F. W. Wehrli, H. Chung, and J. C. Ford, Magn. Reson. Med., 1993, 29, 528. 70. S. N. Hwang and F. W. Wehrli, Int. J. Imaging Syst. Technol., 1999, 10, 186. 71. H. Gubdjartsson and S. Patz, Magn. Reson. Med., 1995, 34, 910.

Biographical Sketch Felix W. Wehrli. b 1941. M.S., 1967, Ph.D., 1970, chemistry, Swiss Federal Institute of Technology, Switzerland. NMR application scientist, Varian AG, 1970±79; Executive Vice President, Bruker Instruments, Billerica, 1979±82; NMR Application Manager, General Electric Medical Systems, Milwaukee, 1982±88. Currently Professor of Radiologic Science and Biophysics, University of Pennsylvania Medical School. Editor-in-Chief of Magnetic Resonance in Medicine 1991±present. Over 100 publications. Current research specialty: NMR imaging of connective tissues, biomaterials, speci®cally trabecular bone.

MRI OF MUSCULOSKELETAL NEOPLASMS

MRI of Musculoskeletal Neoplasms Johan L. Bloem Leiden University Medical Center, The Netherlands

1

INTRODUCTION

Staging of disease is the single most important reason for performing MRI in patients with musculoskeletal tumors. Other indications for MRI in these patients are detection, speci®c diagnosis, chemotherapy monitoring, and detection of recurrence. The last two indications are becoming increasingly important. After a short discussion about technique, this article deals with clinical applications of MRI in patients with primary musculoskeletal tumors, tumor-like lesions and metastases.

2

TECHNIQUE

When a bone lesion is potentially malignant, imaging studies (to allow local staging) and biopsy are needed. Imaging studies are performed prior to histologic biopsy, because they can be used to plan the biopsy procedure. Furthermore, biopsy induces reactive changes (edema, hemorrhage), which interfere with accuracy of staging. Cytologic biopsy is a minimally invasive procedure that can be useful in soft tissue tumors, prior to MRI, to differentiate between broad categories such as lymphoma or sarcoma metastasis. Conventional radiographs have to be present when MRI is planned and executed. This should be a golden rule and will avoid both major disasters and silly mistakes. High- or low-frequency (low or high ®eld) magnetic resonance (MR) systems can be used for imaging musculoskeletal tumors. It is possible to stage bone tumors accurately at 0.5 T. Whenever possible, dedicated coils should be used in order to increase spatial resolution. It is important to use both T1- and T2-weighted sequences. T1-weighted sequences are used for intramedullary staging and T2-weighted sequences are used for de®ning soft tissue extension and cortical involvement. A large number of pulse sequences, such as STIR (short  inversion±recovery), gradient echo (GE) imaging, fast or turbo spin echo (TSE), chemical shift imaging, magnetization transfer contrast added to various pulse sequences, etc., are available. For imaging of musculoskeletal tumors, conventional spin echo or TSE sequences are best used to obtain T1-weighted images. For T2-weighting conventional spin echo sequences, or preferably TSE sequences, are used. Contrast between tumor and normal tissue, especially fat containing tissue, is greatly enhanced by combining TSE with fat selective presaturation. It is important to make sure that enough T2 weighting is achieved. At 0.5 T, a TSE sequence with a TR of 3000±4000 ms and an effective TE of 150 ms has proven to be as good as the conventional T2weighted spin echo sequences in our practice. When fat sup-

1

pression is used, the TE should be shortened. Contrast between tumor and muscle on T2-weighted SE or TSE images is superior to that on native T*2-proton density weighted, or enhanced GE images. STIR±TSE sequences are very sensitive, but the speci®city and low signal-to-noise ratio are low. When T1 or T2-weighted images are mentioned in this article, I refer to both conventional spin echo and TSE sequences. Routinely, multiple imaging planes are used: the transverse plane (T2-weighted) for soft tissue extension, and sagittal or coronal planes (T1-weighted) for intraosseous extension (Figure 1). The sagittal and coronal planes can be angulated to be parallel to the long bones. Additional planes are used when necessary, for instance the transverse plane for intraosseous staging to rule out partial volume effects of cortical bone which may occur in the sagittal and coronal plane in thin tubular bones in children. Slice thickness varies between 2 and 10 mm and the number of excitations will be 1±4, depending on various parameters such as slice thickness, TR, TE, type of coil, and ®eld strength used. Echo train length will vary between 3 and 8. Gadolinium enhanced studies are performed following intravenous injection of 0.1±0.2 mmol Gd chelate/kg body weight with a T1 spin echo or TSE technique and dynamic magnetization prepared GE imaging. The T1-weighted sequences are combined with fat selective presaturation. It is not advisable to combine STIR sequences with gadolinium, since suppression of signal from enhancing tissue is possible. Out-of-phase GE images may further enhance contrast as compared with inphase GE images. Gadolinium can be used in patients with poor renal function, and can be removed from the body with dialysis.

3 MRI CHARACTERISTICS Bone marrow and fat exhibit a high signal intensity (white) on T1-weighted sequences, whereas cortex, ®brous cartilage, and ligaments are seen as signal void (black) structures. Hyaline cartilage and muscle have an intermediate signal intensity (gray). Relative signal intensities of muscle, fat, and bone marrow decrease somewhat on T2-weighted spin echo sequences. This is in contrast to the increase of signal intensity of structures containing a large amount of water, such as hyaline cartilage. Fat displays a high signal intensity on TSE images. Distribution of areas with abnormal signal intensity is important, not only for differentiating tumors and tumor-like lesions, but also in identifying normal variants. The low to intermediate signal intensity areas of red bone marrow in the axial skeleton and femur diaphysis in adults and also in the periphery of the skeleton in adolescents and children on T1weighted images may be confusing.1 On STIR sequences, the high signal intensity of red marrow is easily differentiated from the absence of signal from the nulled yellow marrow. Low signal intensity areas within bone marrow on GE images are due to susceptibility effects secondary to the presence of bone trabeculae and cortex, and iron in red bone marrow. Like most pathologic tissues, musculoskeletal tumors usually have prolonged T1 and T2 relaxation times (Figure 1). Therefore, osteolytic tumors have a relatively high signal intensity on T2-weighted images and a relatively low signal intensity on T1-weighted images. Because of their low spin

2 MRI OF MUSCULOSKELETAL NEOPLASMS

Figure 1 Osteosarcoma in an 18-year-old male imaged with a 0.5 T MRI system. (a) Coronal T1-weighted (600/20) MR image displays the intraosseous tumor margins very well. Medial soft tissue extension and periosteal reaction are also depicted. The tumor has an inhomogeneous low to intermediate signal intensity. (b) Axial T2-weighted TSE (3570/150) MR image. Contrast between the high signal intensity of viable tumor and the low signal intensity of muscle is excellent. However, both tumor and fat have a similar high signal intensity. Ossi®ed periosteal reaction and ossi®ed tumor display a signal void

density and short T2 relaxation time, osteosclerotic components or calci®cations within the tumor have a low signal intensity on both T1- and T2-weighted images. Small calci®cations detected by plain radiographs or computerized tomography (CT) often cannot be detected by MRI. Liquefaction or lacunae caused by necrosis within a tumor can often be identi®ed because the T1 and T2 relaxation times are even longer than those of viable tumor. Cortical bone and periosteal reaction can be evaluated not only with CT, but also with MRI. Normal cortical bone is represented as a signal void area, whereas intracortical lesions almost invariably show areas of increased signal intensity relative to cortex, on T1-weighted, gadolinium enhanced and T2weighted images. Likewise, mineralized periosteal reaction is seen as a solid or layered signal void area, whereas the nonmineralized cellular cambium layer of the periosteum has a high signal intensity on T2-weighted images. Only in selected cases does MRI allow a speci®c diagnosis to be made. The inhomogeneity of the tumor consisting of viable tumor with different histologic components, necrosis, hemorrhage, and reactive changes make it impossible to differentiate histologic types by relaxation times. MRI increases

speci®city only when unusual features such as high signal intensity on T1-weighted images, low signal intensity on T2weighted images, or speci®c morphology or enhancement patterns are observed. Examples of lesions with a high signal intensity on T1-weighted sequences are fat and subacute (one week to several months old) hematoma. The signal intensity of hematoma depends on the sequential degradation of hemoglobin, cell lysis, and ®eld strength of the MR system.2 The high signal intensity representing methemoglobin is initially seen in the periphery, and subsequently extends towards the center of the hematoma. Interstitial hemorrhage is more diffuse than hematoma and is accompanied by edema resulting in aspeci®c prolonged T1 and T2 relaxation times, irrespective of age. Although hemorrhage and hematoma may occur in any tumor or hemophiliac pseudotumor, telangiectatic osteosarcoma and aneurysmal bone cyst are the lesions to consider when hemorrhage is found.3 Telangiectatic osteosarcoma will display malignant features such as indistinct margins and inhomogeneity, whereas aneurysmal bone cyst often contains multiple compartments and displays well-de®ned margins. Layering or a ¯uid level, as can be demonstrated with MRI, is more often

MRI OF MUSCULOSKELETAL NEOPLASMS

3

Figure 2 Hemangioma in the adductor compartment of the left thigh imaged with a 1.5 T MRI system. (a) The high signal intensity and morphology on this axial T1-weighted (600/20) spin echo image are typical of hemangioma. A soft tissue sarcoma would occupy more space. (b) After intravenous administration of 0.1 mmol Gd-DTPA/kg body weight, diffuse enhancement resulting in an increase in signal intensity is appreciated. This is consistent with the presence of well-vascularized cellular tissue. (c) On the coronal T1-weighted (600/20) spin echo image, the same features are visualized. The signal void of rapid ¯ow within the hemangioma is shown to best advantage

seen in aneurysmal bone cyst than in telangiectatic osteosarcoma. These ¯uid levels may also be encountered in chondroblastoma, giant cell tumor, especially when a secondary aneurysmal bone cyst is present, and occasionally in other tumors as well.3 Another reason for hematoma is surgical intervention. Since hematoma and edema may be quite extensive following histological needle or open biopsy, imaging studies are preferably performed prior to invasive procedures. High signal intensity may also be encountered in lipoma. On T2-weighted images the signal intensity will remain identical to that of subcutaneous fat. Liposarcomas may contain areas consisting of well-differentiated fat; these areas have the same appearance on MRI as benign lipomas. Liposarcomas, however, invariably contain large areas (myxoid liposarcoma) or some areas (lipoblastic liposarcoma) of poorly differentiated fat and mesenchymal tissue, which have the same MRI characteristics as other high-grade malignancies.

The signal intensity of hemangioma and arteriovenous malformations is variable. The signal intensity may be high on T1weighted sequences due to adipose tissue within the lesion and due to slow ¯ow through dilated sinuses and vessels (Figure 2). Because of slow ¯ow, the serpiginous vascular channels typically display a high signal intensity on T2-weighted images. Signal void represents high ¯ow in vascular channels. Fat predominates in asymptomatic vertebral hemangiomas. Vascularity is often more pronounced in symptomatic hemangiomas.4 A low signal intensity on T2-weighted sequences, secondary to a low spin density and/or short T2 relaxation time, may indicate the presence of calci®cation, osteoid, collagen, ®brosis, hemosiderin, or bone cement. Low signal intensity areas on T2weighted images may be present in pigmented villonodular synovitis, giant cell tumor of tendon sheath, elasto®broma, ®bromatoses, ®brous dysplasia, neuroma (Morton neuroma), etc.3 The signal intensity re¯ects the amount of collagen and

4 MRI OF MUSCULOSKELETAL NEOPLASMS ®brous tissue present. Cellular areas will be seen as high signal intensity areas on T2-weighted images, whereas the paucicellular areas will exhibit a low signal intensity on all pulse sequences. Low signal intensity masses located within ligaments (Achilles tendon) representing xanthomas may be encountered in patients suffering from familial hypercholesterolemia. This is caused by deposition of cholesterol in combination with dense collagen ®bers. The ¯ow void phenomenon, representing rapid ¯ow in vascular channels, such as found in arteriovenous malformations, may also be a cause for low signal intensity on T2-weighted images. This is in contrast to the characteristics of cavernous or capillary hemangioma. The morphology, distribution, and localization of disorders can also be used in MRI to characterize a lesion. Indistinct margins, a soft tissue lesion with size larger than 3 cm, and heterogeneity indicate malignancy, whereas well-de®ned margins and homogeneity are more indicative of a benign lesion.3,5 Of course there are exceptions to these general statements. Aggressive ®bromatosis is a benign lesion characterized by in®ltrative growth and a very high recurrence rate. Other benign lesions such as active eosinophilic granuloma may also present with indistinct margins, just as malignant lesions may present with well-de®ned margins. Edema, characterized by an increased signal intensity on T2weighted images, a decreased signal intensity on T1-weighted images, and indistinct margins, is a poor indicator of malignancy. Extensive soft tissue and bone marrow edema is frequently encountered in osteoid osteoma, osteoblastoma, and chondroblastoma. Edema in these benign entities is often more pronounced than in malignant tumors. Peritumoral high signal intensity in the soft tissue of patients with primary malignant tumors may represent only edema or a reactive zone also containing tumor. Edema is found in and around many lesions, including eosinophilic granuloma, fracture, bone bruise, necrosis, and transient osteoporosis.3 Neuro®broma is easily recognized on both CT and MRI when it presents as the classic dumb-bell tumor of the vertebral canal. Plexiform neuro®bromatosis extends along neural bundles in a lobulated fashion, and is thus also easily recognized. When located in the retroperitoneum or pelvis, these plexiform neuro®bromas may be quite extensive. The signal intensity of neuro®broma is determined by its prolonged T1 and T2 relaxation times, low to intermediate signal intensity on T1-weighted images, and a very high signal intensity on T2-weighted images. When present, collagen and ®brous tissue decrease the signal intensity (usually centrally). The distribution of areas with abnormal signal intensity is important, not only for differentiating between tumors and tumor-like lesions, but also for identifying normal variants. The low to intermediate signal intensity areas of red bone marrow in the axial skeleton and femur diaphysis in adults and also in the periphery of the skeleton in adolescents and children on T1weighted images may be confusing. When reviewing an MR examination, clinical data and radiographs are of paramount importance. Many tumor-like lesions, such as those due to Paget's disease, pseudoaneurysms, polymyositis, osteomyelitis, and stress fractures, can be identi®ed in the proper clinical setting.

Osteomyelitis will, in contrast to Ewing's sarcoma, usually display only reactive changes in the surrounding soft tissue. However, a soft tissue abscess may be encountered. Stress fractures present as irregular or band-like (perpendicular to cortical bone) areas of low signal intensity on T1-weighted images. The high signal intensity seen on T2-weighted or STIR sequences represents edema.

4 GADOLINIUM ENHANCED MRI The pharmokinetics of gadolinium chelates are similar to iodinated contrast agents. This paramagnetic contrast agent acts indirectly by facilitating T1 and T2 relaxation processes through an alteration of the local magnetic environment. Shortening of both relaxation times is a function of the concentration of the gadolinium complex. Initially, an increase in the gadolinium complex concentration results in a predominantly shortened T1 relaxation time and an ensuing increased signal intensity on T1weighted images (Figure 2). With an increasing concentration of the gadolinium complex, the signal intensity drops because of the dominant effect on the T2* relaxation time (dephasing). With the concentrations used in clinical practice (0.1±0.2 mmol/kg body weight for intravenous administration), the shortening of T1 relaxation time supervenes. Only in areas where a high concentration is reached, such as in the urinary bladder or early in a dynamic sequence during bolus administration in an artery, can a signal void occur. Well perfused, often cellular, tumor components are able to accumulate a high concentration of gadolinium, which results, through shortening of the T1 relaxation time, in a high signal intensity on short TR/TE (600/20) images, as opposed to some tumor components, such as sclerosis, old ®brosis and liquefaction, which are not well perfused and therefore do not show a high signal intensity on the postcontrast images. In general, the presence or absence of late enhancement is not a very good indicator of malignancy. Still, gadolinium may increase tissue characterization. Although the intravenous administration of Gd-DTPA (DTPA, diethylenetriaminepentaacetic acid) may assist in differentiating viable tumor from liquefaction and edema, neovascularity in necrotic areas may also be enhanced. Differentiation between viable and necrotic tumor is, as a consequence, not always possible. Dynamic MRI may assist in differentiating tumor from reactive tissue, since viable tumor usually enhances faster than reactive tissue. Dynamic sequences, obtained during gadolinium complex administration, with a temporal resolution of at least 3 s can be used to help to differentiate benign from malignant soft tissue tumors. Peripheral enhancement starting within 6 s after arterial enhancement without further increase in enhancement suggests malignancy.6 Static short TR/TE spin echo images may suggest a speci®c diagnosis in selected cases. Well-differentiated cartilaginous tumors containing large paucicellular cartilage ®elds or cartilage ®elds with mucoid degeneration demonstrate no or only slight enhancement. Only the margin and cellular septae in the periphery and within the tumor enhance. Because of the lobulated gross anatomy of enchondroma and low grade (grade I and II) chondrosarcoma, the postcontrast MRI scans of these tumors exhibit a septal-like or curvilineair (serpengineous) enhancement pattern7 (Figure 3). Mature enchondromas do not

MRI OF MUSCULOSKELETAL NEOPLASMS

5

Figure 3 Chondrosarcoma grade II imaged with a 1.5 T MRI system. (a) The tumor originating from the petrous bone has an atypical low signal intensity on this axial T1-weighted (550/20) spin echo image. (b) The same sequence is repeated 5 min after intravenous administration of 0.1 mmol Gd-DTPA. A serpentiginous enhancement pattern consistent with the presence of ®brovascular septations of well-differentiated chondrosarcoma is visualized. Compression on the temporal lobe and the brainstem is present

enhance. Enchondromas and osteochondromas may, however, enhance in the adult population 10 s or later after arrival of contrast agent in the artery. Central or peripheral chondrosarcoma grade 1 typically enhances within 10 s of arterial enhancement. Benign cartilaginous lesions in children or in association with sursae (mechanical stress) may also enhance early. Only high-grade cartilaginous tumors such as chondrosarcoma grade III and mesenchymal chondrosarcoma exhibit the aspeci®c homogeneous or inhomogeneous enhancement displayed by most tumors. Other cartilage containing tumors such as chondroid osteosarcoma may also show areas with curvilineair enhancement. Usually other criteria such as morphology will assist in differentiating these lesions. A chondroid osteosarcoma is less homogeneous than a well-differentiated cartilaginous tumor and often contains large areas with osteoid, which can be recognized. The curvilinear enhancement of well-differentiated cartilaginous tumors must be differentiated from thin peripheral enhancement of the pseudocapsule of lesions such as eosinophilic granuloma and solitary bone cyst. 5 LOCAL TUMOR STAGING Adequate, wide or radical surgery has been a prerequisite for proper treatment of patients with primary malignant musculoskeletal tumors. The choice of therapy depends on a number of factors such as the grade of malignancy, the response to

neoadjuvant therapy, local tumor extension, and the presence or absence of regional or distant metastases. The surgical staging system developed by Enneking takes all these factors, with the exception of the response to neoadjuvant therapy, into account. The biologic aggressiveness, indicated by the grade, is the key factor in the selection of the surgical margin required to achieve control. Four different surgical procedures, which imply four different margins, are recognized: intralesional, marginal, wide, and radical. The exact anatomic location of tumor is the key factor in selecting how a required tumor-free margin can be accomplished. 5.1

Bone Marrow Involvement

For extremities, a plane through the long axis of a long bone is chosen, i.e. sagittal or coronal. MRI is superior to technetium bone scanning because tumor related hyperemic osteoporosis is not depicted on MRI and is therefore not a source of false-positive readings. MRI has an almost perfect correlation (r = 0.99) with pathologic/morphologic examination; CT has a less substantial correlation (r = 0.93); 99mTc methylene diphosphonate (MDP) scintigraphy has a weak correlation (r = 0.69).3,8 The osseous tumor margin may be obscured by the presence of intraosseous edema. Often a double margin can be visualized. The margin nearest to the center of the tumor represents the true tumor margin, whereas the outer margin represents the margin of the edematous reactive zone towards normal bone

6 MRI OF MUSCULOSKELETAL NEOPLASMS marrow. Differentiation can be facilitated by using gadolinium. Tumor and the edematous reactive zones display different enhancement patterns: a well-vascularized tumor will enhance more rapidly than the reactive zone, but usually (for instance in osteosarcoma) the intraosseous reactive zone will enhance more than the tumor itself on images taken 5 min after contrast injection. Furthermore, enhancement of the edematous reactive zone is, as a rule, more homogeneous than that of tumor. The intraosseous reactive zone is usually not a clinical problem, because it disappears after one or two cycles of chemotherapy. The true tumor margin is then easily identi®ed. Caution is needed in the diagnosis of skip metastases. Skip metastases, as may be encountered in patients with osteosarcoma and Ewings's sarcoma, are detected with bone scintigraphy unless the size is below the detection threshold. Skip lesions of less than 5 mm may thus pose a diagnostic problem. Imaging of the entire bone on T1-weighted images by using a large ®eld of view may be of help.

5.2 Cortical Involvement Destruction of cortical bone is not a diagnostic problem since it can be evaluated on plain radiographs and, if necessary, with tomography or CT. MR images are also able to visualize the status of cortex. Invasion of cortex by tumor is best shown on T2-weighted images as a disruption of the cortical line and replacement of cortex by the high signal intensity of tumor. The sensitivity and speci®city of MRI (92% and 99% respectively) were not found to be signi®cantly higher than the sensitivity and speci®city (91% and 98% respectively) of CT.3,8 MRI is superior to CT in visualizing sclerotic osteosarcomas, because the signal intensity of osteosclerotic tumor is still slightly higher than the low signal intensity of normal cortex. The high density of osteosclerotic tumor and cortex may be indistinguishable on CT.

5.3 Involvement of Muscular Compartments MRI is signi®cantly superior (sensitivity 97%, speci®city 99%) to CT in identifying muscle compartments containing tumor.3,8 The superior performance of MRI is based on the display of tumor relative to muscular compartments with, compared to CT, superior contrast in the ideal imaging plane. The presence of peritumoral edema is, as a rule, easily identi®ed because of its slightly different signal intensity compared to that of tumor, and especially because of the fading margins of edema as opposed to the distinct margin of the (pseudo)capsule of the tumor. However, accurate delineation of the tumor± edema interface can be rather dif®cult. Differences in signal intensity on multiple echoes and enhancement following administration of gadolinium may be used to differentiate tumor from edema.

5.4 Vascular Involvement Vessels are more often displaced than encased by tumor. The relationship between tumor and neurovascular bundle is easily evaluated on T2-weighted images because normal ¯ow in

a vessel results in low or absent signal intensity, the so-called `¯ow void phenomenon.' The lumen of the vessel may have a higher signal intensity due to paradoxal enhancement, evenecho rephasing, or slow ¯ow caused by compression. CT (sensitivity 36%, speci®city 94%) and MRI (sensitivity 92%, speci®city 98%) provide more information than angiography (sensitivity 75%, speci®city 71%) because large vessels are, especially with MRI, well visualized in relation to the tumor.3,8

5.5

Joint Involvement

CT (sensitivity 94%, speci®city 90%) and MRI (sensitivity 95%, speci®city 98%) are both able to demonstrate joint involvement with high accuracy.3,8 Joint involvement is sometimes more accurately demonstrated on MR images than on CT, because the articular surfaces may be parallel to the transverse CT plane. Cartilage is an effective barrier that is not easily crossed by tumor. Osteosarcoma and giant cell tumor are the tumors that are able to cross cartilage. When assessing possible joint involvement, the number of falsepositive readings is much higher than the number of falsenegative readings. When in doubt, the joint usually is not affected. Joint effusion with or without hemorrhage, often but not always a secondary sign of a contaminated joint, is easily identi®ed on CT and MRI.

6

RECURRENCE

Detection of recurrent or residual tumor following initial treatment is a challenging problem for diagnostic imaging. A large recurrent tumor mass may be detected at clinical examination, with radiographs or by CT. Small recurrent tumors are dif®cult to de®ne in relation to posttherapy changes caused by surgery, radiation therapy, and chemotherapy. The matter is further complicated when ®xation devices have been used in reconstructive surgical procedures. These devices cause susceptibility artifacts on MRI. Even when no hardware is used, susceptibility artifacts, caused by metallic particles left behind after instrumentation, can degrade the MR images. Despite these drawbacks, MRI can be used to detect recurrence (Figure 4). A recurrence is very likely (sensitivity 96± 100%) when, following surgery, with or without chemotherapy, a mass is seen on MRI which is characterized by a high signal intensity of Gd-DTPA enhanced or T2-weighted images.9 Cystic masses without tumor have a high signal intensity on T2weighted images, but do not enhance after intravenous administration of Gd-DTPA. A tumor recurrence is very unlikely when, following surgery, no enhancement or a low signal intensity on T2-weighted images is found. However, knowledge of the signal intensity of the primary tumor prior to therapy is crucial, especially when the primary tumor was characterized by a low signal intensity on T2-weighted images. When an equivocal lesion is detected, follow-up studies, biopsy or other imaging studies such as radiographs or ultrasound may be helpful. Radiation therapy adds to the confusion because it may induce in¯ammatory, reactive changes that may be indistinguishable from tumor recurrence. These reactive

MRI OF MUSCULOSKELETAL NEOPLASMS

7

Figure 4 MRI images obtained with a 1.5 T imager of a patient with enchondromatosis and recurrent chondrosarcoma after resection and reconstructive surgery. (a) Coronal large ®eld of view, T1-weighted (600/20) spin echo images show cartilaginous tumors in both proximal tibias. Marked susceptibility artifacts, secondary to metallic hardware left by the orthopedic surgeon, are seen in the left femur and proximal right tibia. (b) Axial T2-weighted (2000/100) spin echo images show a lobulated soft tissue recurrence posterior to the artifacts within the femur. (c) The soft tissue mass has an atypical low signal intensity on this T1-weighted (600/20) image. (d) After injection of Gd-DTPA the serpentiginous enhancement pattern indicative of recurrent well-differentiated chondrosarcoma is easily appreciated, despite the marked susceptibility artifacts

changes are characterized by a high signal intensity on static Gd-DTPA enhanced and T2-weighted images, and may persist for more than a year. However, reactive lesions usually con®ne themselves to the space available and, unlike tumor recurrence, do not present themselves as space occupying expansile masses. Dynamic MRI can be used to differentiate reactive tissue from viable tumor. Viable tumor typically enhances 6 s after arrival of gadolinium complex in the artery. When detection of early recurrence is of vital importance, or when there is a high risk of recurrence, a baseline study obtained 3±6 months after surgery may be of help. 7 MONITORING CHEMOTHERAPY Preoperative (neoadjuvant) chemotherapy of bone sarcomas has increased the feasibility of limb salvage procedures. As a consequence of chemotherapy, a soft tissue tumor mass may shrink and, in combination with encapsulation of the residual extramedullary tumor mass, improve surgical conditions (down staging). However, identi®cation of good and poor respondents is a challenging and controversial exercise.10 Assessment of viable and necrotic tumor at histology is the gold standard.

Spontaneous necrosis of up to 50% is not uncommon in highgrade malignancies. Therefore a histologic good response is de®ned as the presence of 10% or less viable tumor tissue. A change of tumor volume and signal intensity in osteosarcomas and Ewing sarcoma may correlate with response to chemotherapy. Despite major limitations, several qualitative and quantitative MR parameters contribute to the differentiation of good and poor respondents during and after preoperative chemotherapy. Increase in tumor volume without hemorrhage and increase of signal intensity on T2-weighted images in patients with osteosarcoma, even after the ®rst cycle of chemotherapy, indicates poor response.11 At present there are no reliable criteria for identifying good respondents on native MRI. Reduction of tumor volume in Ewing sarcoma is characteristically seen in all patients with Ewing sarcoma after successful, or unsuccessful, chemotherapy.10 A 75% decrease in tumor volume, or complete absence of a residual soft tissue mass after chemotherapy, are consistent with good response. Resolution of the soft tissue mass in Ewing sarcoma is frequently accompanied by reactive subperiosteal bone formation, which may result in the development of an inhomogeneous cuff of tissue encircling the original cortex.

8 MRI OF MUSCULOSKELETAL NEOPLASMS Progressive ossi®cation of the periosteal mass may re¯ect healing of these tumors, but the presence of minimal residual disease in this peripheral area cannot be excluded with any imaging modality. A well-de®ned rim of low signal intensity forming a margin for the extramedullary tumor compartment represents a ®brous pseudocapsule continuous with the periosteum; however, there is no correlation with the percentage of tumor necrosis.10,11 Although this qualitative sign cannot be used as a differentiating criterion between good and poor respondents, the improved tumor demarcation can facilitate the surgical resection.10,11 Following chemotherapy, the amount of viable residual tumor can be assessed by dynamic gadolinium-enhanced images. Viable tumor enhances within 6 s after arterial enhancement. Non-viable tumor components enhance later or not at all.12

8 BONE METASTASES Bone metastases primarily occur in the red bone marrow of the axial skeleton. We therefore focus here on the vertebral column. Conventional radiographs have a very low yield in depicting skeletal metastases when these are still located in the bone marrow. The origin of metastatic deposits is within the bone marrow of the vertebral body. Metastatic tumor may ®ll the marrow and leave cortical bone intact. Detection of metastatic disease can occur at a much earlier stage when destruction of cortical bone is present. The contours of collapsed vertebral bodies can suggest the nature of the underlying disease, but conventional radiography alone is unreliable in differentiating between benign and malignant causes of vertebral body collapse. MRI can be performed using several pulse sequences; spin echo, TSE, STIR, GE, and out-of-phase chemical shift imaging. Each pulse sequence has its own imaging features and will exhibit normal bone marrow and pathology in a characteristic manner. The signal intensities re¯ect the histology of the tissue. Typically, metastases display a low signal intensity on T1-weighted sequences and a high signal intensity on T2weighted sequences. There are exceptions, such as in sclerotic metastases which exhibit a low signal intensity on all pulse sequences, and lipoblastic metastases which have a relatively high signal intensity on T1-weighted images. In the STIR pulse sequence, yellow marrow signal is nulled, such that bone marrow appears black. T1 and T2 values other than those for fat are additive. Lytic metastases, because of their high water content, produce an increase in both T1 and T2 relaxation times. This will enhance the contrast between lytic metastases and bone marrow. Often lytic metastases are better depicted on STIR images than on T1-weighted spin echo images. Early edema (10±14 days) after radiation therapy can typically be seen earlier on STIR images than on T1- or T2weighted spin echo images. Contrast between normal bone marrow and tumor can also be increased by using the difference in resonance frequency between aliphatic and water protons. Out-of-phase images will increase image contrast in the case of lytic bone marrow metastases. Fat suppression with presaturation pulses may further

increase contrast on T2-weighted sequences. Fat presaturation may be used in combination with opposed-phase imaging. Gadolinium will further increase sensitivity when used in combination with fat suppression. It may also increase the speci®city of MRI, as metastatic deposits in vertebral bodies will exhibit diffuse enhancement of signal intensity on T1weighted images, whereas osteoporotic collapsed vertebral bodies will show band-like enhancement. Some authors, however, have found that the use of Gd-DTPA can be disappointing in the differential diagnosis of malignant and benign tissues. T1- and T2-weighted (STIR or fat suppression) images will constitute a suf®cient routine imaging protocol. Additional imaging in orthogonal planes will render valuable information in the case of soft tissue involvement. Gadolinium should be reserved for selected cases such as differential diagnosis between malignant and benign collapsed vertebrae and where there is suspected leptomeningeal tumor spread. MRI is more sensitive in the detection of vertebral metastases than is bone scintigraphy.13 This is not surprising since MRI visualizes bone marrow directly rather than depending on secondary signs such as new bone formation. Although the MR characteristics of metastatic disease do not always allow reliable differentiation between benign and malignant disease, MRI is still more speci®c than bone scintigraphy in distinguishing between benign and malignant disease. For instance, by means of MR examination it is usually possible to make a reliable distinction between degenerative bone disease in the vertebral column and malignant in®ltration. Morphologic characteristics of the lesion and adjacent disk must be taken into account. Loss of height of the intervertebral disk can be seen in degenerative disease and in diskitis, whereas the shape and height of the disk is usually preserved in metastatic diseases. In diffuse bone marrow lesions, the intervertebral disk may show a high signal intensity relative to the decreased signal intensity of abnormal bone marrow on T1-weighted SE images. The morphology of vertebral bodies and the signal intensity changes of bone marrow, particularly in relation to the vertebral endplates, may assist in differentiating between benign and malignant compression fractures. In old osteoporotic compression fractures the signal intensity is typically normal (fat), whereas in malignant compression fractures an abnormal signal intensity due to replacement of bone marrow is seen. In the (sub)acute phase, an inhomogeneous increase in signal intensity on STIR or T2-weighted images, or a sharply delineated isointense vertical band of preserved normal (fatty) bone marrow along the dorsal aspect of the compressed body, indicates a benign fracture. The abnormal signal intensity in benign fractures may have the shape of a horizontal band. Fractures at multiple levels with preservation of normal bone marrow, vertebral body fragmentation, and disk rupture also indicate benign disease. Signs in favor of malignancy are: homogeneous signal intensity changes; convex anterior, and especially posterior, contour; cortical destruction; multiple levels with abnormal signal intensity, but without fracture; and abnormal signal intensity in posterior elements. Paraspinal masses are more conspicuous in malignant disease, but can be seen in both traumatic and malignant cases. Care must be taken in interpreting MR examinations in recently collapsed vertebral

MRI OF MUSCULOSKELETAL NEOPLASMS

bodies as they can show signal intensities indistinguishable from metastatic disease. Currently, bone scintigraphy remains the screening procedure of choice because it is readily available, and it allows imaging of the entire skeleton in a time effective way. However, secondary to increased sensitivity, speci®city, and faster pulse sequences, the role of MRI is increasing. MRI can visualize metastatic disease when bone scintigraphy is falsely negative. Thus MRI is currently indicated when a strong clinical suspicion is combined with a negative bone scan. MRI may also elucidate the true nature of hot spots in the vertebral column in cancer patients, as it can often assist in making a distinction between malignant and benign disease. MRI can also be helpful in the guidance of biopsies. In order to rule out compressive myelopathy or to establish soft tissue extension of tumor tissue, multiplanar MRI offers unique imaging features. The use of MRI is helpful in determining the local extent of metastatic disease when planning palliative surgery or radiation therapy. 9 RELATED ARTICLES

9

5. M. J. Kransdorf, J. S. Jelinek, and R. P. Moser, Radiol. Clin. North Am., 1993, 31, 359. 6. H. D. van der Woude, K. L. Verstraete, P. C. W. Hogendoorn, A. H. M. Taminiau, J. Hermans, and J. L. Bloem, Radiology, 1998, 208, 821. 7. M. J. A. Geirnaerdt, J. L. Bloem, F. Eulderink, P. C. W. Hogendoorn, and A. H. M. Taminiau, Radiology, 1993, 186, 813. 8. J. L. Bloem, A. H. M. Taminiau, F. Eulderink, J. Hermans, and E. K. J. Pauwels, Radiology, 1988, 169, 805. 9. D. Vanel, L. G. Shapeero, T. de Baere, R. Gilles, A. Tardivon, J. Genin, and J. M. Guinebretiere, Radiology, 1994, 190, 263. 10. H. D. van der Woude, J. L. Bloem, and P. C. W. Hogendoorn, Skeletal Radiol., 1998, 27, 145. 11. H. C. Holscher, J. L. Bloem, D. Vanel, J. Hermans, M. A. Nooy, A. H. Taminian, and M. Henry-Anar, Radiology, 1992, 182, 839. 12. H. D. van der Woude, J. L. Bloem, K. L. Verstraete, A. H. M. Taminiau, M. A. Nooy, and P. C. W. Hogendoorn, Radiology, 1995, 165, 593. 13. P. R. Algra and J. L. Bloem, in `MRI and CT of the Musculoskeletal Systems', ed. J. L. Bloem and D. J. Sartoris, Williams & Wilkins, Baltimore, 1992, Chap. 16.

Acknowledgements

Imaging of Trabecular Bone; Skeletal Muscle Evaluated by MRI.

The contributions of M. Geirnaerdt, H. C. Holscher, H. J. van der Woude, A. H. M. Taminiau, P. Hogendoorn, F. Eulderink, M. A. Nooy, and H. M. Kroon, are gratefully acknowledged.

10

Biographical Sketch

REFERENCES

1. S. G. Moore and K. L. Dawson, Radiology, 1990, 175, 219. 2. J. M. Gomori and R. I. Grossman, RadioGraphics, 1988, 8, 427. 3. J. L. Bloem, H. C. Holscher, and A. H. M. Taminiau, in `MRI and CT of the Musculoskeletal System', ed. J. L. Bloem and D. J. Sartoris, Williams & Wilkins, Baltimore, 1992, Chap. 15. 4. J. D. Laredo, E. Assouline, F. Gelbert, M. Wybier, J. J. Merland, and J. M. Tubiana, Radiology, 1990, 177, 467.

Johan L. (Hans) Bloem. b 1954. M.D., 1979, Ph.D., 1988, Leiden University, The Netherlands. Visiting professor, Charles Gairdner Hospital, Perth; Thomas Jefferson University, Philadelphia. NMR program at Leiden University, 1983±present. Currently, Professor and Chairman of Radiology at Leiden University. Approx. 100 publications; editor of MRI and CT of the Musculoskeletal System. Research interest: MRI of the musculoskeletal system.

PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

Peripheral Joint Magnetic Resonance Imaging Paul S. Hsieh Kaiser Permanente Medical Center, San Diego, CA, USA

and John V. Crues III RadNet Management, Inc., Los Angeles, CA, USA

1 INTRODUCTION MRI is an important tool in the radiological evaluation of peripheral joint disease. Previously available noninvasive imaging techniques include X-ray and ultrasound imaging. X-ray techniques (plain radiographs and computerized tomography) are able to image cortical bone but are not sensitive in imaging medullary bone disease or intra-articular soft tissues.1,2 Ultrasound can evaluate soft tissue structures but is limited in its ability to visualize internal structures in joints and detect some soft tissue diseases.3 Nuclear medicine bone scintigraphy has also been used to detect articular pathology, but is not commonly used because of its lack of speci®city and poor spatial resolution.4,5 MRI is capable of demonstrating the anatomy of the various components of peripheral joints in exquisite detail, including muscles, tendons, ligaments, nerves, blood vessels, fat, and osseous structures.6,7 In addition to its high spatial resolution, MRI displays excellent contrast between musculoskeletal soft tissues because differences in chemical structures lead to different T1 and T2 values. These differences in relaxation times can be exploited to generate images with striking contrast between normal tissues and between normal and pathologic tissues. Intra-articular injection of contrast material (arthrography) allows X-ray techniques to visualize the surfaces of intra-articular structures, but X-ray arthrography is both invasive and insensitive to numerous pathologic conditions that do not manifest with surface irregularities.8 MR arthrography is now widely used for the evaluation of speci®c articular pathology because it combines the sensitivity of arthrography for surface abnormalities with high soft-tissue contrast.9±11 For these reasons, many consider MRI to be the modality of choice for noninvasive imaging of peripheral joint disease.

2 BASIC TECHNIQUES Most MRI of peripheral joints is performed with twodimensional spin echo pulse sequences.6 This is a basic (90± 180 ) sequence, with the time between successive 90 pulses denoted TR and that between 90 pulse and the signal acquisition denoted TE. A slice selection gradient is applied during each rf pulse to limit the excitation to the desired anatomic plane of interest. Perpendicular gradients are also used to provide frequency and phase encoding. The basic pulse sequence

1

is repeated 128, 192, or 256 times, each time with a different phase-encoding gradient. The signals or echoes from each acquisition are collected and processed using a two-dimensional Fourier transform to generate the ®nal medical image. Many varieties of the basic scheme are now in clinical use.12 By adjusting the TR and TE, the signal can be manipulated to have lesser or greater degrees of T1 or T2 contrast. Hence, an image acquired with a sequence with a short TR (~600 ms) and short TE (~15 ms) is referred to as a T1-weighted image. Similarly, if the TR is long (~2000±4000 ms) and the TE is long (~80 ms), this is referred to as a T2-weighted image. If the TR is long but the TE is short, then many refer to this as a proton density-weighted image. Most structures in peripheral joints have fairly characteristic T1 and T2 values that make them easy to distinguish on the spin echo images. For instance, fat has a short T1, so it has bright signal intensity on T1-weighted images. Ligaments, tendons, and bone cortex have a paucity of mobile protons and therefore demonstrate low signal intensity on all pulse sequences. Muscle tissue also has a fairly long T1 and therefore looks dark on a T1-weighted image, although not as dark as tendon or ligament. Bone marrow can have intermediate to bright intensity on T1-weighted images depending on the fat content (nonhematopoietic or `yellow' marrow has more fat than hematopoietic or `red' marrow). Fluid within joints has a long T1 and a long T2; hence it will appear dark on T1weighted images and bright on T2-weighted images. Most pathologic tissues have increased edema compared with their normal counterparts. This increased water content causes prolongation of their T1 and T2 values. Hence, on T1weighted images, abnormal tissues tend to look dark, whereas on T2-weighted images, these tissues generally look abnormally bright. Most lesions are fairly obvious on T2-weighted images. However, the presence of bright signal on a T2-weighted image is a nonspeci®c ®nding: many disease processes (i.e., tumor, infection, trauma, etc.) can cause this appearance and other clues (such as lesion morphology and location) must be sought in order to make a more speci®c diagnosis. In particular, abnormalities within ligamentous, tendinous, and other ®brocartilaginous structures (such as the menisci in the knee and the labra in the shoulder) manifest themselves as foci of abnormally increased signal within a usually homogeneously dark structure.4,13 Normally, there are few mobile water molecules within these structures capable of generating any signi®cant signal. The water that does exist within them is bound to the large macromolecules such as cartilage and is incapable of free translation and rotation.14 The protons within these bound water molecules have very short T2 values and therefore do not generate any detectable signal. However, if the collagen matrix undergoes degeneration with microscopic tears, then small amounts of water can be trapped in the interstices. These water molecules are more mobile and therefore have slightly longer T2 values, which can be detected on short TE images (i.e., T1-weighted and proton density-weighted images). The T2 values are still too short to generate signal on longer TE images (i.e. true T2-weighted images). There may also be a secondary T1 shortening effect that might contribute signal on T1-weighted images. With degenerative disruption of macromolecules, water protons may be exposed to protons deep in the macromolecules. This close proximity between the two sets of protons allows for some magnetization exchange to occur,

2 PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

Figure 1 A sagittal proton density-weighted image of a knee (TR 2000, TE 25) demonstrating tear within the lateral meniscus. These are visualized as increased signal (white arrowhead) within the normal low signal of the menisci. A larger area of bulk ¯uid is seen anteriorly (white arrow) representing an intrameniscal cyst

with resulting shortening of the T1 time.14 The hallmark of this type of pathologic process is increased signal on the proton density-weighted images but not on the T2-weighted images.13 In the presence of more signi®cant trauma, macroscopic amounts of free ¯uid can be present. Clinical examples would include complete rotator cuff tears and large meniscal tears with development of intrameniscal cysts.4,13 In these settings, the abnormal area would show increased signal on both the proton density- and T2-weighted images (see Figures 1 and 2).

Figure 2 A sagittal T2-weighted image (TR 2000, TE 80) of the same knee as in Figure 1 clearly showing the bulk ¯uid (white arrow) of the intrameniscal cyst. The T2 of the bulk ¯uid is prolonged, making this ¯uid visible. However, the meniscal tears (white arrowhead) are considerably less apparent because their T2 values are not as prolonged

Another commonly used pulse sequence is the STIR or short tau inversion recovery sequence. This pulse sequence consists of (180 ±±90 ±180 ), where  is the delay time between the initial 180 inversion pulse and the 90 pulse. In a STIR sequence,  is set to the null point of fat, i.e. the length of time it takes for the fat to recover to a point of zero longitudinal magnetization after an initial inversion. At a typical ®eld strength of 1.5 T, the appropriate value of  is approximately 160 ms. In a STIR sequence, the signal intensity of the tissue is directly related to its T1 and T2 values. Tissues with short T1 and T2 values such as fat show little or no signal, whereas tissues with long T1 and T2 values have bright signal. Because most pathologic tissues have prolonged T1 and T2 relaxation times they are bright on STIR images even more so than on T2-weighted images. This is especially valuable when imaging at mid and low magnetic ®elds (0.2±0.5 T ). However, as with a T2-weighted image, bright signal is not speci®c for any one disease process and other information (such as morphology and location of the lesion) must be considered to make a more speci®c diagnosis. Another set of pulse sequences commonly used in musculoskeletal MRI is the RARE sequence,15 which is also known as fast spin echo (FSE) or turbo spin echo.16 In this pulse sequence, an initial 90 pulse is followed by a rapid succession of 180 pulses to generate a series of echoes known as an echo train. Typically 4, 8, or 16 echoes are generated per excitation, each preceded by a different phase-encoding gradient. Hence, several lines of k-space are acquired per excitation, resulting in a considerable saving in scanner time. However, the TE values are not uniform within the image: some of the echoes making up the image will have shorter TEs, whereas others will have longer TEs. This can result in a variety of image artifacts including blurring and loss of resolution and changes in image contrast.17 Also, because some pathological processes (such as meniscal tears in knees) demonstrate increased signal on short TE sequences but not on long TE sequences, it is possible that the mix of long and short TEs in a FSE image might make it less sensitive to subtle lesions than a standard spin echo image.18 For this reason, at our institution the following compromise is used: for the key anatomic plane of a joint (which is the sagittal plane for knees and the oblique coronal plane for shoulders), a slower standard spin echo pulse sequence is used to generate proton density and T2-weighted images. For the other anatomic planes, FSE is used. In this way, a reasonable balance between scanning time and diagnostic accuracy is maintained. However, as interpreters get more experience with the vagaries of these sequences on individual scanners, more sequences are being converted to the faster techniques. Spectral fat saturation is often advocated with the use of FSE imaging in the musculoskeletal system to eliminate increased signal from fat on FSE T2-weighted images.19 Gradient echo pulse sequences are only infrequently used in MRI of peripheral joints. They are fairly sensitive for the same pathological processes that are detected by short TE spin echo sequences. However, the soft tissue contrast is worse with gradient echo sequences than with spin echo sequences. This is because the contrast in gradient echo sequences is dependent on differences in T2* rather than in T2.20 In clinical MR imaging, most of the T2* effect is caused by ®eld inhomogeneities and other factors not dependent on the biomedical make-up of

PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

the tissue. Hence, the signal from the tissues undergoes T2 decay before the differences in the true tissue T2s can manifest themselves. For this reason, most soft tissues have very similar appearances and signal intensities on gradient echo images, regardless of the tissue type or the degree of involvement by pathology. Nevertheless, some investigators use gradient echo sequences because the short TRs allow 3-dimensional Fourier transform acquisitions in a reasonable time period for isotropic high-resolution imaging. This may be valuable in imaging the labrum and articular cartilage.13,21 In certain settings, intravenous or intraarticular gadolinium may prove useful for diagnosis. Gadolinium is a paramagnetic metal which can be bound to an organic chelating agent such as diethylenetriaminepentaacetic acid (DTPA). In the doses used in clinical practice, the main effect of intravenous gadolinium is to shorten the T1 relaxation times of the perfused tissues. This causes them to appear brighter on T1-weighted images. Some investigators have found this useful in evaluating possible recurrences of soft-tissue neoplasms following surgical resection.22,23 Intravenous gadolinium can also be useful in determining if a soft-tissue mass with long T1 and long T2 is cystic or solid. A cystic lesion would not show any signal enhancement following contrast administration whereas a solid lesion would. Similarly, in the setting of a soft-tissue infection, it can be very dif®cult to tell the difference between an area of phlegmonous and in¯amed solid tissue and a drainable ¯uid collection. On a postgadolinium image, the phlegmon should enhance, whereas the ¯uid collection should not enhance.24,25 Gadolinium can also be injected into a joint space in the form of a dilute mixture of saline and gadolinium-DTPA.9,10 This forms a positive contrast on T1-weighted images that distends the joint capsule and helps delineate adjacent structures. This type of MR arthrography has proven particularly useful in the shoulder, where subtle abnormalities in the cartilaginous labra can be better appreciated following contrast injection.26 However, this converts the MR study from a noninvasive procedure to an invasive procedure, with resultant discomfort and potential risk to the patient. (Some radiologists also inject saline without gadolinium, which produces a similar arthrographic effect on T2-weighted images).26 MR knee anthrography is helpful in some patients who have had previous meniscal repair.27

3 DISEASE ENTITIES 3.1 Trauma MR has proven to be a valuable tool in the evaluation of acute trauma to peripheral joints as well as in evaluation of chronic trauma and degeneration. For instance, within the knee joint (the most commonly imaged peripheral joint), MR can be used to detect partial and complete tears of key anatomic structures including the medial and lateral collateral ligaments, the anterior and posterior cruciate ligaments, and the medial and lateral menisci.6 Injuries to these structures can be detected by identifying abnormally increased signal within these normally low-signal structures. Other important radiographic signs include alterations in the contour and morphology of these structures, and the presence of edema in the surrounding soft tissues (manifested as high signal intensity on T2-weighted

3

Figure 3 A sagittal proton density-weighted image (TR 2200, TE 20) of the ankle showing a tear of the achilles tendon. The tear is the area of increased signal (arrow) within the normally dark substance of the tendon (arrowheads). Notice the focal thickening of the tendon at the site of the tear

images). These principles apply to evaluation of ligamentous and tendinous, and muscular structures at all joints, including the rotator cuff of the shoulder and the achilles tendon in the ankle (see Figures 1±3).13,28,29 MRI has also proven to be sensitive in the detection of radiographically occult fractures.28,30 Often these fractures produce a characteristic linear pattern of bone marrow edema, which can be visualized as a region of low signal intensity within the marrow on T1-weighted images and high signal intensity within the marrow on T2-weighted images. Detection of these fractures can have important therapeutic implications, so MRI should be strongly considered in the setting of a patient with clinical ®ndings suspicious for fracture with normal radiographs. Many investigators feel that MR is at least as sensitive for detection of such fractures as the other major technique, bone scinitigraphy, and is more speci®c (see Figure 4).31,32 Traumatic lesions of the articular cartilage and subchondral bone can also be detected with MRI. In an acute setting, these are known as osteochondral fractures, whereas in the setting of chronic repetitive microtrauma, these lesions are referred to as osteochondritis dissecans. Pertinent MR ®ndings which may be seen include disruption or thinning of the articular cartilage, alterations of the contour of low-signal subchondral bony plate (which may be accompanied by abnormal high signal within the normal low signal of cortical bone), and edema in the adjacent bone marrow.33 3.2

Infection

Another common use of MRI is in the evaluation of suspected infections of peripheral joints.24,34 MR can detect the presence of abnormal ¯uid within a joint space, but cannot determine if the ¯uid is sterile or infected (i.e., if the patient has a bland joint effusion or a septic arthritis). MR is also useful in the evaluation of osteomyelitis. The principal ®nding in

4 PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

Figure 4 A coronal T1-weighted image (TR 600, TE 15) of the hips showing a fracture of the right femoral neck (straight arrow). The abnormal low signal in the marrow is caused by edema. For comparison, the marrow signal in the left femoral neck is normal (curved arrow)

osteomyelitis is bone marrow edema (manifested by the usual low signal intensity in the marrow on the T1-weighted images and high signal intensity on the T2-weighted image). Osteomyelitis is also often accompanied by in¯ammatory changes in the adjacent soft tissues. If an area of soft tissue edema is identi®ed, then MR with intravenous gadolinium administration can help to determine if there is a component of drainable ¯uid within the in¯amed phlegmonous region.24,25 This is extremely valuable in the evaluation of the diabetic foot.34 As discussed above, a ¯uid collection should not enhance, whereas the phlegmonous component should (see Figure 5).

Figure 6 A coronal T1-weighted image (TR 500, TE 21) of the wrist showing avascular necrosis of the proximal pole of the scaphoid bone (straight arrow), with abnormally low signal intensity within the marrow. The distal pole of the scaphoid bone has relatively normal signal intensity within the marrow (curved arrow)

3.3 Arthritis

including the cartilage loss, subchondral sclerosis and cyst formation, and osteophytes.35 In rheumatoid arthritis, MR may be helpful in delineating the exact extent of the in¯ammed pannus tissue. On standard MR images, the pannus can look similar to joint ¯uid, but after the administration of intravenous gadolinium, the pannus should enhance intensely. MR is also sensitive for early detection of erosions, which may also be helpful in patients with rheumatoid arthritis or other erosive arthritides.36

The role of MRI in evaluation of arthritis patients is fairly limited. MR can detect the ®ndings seen in osteoarthritis,

3.4

Ischemic Disease

MR is sensitive in the evaluation of ischemic bone disease including avascular necrosis and bone infarcts.37,38 The devascularized portion of bone becomes edematous initially. Later, as ®brovascular tissue replaces granulation tissue the T2 relaxation time shortens (Figure 6). Some investigators have found that dynamic gadolinium enhancement studies are useful in assessing the prognosis of involved bone.39,40 For instance, with avascular necrosis of the femoral head and scaphoid fractures, patients with lesions which enhanced following intravenous gadolinium administration may have an improved prognosis over patients whose lesions did not enhance. Further work remains to be done to see if this is true in other areas of the body. 3.5 Figure 5 A coronal T1-weighted image (TR 800, TE 20) of the hips showing marrow edema caused by osteomyelitis in the left femoral neck (white arrow). The signal intensity is similar to that of the edema caused by the fracture in Figure 4, but the abnormal area is more extensive and ill-de®ned. This is more compatible with an infectious process. Edema is a nonspeci®c ®nding; clinical history and morphologic clues are often necessary to distinguish between various disease entities

Neoplasms

MR has proven valuable in staging of osseous and soft tissue neoplasms.22,41±45 With MRI, one can evaluate the anatomic extent of the tumor and the integrity of adjacent neurovascular bundles. MR is helpful in determining whether a tumor of soft tissue origin involves adjacent bone and if a tumor of bony origin involves adjacent soft tissue structures. The extent of the bone marrow involvement can also be evaluated with MRI. Most neoplasms have prolonged T1 and T2

PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

values, resulting in the typical dark appearance on T1-weighted images and bright on T2-weighted images. The contrast between the tumor and bone marrow or subcutaneous fat is most marked on T1-weighted or STIR images because fat has such a short T1 in contrast with the long T1 of the tumor. On the other hand, T2-weighted images are better at demonstrating the difference between the tumor and normal muscle or normal neurovascular structures. These tumors are often surrounded by a halo of edema, which also has similar signal characteristics to the main tumor. In this case, the area of signal abnormality on the images is larger than the size of the actual tumor. In this setting, evaluation with intravenous contrast and fat-saturated T1-weighted images may be valuable.46,47 MRI is not very speci®c for most tumor cell types. The only major exceptions are lipomas, which are nearly entirely composed of fat.48 These can be identi®ed by the characteristic signal intensity equal to normal fat on all pulse sequences, as well as by secondary signs such as chemical shift artifact. If a fat suppression pulse sequence is used (where rf energy is applied to the fat peak prior to image acquisition), or a STIR series is acquired, the signal from lipomas should disappear. If a tumor demonstrates all of these signal characteristics, and does not contain any signi®cant amounts of nonfatty tissue, then the diagnosis of benign lipoma can be con®dently made. Most soft tissue tumors have a nonspeci®c appearance, and the main role of MR is to evaluate for anatomic extent. This is useful in planning biopsies, surgical resections, and/or radiation therapy. MRI is also helpful in the evaluation of suspected recurrence following resection. In this setting, some investigators have found that intravenous gadolinium is helpful in distinguishing recurrent tumor from normal postoperative reaction.29,49,50 Examples of bone and soft tissue tumors are shown in Figures 7±10.

4 FUTURE DIRECTIONS There are several technical advances that may have promising applications in musculoskeletal MRI. One of these is MR microscopy. With current clinical MRI, the spatial resolution is on the order of 0.5±1.0 mm per pixel. With specialized gradients and other MR microscopy techniques, this can be improved by a factor of approximately 10. If this is done, then it becomes possible to evaluate structures such as articular cartilage in much greater detail. Instead of appearing as a thin stripe of signal, 2 or 3 pixels thick, the articular cartilage will be a broad band, 20 or 30 pixels thick. (See Figure 11). It may then be possible to evaluate subtle pathology within the various sublayers of articular cartilage.51±53 Any technique that can determine the presence or absence of early changes in rheumatological disease is also helpful, particularly for evaluating the ef®cacy of various treatments. Other potentially important advances are the various diffusion and perfusion imaging techniques. The musculoskeletal system is well suited for application of these techniques because these body parts can be kept fairly immobile during scanning, which is a requirement for successful use of these techniques. Although this has not yet been proven, it is possible that some pathological processes may manifest themselves

5

Figure 7 A coronal T1-weighted image (TR 800, TE 20) showing a large destructive mass in the lateral femoral condyle (arrow), caused by metastatic renal cell carcinoma

as alterations in either local perfusion or alterations in the local diffusion coef®cients before one sees evidence of gross edema and prolongation of T1 and T2 relaxation times.54 Magnetization transfer contrast is another technique that may prove useful in musculoskeletal application, particularly in

Figure 8 A transaxial T2-weighted image (TR 2117, TE 80) of the same lesion shown in Figure 7, showing increased signal intensity within the mass (arrow) caused by T2 prolongation

6 PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

Figure 9 A coronal T1-weighted image (TR 600, TE 20) of the right thigh showing a round mass in the medial soft tissues (white arrow). The mass (a malignant ®brous histiocytoma) is dif®cult to identify because it has similar signal intensity to normal muscle on this sequence Figure 11 A T1-weighted MR microscopy image showing the articular cartilage as a thick band of intermediate signal intensity (arrows)

Figure 10 A coronal STIR image (TR 2200, TE 35,  160) showing the same lesion as that shown in Figure 9 (arrow). The mass is much more conspicuous on the STIR image

articular cartilage. Some investigators have demonstrated a large magetization transfer effect due to the high degree of interaction between water and collagen macromolecules in articular cartilage. Speci®cally, it has been shown that the majority of the magnetization transfer effect is caused by interactions between water and the collagen matrix, and not between water and the proteoglycan component of cartilage.55 This ®nding may prove useful in evaluation of subtle cartilaginous pathology, perhaps in conjuction with MR microscopy. MR spectroscopy has existed for many years. Some work has been done using spectroscopy to evaluate metabolic diseases of muscles, but none of these techniques is in routine clinical use as yet. In recent years several manufacturers have developed smaller, relatively inexpensive scanners designed to scan extremity joints. These scanners are often 50 to 80% less expensive than traditional whole-body scanners and can be installed in standard clinic examining rooms (e.g. Esaote, Genoa, Italy). Though these scanners typically operate at low magnetic ®elds (0.2 T) and produce relatively noisy images, the price, convenience, and comfort is highly attractive to many patients. These devices are currently the fastest growing segment of the MR market.

PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING

5 CONCLUSIONS In summary, MRI is a powerful tool in noninvasive evaluation of abnormalities of peripheral joints. The combination of high spatial resolution and sensitivity to local alterations in water content and T1 and T2 relaxation times makes it ideal for demonstrating pathological processes. MR has a role in the evaluation of many disease processes including trauma, infection, vascular compromise, and neoplasm. Additional techniques that may prove clinically useful in the future include MR microscopy, diffusion and perfusion imaging, magnetization transfer contrast, MR spectroscopy, and speciality scanners.

6 RELATED ARTICLES Gadolinium Chelate Contrast Agents in MRI: Clinical Applications; Gadolinium Chelates: Chemistry, Safety, and Behavior; MRI of Musculoskeletal Neoplasms; Skeletal Muscle Evaluated by MRI.

7 REFERENCES 1. J. Tehranzadeh, W. Mnaymneh, C. Ghavam, G. Morillo, and B. J. Murphy, J. Comput. Assist. Tomogr., 1989, 13, 466. 2. W. P. Chan, P. Lang, M. P. Stevens, K. Sack, S. Majumdar, D. W. Stoller, C. Basch, and H. K. Genant, Am. J. Roentgenol., 1991, 157, 799. 3. P. Dragorot and C. Claussen, ROFO, 1980, 133, 185. 4. J. V. Crues and D. W. Stoller, in `Magnetic Resonance Imaging of the Knee', 2nd edn., eds., J. Mink, M. Reicher, J. V. Crues, and A. Deutsch, Raven Press, New York, 1993, pp. 91±140. 5. I. P. C. Murray, J. Dixon, and L. Kohan, Clin. Nucl. Med., 1990, 15, 828. 6. J. H. Mink, M. A. Reicher, J. V. I. Crues, and A. L. Deutsch, `Magnetic Resonance Imaging of the Knee', 2nd edn. eds. J. Mink, M. Reicher, J. V. Crues, and A. Deutsch, Raven Press, New York, 1993, p. 474. 7. D. W. Stoller, `Magnetic Resonance Imaging in Orthopedic and Sports Medicine', Lippincott, Philadelphia, PA, 1993, 1127. 8. D. Resnick and M. Niwayama, eds. `Diagnosis of Bone and Joint Disorders', 2nd edn. W. B. Saunders, Philadelphia, PA, 1988, Vol. 2, 4174+. 9. B. Flannigan, S. Kursunoglu-Brahme, S. Snyder, R. Karzel, W. Del Pizzo and D. Resnick, Am. J. Roentgenol., 1990, 155, 829. 10. J. Hodler, S. Kursunoglu-Brahme, S. J. Snyder, V. Cervilla, R. P. Karzel, M. E. Schweitzer, B. D. Flannigan, and D. Resnick, Radiology, 1992, 182, 4316. 11. J. W. Helgason, V. P. Chandnani, and J. S. Yu, Am. J. Roentgenol., 1997, 168, 1473. 12. J. V. I. Crues, in `MRI of the Knee', eds. J. H. Mink et al., Raven Press, New York, 1993, p. 1. 13. J. V. Crues and R. Ryu, in `Magnetic Resonance Imaging', eds. D. Stark and W. G. Bradley, Mosby, St. Louis, MO, 1992, Vol. 2, pp. 2424±2458. 14. S. Koenig, R. D. Brown III, and R. Ugolini, Magn. Reson. Med., 1993, 29, 77. 15. J. Hennig and H. Friedburg, Magn. Reson. Imaging, 1988, 6, 391. 16. J. Listerud, S. Einstein, E. Outwater, H. Y. Kressel, Magn. Reson. Q., 1992, 8, 199.

7

17. R. T. Constable and J. C. Gore, Magn. Reson. Med., 1992, 28, 9. 18. D. A. Rubin, J. B. Kneeland, J. Listerud, S. J. Underberg-Davis, and M. K. Dalinka, Am. J. Roentgenol., 1994, 162, 1131. 19. J. A. Carrino, T. R. McCauley, L. D. Katz, R. C. Smith, and R. C. Lange, Radiology, 1997, 202, 533. 20. E. M. Haacke and J. A. Tkach, Am. J. Roentgenol., 1990, 155, 951. 21. M. P. Recht, G. A. Piraino, J. P. Schils, and G. H. Belhobek, Radiology, 1996, 198, 209. 22. M. J. Geirnaerdt, J. L. Bloem, F. Eulderink, P. C. Hogendoorn, and A. H. Taminiau, Radiology, 1993, 186, 813. 23. K. Herrlin et al., Acta Radiol., 1990, 31, 233. 24. B. C. Dangman, F. A. Hoffer, F. F. Rand, and E. J. O'Rourke, Radiology, 1992, 182, 743. 25. M. F. Reiser and M. Naegele, J. Magn. Reson. Imaging, 1993, 3, 307. 26. P. F. J. Tirman, A. E. Stauffer, J. V. Crues III, R. M. Turner, W. M. Nottage, W. E. Schobert, B. D. Rubin, D. L. Janzen, and R. C. Linares, Arthroscopy, 1993, 9, 550. 27. A. L. Deutsch, J. H. Mink, J. M. Fox, M. J. Friedman, and S. M. Howell, Magn. Res. Q., 1992, 8, 23. 28. A. L. Deutsch, J. H. Mink, and R. Kerr, `MRI of the Foot and Ankle'. New York, Raven Press, 1992, p. 378. 29. J. H. Mink, in `MRI of the Foot and Ankle', eds. A. L. Deutsch, J. H. Mink, and R. Kerr, Raven Press, New York, 1992. 30. T. C. Lynch, J. V. Crues III, F. W. Morgan, W. E. Sheehan, L. P. Harter, and R. Ryu, Radiology, 1989, 171, 761. 31. G. A. Bogost, E. K. Lizerbram, and J. V. Crues III, Radiology, 1995, 197, 263. 32. A. Vellet, P. Marks, P. Fowler, and T. Munro, Radiology, 1991, 178, 271. 33. G. M. Blum, P. F. J. Tirman, and J. V. Crues III, in `MRI of the Knee', 2nd. edn., eds. J. H. Mink, M. A. Reicher, J. V. Crues III, and A. L. Deutsch, Raven Press, New York, 1993, pp. 295±332. 34. A. Wang, D. Weinstein, L. Green®eld, L. Chiu, R. Chambers, C. Stewart, G. Hung, F. Diaz, and T. Ellis, Magn. Reson. Imaging, 1990, 8, 675. 35. C. P. Sabiston, M. E. Adams, and D. K. Li, J. Orthop. Res., 1987, 5, 164. 36. M. O. Senac, Jr., D. Deutsch, B. H. Bernstein, P. Stanley, J. V. Crues III, D. W. Stoller, and J. Mink, Am. J. Roentgenol., 1988, 150, 873. 37. H. J. Mankin, N. Engl. J. Med., 1992, 326, 1473. 38. D. G. Mitchell, M. E. Steinberg, M. K. Dalinka, V. M. Rao, M. Fallon, and H. Y. Kressel, Clin. Orthop., 1989, 244, 60. 39. H. Tsukamoto, Y. S. Kang, L. C. Jones, M. Cova, C. J. Herold, E. McVeigh, D. S. Hungerford, and E. A. Zerhouni, Invest. Radiol., 1992, 4, 275. 40. M. Cova, Y. S. Kang, H. Tsukamoto, L. C. Jones, E. McVeigh, B. L. Neff, C. J. Herold, J. Scott, D. S. Hungerford, and E. A. Zerhouni, Radiology, 1991, 179, 535. 41. M. J. A. Geirnaerdt, J. Hermans, J. L. Bloem, H. M. Kroon, T. L. Pope, A. H. M. Taminiau, and P. C. W. Hogendoorn, Am. J. Roentgenol., 1997, 169, 1097. 42. J. S. Jelinek, M. J. Kransdorf, B. M. Shmookler, A. J. Aboula®a, and M. M. Malawer, Radiology, 1993, 186, 455. 43. J. S. Jelinek, M. J. Kransdorf, B. M. Shmookler, A. A. Aboula®a, and M. M. Malawer, Am. J. Roentgenol., 1994, 162 919. 44. J. S. Jelinek, M. D. Murphey, M. J. Kransdorf, B. M. Shmookler, M. M. Malawer, and R. C. Hur, Radiology, 1996, 201, 837. 45. D. G. Varma, A. G. Ayala, S. Q. Guo, L. A. Moulopoulos, E. E. Kim, and C. Charnsangavej, J. Comput. Assist. Tomogr., 1993, 17, 414. 46. S. L. Hanna, B. D. Fletcher, D. M. Parham, and M. F. Bugg, J. Magn. Reson. Imaging, 1991, 1, 441. 47. P. Lang, G. Honda, T. Roberts, et al., Radiology, 1995, 197, 83.

8 PERIPHERAL JOINT MAGNETIC RESONANCE IMAGING 48. T. H. Berquist, R. L. Ehman, B. F. King, C. G. Hodgman, and D. M. Iistrup, Am. J. Roentgenol., 1990, 155, 1251. 49. R. Erlemann, J. Sciuk, A. Bosse, J. Ritter, C. R. Kusnierz-Glaz, P. E. Peters, and P. Wuisman, Radiology, 1990, 175, 791. 50. R. Erlemann, M. Reiser, P. Peters, P. Vasallo, B. Nommensen, C. R. Kusnierz-Glaz, J. Ritter, and A. Roessner, Radiology, 1989, 171, 767. 51. J. Rubenstein, M. Recht, D. G. Disler, J. Kim, and R. M. Henkelman, Radiology, 1997, 204, 15. 52. J. D. Rubenstein, J. G. Li, S. Majumdar, and R. M. Henkelman, Am. J. Roentgenol., 1997, 169, 1089. 53. K. B. Lehner, H. P. Rechl, J. K. Gmeinwieser, A. F. Heuck, H. P. Lukas, and H. P. Kohl, Radiology, 1989, 170, 495. 54. D. Le Bihan, Radiology, 1998, 207, 305. 55. D. K. Kim, T. L. Ceckler, V. C. Hascall, A. Calabro, and R. S. Balaban, Magn. Reson. Med., 1993, 29, 211.

Biographical Sketches Paul S. Hsieh. b 1962. B.S. (Mathematics), MIT, 1984; M.D. University of Michigan, 1989; residency in Diagnostic Radiology, Mallinckrodt Institute of Radiology, 1993; MRI training with John V. Crues, 1994; Faculty in Musculoskeletal Radiology at the Mallinckrodt Institute of Radiology 1994±1997. Currently staff radiologist, Kaiser Permanente, San Diego, CA. Research interests: musculoskeletal MRI. John V. Crues, III. b 1949. A.B., 1972, Harvard; M.S., 1975, physics, with Charles Slichter, University of Illinois; M.D., 1979 Harvard Medical School, residency in Internal Medicine 1982 and Radiology 1985 at Cedars-Sinai Medical Center. Currently Director of Magnetic Resonance at Cedars-Sinai Medical Center. Former President of the International Society for Magnetic Resonance in Medicine. Currently Medical Director of RadNet Management, Inc. and President, ProNet Imaging in Los Angeles, CA. Approx. 200 publications. Current research specialty: musculoskeletal MRI, picture archiving and communication systems.

PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS

Peripheral Muscle Metabolism Studied by MRS Peter A. Martin, Henry Gibson, and Richard H. T. Edwards Magnetic Resonance Research Centre and Muscle Research Centre, The University of Liverpool, Liverpool, UK

1 INTRODUCTION Muscle is a unique biological machine providing the main method of generating motive force, work, and power: it constitutes some 40% of total body cell mass in normal man. It is easily accessible for study and capable of great versatility in performance from a delicate touch to a powerful punch or from running a sprint to a marathon. Muscle metabolic rates can rapidly increase by over 100-fold. The body has nearly 600 muscles divided into three groups, skeletal (striated), visceral (non-striated), and cardiac; cardiac muscle is dealt with elsewhere (see NMR Spectroscopy of the Human Heart). Here we are only concerned with the muscles of the limbs, i.e., `peripheral' skeletal muscle, which was one of the ®rst tissues to be studied by magnetic resonance spectroscopy (MRS). Diseases of muscle (myopathies) are rare and often chronic and disabling, largely due to wasting of muscle tissue, which may be progressively replaced by fat or ®brous tissue, such as in the inherited X-linked muscular dystrophies. This creates a problem in trying to monitor muscle biochemistry because of serious partial volume effects, i.e., any volume of muscle under study contains an abnormally high proportion of fat or ®brous tissue as is readily seen on muscle biopsy. Quite different are the speci®c cellular enzyme and membrane defects of muscle, where this fat replacement is absent or less evident. They represent `Nature's experiments' where the impairment is due to interference with some essential metabolic pathway. They constitute an interesting group of disorders which are particularly amenable to study by in vivo MRS. Magnetic resonance imaging (MRI) has been con®ned largely to the study of anatomy and gross pathology (see Skeletal Muscle Evaluated by MRI), whereas MRS has mostly been con®ned to metabolic studies of phosphorus (31P)-containing compounds, although some work has been done using hydrogen (1H)1,2 and carbon-13 (13C).3 Fortunately, the phosphorus-containing compounds visible by MRSÐphosphocreatine (PCr), adenosine triphosphate (ATP) and inorganic phosphate (Pi)Ðare of great interest since they are involved in the energy metabolism of the muscle cells enabling MRS to be used to study muscle energetics noninvasively. 2 MUSCLE STRUCTURE AND FUNCTION Human muscle comprises at least two populations of ®bers with different functional and metabolic characteristics. In

1

health, the distribution of ®bers in particular muscles re¯ects the physiological function of the muscle, whether for postural control (e.g. soleus) or for rapid movement (e.g. biceps). Muscles are organized in bundles (fascicles) and form into groups with a particular gross structure characteristic of that muscle (e.g. pennation angle, which can be determined by MRI4). Muscles work together as `agonists' and `antagonists' to achieve a particular force or movement. The protein chemistry and ®ne structural features responsible for force generation in muscle are beyond the resolution of whole body MR systems, and are, thus, not covered here. The muscle ®bers (cells) have an outer membrane, the sarcolemma, ®lled with a ¯uid, sarcoplasm, containing the endoplasmic reticulum, mitochondria, and the many peripherally located nuclei. The myo®brils, which constitute a large component of the cell, consist of a matrix of interdigitating actin and myosin protein chains. Muscle performance is achieved by the action of actin and myosin chains sliding over each other, shortening the length of the sarcomere and thus the muscle. Muscles contract in response to a nerve impulse and each muscle cell must therefore have a neuronal connection. However a single neurone will have from a few to several hundred branches, each connecting with muscle cells that all contract together. This group forms a motor unit and the recruitment of an increasing number of motor units is one of the main factors governing the control and precision of any force generated. Initiation of a contraction depends on a nerve signal being transmitted to the muscle via a chemical messenger acetylcholine. This causes an `action potential' to be produced in the form of a region of electrical depolarization on the sarcolemmal membrane. The action potential rapidly travels the length of the muscle cell releasing calcium from the lateral cisternae of the endoplasmic reticulum. The calcium release activates actomyosin ATPase and triggers myo®brillar cross bridge interaction and thus force generation. A single action potential produces a single twitch involving the entire muscle ®ber. Gradation of strength depends on changing the number of muscle ®bers that are active. For a prolonged contraction of greater force, multiple stimuli are used, and if the frequency of stimulation is high enough the individual twitches fuse together to produce a smoothly sustained tetanic contraction, which lasts longer and is several-fold stronger than a single twitch. The two basic types of skeletal muscle ®bers contract at different speeds; the `red', slow-twitch (or type I) ®bers are best suited to prolonged aerobic exercise, whereas the `white' fast-twitch (or type II) ®bers are best adapted to high intensity exercise that is largely anaerobic; intermediate ®ber types also occur, and their prevalence depends on the activity history of the muscle (see Table 1). The importance of the recognition of two main ®ber type populations in human muscles is that regional metabolic inhomogeneity can develop within the muscle as a consequence of particular forms of muscular activity in which individual populations of muscle become fatigued to different extents, resulting in their discrete appearance in the spectra. [e.g. split inorganic phosphate (Pi) peak]. Another consideration which can confuse this interpretation is that the sensitive volume of tissue studied may include more than one agonist muscle group, of which only one may be active.

2 PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS Table 1 Summary of broad characteristics of main ®ber types in human skeletal muscle Characteristic

Slow twitch, Type I

Fast twitch, Type II

Contraction time Relaxation time Myosin ATPase activity Fatigability Phosphocreatine content Oxidative enzyme activity Capillary density Mitochondria Glycogen content Fat content Myoglobin content

Long Long Low Low Low High High Numerous No difference High High

Short Short High High High Low Low Few No difference Low Low

3 MUSCLE AS A BIOLOGICAL MACHINE When considering muscle function it is important to recognize the relationships between energy, force, work, and power. In the human body energy, measured in joules (J), is available for muscle metabolism via either the long- or the short-term energy supply processes. The maximum force, measured in newtons (N), that a muscle can generate (its strength), depends on the cross-sectional area of active muscle ®bers. Comparison of working muscles requires that they are doing the same amount of work, i.e. force  distance, or generating the same power output (force  distance/time) which is measured in watts (W). The work done by the muscle will equal the energy expended only if the muscle is 100% ef®cient. Most muscular activity is about 20% ef®cient, the remainder of the energy being dissipated as heat. The maximum power output of the muscle depends on its maximum force and its velocity of shortening. During isometric exercise, when the muscle contracts without altering its length, external power output is zero; however the force±time integral can be used as an indicator of muscular activity.5

the 1.5±2 T horizontal bore, whole body magnet systems, necessitate that the subject should lie down in the magnet. MR systems have magnets that are designed to have good homogeneity, but they also produce strong fringe ®elds. Commercially available ergometers are made of ferromagnetic materials or involve electrical motors or dynamos, which means that they will not work properly in the close proximity of strong magnetic ®elds of the MR system; such ergometers will in turn destroy the homogeneity of the magnetic ®eld. In the future, we may expect both these problems to be resolved as new magnet designs are becoming available that will enable the subject to stand up within the magnet system and carry out exercise routines such as running or cycling, which are more in keeping with normal human activity. Secondly, improved magnetic shielding is reducing the stray ®eld of the magnet systems to such an extent that conventional ergometers can be used within 1 m or so of the MR magnet. However it is unlikely that the need for specially designed nonmagnetic ergometers will disappear; indeed the opportunities opened up by this new ®eld will require new and more sophisticated designs. Figure 1 shows an example of an apparatus that has been used for the study of isometric exercise in the quadriceps muscle during both voluntary and electrically stimulated exercise studies.6 The subject lies supine with the knees bent and resting at an angle of 120 over a polystyrene foam block. The ankle is restrained by a strap which is attached to a force transducer while a wide belt across the waist prevents subjects from using their back and abdominal muscles to aid in the exercise. Stimulation electrodes strapped on with crepe bandages are applied both superior and inferior of the quadriceps to allow electrical stimulation of the muscle. The electrodes are connected via screened leads and a radiofrequency ®lter to a

Ankle strap

Aluminum bar

4 PRACTICAL ASPECTS OF ERGOMETRY Of vital importance in the study of any particular muscle is the need to be able to measure accurately and objectively its force, work, and power output. In the body, muscles in vivo form part of a complex machine, with their diverse attachments to bones and the agonist/antagonist arrangement by which they work together or against each other respectively. Sensible force measurement depends on the careful design of the ergometers used. The `man±machine interface' thus becomes of paramount importance to provide an objective work standard against which NMR measurements may be correlated. Exercise can be dynamic such as running or walking or isometric (static) as in holding various body postures or a heavy weight. Although much human activity consists of dynamic exercise the constraints of space within the magnet system mean that isometric exercise is more easily studied. Although many ergometers are available, their use in conjunction with MR systems is limited, not least because human muscle studies by MRS carried out in

Strain gauge

Mirror LED display Surface coil

Restraining strap

Stimulation electrodes (secured by bandages)

Figure 1 Apparatus for the study of the quadriceps muscles

PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS

commercial stimulator and consist of 2 cm2 of copper over a conducting gel pad, 5 cm  10 cm, cut from a commercial de®brillator pad. The force transducer consists of an aluminum bar and strain gauge connected to a preampli®er, the output of which passes out of the scan room via a radiofrequency ®lter. Visual and audio feedback of the measured force are provided to the subject from a LED bar graph and tone generator to enable the subject to make voluntary contractions to prescribed target forces. Red, yellow and green LEDs under control from the spectrometer's computer are used to tell the subject when to exercise and to provide warning of forthcoming electrical stimulation. The limb may be made ischemic (blood supply stopped) by simply in¯ating a thigh sphygmomanometer cuff to 100 mmHg above systolic blood pressure, thereby providing a closed system trapping metabolites and preventing oxidative recovery processes. Measurement of myographic activity is also possible without deterioration of the magnetic homogeneity.7

13C

The spectacular improvement in MRI has not been matched by a comparable improvement in in vivo MRS over the 15 or so years since the introduction of the ®rst small-bore in vivo spectroscopy systems. Clinical applications of spectroscopy have been slow to appear, and it is still not a routine diagnostic tool. An illustration of the range and quality of spectra that can be obtained is shown in Figure 2. This shows a set of spectra from the forearm of a normal boy and a dystrophic boy with Duchenne muscular dystrophy.8 The 1H spectrum for the dystrophic forearm shows a decrease in the water peak and an increase in the fat peak. The 13C spectrum shows a lot more peaks, but at the resolution achievable with the 1.5±2.3 T of most in vivo systems, this is dif®cult to interpret with many overlapping peaks. The peaks due to CH3 and CH2 groups are clearly visible, however, and the dystrophic forearm shows a general increase in the 13C signal, which is most obvious in fat signals in the CH2 region. The phosphorus spectrum shows a large peak due to PCr and three peaks due to ATP, as well as a small peak due to Pi. The dystrophic 31P spectrum shows a decrease in the total signal of phosphorus metabolites as the fat displaces the muscle. The utility of MRS for muscle studies depends primarily on the quality of the spectra that can be obtained. Quantitative analysis requires as good a signal-tonoise ratio (S/N) as possible for all the peaks of interest. Qualitative studies can tolerate a poorer S/N but here extra care needs to be taken in the interpretation of results. Many factors interact to in¯uence the spectral quality, e.g. the ®eld strength and homogeneity of the magnet, the size and location of the muscle to be studied, the magnetic sensitivity of the nucleus, e.g. 31P, 13C, 1H, the type of radiofrequency coil to be used, the MR localization technique, and the type of muscular activity to be studied. The ®rst localization technique used in muscle studies was `surface coil localization'.9 This is suitable for all nuclei and relies entirely on the localized but inherently nonuniform response of a simple loop antenna (the surface coil) to restrict the region from which signals would be obtained (see Surface Coil NMR: Detection with Inhomogeneous Radiofrequency Field Antennas); when both transmitter and receiver surface

spectra

Normal

Dystrophic

180 140 100 60 20

1H

180 140 100 60 20 ppm

spectra

Normal

5 MRS TECHNIQUES FOR MUSCLE STUDIES

3

Dystrophic

15

10

5

0

–5 15 10

31P

5

0

–5 ppm

spectra

Normal

Dystrophic

10

0

–10 –20

10

0

–10 –20 ppm

Figure 2 Resting 13C, 1H, and 31P NMR spectra of the forearm. The spectra on the right are from a 9-year-old with Duchenne's dystrophy, and on the left are from an age-matched control with similar skin thickness

coils are used they have a sensitive volume which forms a hemisphere penetrating to a depth of roughly one coil radius. This was later supplemented by the use of magnetic ®eld localization techniques of which the most widely used for muscle studies has been the topical magnetic resonance (TMR) method.10 This used a static, high order (Z4) gradient to spoil deliberately the magnet homogeneity over all but the chosen small diameter, central region of the magnet; any signals coming from this uniform central volume would be sharp peaks, whereas any signals coming from outside this volume would appear as a broad hump in the baseline, which could be removed from the ®nal spectra via a convolution difference technique. Topical magnetic resonance had the big advantage that good localization (superior to that of a surface coil alone) can be obtained from a single acquisition, making it suitable for studies requiring time resolution of the order of 1 or 2 s. It had the disadvantage of relying on static magnetic ®eld gradients, which could not be switched rapidly on and off, and which only produced a uniform ®eld at the center of the magnet, requiring the muscle of interest to be located there. This put

4 PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS Table 2

31

P relaxation times in human skeletal muscle Field (T)

-ATP

-ATP

-ATP

1.5 1.5

3.603 22.2a

4.310 8.1a

4.755 16.1a

T1(s) T2(ms)

PCr 5.517 424.3

Pi 4.017 204.7

Ref. 16 14

a

These measured values are probably lower than the true in vivo values due to the effects of J coupling which were not taken into account in this study17

physical and anatomical constraints on what could be studied. Thus, the TMR method rapidly fell out of favor once magnets with switched gradients became available holding the promise of combining imaging and spectroscopy in studies and pulsed gradient localization techniques, allowing localization of any volume within the (larger) homogenous region in the center of the magnet. An alternative localization strategy that has been used by some groups is to use one-dimensional, `Rotating Frame' chemical shift imaging;11 this has not been widely adopted, however, largely because of the T2 distortion to which it is susceptible. This shows itself primarily as a decrease in the area of the -ATP peak which is a particular disadvantage since this peak is often used as an internal reference for quantitation. Of the many gradient localization techniques that have been introduced over the last 10 years, few of them have found practical application in muscle studies. This is because most muscle studies have used 31P spectroscopy. Carbon-13 spectroscopy has been done at natural abundance,12,13 but generally the most useful results are obtained with the introduction of 13C labeled compounds, but with greatly increased cost. Water-suppressed 1H spectroscopy has not yet been widely applied to muscle and for the most part unsuppressed 1 H spectra are only of limited value. Compared with protons, phosphorus nuclei tend to have short T2 relaxation times,14 which makes them unsuitable for any localization technique that relies on long echo time, spin, or stimulated echoes, e.g. Pixel-RESolved Spectrocopy (PRESS) and the STimulated Echo Acquisition Method (STEAM). One simple localization technique which has found some application in muscle specTable 3

troscopy is Depth-REsolved Surface coil Spectroscopy (DRESS). This consists of a conventional slice selection followed by collection of an FID signal, though the loss of some signal, particularly from -ATP, is a disadvantage, and baseline distortions caused by the long delays between excitation and data acquisition make subsequent processing and any attempts at accurate quantitation dif®cult.15 The best true volume localization technique for 31P spectroscopy is the Image-Selected In vivo Spectroscopy technique (ISIS). This has the advantage that it does not use spin echoes and thus is capable of giving good, undistorted spectra suitable for quantitation, but ISIS suffers from the serious disadvantage that it is a differencing technique requiring a minimum of eight signals for full volume localization. This makes it highly susceptible to motion artifacts and thus completely unsuitable for exercise studies. The quality of localization is affected by T1 effects and thus, for muscle studies, it should ideally be run at repetition times (TR) of 15 s, making it unsuitable for short-duration time course studies. Because of the problems associated with volume localization techniques most muscle studies still use surface coil localization alone. This presents some dif®culties for quantitation since the nonuniform response of the coil results in a variation of the ¯ip angle throughout the sensitive volume causing variation in signal strength due to T1 and T2 effects (see Table 2). This has been ameliorated a little with the introduction of specially designed rf pulses designed to give uniform ¯ip angles over a larger proportion of the coil's sensitive volume (see Surface Coil NMR: Detection with Inhomogeneous Radiofrequency Field Antennas).

Energy sources for muscular activity Short-term (anaerobic) energy sources

ATP hydrolysis:

myosin-ATPase Adenosine triphosphate (ATP) + H2O ÐÐÐÐÐÐÐ ! adenosine diphosphate (ADP) + inorganic phosphate (Pi) + energy

Creatine kinase reaction:

Anaerobic glycolysis:

creatine kinase + Phosphocreatine + ADP + 0.9 HÐÐÐÐÐÐÐ ! creatine + ATP Glycogen (glycosyl unit) + 3 Pi + 3 ADP ÐÐÐÐÐÐÐ ! 2 lactate + 3 ATP Long-term (aerobic) sources

Oxidative phosphorylation:

Glycogen/Glucose/Free fatty acids/Free amino acids ÐÐÐÐÐÐÐ ! NADH NADH + 1.5 H+ + 0.5O2 + 3ADP + 3Pi ÐÐÐÐÐÐÐ ! H2O + NAD+ + 3ATP

PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS

In vivo spectroscopy is currently only capable of `seeing' narrow lines produced from long T2 metabolites, i.e. those in solution in the cytoplasm. These metabolites become MR invisible or only give broad lines, which appear as a hump in the baseline, when in a viscous medium or when bound to the mitochondrial matrix. This is an advantage in that only the cytoplasmic metabolites are important in assessing the muscle energy metabolism during exercise. In vivo quanti®cation has always been dif®cult. As a result, many muscle studies report their results in terms of ratios of peak areas. An approximation to quantitation can then be made if appropriate assumptions are made, e.g., the total MRS visible phosphorus concentration or the ATP concentration. This enables estimates to be made of the 31P-containing metabolite concentrations. A common reference method is by comparison with -ATP concentration. Improved quantitation can be obtained if some form of external or internal standard is used. Internal standards have to be inherent in the muscle under study and thus one method is to collect a 1H spectrum and to use the water peak as a reference. One problem with this is that the water content of muscle varies, especially during exercise. Assumptions about the water concentration cannot therefore be used. To overcome this, several water spectra at different TRs must be obtained and the true H2O concentration calculated, but again this is not suitable for exercise studies. An acceptable method is to use an external phantom, often attached to the coil as a reference, and the 31P concentration in vivo is determined by comparison with that from a series of different concentration phosphate solutions external to the limb studied or suitably substituted for the limb after the study.

6 METABOLIC PATHWAYS Most MRS studies of muscle concern the biochemical pathways (see Table 3) for the supply and utilization of energy. The ®rst reaction to be considered is the hydrolysis of ATP to ADP, catalyzed by myosin ATPase, which produces the energy-driving muscular contraction. Normal muscle only contains a small amount of cytosolic ATP and so can only sustain contractile activity for a very short period of time before fresh supplies of ATP must be made available. These supplies are obtained not from stores of ATP itself but via a reservoir of energy in the form of PCr which is used in the cytosol by the creatine kinase reaction to recycle ADP rapidly back to ATP again. The resulting creatine (Cr) is transported to the mitochondria of the cell where the reverse reaction occurs catalyzed by creatine kinase and Cr is converted back to PCr at the expense of mitochondrial ATP; this is the oxidative phosphorylation reaction or mitochondrial respiration. Under aerobic conditions, i.e. when the O2 supply from the blood is maintained, mitochondrial respiration supplies most of the muscles' energy requirements in the form of ATP. This results in production of one-tenth of a mole of H+, from carbonic acid, per mole of ATP. When the energy demand exceeds the mitochondrial capacity or when anaerobic (ischemic) conditions exist, only PCr and glycogenolysis can supply the ATP for contraction. When anaerobic metabolism takes place, large amounts of lactate are therefore produced from glycogenolysis, giving two-thirds of a mole of H+ per mole of ATP leading to a large decrease in intracellular pH.18 This fall in pH may

5

have profound effects on pH-sensitive metabolic processes (see e.g., Table 1). Phosphocreatine hydrolysis consumes protons whereas PCr synthesis produces protons. Fortunately it is easy to measure pH by 31P MRS, and after taking into account the effects of buffering and proton ef¯ux, i.e. loss of protons from the cell, the rate of proton production can be calculated. Thus the effect of pH on ATP ¯uxes can be allowed for. Furthermore, it is believed that pH affects cross bridge kinetics19 leading to a reduction in force generation, i.e. fatigue.

7 7.1

THE STUDY OF MUSCLE USING

31

P MRS

Estimation of pH

Inorganic phosphate (Pi) in vivo has a pKa of about 6.75 at physiological pH and exists in an equilibrium between two 2ÿ forms H2POÿ 4 and HPO4 . These give rise to two separate phosphate resonances, 2.3 ppm apart, which undergo rapid chemical exchange (109±1010 sÿ1), resulting in a spectrum containing a single resonance whose frequency depends on the relative amounts of the two moieties. Since the equilibrium between the two forms of Pi is pH-dependent, the chemical shift of the Pi peak can be used as an indicator of pH by measuring its chemical shift either relative to an external standard such as methylenediphosphonate or relative to internal standards such as the pH-insensitive resonances due to the phosphorus PCr peak or the proton water signal. In the normal, resting, human forearm muscle of Figure 2, the chemical shift of Pi from PCr is 5.00 ppm corresponding to a pH of 7.15, whereas during exercise the chemical shift of Pi falls to 4.66 ppm corresponding to a pH of 6.88, due to the presence of lactate. The pH may be calculated from the equation. pH ˆ 6:75 ‡ log‰… ÿ 3:27†=…5:69 ÿ †Š

…1†

where  is the chemical shift difference, in ppm, between the Pi and PCr peaks. 7.2

The Buffering Capacity of Muscle

A knowledge of the buffering capacity is necessary for determination of ATP ¯uxes since the H+ ions are involved in the equilibria (see Table 3). During the initial part of aerobic exercise the assumption can be made that the proton ef¯ux can be neglected, enabling the glycogenolytic rate to be estimated in the same way as for ischemic exercise. This assumption falls down when the buffering capacity has been exhausted and the pH starts to fall. The cytosolic buffering capacity of skeletal muscle depends on Pi (pK = 6.57), bicarbonate (pK = 6.1), and other buffers, largely imidazole groups in histidine residues. It has been shown20 that in a closed system where the total CO2 is constant (e.g. during ischemia) the buffer capacity is given by: For Pi: ˆ 2:3 ‰PiŠ=f‰1 ‡ 10…pH ÿ 6:75† Š ‰1 ‡ 10…6:75 ÿ pH† Šg

…2†

where is measured in slykes (i.e. mmol Lÿ1 per pH unit).

6 PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS For bicarbonate: ˆ 2:3 S pCO2 10…pH ÿ 6:1† =f‰1 ‡ 10…pH ÿ 6:1† Š ‰1 ‡ 10…6:1 ÿ pH† Šg

…3†

where S is the solubility of CO2. Taking pCO2 as 5 kPa and S as 0.3 mmol Lÿ1 kPaÿ1 then at its resting pH in closed muscle is less than 5 slykes. For other buffers: = 20±30 slykes; inferred by analysis of 31P MRS data and from measurements in muscle homogenates.20 7.3 Muscle at Rest Many studies have been performed on resting muscle but these have frequently been hampered by the wide range of physiological states of the muscle under study, largely due to the intrinsic variations in the level of training in the populations under study. One of the ®rst and most important ®ndings from studies of resting muscle has been the observation that the concentration of PCr is consistently higher when measured by 31P MRS than when measured by freezeclamped needle biopsy.8 This is almost certainly due to the rapid hydrolysis of PCr to Pi during the freeze-clamping process and this interpretation can be supported by the observation that the total phosphorus concentration [PCr + Pi] obtained by both methods is approximately the same. No such discrepancies between the techniques have been observed with ATP, possibly due to equilibration of the creatine kinase reaction during the extraction procedures. Most quantitative 31P MRS studies on resting muscle have relied not on true quantitation, but on the assumption that the ATP levels are reasonably constant. This assumption requires care as there is evidence that changes in the resting levels of metabolites depend on the prior history of the muscle. If the muscle has been involved in exercise involving lengthening contractions,21 the PCr/Pi ratio is signi®cantly reduced up to 1 h after exercise; the reduction continues with the ratio reaching a minimum at 1 day postexercise and remaining low for between 3 and 10 days postexercise. Similarly, abnormal spectra have been reported for up to 2±3 days after short-term exercise, even when no muscle ®ber damage is thought to have occurred. Much evidence is now accumulating to show that type I and type II ®bers have different PCr/Pi ratios, with type II ®bers having elevated PCr and ATP compared with type I. The resting ratios and changes discussed above may be related to the different ®ber-type ratios found in normal subjects.22

7.4 Exercise and Fatigue While much information about muscle energy status can be obtained at rest, it is during exercise that the most dramatic changes are seen owing to the high metabolic exchanges associated with muscular activity. Particular interest lies in the study of individuals with defects of metabolism, which can give information about metabolic processes not normally accessible. This is discussed further in Section 8. Metabolic requirements may be determined at various workloads and metabolic processes may be related to physiological changes in function. Muscle fatigue, the decline in force or power output with prolonged activity, has received much attention in this respect, but is complicated by the type of muscular activity

undertaken and the many cellular physiological factors that appear to be interrelated with the chemical changes taking place. For this reason, and the importance of maintaining and improving muscle performance in sports and disease, the various known mechanisms contributing to fatigue are described below. The central contribution to fatigue can be measured by comparing force output between alternate periods of voluntary contractions with that produced from electrical stimulation of a peripheral nerve. If the voluntary force output falls more than the stimulated force output the difference is due to fatigue of central motor control mechanisms rather than of the muscle itself.23 Except in the rare neuromuscular disease myasthenia gravis, fatigue due to failure of the neuromuscular transmission is rare. Much of the research of the last half century in this ®eld has been directed toward gaining an understanding of the extent to which impaired energy supply or electromechanical coupling failure is the dominant problem in a particular form of muscular activity.24,25 Fatigue of the muscle itself can be classi®ed according to the response to different frequencies of electrical stimulation. Fatigue that is produced more with high-frequency electrical stimulation (HFF) occurs in myasthenia gravis. Lowfrequency fatigue (LFF), in which there is selective impairment of force generation at low frequency, i.e., reduced 20: 50 Hz force ratio, can occur for a long time after ischemia26 or an eccentric muscular contraction27 when force generation at high frequency has recovered. From the energetics point of view HFF and LFF can be distinguished by the fact (from needle biopsy and MR studies early in recovery from severe ischemic exercise) that with HFF, recovery of force may occur before full recovery of PCr/ -ATP (or Pi/PCr), due to rapid recovery of membrane excitation (compound muscle action potential; CMAP). With LFF, force is still reduced 1 h after exercise when CMAP and PCr/ -ATP have recovered.28 7.4.1

ATP Turnover in Ischemic Exercise

In ischemic (anaerobic) exercise (i.e. when the blood supply to the muscle has been occluded), ATP is produced from two sources, the hydrolysis of PCr catalyzed by creatine kinase (= PCr depletion) and glycogenolysis, the breakdown of glycogen to lactic acid.29 Knowledge of [PCr] and pH can be used to estimate the rate of ATP synthesis. Since decay of the PCr peak can be measured directly, the rate of ATP synthesis from the hydrolysis of PCr (D) can also be measured directly: D ˆ ÿ‰PCrŠ=t

…4†

Hydrolytic ATP synthesis also results in a net proton consumption at the rate: net proton consumption ˆ D=‰1 ‡ 10…pH ÿ 6:75† Š

…5†

The production of ATP via glycogenolysis (L) can be estimated from the effect of lactic acid production on pH. Since during ischemic exercise protons cannot escape from the system, glycogenolysis (from glycosyl units) produces 2 mol of lactic acid thereby releasing 3 mol of ATP.29 Thus:

7

PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS L ˆ …3=2†fD=‰1 ‡ 10…pH ÿ 6:75† Š ÿ pH=tg

where is the cytosolic buffering capacity of the muscle (derived as above). Since the total rate of ATP synthesis (F) is: F ˆD‡L

PCr

…6† Pi

Normal subject

…7†

18.4

7.11

15.9

Exercise 13.5 Exercise Exercise Cuff on 8.2 Exercise Exercise

7.18

…8†

4.4

7.18

ATP Turnover in Aerobic Exercise

In aerobic (oxidative) exercise ATP is produced from PCr and the oxidative synthesis of ATP in the mitochondria. As before, the rate of depletion of PCr and thus the rate of hydrolytic ATP production is readily accessible by MRS. However the presence of oxidative ATP synthesis and the fact that the system is no longer closed, i.e. there is a net ef¯ux of protons from the cells, complicates the assessment of total ATP production. One strategy has been to compare aerobic with ischemic exercise, making use of the power output relationships derived from graded ischemic exercise to provide the `missing data' from aerobic exercise studies at the same power output. At the start of aerobic exercise, proton ef¯ux can be neglected and the glycogenolytic rate estimated as for ischemic exercise. At the end of ischemic exercise, when the pH is falling, estimates of glycogenolytic ATP synthesis rates must be corrected for the large proton ef¯ux from the system. This can be done by assuming the proton ef¯ux rate has a linear relationship to pH and [PCr] and thus can be inferred from the initial phase of recovery from exercise. 7.5

20.5

Cuff off

7.11

7.4.2

Recovery

The study of postexercise recovery provides an opportunity of studying the kinetics of recovery of muscle metabolites, and in particular it can be used to provide information on the pHdependence of proton ef¯ux, which can also be applied as a correction during the analysis of aerobic exercise.30 After the completion of exercise, the accumulated ADP continues to stimulate mitochondrial respiration to resynthesize ATP; as a consequence, the cytoplasmic PCr pool is replenished through the creatine kinase equilibrium. The PCr replenishment during recovery from exercise has been shown to depend entirely on mitochondrial respiration by the absence of any metabolic recovery when the muscle is made ischemic after exercise by rapid in¯ation of a sphygmomanometer cuff.31 Thus, we can consider the rate of PCr resynthesis to be a good index of mitochondrial function.32 The recovery of PCr after exercise appears to follow an exponential time course and experimentally is usually treated as such although it has been shown to be biphasic,33 showing an initial rapid rise followed by a much slower rate of recovery back to resting level. The rate of PCr resynthesis is dependent on the extent of intracellular acidosis, which in turn depends on the work rate during exercise. The ®nal rate of PCr recovery is dependent on the rate of pH recovery and a direct linear relationship has been shown between the value of intracellular pH at the end of exercise and the rate of PCr recovery.34

ATP b

7.05 7.11

the total turnover of ATP is: F ˆ D ‡ …3=2†fD=‰1 ‡ 10…pH ÿ 6:75† Š ÿ pH=tg

g a

7.18

2.0 10

0

Time (min)

Exercise

–10 –20 ppm PCr

Phosphofructokinase deficient

7.24

7.24 g a PME Pi

ATP b 18.7

7.18 16.6

Cuff off

7.18 14.1

Exercise 11.2 Cuff on

7.18 7.18

8.8

7.24

6.4

Exercise

Exercise

Time (min)

Exercise 4.0

10

0

–10 –20 ppm

Figure 3 A comparative 31P NMR exercise study of a patient with phosphofructokinase de®ciency and a normal subject

At the end of exercise, intracellular pH continues to fall until a minimum which, depending on the work rate during the exercise, occurs approximately 1 min after the end of the exercise period. The higher the work rate the later pH recovery begins (see Figure 3). The mechanisms that control intracellular pH are not well known though it is likely that active transport mechanisms are involved.35 During recovery [Pi] generally mirrors [PCr]. The Pi accumulated during exercise is transported into the mitochondria where it is used in the phosphorylation of ADP. Like PCr, Pi recovery is biphasic.36 After the end of exercise and during the period in which pH is still falling Pi recovery is fast. Subsequently, as the pH recovers, Pi recovery slows down, and [Pi] decreases to undetectable levels for several minutes before it again reappears and recovers to resting levels. During this time a temporary decrease in the total [PCr + Pi] occurs, which otherwise remains constant throughout exercise. When exhaustive exercise has been carried out, such as to show a reduction of [ATP], e.g. strenuous aerobic exercise, then all recovery processes are substantially impaired.37 8 STUDIES OF MUSCLE DISEASE Much of what is known about muscle energy metabolism in disease has come from needle biopsy studies. MRS affords a noninvasive approach to the study of metabolism in disease, but the diagnostic value of MRS in muscle is still yet to be

8 PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS Table 4

Diseases studied by MRS of skeletal muscle (after Barbiroli39)

Disease

Nucleus

Cause or defect

Conclusions from MRS

Ref.

Congenital heart disease

31

Impaired O2 cyanosis

Resting pH and Pi elevated; abnormal PCr depletion and acidi®cation during exercise; prolonged recovery times

40

Duchenne/Becker dystrophy

31

Inherited lack of dystrophin

Progressive replacement of muscle tissue by fat; high resting Pi/PCr ratio and slightly alkaline pH; PDE peak increases with age; altered energy metabolism in carriers

41±43

Mitochondrial myopathy

31

Various enzyme defects

Abnormal transfer function: slow recovery of PCr after exercise

44,45

Encephalomyopathy

31

Mitochondrial enzyme defects

Abnormal transfer function and slow recovery of PCr and pH; abnormal brain energy metabolism

46,47

Glycogenosis

31

Phosphofructokinase de®ciency

Limited acidosis, abnormal build up of PME peak during exercise

48,49

Phosphoglycerate mutase

Limited acidosis and abnormally raised PME peak during exercise

50

51,48

P

P, 1H

P P

P

McArdle's disease

31

Myophosphorylase de®ciency

Lack of acidosis during sustained aerobic or ischemic exercise

Sickle cell anemia

31

Muscle ischemia

Reduced total mals

P compared with nor-

52

Malignant hyperthermia

31

Anesthetic-induced hyperthermia

High resting Pi/PCr ratio; slow postexercise recovery of PCr/Pi ratio

53

Peripheral vascular disease

31

Relative ischemia

Acid shift and a fall in PCr/Pi with exercise

54

P P P P

shown. The advantage of MRS here is in continually providing information of metabolic levels during exercise (whereas repeated biopsy samples were previously required), although resting spectra can provide valuable metabolic information in certain myopathies such as in dystrophy38 and muscle injury.21 The question of whether metabolic abnormalities are secondary to other disease processes is becoming of increasing interest clinically in view of the consequences of disease with respect to muscle pain, performance, and fatigue. Some examples are shown in Table 4. A further application has been to study the effects of therapy, for example, in altering muscle energetics, where enzyme defects have limited the energy supply for contraction, such as glucose infusion in McArdle's disease (myophosphorylase de®ciency in which glycolysis is impaired),55 drug trials in Duchenne dystrophy,38 and the consequences of insulin infusion on phosphate metabolism.56 The early applications of MRS in disease, in particular to patients with clear metabolic defects, provided a novel means to study the normal physiological mechanisms limiting exercise. Such patients can be considered to represent `Nature's experiments' allowing examination of metabolic processes under conditions that would normally not be possible in human

31

muscle. McArdle's disease57 represents an example where a defect in metabolism may manifest itself as a failure of membrane excitation, highlighting the importance of glycolysis for the maintenance of membrane excitability. No lactate is produced in these individuals and hence no acidic shift in the Pi peak occurs, making it possible to measure initial rates of PCr resynthesis and Km for ADP control of oxidative phosphorylation.51 Figure 3 illustrates the metabolic changes seen during exercise in a patient with phosphofructokinase de®ciency,58 which is arguably a more signi®cant metabolic defect than McArdle's disease, in that oxidation of blood-borne glucose is not possible. Accumulation of fructose 6-phosphate traps Pi resulting in a relatively small rise in Pi, and again no change in pH occurs. In these patients fatigue occurs rapidly, but this cannot be attributed to Pi or H+. Another group of the metabolic myopathies is represented by the mitochondrial abnormalities. Those patients affected show abnormalities in oxidative phosphorylation and consequently a high Pi and low PCr,59 and slow PCr resynthesis.60,61 A number of these patients have been studied with various defects in mitochondrial function, with slowed PCr and ADP recovery, although the most reliable indication comes from the resting spectrum.

PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS

Abnormalities in MRS parameters in Duchenne muscular dystrophy have been discussed earlier (see Section 5). The question of whether there is a reduction in energy state of the dystrophic tissue has been dif®cult to answer owing to the increasing proportion of muscle tissue replaced with fat and connective tissues as the disease progresses,38 and highlights an inherent problem in acquiring a signal from an inhomogenous tissue. Attempts to correct the acquired phosphorus signal by dilution effects from noncontractile tissue have been made from biopsy samples and indeed several studies have shown a reduced PCr/ATP ratio in dystrophic muscle. Moreover, a reduction in PCr/Pi ratio suggests impairment of mitochondrial function62 which is likely to be secondary to the dystrophic process, probably owing to disuse of the muscle, rather than contributing to the damage process, as greater changes are seen in patients with enzyme defects in whom histological evidence of damage is not apparent. Indeed, 31P MRS has been reported to be suf®ciently sensitive to show abnormalities in metabolism in patients with minimal or no muscle weakness.39 The most marked changes in muscle metabolism demonstrated in disease are probably secondary in origin, particularly where impairment of blood ¯ow is evident. Examples include peripheral vascular disease,54 sickle cell anemia,52 and congenital heart disease.40 It is likely that many more disease states where some degree of fatigue or muscle impairment occurs may also demonstrate abnormalities in metabolism of muscle. Although MRS has found application in the study of muscle disease for purely scienti®c purposes, it has yet to ®nd any great application as a routine diagnostic test for muscle disease. This is primarily due to the specialized nature of the MRS examination and the great expense involved when many less expensive procedures are available, e.g. needle biopsy, followed by chemical and genetic analysis. Furthermore, these other procedures are capable of providing information on a greater range of metabolites or gene products than is currently possible by MRS. There is little doubt however that MRS will continue to provide a useful tool for the scienti®c elucidation of muscle biochemistry in the future, particularly when combined with the use of labeled compounds such as 13C-labeled glucose as tracers.

9 RELATED ARTICLES Animal Methods in MRS; Body Fat Metabolism: Observation by MR Imaging and Spectroscopy; MRI in Clinical Medicine; MRI of Musculoskeletal Neoplasms; NMR Spectroscopy of the Human Heart; Proton Decoupling in Whole Body Carbon-13 MRS; Quantitation in In Vivo MRS; Skeletal Muscle Evaluated by MRI; Surface Coil NMR: Detection with Inhomogeneous Radiofrequency Field Antennas; Tissue and Cell Extracts MRS; Tissue Behavior Measurements Using Phosphorus-31 NMR; Whole Body Studies: Impact of MRS.

10

REFERENCES

1. J. W. Pan, J. R. Hamm, D. L. Rothman, and R. G. Shulman, Proc. Natl. Acad. Sci. U.S.A., 1988, 85, 7836.

9

2. J. W. Pan, J. R. Hamm, H. P. Hetherington, D. L. Rothman, and R. G. Shulman, Magn. Reson. Med., 1991, 20, 57. 3. R. Taylor, T. B. Price, D. L. Rothman, R. G. Shulman, and G. I. Shulman, Magn. Reson. Med., 1993, 27, 13. 4. M. V. Narici, L. Landoni, and A. E. Minetti, Eur. J. Appl. Physiol., 1992, 65, 438. 5. M. Boska, NMR Biomed., 1991, 4, 173. 6. P. A. Martin, H. Gibson, S. Hughes, and R. H. T. Edwards, Proc. Xth Ann Mtg. Soc. Magn. Reson. Med., San Francisco, 1991, p. 547. 7. P. Vestergaard-Poulsen, C. Thomsen, T. Sinkjaer, M. Stubgaard, A. Rosenfalck, and O. Henriksen, Electroencephalogr. Clin. Neurophysiol., 1992, 85, 402. 8. R. H. T. Edwards, M. J. Dawson, D. R. Wilkie, R. E. Gordon, and D. Shaw, Lancet, 1982, 725. 9. D. I. Hoult, S. J. W. Busby, D. G. Gadian, G. K. Radda, R. E. Richards, and R. J. Seeley, Nature (London), 1974, 252, 285. 10. R. E. Gordon, P. E. Hanley, and D. Shaw, Prog. Nucl. Magn. Reson. Spectrosc., 1981, 15, 1. 11. J. F. Dunn, G. K. Kemp, and G. K. Radda, NMR Biomed., 1992, 5, 154. 12. T. B. Price, D. L. Rothman, M. J. Avison, P. Buonamico, and R. G. Shulman, J. Appl. Physiol., 1991, 70, 1836. 13. R. Taylor, T. B. Price, L. D. Katz, R. G. Shulman, and G. I. Shulman, Am. J. Physiol., 1993, 265, 224. 14. C. Thomsen, K. E. Jensen, and O. Henriksen, Magn. Reson. Imag., 1989, 7, 557. 15. P. A. Bottomley, T. B. Foster, and R. D. Darrow, JMRI, 1984, 59, 338. 16. C. Thomsen, K. E. Jensen, and O. Henriksen, Magn. Reson. Imag., 1989, 7, 231. 17. W. I. Jung, K. Straubinger, M. Bunse, S. Widmaier, F. Schick, K. Kuper, G. Dietze, and O. Lutz, Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, 30, 138. 18. K. Sahlin, Acta Physiol. Scand., Suppl., 1978, 455, 1. 19. A. Fabiato and F. Fabiato, J. Physiol. London, 1978, 276, 233. 20. G. K. Kemp, D. J. Taylor, P. Styles, and G. K. Radda, NMR Biomed., 1993, 6, 73. 21. K. McCully, Z. Argov, B. P. Boden, R. L. Brown, W. J. Bank, and B. Chance, Muscle Nerve, 1988, 11, 212. 22. M. J. Kushmerick, T. M. Moerland, and R. W. Wiseman, Proc. Natl. Acad. Sci. U.S.A., 1992, 89, 7521. 23. B. Bigland-Ritchie, D. A. Jones, G. P. Hosking, and R. H. T. Edwards, Clin. Sci., 1978, 54, 609. 24. `Human Muscle Fatigue: Physiological Mechanisms', eds. R. Porter and J. Whelan, Ciba Foundation Symposium 82, Pitman Medical, London, 1981. 25. `Neuromuscular Fatigue', eds. A. J. Sargeant and D. Kernell, R. Neth. Acad. Arts & Sci., Amsterdam, 1993. 26. R. H. T. Edwards, D. K. Hill, D. A. Jones, and P. A. Merton, J. Physiol. London, 1977, 272, 769. 27. D. J. Newham, K. R. Mills, B. M. Quigley, and R. H. T. Edwards, Clin. Sci., 1983, 64, 55. 28. R. G. Miller, D. Giannini, H. S. Milner-Brown, R. B. Layzer, A. P. Koretsky, D. Hooper, and M. W. Weiner, Muscle Nerve, 1987, 10, 810. 29. G. K. Kemp, C. H. Thompson, P. R. Barnes, and G. K. Radda, Proc. Ist Ann Mtg. Int. Soc. Magn. Reson. Med., Dallas, 1994, 31, 248. 30. G. K. Kemp, D. J. Taylor, and G. K. Radda, NMR Biomed., 1993, 6, 66. 31. D. J. Taylor, P. J. Bore, P. Styles, D. G. Gadian, and G. K. Radda, Mol. Biol. Med., 1983, 1, 77. 32. G. K. Kemp, D. J. Taylor, C. H. Thompson, P. Styles, L. J. Hands, B. Rajagopalan, and G. K. Radda, NMR Biomed., 1993, 6, 302.

10 PERIPHERAL MUSCLE METABOLISM STUDIED BY MRS 33. D. L. Arnold, P. M. Matthews, and G. K. Radda, Proc. IIIrd Ann Mtg. Soc. Magn. Reson. Med., New York, 1984, 1, 307. 34. D. Bendahan, S. Confort-Gouny, G. Kozak-Reiss, and P. J. Cozzone, FEBS Lett., 1990, 272, 155. 35. I. H. Nadshus. Biochem. J., 1988, 250, 1. 36. S. Iotti, R. Funicello, P. Zaniol, and B. Barbiroli, Biochem. Biophys. Res. Commun., 1991, 176, 1204. 37. D. J. Taylor, P. Styles P. M. Matthews, D. L. Arnold, D. G. Gadian, P. J. Bore, and G. K. Radda, Proc. Vth Ann Mtg. Soc. Magn. Reson. Med., Montreal, 1986, 3, 44. 38. R. D. Grif®ths, E. B. Cady, R. H. T. Edwards, and D. R. Wilkie, Muscle Nerve, 1985, 8, 760. 39. B. Barbiroli, Magn. Reson. Spectrosc. Biol. Med., 1992, 20, 369. 40. I. Adatia, G. J. Kemp, D. J. Taylor, G. K. Radda, B. Rajagopalan, and S. G. Haworth, Clin. Sci., 1993, 85, 105. 41. R. J. Newman, P. J. Bore, L. Chan, D. L. Gadian, P. Styles, D. Taylor, and G. K. Radda, Br. Med. J., 1982, 284, 1072. 42. D. Younkin, P. Berman, J. Sladky, C. Chee, W. Bank, and B. Chance, Neurology, 1987, 37, 165. 43. B. Barbiroli, R. Funicello, A. Ferlini, P. Montagna, and P. Zaniol, Muscle Nerve, 1992, 15, 344. 44. D. L. Arnold, D. J. Taylor, and G. K. Radda, Ann. Neurol., 1985, 18, 189. 45. Z. Argov, W. J. Bank, J. Maris, P. Peterson, and B. Chance, Neurology, 1987, 37, 257. 46. D. J. Hayes, D. Hilton-Jones, D. L. Arnold, G. Galloway, P. Styles, J. Duncan, and G. K. Radda, J. Neurol. Sci., 1985, 71, 105. 47. B. Barbiroli, P. Montagna, P. Cortelli, P. Martinelli, T. Sacquegna, P. Zaniol, and E. Lugaresi, Cephalalgia, 1990, 10, 263. 48. D. Duboc, P. Jehenson, S. Tran Dinh, C. Marsac, A. Syrota, and M. Fardeu, Neurology, 1987, 37, 663. 49. Z. Argov, W. J. Bank, J. Maris, J. S. Leigh, Jr., and B. Chance, Ann. Neurol., 1987, 22, 46. 50. Z. Argov, W. J. Bank, B. Boden, Y. I. Ro, and B. Chance, Arch. Neurol., 1987, 44, 614. 51. G. K. Radda, Biochem. Soc. Trans., 1986, 14, 517. 52. S. L. Norris, J. R. Gober, L. J. Haywood, J. Halls, W. Boswell, P. Colletti, and M. Terk, Magn. Reson. Imag., 1993, 11, 119. 53. J. Olgin, H. Rosenberg, G. Allen, R. Seestedt, and B. Chance, Anesth. Analg. (Cleveland), 1991, 72, 36. 54. M. A. Zatina, H. D. Berkowitz, G. M. Gross, J. M. Maris, and B. Chance, J. Vasc. Surg., 1986, 3, 411.

55. S. F. Lewis, R. G. Haller, J. D. Cook, and R. L. Nunnally, J. Appl. Physiol., 1985, 59, 1991. 56. D. J. Taylor, S. W. Coppack, T. A. D. Cadoux-Hudson, G. J. Kemp, G. K. Radda, K. N. Frayn, and L. L. Ng, Clin. Sci., 1991, 81, 123. 57. B. McArdle, Clin. Sci., 1951, 10, 13. 58. R. H. T. Edwards, Muscle Nerve., 1984, 7, 599. 59. D. G. Gadian, G. K. Radda, B. D. Ross, J. Hockaday, P. Bore, D. Taylor, and P. Styles, Lancet ii, 1981, 774. 60. G. K. Radda, P. J. Bore, D. G. Gadian, B. D. Ross, P. Styles, D. J. Taylor, and J. Morgan-Hughes, Nature, (London), 1982, 295, 608. 61. R. H. T. Edwards, R. D. Grif®ths, and E. B. Cady, Clin. Physiol., 1985, 5, 93. 62. B. Chance, S. Eleff, J. S. Leigh, Jr., D. Sokolow, and A. Sapega, Proc. Natl. Acad. Sci. U.S.A., 1981, 78, 6714.

Biographical Sketches Peter A. Martin. b 1954. B.Sc. (Loughborough), Ph.D. (Dunelm). Applications scientist at Oxford Research Systems Limited, 1980±85. Operations manager of Liverpool University's Magnetic Resonance Research Centre 1986±present. Research interests include the applications of neural networks to spectral analysis, proton spectroscopy of the brain, and the quantitation of muscle physiology via MRI and MRS. Henry Gibson. b 1961. B.Sc. (Liverpool), M.Sc. (King's College, London), Ph.D. (Liverpool). Honorary nonclinical lecturer in medicine, Research interests; quanti®cation of muscle physiology via MRI and MRS and ultrasonography of muscle. Richard H. T. Edwards. b 1939. B.Sc., M.B., B.S., Ph.D. (London), F.R.C.P., Codirector of the Jerry Lewis Muscle Research Centre, Royal Postgraduate Medical School. Honorary Consultant Respiratory Physician, Hammersmith Hospital; Professor of Human Metabolism, Hospital Medical School and Head of Department of Medicine, University College, London; Professor and Head of Department of Medicine, University of Liverpool, 1984±present. Director, Magnetic Resonance Research Centre, 1986±present.

SKELETAL MUSCLE EVALUATED BY MRI

Skeletal Muscle Evaluated by MRI James L. Fleckenstein University of Texas Southwestern Medical Center, Dallas, TX, USA

1 INTRODUCTION Clinical evaluation of skeletal muscle has long been hampered by dif®culty in assessing the morphology and functional integrity of skeletal muscles. Proton magnetic resonance imaging (MRI) represents a major advance in the diagnosis and management of patients with muscle disease by probing beyond the relatively bland surface of skin to identify focal muscle structural lesions, to determine their extent, and to characterize their composition, and guide invasive procedures and monitor therapies. The purpose of this chapter is to review advances made by MRI in understanding the quality of muscle in health and disease.

2 MR TECHNIQUES MRI is used to examine patients' muscular anatomy noninvasively and determine which muscles are abnormal in size and shape. MRI further characterizes the quality of muscle by discriminating between the mesenchymal alterations of muscle fat and edema, depending on the pulse sequences used.1±4 Fat is detected on T1-weighted images (short TR, short TE) by high signal intensity due to its short T1 time constant. Because fat has a long T2 time constant, it also manifests high signal intensity on T2-weighted (long TR, long TE) sequences. Increased tissue water leads to increased spin density and elevated T1 and T2 relaxation times; hence, muscle edema is detectable using MRI.1,2 The long T1 can be manifested by decreased signal intensity on heavily T1-weighted spin echo images. However, in cases in which the sequence is also sensitive to changes in proton density and T2, the change in T1 is frequently not suf®cient to result in a net change in signal intensity.2 Using conventional T1-weighted and T2-weighted spin echo sequences, dif®culty is sometimes encountered in differentiating intramuscular fat from edema, especially when they coexist. This is because both edema and fat are hyperintense to muscle on long TR/long TE sequences. This accounts in part for why fat-suppression techniques improve detection of muscle edema. Multiple fat suppression sequences have been developed that improve detection of muscle edema.1±3 One of these employs an inversion pulse that nulls signal from tissue having a T1 time equal to that of fat [short inversion time inversion recovery (STIR)]. STIR has the additional advantage of producing heightened lesion conspicuity due to additive effects to signal intensity caused by edema-associated increases in lesion spin density and T1 and T2 times.3 Other sequences employ fre-

1

quency selective pulses to null the fat signal. These techniques generally require a longer TE to achieve the same degree of lesion conspicuity as STIR sequences at the same TR. Important from a ®nancial perspective, a variety of fat-suppression sequences can be incorporated into fast scan techniques, so that high sensitivity to muscle edema can be realized with short scan times. This makes muscle imaging an economically viable, as well as informative, application of MRI.

3

NORMAL MUSCLE ANATOMY AND PHYSIOLOGY

Like computerized tomography and ultrasound, MRI aids physiological studies of muscle size by providing quantitative morphological data regarding the mass of muscle performing work. While cross-sectional area (CSA) has most frequently been used for this purpose, the safety and high spatial resolution of MRI allow for accurate determinations of muscle volume, eliminating errors inherent in modeling volume estimates from single CSA measurements. Incorporation of the muscle ®ber pennation angle into estimates of size has further de®ned morphological determinants of muscle strength.5 The ability of MRI to accurately de®ne muscle fascicle architecture also enhances MRI assessment of muscle morphology.6 A variety of these techniques has been used to monitor changes in muscle volume during training7 and detraining.8 Compositional alterations of muscle that can potentially be quantitated with MRI include ®ber type, water content and compartmentation, and fat content.9±25 MRI of muscle ®ber type has been studied in both animals and humans, although with con¯icting results. In rat soleus muscle, which is nearly all type I (slow twitch, oxidative) ®bers, T1 and T2 relaxation times are longer than in gastrocnemius muscle, which has a greater proportion of type II (fast twitch, glycolytic) ®bers. This difference correlates with a greater extracellular water content in the soleus muscle.9 Muscles having such markedly disparate ®ber type proportions can easily be differentiated from other muscles on MRI (Figure 1).2 Although human muscle has much less variability in ®ber type proportion than does animal muscle, a study in humans indicated that the percentage of type II ®bers positively correlated with T2 times.10 This result is the opposite of what would be predicted from the animal data and so additional studies are needed to determine the accuracy and validity of this MRI application. MRI measurement of muscle water content can be applied not only to studies of ®ber typing, but also to changes that occur as a result of muscular contraction,11±22 diuresis,11 and denervation atrophy.23,24 As an example, the effects of exercise on the MRI appearance of exercise muscle will be examined in detail. During low-intensity muscular contractions, increases of muscle water primarily occur in the extracellular space; during maximal intensity exercise small increases also occur in the intracellular water space.26 These changes in muscle water content/compartmentation are associated with transient increases in muscle spin density and T2 times, while T1 times are relatively less affected.12 Interestingly, postexercise hyperemia is neither required12,13 nor suf®cient14 for the effect to be observed (Figure 1). Animal studies suggest that increases in muscle extracellular water content underlie an increase in the

2 SKELETAL MUSCLE EVALUATED BY MRI

Figure 1 Effects of ®ber type heterogeneity and ischemic muscle contraction visible by MRI. Coronal 2000/60 images of rabbit thighs before (a) and after (b) ischemic muscle stimulation. Note on the prestimulation image that a single muscle (semitendinosus, arrows) has a higher signal intensity than nearby muscles. This is a normal ®nding in this breed of rabbit and re¯ects a high proportion of oxidative ®bers. After stimulation of the left sciatic nerve, signal intensity increases markedly within the left lateral thigh muscles (arrowhead). Because aortic ligation and death preceded the stimulation, blood ¯ow could not have contributed to signal intensity changes in the stimulated muscles, supporting data in humans that blood ¯ow is not critical in mediating transient MRI changes of signal intensity due to exercise. (Reproduced by permission of the Radiological Society of N. America from J. L. Fleckenstein, R. G. Haller, L. A. Bertocci, R. W. Parkey, and R. M. Peshock, Radiology, 1992, 183, 25)

number of proton spins having long T2 decay times.9,11 This has led some investigators to propose that an increase in this water space is the primary determinant of the changes visible on MRI.12,13,15 A role for changes in intracellular water content has also been proposed16 and debated.12 These MRI-visible alterations in muscle water are speculated to result from increased muscle osmolality due to accumulation of lactate and other ions which cause osmotic shifts of water between intracellular and extracellular water compartments. This conclusion is supported by absence of the normal postexercise muscle T2 variations in patients in whom muscle lactate accumulation is absent due to defective muscle phosphorylase (glycogenosis V, McArdle's disease)14 and by a high correlation between the magnitude of T2 change and fall in pH in healthy volunteers.17 Magnetization transfer contrast techniques, exploiting selective saturation of a pool of `bound' water protons, were applied in two studies to improve the understanding of water compartmentation during MRI of exercise in humans.15,18 However, the studies provided con¯icting results and conclusions. Another area of apparent controversy regards changes in the T2 time as a function of work performed. While some studies suggested that T2 increases in direct proportion to work intensity, the linearity of this relationship has been questioned.19 Taking these data together it is likely that T2 varies linearly with work in some regimens but not in others. Recognition that a limit exists in the magnitude of T2 response (~30%)19 suggests the existence of a limit to the amount of water that muscle can imbibe from the vasculature during exertion and/or to the magnitude of changes of muscle water compartmentation/binding that can occur during exercise.

Figure 2 Finger-speci®c components of the ¯exor digitorum super®cialis. Four discrete parts of this muscle can be demonstrated by selective exercise of individual ®ngers. From left to right are the index, long, ring, and small components of the ¯exor digitorum super®cialis at the mid-forearm. The bones are the radius and ulna. (Reproduced by permission of Raven Press from J. L. Fleckenstein et al.2)

Although the precise mechanism(s) involved in exerciseinduced changes in muscle relaxation times during exercise remains unknown, the changes in image contrast between strongly recruited muscle and less active muscle have been exploited in a number of interesting practical applications: diagnosis of disorders of muscle energy metabolism;14 chronic exertional compartment syndromes;20 identi®cation of muscle recruitment patterns relevant to MR spectroscopy studies of exercise (Figure 2);21,22 assessment of manufacturers' claims of muscle recruitment patterns using commercially available exercise equipment (Figure 3).

4

MRI OF MUSCLE PATHOLOGY

MRI has been applied to the evaluation of a broad range of neuromuscular and orthopedic muscle disorders. One of the critical issues that faces a clinician evaluating a patient with neuromuscular disease is determining whether the disease is primarily neurogenic or muscular in origin. This is of particular importance in pediatrics because patients with spinal muscular atrophies (SMA) may present with similar clinical ®ndings to patients with muscular dystrophies, particularly of the Duchenne type (DMD). Early attempts to differentiate SMA from DMD employed ultrasound of the extremities and reported that in SMA the overall volume of muscle was decreased, compared with that of subcutaneous fat.27,28 As an additional diagnostic clue, MRI has disclosed selective sparing of speci®c muscles in DMD, particularly of the gracilis, semimembranosus, and sartorius.25,29 A more recent study sought to distinguish SMA of the Kugelberg±Welander type (KW) from DMD.30 It was reported that in KW, muscle deterioration tended to be more diffuse than in DMD and the previous ®nding of generalized muscle atrophy in SMA was corroborated. A tendency toward relatively selective involvement of type II muscles in DMD was also observed. While more work needs

SKELETAL MUSCLE EVALUATED BY MRI

Figure 3 MRI of `¯y' exercise. Using a commercially available exercise device this healthy subject performed arm abduction against resistance within 1 min of scan acquisition. Note that the pectoralis major (arrowhead), subscapularis (curved arrow), and coracobrachialis (small arrow) are strongly stressed while the pectoralis minor is relatively unstressed (arrow). Therefore, the device does not stress the pectoralis group homogeneously, despite claims of the manufacturer to the contrary

to be performed to assess the capability of MRI to distinguish between neurogenic and primary muscle diseases, results to date indicate that MRI may be helpful in this distinction. Muscle dysfunction that results from peripheral neuropathy has also been evaluated with MRI.24 MRI was found able to detect edema-like changes in muscles affected by traumatic or compressive peripheral nerve lesions (Figure 4). These changes were anticipated based on results from an animal study in which proton relaxation times were shown to be prolonged in denervated muscle due to ®ber atrophy and a resultant increase in the extracellular water space.23 Like electromyography, MRI was limited in its ability to detect muscle abnormalities in the ®rst few weeks of denervation. On the other hand, denervation

Figure 4 MRI of subacute leg muscle denervation: edema-like change. Lateral collateral ligament injury (arrow, 500/40) (a) and subsequent scar formation resulted in compression of common peroneal nerve (not shown). Note that signal intensity of denervated anterior leg muscles mimics edema, being normal on coronal T1-weighted image (arrowhead) (a), and increased on axial T2-weighted image (arrowhead, 2000/60) (b) and STIR (arrowhead, 1500/30/100) (c)

3

Figure 5 Muscle atrophy and hypertrophy in muscular dystrophy. Symmetric, proximal diminution in muscle volume is evident on a 500/ 30 sequence. Note sparing of the gracilis (g) and sartorius (s). (Reproduced by permission of Raven Press from J. L. Fleckenstein et al.2)

was readily visible on edema-sensitive sequences when denervation had occurred prior to 1 month before the MRI scan. While edema-like change dominated the appearance of muscle in the ®rst year of denervation, fatty change of the muscle was observed in more long-standing denervation. Muscle hypertrophy is a relatively rare result of denervation, and while the ®nding may be prominent in patients with a remote history of poliomyelitis, it may also be observed relatively early after insult to peripheral nerves.2 Unlike disease of nerve and muscle, neuromuscular junction dysfunction, such as seen in myasthenia gravis, has yet to be reported to have an abnormal appearance on MRI, or any other imaging modality. On the other hand, primary myopathies, including dystrophies, idiopathic in¯ammatory myopathies, metabolic myopathies, and congenital myopathies, display various muscle abnormalities on MRI.25,29,31,32 These ®ndings, including edema-like change and fatty in®ltration of muscles, tend to be nonspeci®c in terms of distribution. For example, selective sparing of the sartorius and gracilis is a feature not only of DMD (Figure 5), but also of polymyositis (Figure 6), congenital myopathies,29 and metabolic myopathies (Figure 7). Selective involvement of the same muscles is a feature of some mitochondrial myopathies2 but may be seen in centronuclear myopathy. The character of the imaging abnormality is also nonspeci®c, in that edema-like change on MRI may be seen in denervation, necrosis, and in¯ammation.2 While not pathognomonic for speci®c disease processes, the MRI abnormalities are useful in directing invasive procedures, such as biopsy.31,32 The objective nature of imaging abnormalities can also be used to monitor response to therapy, which is otherwise limited by the subjective aspects of the patients' sense of well being and by limitations in assessing muscular strength. In the ®eld of orthopedics, muscle injuries are extremely common and include muscle strains and contusions, delayed onset muscle soreness, and chronic overuse syndromes.33±37 Associated injuries and sequelae of injuries are important determinants of the prognosis of muscle injuries. Because MRI is

4 SKELETAL MUSCLE EVALUATED BY MRI

Figure 6 Chronic polymyositis. The extensive high signal intensity throughout the thigh muscles on 500/30 (a), and 2000/60 (b), images implies fatty change. STIR suppresses signal intensity from fat, while high signal intensity identi®es regions of coexistent muscle edema (arrow) (c). Note that muscle edema is easy to identify only with fat suppression. Such edematous areas are of particular interest during biopsy when in¯ammatory myopathy is suspected. Note sparing of multiple muscles, including sartorius (arrow) (a) and gracilis (arrowhead) (a)

sensitive to both muscle trauma and associated abnormalities, it has been aggressively applied to these orthopedic issues. Muscle strain is de®ned as an indirect injury to muscle caused by excessive stretch. Although various clinical schemes of grading severity of muscle strains have been advanced, it is acknowledged that clinical evaluation of muscle strains is dif®cult, even more so than injuries of tendons or bones. MRI aids in assessing integrity of muscles, myotendinous junctions, fascia, and the tendoosseous unit. MRI identi®es edema within and/or around injured muscles, depending on the stage of healing.33,34 The myotendinous junction is frequently the point of rupture and the extent of associated fascial or tendinous tear can be addressed by MRI (Figures 8±10). When a fascial or tendinous tear is small (Figure 8), the injury can safely be managed conservatively. However, when a rupture is complete, or nearly complete (Figures 9 and 10), early surgery may be indicated; a lapse of even a short time may cause an inferior functional result, due to muscle ®brosis and retraction. Fluid collections also frequently accompany strains. These can themselves be a cause of swelling and weakness in the absence of fascial tear. The use of MRI to distinguish focal hematomas from swollen, edematous muscles can guide clinical management; the former may bene®t from drainage while the latter are often treated with wrapping procedures for compression and support of the injured area.4 Since recurrent muscle strains can devastate elite athletes, it is noteworthy that MRI in muscle injuries can provide information regarding the prognosis of muscle strain. Two studies have reported on MRI ®ndings that are associated with poor outcome. These studies indicated that the occurrence of focal ¯uid collections (Figure 10), relatively large volume of abnormality (Figure 11) and ®brosis (Figure 12), correlate with either recurrence of muscle strain, delayed convalescent inter-

Figure 7 Glycogenosis: T1W (500/30) (a) and T2W (2000/60) (b) images of the thighs in a patient with phosphofructokinase de®ciency demonstrate marked replacement of most thigh muscles by high signal intensity fat. STIR (c) suppresses signal intensity from fat, while areas of high signal intensity identify regions of coexistent muscle edema in the vastus lateralis (arrows). Such edematous areas should be avoided during biopsy when glycogenoses are considered because muscle necrosis may produce a small amount of fetal myophosphorylase in patients who usually lack the adult form of that enzyme (McArdle's disease, Glycogenosis V)

val, or both.35,36 It is interesting that MRI alterations of strained muscles typically persist for longer than any other clinical evidence of injury.33,34 One could speculate that reinstitution of exercise in this setting might be harmful to the healing muscle, but this has not yet been studied. A more immediately practical implication of the delayed disappearance of edema from muscle after injurious exercise is that one may detect evidence of previous muscle injury on MRI after the patient forgets the inciting event. This suggests a potential

Figure 8 Important MRI ®ndings in muscle injury: partial tendon tear. Associated injuries to fasciae and tendons are important to quantitate since small ®brous tears may require no operative intervention. Note partial biceps femoris tendon tear (arrow, 600/20) (a) and STIR (b), and the superior delineation of perifascial ¯uid using STIR. The lesion was conservatively managed. (Reproduced by permission of Raven Press from J. L. Fleckenstein et al.2)

SKELETAL MUSCLE EVALUATED BY MRI

Figure 9 Important MRI ®ndings in muscle injury: complete avulsion. When muscle tearing is complete, myotendinous avulsion occurs. Complete myotendinous avulsion of a juvenile football player's gluteus medius is easily appreciated on the coronal spin densityweighted image (arrow, 2000/30). The patient was treated by immobilization in a body cast for 12 weeks. (Reproduced by permission of Raven Press from J. L. Fleckenstein et al.2)

source of serendipitously observed MRI abnormalities. On the other hand, this radiographic ®nding may be the only clue to the true origin of the patient's musculoskeletal complaints.4 Other sequelae of muscle injury detectable by MRI include muscle calci®cation and ossi®cation. Calci®c myonecrosis is a delayed complication of muscle injury in which progressive calci®cation of an injured muscle is associated with slow development of a mass lesion of the extremities. This condition is a rare complication but its radiology is well described, allowing for a con®dent preoperative diagnosis.37 Muscle ossi®cation (myositis ossi®cans) is also detectable by MRI but its appearance is nonspeci®c early in its development, mimicking

5

Figure 11 Important MRI ®ndings in muscle injury: large volume edema. Factors that suggest a longer convalescent period after injury include a large volume of muscle edema, such as in an axial STIR image of an acutely strained rectus femoris muscle (arrow). (Courtesy of S. C. Schultz, Fort Worth, TX)

neoplastic disease. As it matures, ossi®ed muscle may be identi®ed by peripheral low signal intensity, corresponding to the outer zone of ossi®cation (see MRI of Musculoskeletal Neoplasms).

5

CONCLUSION

MRI circumvents traditional obstacles in the clinical evaluation of muscle physiology and pathology. MRI detection of muscle size, ®ber composition and orientation, and water shifts are promising research arenas in the ®eld of exercise physiology. The high sensitivity of MRI in detecting muscle edema

Figure 10 Important MRI ®ndings in muscle injury: ¯uid collection. Rupture and retraction of the posterior head of the rectus femoris in an elite kicker with recurrent muscle strain reveals a `ganglion-like' ¯uid collection within a portion of the rectus femoris on axial STIR image (arrow) (a), corresponding to ¯uid collecting between the retracted posterior head of the rectus femoris and its origin (arrows, sagittal STIR) (b). This injury was treated by resection of the avulsed muscle head

6 SKELETAL MUSCLE EVALUATED BY MRI

Figure 12 Important MRI ®ndings in muscle injury: scar. MRIvisible scar formation in myotendinous tears is associated with recurrence of strains. Scar is characterized by excessive deposition of signal poor tissue and muscle atrophy (arrow). (Courtesy of S. C. Schultz, Fort Worth, TX)

and fat allows improved delineation of the distribution and composition of neuromuscular and orthopedic disorders. This sensitivity can be used to substantiate muscle as the source of musculoskeletal pain, weakness, or stiffness in a broad range of patients and on the basis of positive, objective ®ndings rather than by exclusion.

6 RELATED ARTICLES Inversion±Recovery Pulse Sequence in MRI; Magnetization Transfer Contrast: Clinical Applications; MRI of Musculoskeletal Neoplasms; Peripheral Joint Magnetic Resonance Imaging; Peripheral Muscle Metabolism Studied by MRS.

7 REFERENCES 1. J. L. Fleckenstein, B. T. Archer, B. A. Barker, J. T. Vaughn, R. W. Parkey, and R. M. Peshock, Radiology, 1991, 179, 499. 2. J. L. Fleckenstein, P. T. Weatherall, L. A. Bertocci, M. Ezaki, R. G. Haller, R. Greenlee, W. W. Bryan, and R. M. Peshock, Magn. Reson. Q., 1991, 7, 79. 3. A. J. Dwyer, J. A. Frank, V. J. Sank, J. W. Reinig, A. M. Hickey, and J. L. Doppman, Radiology, 1988, 168, 827. 4. J. L. Fleckenstein and F. G. Shellock, Top. Magn. Reson. Imaging, 1991, 3, 50. 5. T. Fukunaga, R. R. Roy, F. G. Shellock, J. A. Hodgson, M. K. Day, P. L. Lee, H. Kwong-Fu, and V. R. Edgerton, J. Orthop. Res., 1992, 10, 928. 6. S. H. Scott, C. M. Engstrom, and G. E. Loeb, J. Anat., 1993, 182, 249. 7. W. J. Roman, J. L. Fleckenstein, J. Stray-Gundersen, S. E. Alway, R. Peshock, and W. J. Gonyea, J. Appl. Physiol., 1993, 74, 750. 8. T. Fukunaga, K. Day, J. H. Mink, and V. R. Edgerton, Med. Sci. Sports Med., 1991, 23, 5110. 9. J. F. Polak, F. A. Jolesz, and D. F. Adams, Invest. Radiol., 1988, 23, 107. 10. S. Kuno, S. Katsuta, T. Inouye, I. Anno, K. Matsumoto, and M. Akisada, Radiology, 1988, 169, 567.

11. E. Le Rumeur, J. de Certaines, P. Toulouse, and P. Rochcongar, Mag. Reson. Imaging, 1987, 5, 267. 12. B. Archer, J. L. Fleckenstein, L. A. Bertocci, R. G. Haller, B. Barker, R. W. Parkey, and R. M. Peshock, J. Magn. Reson. Imaging, 1992, 2, 407. 13. J. L. Fleckenstein, R. C. Canby, R. W. Parkey, and R. M. Peshock, Am. J. Roentgenol., 1988, 151, 231. 14. J. L. Fleckenstein, R. G. Haller, S. F. Lewis, B. T. Archer, B. R. Banker, J. Payne, R. W. Parkey, and R. M. Peshock, J. Appl. Physiol., 1991, 71, 961. 15. X. P. Zhu, S. Zhao, and I. Isherwood, Br. J. Radiol., 1992, 65, 39. 16. F. G. Shellock, T. Fukunaga, J. H. Mink, and V. R. Edgerton, Am. J. Roentgenol., 1991, 156, 765. 17. E. R. Weidman, H. C. Charles, R. Negro-Vilar, M. J. Sullivan, and J. R. MacFall, Invest. Radiol., 1991, 26, 309. 18. K. T. Mattila, M. E. Komu, S. K. Koskinen, and P. T. Niemi, Acta Radiol., 1993, 34, 559. 19. J. L. Fleckenstein, D. Watumull, D. D. McIntire, L. A. Bertocci, D. P. Chason, and R. M. Peshock, J. Appl. Physiol., 1993, 74, 2855. 20. A. Amendola, C. H. Rorabeck, D. Vellett, W. Vezina, B. Rutt, and L. Nott, Am. J. Sports Med., 1990, 18, 29. 21. J. L. Fleckenstein, L. A. Bertocci, R. L. Nunnally, R. W. Parkey, and R. M. Peshock, Am. J. Roentgenol., 1989, 153, 693. 22. J. L. Fleckenstein, D. Watumull, L. A. Bertocci, R. W. Parkey, and R. M. Peshock, J. Appl. Physiol., 1992, 72, 1974. 23. J. F. Polak, F. A. Jolesz, and D. F. Adams, Invest. Radiol., 1988, 23, 365. 24. J. L. Fleckenstein, D. Watumull, K. E. Conner, M. Ezaki, R. G. Greenlee, Jr., W. W. Bryan, D. P. Chason, R. W. Parkey, R. M. Peshock, and P. D. Purdy, Radiology, 1993, 187, 213. 25. W. A. Murphy, W. G. Totty, and J. E. Carroll, Am. J. Roentgenol., 1986, 146, 565. 26. G. Sjogaard, R. P. Adams, and B. Saltin, Am. J. Physiol., 1985, 248, R190. 27. J. Z. Heckmatt, N. Pier, and V. Dubowitz, J. Clin. Ultrasound, 1988, 16, 171. 28. A. Lamminen, J. Jaaskelainen, J. Rapola, and I. Suramo, J. Ultrasound Med., 1988, 7, 505. 29. A. E. Lamminen, Br. J. Radiol., 1990, 63, 946. 30. D. Suput, A. Zupan, A. Sepe, and F. Demsar, Acta Neurol. Scand., 1993, 87, 118. 31. A. Pitt, J. L. Fleckenstein, R. G. Greenlee, D. K. Burns, W. W. Bryan, and R. G. Haller, Mag. Res. Imaging, 1993, 11, 1093. 32. J. H. Park, S. J. Gibbs, R. R. Price, C. L. Partain, and A. E. James, Jr., Radiology, 1991, 179, 343. 33. J. L. Fleckenstein, P. T. Weatherall, R. W. Parkey, J. A. Payne, and R. M. Peshock, Radiology, 1989, 172, 793. 34. F. G. Shellock, T. Fukunaga, J. H. Mink, and V. R. Edgerton, Am. J. Roentgenol., 1991, 156, 765. 35. S. J. Pomeranz and R. S. Heidt, Jr., Radiology, 1993, 189, 897. 36. A. Greco, M. T. McNamara, R. M. B. Escher, G. Tri®lio, and J. Parienti, J. Comput. Assist. Tomogr., 1991, 15, 994. 37. D. L. Janzen, D. G. Connell, and B. J. Vaisler, Am. J. Roentgenol., 1993, 160, 1072.

Biographical Sketch James L. Fleckenstein. b 1957. B.S. 1979, M.D. 1984, University of Washington. Medical internship, 1984±1985, University of Texas. Radiology residency 1985±1989, University of Texas. Assistant instructor, Radiology, University of Texas, 1989±1990. Director of Neuro MRI, Division of Neuroradiology, and Associate Professor of Radiology at University of Texas Southwestern Medical Center, 1991± present. Approx. 200 publications. Research specialties: MRI of muscle in health and disease.

HIGH-FIELD WHOLE BODY SYSTEMS

High-Field Whole Body Systems Hoby P. Hetherington Brookhaven National Laboratory, Upton, NY, USA

and Gerald M. Pohost University of Alabama at Birmingham, AL, USA

1 INTRODUCTION Throughout the history of nuclear magnetic resonance research there has been a push toward the use of higher ®eld magnets for their advantages in signal-to-noise ratio (SNR), spectral resolution, and simpli®cation of the complex multiplet structures present in J-coupled spin systems. Although the advantages of high ®eld systems have been demonstrated for use in analytical work and structural determinations, their application to in vivo spectroscopy in humans has been limited by technical demands. Additionally, because of ®eld-dependent changes in the relaxation properties of the molecules being studied, signi®cant methodological limitations have occurred when lower ®eld (1.5 T) approaches have been applied at high ®eld. Recently, the development of new hardware and alternative methodological approaches has enabled these obstacles to be overcome and many of the advantages of in vivo high-®eld spectroscopy and imaging to be realized. In this section we will provide a brief summary of the hardware and methodological advances that have led to the realization of the bene®ts of high ®eld in vivo spectroscopy in humans, and present several examples of their application. The ®rst section will deal with some of the limitations of conventional rf coil design and an alternative approach for the design of large highly homogeneous rf coils. The second section will describe some of the advantages and the limitations of spectroscopy at high ®eld. The ®nal section will summarize the advantages of both anatomical and functional imaging and how performing these measurements at high ®eld when coupled with the correct methodology yields improvements in both contrast, resolution and sensitivity to physiological alterations.

2 HARDWARE LIMITATIONS Although homogeneous volume rf coils and ef®cient surface coils for low-®eld human studies or high-®eld small animal research are widely available, the performance of these coils becomes very limited as their dimensions approach the wavelength of the rf being applied. For volume coils of the `birdcage design', self-resonance for head coils may occur substantially before reaching the desired ®eld strength. Because of the propagation velocity in the conductors used in these coils and the length of the current-carrying elements (as a fraction of

1

the wavelength), phase changes along the conductors can occur that degrade the homogeneity and sensitivity of the rf coil. Although using distributed lumped element approaches can reduce some of these effects, an alternative approach using transmission line segments and a resonant cavity has seen substantial success. The utility of transmission lines for high-frequency NMR probes has been recognized since the mid 1970s by Schneider and Dullenkopf.1 As the ®rst 4 T systems became available Barfuss2 used capacitively shortened half wave slotted tubes to construct a 170 MHz body coil. Roschmann3 improved on the high-®eld design by replacing the unshielded lumped element slotted tubes with half wave coaxial transmission line segments. This design minimized the electric ®eld losses, thereby improving the sensitivity and ef®ciency of the coil. However, these initial coils had no provision for tuning or matching on each subject, thereby prohibiting maximum performance to be achieved under all conditions. Vaughan4 has overcome these limitations by providing a capacitively coupled matching network and a tuning mechanism to adjust the capacitance of all of the transmission line segments, permitting in situ tuning and matching. Initial theoretical predictions of the homogeneity of high®eld rf coils suggested that enhanced skin depth effects would cause an attenuation of the rf ®eld resulting in severe inhomogeneities. However, Vaughan4 has shown that the effects of dielectric resonance tend to offset these skin depth effects to provide a highly homogeneous rf ®eld within the human head at 4 T. An additional concern is the perturbation of the rf ®eld by the geometry of the head and the differing electrical properties of the layers surrounding the brain. In most cases these effects can be minimized for a typical head by calculating the B1 ®eld distribution and/or mapping it, and then adjusting the capacitance and inductance of the coil elements to compensate for the inhomogeneity induced by the cranial geometry and electrical properties. Excellent homogeneity is obtained throughout the slice. The high sensitivity of the coil supports 512  512 resolution for this 3 mm slice, yielding an inplane resolution of 0.47 mm (Figure 1).5 Recently, using the same design with multiple drive points, a highly homogeneous 8 T head coil has been constructed.6 With the improvement in SNR afforded by higher ®eld strengths, there has been renewed interest in performing spectroscopy studies of lower gyromagnetic ratio nuclei, such as 31P and 13C. For purposes of spatial co-registration, 1H decoupling and or heteronuclear acquisition methods, special multiply tuned coil systems have been developed. Vaughan has extended the single-tuned cavity resonator design to a doubletuned coil. Speci®cally, the 1H and X nuclei are driven 90 out of phase to each other, using alternate elements around the coil.7 This design achieves high homogeneity, high SNR, and (because of the high degree of isolation between the 1H and X nucleus elements) heteronuclear decoupling is readily achieved without SNR loss. Alternatively, for studies investigating peripheral regions, Adrianny has described a combined surface coil with orthogonal quadrature decoupling coils.8 Isolation between the surface coil and quadrature decoupling is achieved through manipulation of the geometry between the two coils. When applied to the occipital lobe, the 1H quadrature decoupling coils provide excellent visualization of the posterior

2 HIGH-FIELD WHOLE BODY SYSTEMS

Figure 1 Axial image (3 mm) of the human brain acquired at 4.1 T using a cavity resonator. The ®eld-of-view is 240  240 mm using a resolution of 512  512 pixels

portion of the head, with relatively homogeneous decoupling throughout the sensitive volume of the surface coil.

3 SPECTROSCOPY Perhaps the most obvious advantage of performing spectroscopy at high ®eld is the predicted theoretical linear improvement in the SNR with ®eld. This improvement (a factor of 2.7 from 1.5 to 4.1 T) translates into a linear increase in accuracy of the measurement of metabolites, or a reduction in necessary volume size to achieve the same SNR by 2.7 or, alternatively, a factor of 7.3 increase in speed of acquisition for an equivalent volume and SNR. However, reductions in volume size, especially for spectroscopic imaging, provide special challenges with respect to volume misregistration arising from the increased spectral bandwidth and limited rf ampli®er power and gradient strength. To achieve these gains, the alterations in relaxation properties (signi®cantly shorter T2 values of metabolites) must also be addressed in the methods used in data collection. Finally, changes in the J coupling of 1 H metabolites must also be added before improvements in spectral resolution and spectral simpli®cation can be realized. 3.1 Improvements in SNR and Spatial Resolution at High Field Hoult has previously predicted that the SNR improvement under sample noise dominant conditions should be linear with increasing ®eld.9 This improvement has been documented for 1 H spectroscopy where the increase in the SNR for N-acetyl-

aspartate (NAA) was found to be increased by a factor of 2.62±3.44 at 4.1 T over that of 1.5 T after corrections for pulse sequence ef®ciencies, volume sizes, scans and differential relaxation properties.10 This improvement in the SNR has allowed high-resolution spectroscopic images of human cerebral NAA to be created with an SNR of 32.5 : 1 from voxel sizes of 0.5 ml (Figure 2), a factor of two smaller than that typically used at 1.5 T. Despite the small spatial extent of the gray matter about the intrahemispheric ®ssure, the 0.5 ml voxel allows resolution of the alterations in metabolite ratios and content in gray and white matter. Speci®cally, the elevated total creatine (Cr)/NAA ratio reported in autopsy results11 for gray matter to white matter is easily visualized in the spectra from parietal white and gray matter (Figure 2). The ratio NAA/Cr has been extensively used as a marker for neuronal loss, such that the resolution of these differences has important applications in resolving true neuronal loss from partial volume effects resulting from inclusion of a mixture of gray and white matter. By optimizing the repetition time, echo time, and using frequency selective suppression of the lipid resonances, ultrahigh-resolution spectroscopic images can be obtained. Proton spectroscopic images employing volumes as small as 175 l have been acquired using a homogeneous volume head coil and are presented in Figure 3. In these images, the 2±3 mM difference in Cr content between gray and white matter is easily visualized in the Cr spectroscopic image. Perhaps more signi®cantly, the improved SNR can also be used to improve the spatial resolution of low sensitivity nuclei such as 31P and 13C. Typically, 31P spectra of the human heart are acquired with voxel resolutions of 20 to 80 ml using single voxel localization methods or, more recently, one-dimensional chemical shift imaging.12±14 Because of the large voxels acquired in these studies, signi®cant contamination from the myocardial blood pool occurs. Owing to this contamination, corrections for the adenosine triphosphate (ATP) content of the blood pool, based on the 2,3-diphosphoglycerate (2,3-DPG) content are required, before correct myocardial PCr/ATP ratios can be calculated. At 4 T, Menon has demonstrated, using a one-dimensional chemical shift imaging method, that voxels of 8±25 ml can be collected from the human heart that are free of blood pool contamination.15 Using the increased SNR available at 4.1 T, complete three-dimensional chemical shift images of the myocardial PCr content can be acquired using three-dimensional spectroscopic imaging methods. Displayed in Figure 4 are representative spectra of human heart acquired with a spectroscopic imaging sequence using 8 ml voxels.16 Evaluation of spectra acquired from the apex, septum, and left ventricular free wall shows that they have minimal contributions from 2,3DPG (Figure 4). Therefore, quantitative measurements of the regional levels of high-energy myocardial phosphates are possible. Gruetter has used the enhanced sensitivity at 4 T to acquire 13C spectra of human brain from volumes of 72 ml using a 72 min acquisition.17 3.2

Relaxation Properties at High Field

Many 1H localization methods have relied heavily on long spin echo sequences (TE = 136±272 ms) to achieve good water and lipid suppression. These methods have been highly successful and provide good sensitivity through the long T2 values of many cerebral metabolites at 1.5 T. Frahm has measured the

HIGH-FIELD WHOLE BODY SYSTEMS

3

(c)

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

1

6

2

7

3

8

4

9

5

10

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

Figure 2 (a) Proton scout image of the human brain and (b) the corresponding NAA image. The NAA image is acquired from a FOV of 240  240 mm using 32  32 encodes. The echo time (TE) of the spectroscopic image is 50 ms. (c) The numbered locations (1±10) on the scout image correspond to the numbered spectra. The resolved resonances are NAA 2.02 ppm, creatine 3.02 ppm, and choline 3.19 ppm. (Reproduced from Hetherington et al.10 )

4 HIGH-FIELD WHOLE BODY SYSTEMS

Figure 3 A scout image (a), a spectroscopic image of the distribution of creatine (CR) in the selected slice (b), and an overlay image (c), where the blue contours indicate the edges between gray and white matter. (d) An image and three representative spectra (175 l nominal volume) and the corresponding locations on the scout image

T2 values of NAA, Cr, and choline to be 450, 240, and 330 ms, respectively.18 At 4.1 T the T2 values of these metabolites become signi®cantly shorter, decreasing by nearly a factor of two to 220, 140, and 190 ms, respectively.10 Similar values have also been reported by Posse and co-workers.19 Therefore, signi®cant losses occur at high ®eld when long echo times are used. This necessitates the use of short-echo localization methods such as stimulated echo acquisition (STEAM)20 and/ or simple spin echo methods.21 Unlike T2 relaxation, T1 relaxation times for 1H metabolites are relatively unchanged. Posse has reported T1 times of 1.6, 1.6, and 1.2 s for NAA, Cr, and choline, respectively, at 4 T. These values compare well with the T1 values of 1.45, 1.55, and 1.15 s, respectively, measured at 1.5 T. Therefore, no additional sequence modi®cations are required for T1-based selection or discrimination methods.

Myocardial 31P metabolites show increases in their T1 values. Menon et al. have measured myocardial T1 values of ATP and PCr to be 2.7 and 5.4 s, respectively. Therefore, the lengthening of the T1 at 4 T for 31P metabolites will tend somewhat to offset the SNR gains for equivalent repetition times by causing greater saturation effects. To take full advantage of the SNR gains possible in 31P spectroscopy, methods utilizing partial tip adiabatic excitations22 and/or combinations of homogeneous transmit and surface coil receivers23 will be of signi®cant bene®t. 3.3

Spectral Resolution

One advantage of high-®eld spectroscopy is the improvement in spectral resolution achievable for J-coupled

HIGH-FIELD WHOLE BODY SYSTEMS

Septum

Apex

Left ventricle

Skeletal muscle

15

10

5

0

–5 –10 –15 –20 ppm

Figure 4 Voxels (8 ml) acquired from a three-dimensional spectroscopic imaging sequence applied to the human heart. The resolved resonances are for PCr, (0 ppm), -, -, and -ATP (ÿ2 to ÿ18 ppm) and inorganic phosphate (Pi) + 2,3-DPG (5±10 ppm) (Reproduced from H. P. Hetherington, D. J. E. Luney, and J. T. Vaughan, Magn. Reson. Med., 1995, 33, 427.)

resonances. This effect stems from two factors, the increase in spectral dispersion (in hertz) from the increased ®eld in the presence of a ®eld invariant coupling constant and a simpli®cation of strong coupling for homonuclear coupled 1H resonances. The advantage of the increased spectral resolution is most

5

easily seen in the 1H spectra of human brain. For 1H spectroscopy, the most dramatic improvement is in the resolution of the C-3 and C-4 resonances of amino acids such as glutamate and glutamine. At 1.5 T, the C-3 and C-4 resonances of glutamate are strongly coupled such that they appear as a broad complex multiplet (Figure 5). However at 4.1 T the C-3 and C-4 resonances experience much weaker coupling such that they resolve into two clearly distinct multiplets. This improved resolution also permits the individual C-4 resonances of glutamate and glutamine to be resolved from each other. Speci®cally the two most down®eld lines of the glutamine triplet are resolved from the two most up®eld lines of the glutamate triplet (Figure 6). Therefore, unlike measurements at 1.5 T, where the strongly coupled resonances of glutamate and glutamine show substantial overlap, the individual contributions of these resonances can be clearly determined at 4 T.24 However, as the nature of the coupling changes, so do the amplitudes of the multiplets as a function of echo time. Although echo times of 20±30 ms can be used for detection of amino acids at 1.5±2.0 T,25,26 an echo time of 20 ms results in substantial loss in sensitivity at 4.0 T.24 However, Pan has demonstrated that SNR losses in amino acids can be avoided at moderate echo times when using a J-refocused acquisition.27 In this sequence, the effects of J-modulation for coupled spins is reversed by transferring magnetization through a multiple quantum evolution period. Figure 7 shows spectra acquired with a 40 ms TE using the J-refocused sequence from human brain at 4 T. In addition to the improvement in spectral resolution of amino acids, the increased separation between coupled reson-

Glutamate 1.5 T

Glu Gln

3.5

3.0

2.5

2.0

1.5

1.0 ppm

Glutamate 4.1 T

3.5

3.5

Figure 5

3.0

2.5

2.0

1.5

1.0 ppm

Spectra of a solution of glutamate acquired at 1.5 and 4.1 T

3.0

2.5

2.0

1.5 ppm

Figure 6 Spectrum acquired for a 2.25 ml voxel from human brain at 4.1 T as part of a spectroscopic imaging study using a TE of 15 ms. The two most up®eld lines of glutamate (Glu) are resolved from the two most down®eld lines of glutamine (Gln). The data set was acquired using 16  16 phase encodes with four phase cycles and a TR of 2000 ms

6 HIGH-FIELD WHOLE BODY SYSTEMS

NAA Control hippocampus -CH Cr

Cr Ino

Tau Ch

Asp

Gln Glu Glx Patient ipsilateral

Patient contralateral 3.5

Figure 7 The J-refocused spectroscopic imaging sequence used to acquire a 1 ml spectrum from the occiptal lobe. NAA, Nacetylaspartate; Glx, glutamate and glutamine C-3; Glu, glutamate C4; Gln, glutamine C-4; Asp, aspartate; CR, creatine; Ch, choline; Tau, taurine; Ino, inositol; -CH, the alpha amino proton resonance of a variety of amino acids

ances at 4 T affords signi®cant advantages for spectral editing sequences. Rothman has demonstrated at 2.1 T that gammaaminobutyric acid (GABA), the primary inhibitory neurotransmitter in mammalian brain, can be detected using spectral editing sequences.28 However, because of the proximity of a macromolecule resonance and similarity in chemical structure, substantial overlap (60%) between the two resonances occurs in the edited spectrum. At 4 T, as a result of the additional separation in hertz between the coupled resonances of the macromolecule (1.7 ppm) and C-3 GABA resonance (1.9 ppm), this overlap can be reduced to 24% when highly selective inversion pulses are used.29 Use of even more highly selective pulses can reduce this overlap to less than 10%. Use

2.5

2.0

1.5

1.0

0.5 ppm

Figure 9 Spectra acquired from the hippocampi of a patient with temporal lobe epilepsy and a normal volunteer as part of a spectroscopic imaging at 4.1 T. The data were acquired using a TE of 50 ms and a TR of 2000 ms. The 240  240 ®eld-of-view was acquired using 32  32 phase encodes with a single acquisition per encode

of these methods has permitted the measurement of acute elevations of brain GABA levels in response to novel antiepileptic drugs in normal controls (Figure 8).30 3.4

Clinical Applications

The impact of the ability to acquire small voxels can be appreciated in the use of 1H spectroscopic imaging in the diagnosis of temporal lobe epilepsy in patients with intractable seizures. It has been demonstrated by a number of laboratories that 1H spectroscopy at 1.5±2.0 T provides a sensitive measure of neuronal loss as expressed in NAA content or NAA/Cr ratio measures.31±33 However, the small size of the hippocampus (1.5 ml  1 ml  3 ml) and the complex anatomy surrounding the hippocampus, cerebellar vermis, temporal gray and white matter, and midbrain, with their varying Cr/NAA ratios (0.5± 1.33, normal gray matter is 0.7) can obscure measurement of neuronal loss. False positives and negatives can occur when signi®cant amounts of the cerebellar vermis or temporal white matter are included.34 However, the small voxels afforded at 4 T (0.5 ml nominal volume) permit separation of these regions and improve the sensitivity of the detection of epileptic foci. Figure 9 shows typical spectra from the ipsilateral and contralateral hippocampi of an epilepsy patient and a healthy volunteer. The affected hippocampus is identi®ed by the marked increase in the ratio of Cr/NAA.

4

Figure 8 Spectra acquired from a 13.5 ml volume in the occipital lobe employing a 4.26 min acquisition. Spectrum acquired without (a) and with (b) a selective inversion pulse applied to the 1.9 ppm gammaaminobutyric acid (GABA) C-3 resonance. (c) The difference spectrum of (a) and (b) displays the characteristic doublet GABA-edited spectrum (outer two lines of the GABA triplet)

3.0

ANATOMICAL AND FUNCTIONAL IMAGING

Previously it has been expected that anatomical images at high ®eld would be degraded by the effects of an increased skin depth effect and the loss in intrinsic contrast by decreasing differences in the T1 values of gray matter, white matter and cerebrospinal ¯uid. Ugurbil and colleagues ®rst demonstrated that excellent gray/white contrast could be achieved in the human brain at 4 T using a modi®ed driven equilibrium sequence exchanging the SNR for contrast.35 The sequence uti-

HIGH-FIELD WHOLE BODY SYSTEMS

Figure 10 High resolution 3 mm image of the human brain acquired at 4.1 T. The ®eld-of-view is 240  240 using a resolution of 512  512. The data were acquired with 3 mm slice thickness as part of an eight slice study using an inversion±recovery time of 1000 ms, a TR of 2500 ms, and a TE of 17.5 ms

lized a 90±180±90 sequence to provide an approximate quadratic dependence on T1. More recently it has been demonstrated that excellent contrast can be achieved using a simple inversion±recovery sequence with a gradient echo readout.5 Displayed in Figure 10 is a 3 mm 512  512 gradient echo image of a human brain acquired at 4.1 T. Such images display excellent resolution of a number of cerebral anatomical features (caudate head, globus pallidus and putamen). The internal capsule is resolved through the high gray/white matter contrast. Resolution of these subcortical structures is important since they are frequently the site of small strokes less than 10 mm in diameter (lacuneae). The major groups of the thalami are resolved and correspond to the medial and lateral thalamic nuclear masses. The high spatial resolution and contrast are critical for visualizing the thalamic masses given that the internal medullary lamina, the intervening structure, is approximately 1 mm thick. Because of the arterial blood ¯ow originating from the posterior aspect of the circle of Willis medial to the temporal lobe, the highest quality coronal images of the temporal lobes are obtained by gating the acquisition to the cardiac cycle. Owing to the strong susceptibility differences in the regions around the temporal lobes and the relatively long echo time used, some degradation of image quality is experienced in the temporal lobe. This can be compensated for in part by using a shorter gradient echo time in the acquisition sequence. Regardless, excellent resolution and detail is observed in the visualization of the hippocampus (Figure 11). Speci®cally, the parahippocampal gyrus, Ammon's horn with the alveus extending into the fornix, the subiculum, and the overlying super®cial medullary stratum are well de®ned. The reliability of lateraliz-

7

Figure 11 High-resolution 3 mm image of the human brain acquired at 4.1 T. The ®eld-of-view is 240  240 using a resolution of 512  512. The data were acquired with a 3 mm slice thickness as part of an eight slice study using an inversion±recovery time of 1000 ms, a TR of 2500 ms, and a TE of 17.5 ms. The data were acquired with a coronal orientation through the hippocampi

ing the focus in temporal lobe epilepsy by MRI will be improved with this level of resolution. Such high-resolution images may also aid in the further study and diagnosis of Alzheimer's disease. Perhaps the most striking qualitative difference between 4.1 T high-resolution images and those acquired on widely available lower ®eld clinical systems is the visualization of small cerebral vessels in all axial images (Figure 12). With ®ner slice separation it should be possible to de®ne the majority of the deep cerebral veins. The ®ne radiation of vessels has not previously been seen routinely in vivo. Visualization of such vessels is most demanding of resolution and contrast available at high ®eld. The contrast is most likely related to the deoxygenated hemoglobin present in veins, decreasing the signal against a background of high signal intensity from white matter. In summary, imaging at 4.1 T depicts the brain with very high resolution, which should allow detection of disease states not heretofore imaged noninvasively. This approach could potentially obviate the need for angiography or biopsy. The best advantage of high-®eld imaging is that of the enhanced sensitivity to changes in deoxyhemoglobin, which forms the mechanism for blood oxygenation-dependent (BOLD) contrast in functional imaging.35 The enhancement of ®eld strength is believed to range from linear to quadratic in strength. Thus the small changes that are seen in functional imaging studies (1±5%) can be measured with the highest sensitivity only at high ®eld. This effect has been used by a number of investigators at 4 T to investigate the effects of various paradigms of brain activity on oxygen consumption and delivery and the localization of those effects. The results suggest that valuable insights into brain physiology and psychiatric and various pathological states are on the horizon. The utility of functional imaging at high ®eld will provide a primary driving force for the proliferation of high ®eld systems.

8 HIGH-FIELD WHOLE BODY SYSTEMS

Figure 12 High-resolution 3 mm image of the human brain acquired at 4.1 T. The ®eld-of-view is 240  240 using a resolution of 512  512. The data were acquired with a 3 mm slice thickness as part of an eight slice study using an inversion±recovery time of 1000 ms, a TR of 2500 ms, and a TE of 17.5 ms. Note the ®ne detail showing deep cerebral veins about the ventricles

5 RELATED ARTICLES Birdcage Resonators: Highly Homogeneous Radiofrequency Coils for Magnetic Resonance; Low-Field Whole Body Systems; MRI at Mid®eld Strength; Structural and Functional MR in Epilepsy; Whole Body Studies: Impact of MRS. 6 REFERENCES 1. H. J. Schneider and P. Dullenkopf, Rev. Sci. Instrum., 1977, 48, 68. 2. H. Barfuss, H. Fischer, D. Hentschel, R. Ladebeck, A. Oppelt, R. Wittig, W. Duerr, and R. Opelt, NMR Biomed., 1990, 3, 31. 3. P. K. Roeschmann, US Patent 4 746 866, 1988. 4. J. T. Vaughan, H. P. Hetherington, J. O. Otu, J. W. Pan, and G. M. Pohost, Magn. Reson. Med., 1994, 32, 206. 5. J. W. Pan, J. T. Vaughan, R. I. Kuzniecky, G. M. Pohost, and H. P. Hetherington, Magn. Reson. Imag., 1995, 13, 915. 6. N. Zhang, M. S. Roos, J. T. Vaughan, S. T. S. Wong, and T. F. Budinger, Proc. IVth Annu. Mtg (Int.) Soc. Magn. Reson. Med., New York, 1996, p. 252. 7. J. T. Vaughan, H. P. Hetherington, and G. M. Pohost, Proc. IInd Annu. Mtg (Int.) Soc. Magn. Reson. Med., San Francisco, 1994, p. 1119. 8. G. Adrianny, J. T. Vaughan, P. Andersen, H. Merkle, M. Garwood, and K. Ugurbil, Proc. IInd Annu. Mtg (Int.) Soc. Magn. Reson. Med., Nice, 1995, p. 921.

9. D. I. Hoult, and P. C. Lauterbur, J. Magn. Reson., 1979, 34, 425. 10. H. P. Hetherington, G. F. Mason, J. W. Pan, S. L. Ponder, J. T. Vaughan, D. B. Twieg, and G. M. Pohost, J. Magn. Reson., 1994, 32, 565. 11. O. A. C. Petroff, D. D. Spencer, J. R. Alger, and J. W. Prichard, Neurology, 1989, 39, 1197. 12. P. A. Bottomley, R. J. Herfkens, L. S. Smith, and T. M. Bashore, Radiology, 1987, 165, 703. 13. P. R. Luyten, A. de Roos, L. J. M. J. Oosterwaal, K. Doornbos, and J. A. den Hollander, Proc. XIth Annu. Mtg Soc. Magn. Reson. Med., Berlin, 1992, p. 74. 14. R. G. Weiss, P. A. Bottomley, C. J. Hardy, and G. Gerstenblith, N. Engl. J. Med., 1990, 323, 1593. 15. R. S. Menon, K. Hendrich, X. Hu, and K. Ugurbil, Magn. Reson. Med., 1992, 26, 368. 16. H. P. Hetherington, D. J. E. Luney, J. T. Vaughan, J. W. Pan, S. L. Ponder, O. Tschendel, D. B. Twieg, and G. M. Pohost, Magn. Reson. Med., 1995, 33, 427. 17. R. Gruetter, G. Adriany, H. Merkle, and P. M. Andersen, Magn. Reson. Med., 1996, 36, 659. 18. J. Frahm, H. Bruhn, M. L. Gyngell, K. D. Merboldt, W. Hanicke, and R. Sauter, Magn. Reson. Med., 1989, 11, 47. 19. S. Posse, C. A. Cuenod, R. Risinger, D. Le Bihan, and R. S. Balaban, Magn. Reson. Med., 1995, 33, 246. 20. J. Frahm, K. D. Merboldt, and W. Hanicke, J. Magn. Reson., 1987, 72, 502. 21. H. P. Hetherington, J. W. Pan, G. F. Mason, S. L. Ponder, D. B. Twieg, G. Deutsch, J. Mountz, and G. M. Pohost, Magn. Reson. Med., 1994, 32, 530. 22. M. Garwood and K. Ugurbil, NMR Basic Princ. Progr., 1992, 110. 23. P. A. Bottomley, C. J. Hardy, and R. G. Weiss, J. Magn. Reson., 1991, 95, 341. 24. G. F. Mason, J. W. Pan, S. L. Ponder, D. B. Twieg, G. M. Pohost, and H. P. Hetherington, Magn. Reson. Med., 1994, 32, 145. 25. T. Michaelis, K. D. Merboldt, W. Hanicke, M. L. Gyngell, H. Bruhn, and J. Frahm, NMR Biomed., 1991, 4, 90. 26. R. Kreis and B. D. Ross, Radiology, 1992, 184, 123. 27. J. W. Pan, G. F. Mason, G. M. Pohost, and H. P. Hetherington, Magn. Reson. Med., 1996, 36, 7. 28. D. L. Rothman, O. A. C. Petroff, K. L. Behar, and R. H. Matson, Proc. Natl. Acad. Sci. USA, 1993, 90, 5662. 29. H. P. Hetherington, B. R. Newcomer, and J. W. Pan, Magn. Reson. Med., 1998, 39, 6. 30. R. I. Kuzniecky, R. H. P. Hetherington, S. Ho, J. W. Pan, R. Martin, F. Gilliam, and E. Faught, Neurology, 1998, 51, 627. 31. D. G. Gadian, A. Connelly, J. S. Duncan, J. H. Cross, F. J. Kirkham, C. L. Johnson, F. Vargha-Khadem, B. G. R. Neville, and G. D. Jackson, Acta Neurol. Scand., 1994, 152, 116. 32. F. Cendes, F. Andermann, P. C. Preul, and D. L. Arnold, Neurology, 1993, 43, 223. 33. J. W. Hugg, K. D. Laxer, G. B. Matson, A. A. Maudsley, and M. W. Weiner, Ann. Neurol., 1993, 34, 788. 34. H. P. Hetherington, R. I. Kuzniecky, J. W. Pan, G. F. Mason, J. T. Vaughan, C. Harris, H. Morawetz, and G. M. Pohost, Ann. Neurol., 1995, 38, 396. 35. K. Ugurbil, M. Garwood, J. Ellerman, K. Hendrich, R. Hinkle, X. Hu, S. G. Kim, R. Menon, H. Merkle, S. Ogawa, and R. Salmi. Magn. Reson. Q., 1993, 9, 259.

LOW-FIELD WHOLE BODY SYSTEMS

Low-Field Whole Body Systems Leon Kaufman, David Kramer, Joseph Carlson and Mitsuaki Arakawa UCSF-RIL, San Francisco, CA, USA

1

to provide a wide range of contrast capabilities with good signal-to-noise ratios.4,5 Furthermore, for water-elevating lesions, for the same sequence, contrast will typically be higher at the lower ®eld. This is due to two factors: the shortened T1 of tissues permits contrast from T2 and proton density N(H) changes to become visible at shorter TR values;6 and the change in T1 and T2 with water content is larger the lower the ®eld.1,2 The impact of these effects will depend on the type of study being performed, but, in general, for the same volume resolution and coverage, and for equivalent contrast, there will be a signal-to-noise ratio difference of 2±3 over a range of ®eld strength of a factor of 25.

1 INTRODUCTION

3

MRI is a medical diagnostic modality. As such, it is of value only if it is available to the people who can bene®t from its use. Because for most subjects MRI use is not limited by risk±bene®t considerations, cost±bene®t analysis is becoming the major determinant of use. In cases like this, as the cost of a study decreases, the utilization (and presumably the bene®t) increases. Decreasing the costs of MRI requires careful attention to the physics and engineering of the discipline. The major determinant of cost is ®eld strength. By reducing the ®eld strength it becomes possible to use nonsuperconducting magnets. Low-®eld systems can be sited in smaller rooms, and this further reduces initial and ongoing costs. Other components where signi®cant savings are achieved include rf transmitters and, in some cases, gradient coils. Once the cylindrical magnet con®guration can be avoided, the range of suitable low-®eld magnets includes some that are open. These, in turn, offer opportunities in terms of interventional studies (where low ®elds have some other practical advantages), and in dealing with dif®cult or ill patients.

Various choices of magnet are available for low-®eld development. Resistive magnets were the early choices, but with the advent of practical permanent magnet designs the latter have become incorporated into the more recent commercially available products. Magnet technology involves two basic components: the driver and the core of the magnet. The driver provides the motive force that establishes the magnetic ®eld. Three types of driving technology are available in the marketplace: resistive, permanent, and superconducting. The Earth's magnetic ®eld is a fourth source that has been discussed in the literature but has not become commercially available. The core is the medium upon which the driver acts. Because a patient has to be able to be located in the magnet (a prerequisite for reimbursement), all magnets have at least a portion of their ®eld in air. Air-core magnets are those where all of the core is air. These may have a return path (the region outside of the core where the magnetic ®eld lines close) that is either air or iron. Iron-core magnets have, except for the patient volume, iron paths both for the drivers to act on and as return paths. The relative merits of these approaches to magnet design are discussed below.

2 PHYSICS OF LOW FIELD STRENGTH OPERATION As ®eld strength decreases, so does the strength of the NMR signal emitted by the body. Nevertheless, imaging is a complex process that involves many factors. With decreasing ®eld strength, noise from the body also decreases. Chemical shift and susceptibility artifacts decrease, permitting narrower bandwidths and consequent reductions in noise. T1 shortens1,2 and T2 lengthens,2 this increasing signal level during imaging. Motion artifacts decrease, so that the fraction of study time spent on motion reduction can be used for data acquisition, and rf power deposition decreases dramatically. Radiofrequency pulses can be tailored to improve section pro®les and, more importantly, the hazard from rf heating is all but eliminated. At lower ®eld, absolute magnetic ®eld inhomogeneities get smaller (for imaging, a magnet with a 30 ppm homogeneity speci®cation at 500 G is equivalent to one with a 1 ppm speci®cation at 1.5 T). This, and decreased susceptibility effects, permit the use of gradient echo techniques where spin echoes may need to be used at higher ®elds.3 The shortened T1 values favor the use of three-dimensional FT techniques, which can be designed

MAGNETS FOR LOW-FIELD MRI

3.1 3.1.1

Drivers Resistive Drivers

In a resistive driver, a current is established on a good conducting material, typically copper or aluminum (during the Manhattan Project silver from the US reserves was used for huge magnets for uranium separation, since copper and aluminum were needed for armaments). The conductor is at or near room temperature, so it offers resistance to the ¯ow of current. This resistance has two effects: power has to be used to maintain the current ¯ow and attendant magnetic ®eld, and heat is generated in the conductor through resistive losses. Consequently, water needs to be used to extract this heat. In addition, the stability of the ®eld depends on power-supply stability, stability coming at a price. The advantages of the resistive driver is that the technology is well known and understood, and the entry is easy, since a great deal of expertise exists or is easily acquired by a company. The disadvantages are high power consumption, large cooling demand, a large conductor mass, and its consequent weight. When an iron core is used, the larger mass of conduc-

2 LOW-FIELD WHOLE BODY SYSTEMS tor forces an increase in the physical size of the core, which then results in further weight increases. 3.1.2

Permanent Drivers

In a permanent magnet, the ®eld originates in blocks of permanently magnetized material. Currently available are ferrite magnets for low-®eld operation and Neomax (a rare earth alloy) for higher ®elds. The attractive aspect of permanent magnet drivers is that they are totally passive, requiring no added services such as electricity, cryogens, or cooling. The disadvantages are that there is a very steep cost/®eld curve, especially when rare-earth materials are needed for higher ®elds; in addition, the weight of the permanent magnet material adds to the weight of an iron core. For the higher ®eld units, a potential problem is that the magnet cannot be discharged in an emergency. 3.1.3

Superconducting Drivers

There are materials, generically known as superconductors, that if kept at a suf®ciently low temperature lose all resistance to current ¯ow. In typical magnets for MRI, this temperature has to be well below 10 K, that is, under 10 C above absolute zero. A convenient way to achieve this is by the use of liquid helium, which has a temperature of near 4 K. The liquid boils away through heat leaking into the magnet from room temperature. Because of cost, it is desirable to minimize the boil-off of the helium. One method used in the early years of commercial MRI was to use a surrounding container with liquid nitrogen, which has a temperature of 77 K, intermediate between the helium and room temperatures. This, of course, adds liquid nitrogen consumption to the services needed for magnet operation. Later, electrical shield coolers were added to magnets to replace the liquid nitrogen `shield'; these coolers usually reached temperatures well below 77 K, thus improving the shielding effect. Also, recirculating coolers that relique®ed the vented helium gas were added to some magnets, even though in the USA this was not usually economic because of the relatively low cost of liquid helium. More recently, electrical coolers that reach 4 K have become commercially available. This opens the way to cryogenless superconducting magnets with low power consumption. It is worth noting that the high-temperature superconducting materials that a few years ago were promoted as permitting magnets cooled by liquid nitrogen have yet to provide the performance required by MRI magnets; however, they have found use as superconducting leads, which bring current from the outside to the main windings during magnet charging and discharging. The advantage of superconductors for MRI is the achievement of relatively higher ®elds compared with the other technologies, with low weight and modest power consumption. The disadvantage is the need for cryogens (or, more recently, cryocoolers that can replace cryogens) and the technical complexity of these systems compared with other drivers, this making entry more dif®cult. In terms of cost, at a certain ®eld strength crossover point superconductors become cheaper than other drives. Once the mid®eld range is reached, they have de®nite advantages in this respect.

3.2

Magnet Core

The driver can be shaped to obtain the desired imaging ®eld and ®eld return path without the aid of signi®cant amounts of iron; this is known as an air-core magnet. Alternatively, the iron can be used for shaping the imaging ®eld and providing a return path; this is an iron-core magnet. Some air-core magnets have an iron return path or shield. The advantage of air-core magnets is low weight. The disadvantages are inef®cient use of conductor, demands on the conductor con®guration and manufacturing tolerances (since the homogeneity depends on the conductor), and large fringe ®elds that require active or passive shielding. All of this, of course, results in increased costs. Iron-core magnets make more ef®cient use of the driver, since the iron provides a medium that offers less `resistance' to the ¯ow of the magnetic ®eld lines. In iron-core magnets fringe ®elds are relatively small, and homogeneity is determined by the shape of the iron. This places fewer demands on the driver's con®guration and tolerances. The main unavoidable disadvantage of these cores is weight. 3.3

Other Factors

There are obviously other nuances about the different drivers and cores that need to be taken into account when designing an MRI system, but the characteristics described above are of a reasonably general nature and not amenable to signi®cant manipulation. Commercially, we have seen the following driver±core combinations become available at one time or another in the lifetime of clinical MRI: (a) resistive, superconducting, and permanent driver air cores, the ®rst two sometimes with iron shielding; and (b) resistive, permanent, and superconducting driver iron core. Iron cores can be 1-(C), 2-, and 4-post designs. Recent magnet assembly technology results in elimination of eddy currents without the need for active gradient coil shielding. The design of these magnets involves tradeoffs among cost, size, weight, gap, homogeneity, fringe ®eld and ®eld strength. Typically, larger gaps, higher homogeneities, smaller fringe ®elds, and higher ®eld strengths require larger, heavier magnets of higher cost.

4

SEQUENCING FOR LOW-FIELD IMAGING

Sequence development for low-®eld imaging needs to take into account, and take advantage of, the short T1 value of solid tissues. Also of importance are reduced chemical shift and susceptibility artifacts. One example involves strategies for the detection of lesions where water content is increased. This is a typical effect seen in tumors, edema, and infarcts. Considering the case where there is no blood, for these lesions N(H), T1, and T2 are all elevated.6 In a spin echo sequence the elevation of N(H) and T2 increases signal, while the elevation of T1 decreases signal. Thus N(H) and T2 effects are diluted by T1 effects. To obtain a lesion that is brighter than the background it is necessary to increase TR to the point that T1 effects become unimportant. This results in what is misnomered a `T2weighted' image. In fact, a more accurate measure would be `not T1-weighted'. To achieve this condition reliably, it is

LOW-FIELD WHOLE BODY SYSTEMS

necessary to operate with a TR of at least 2T1, preferably 2.5± 3T1. Since T1 increases with ®eld strength, this means that it is necessary to operate with a longer TR at the higher ®eld. For instance, for brain tissue the increase between 0.064 T and 1.5 T is approximately a factor of 4, from 250 ms to about 1 s for white matter, and from 300 ms to 1.2 s for gray matter. Thus, for a spin echo sequence, the desired lesion contrast is easier to achieve, and is more reliably achieved, at the lower ®eld. This has signi®cant consequences in terms of the signal-to-noise ratio, since shorter TR values can lead to a larger number of data averages for ®xed imaging time. For `watery' lesions, the ability to use short TR values opens an avenue for imaging with three-dimensional FT sequences. The TR that can be achieved in three-dimensional FT imaging is limited by practical considerations of time, since this time increases linearly with the number of slices obtained. If gradient echo instead of spin echo imaging is used, a ¯ip angle smaller than 90 results in a decrease of T1 effects. Lowering the angle at ®xed TR has the same effect as lengthening TR for a 90 angle. Unfortunately, this cannot be carried to extremes, but this process is more effective the larger the TR/T1 ratio. Thus, from a practical point of view, it is only achievable for short T1 values, where it is highly effective. Furthermore, to obtain the desired contrast level it is desirable to operate with a relatively long TE, to take advantage of T2 effects. Gradient echo imaging (used in conjunction with partial ¯ip angles) is highly sensitive to susceptibility and inhomogeneity effects, this sensitivity increasing with TE and with ®eld strength. At low ®eld the gradient echo technique shows no noticeable artifacts from susceptibility, chemical shift or inhomogeneity artifacts, permitting a late echo (TE = 53 ms) with a long sampling window. A relatively long TR value (110±145 ms in our case), combined with a ¯ip angle of 12± 20 , at 640 G, reduces T1 effects so that the elevated T2 and N(H) of lesions makes them appear bright, with cerebrospinal ¯uid (CSF) still being of low intensity4 (Figure 1). Thus, the combination of short T1 and the low inhomogeneity and susceptibility effects that accompany low-®eld operation permit for this application the use of gradient echo three-dimensional FT imaging, which under these conditions results in signal-tonoise ratios at low ®elds comparable to those of mid- and high-®eld systems when taking into account the imaging time and the number of sections obtained.4 At this ®eld strength three-dimensional FT imaging is the preferred method for obtaining images of thin sections.5 Figures 2 and 3 show a pituitary with section thicknesses of 4.5 and 2.25 mm, respectively. A hybrid technique consisting of a multislab two-dimensional FT sequence with two-slice encodings within a slab, coupled with partial ¯ip angle gradient echo imaging, has permitted us to develop a sequence that takes advantage of low ®eld operation characteristics and doubles the signal-to-noise ratio over that of conventional spin echo.3 The multislice double echo spin echo imaging technique introduced by Crooks7 quickly became the `gold standard' of lesion detection.8 At low ®eld it is possible to increase its ef®ciency dramatically by the following steps: 1. Take advantage of short T1 values by using a short TR and slice doubling9 to maintain high coverage. The shortened

3

Figure 1 Transaxial view of a patient with multiple sclerosis lesions, obtained at 640 G. Lesions are brightest, and, in descending order of intensity, gray matter, white matter and CSF. Three-dimensional FT gradient echo with TR = 145 ms, TE = 53 ms, and ¯ip angle = 20

TR is traded off against increased data averaging to increase the signal-to-noise ratio. 2. Lengthen the echo window. 3. Use gradient reversal instead of rf refocused echoes to reduce dead time in the sequence. This permits the number of slices obtained to be high, even though the windows are long.

Figure 2 Three-dimensional FT gradient echo imaging with TR = 68 ms, TE = 24 ms, and ¯ip angle = 60 , at 640 G. Section thickness 4.5 mm

4 LOW-FIELD WHOLE BODY SYSTEMS

Figure 3 As Figure 2, but with a ¯ip angle of 45 and a section thickness of 2.25 mm

4. Take advantage of the gradient reversal echo to use partial ¯ip angle excitations, in order to maximize further the signal-to-noise ratio and the contrast.

Figure 5 Two-dimensional FT, multislice gradient echo images through the head of a subject with white matter lesions. Note the low level of susceptibility artifacts at 640 G

Figure 4 (a, b) Spin echo imaging 640 G in a patient with multiple sclerosis; TR = 2 s, and TE = 30 and 105 ms in (a) and (b), respectively. (c, d) Gradient echo images; ¯ip angle = 60 , TR = 0.8 s, and TE = 20 and 70 ms in (c) and (d), respectively. The spatial resolution and number of sections are the same for both sequences. The earlier TE values of the set (c, d) reduce contrast, but the display system can be set to display any TE in a real-time basis. Comparing (a, b) and (c, d), the signal-to-noise ratio has essentially doubled, in the latter

Compared with a spin echo sequence, for equal imaging time and number of slices, this sequence has comparable contrast and twice the signal-to-noise ratio3 (Figures 4 and 5). A signi®cant bene®t is derived in low-®eld operation from the fact that, while at low ®eld solid tissues have shorter T1 values, CSF has a T1 that is constant with ®eld. Consequently, at the short TR where lesion contrast is achieved, the signal from CSF is small. At high ®eld, because TR has to be long, the CSF signal is large, obscuring lesions that may be contiguous with CSF spaces. To summarize, low ®eld strengths offer many advantages for the detection of water-elevating lesions, including increased sensitivity to water changes, the ability to operate with relatively short TR values, and the avoidance of confusing effects from bright CSF.

LOW-FIELD WHOLE BODY SYSTEMS

Figure 6

MRA at 640 G, using phase contrast imaging

5

Figure 7 Subject with silicone breast implants. Top row: TR = 110 ms, TE = 40 ms, ¯ip angle = 45 , magnitude image. Bottom row: phase image from the same data set. At 640 G, silicone and water are nearly out of phase (silicone, bright; water, dark) and fat is intermediate

Another example of low-®eld sequencing involves magnetic resonance angiography (MRA). In MRA the clinically most successful techniques take advantage of wash-in effects.10 The blood entering the volume has not been excited previously and thus produces a great deal of signal following the ®rst one or two excitations.11 Meanwhile, it is desirable to saturate the signal from stationary tissues by the same repeated excitations that produce signal from blood. This saturation is more effective the longer the T1, since recovery time is longer. Because high-®eld operation entails long T1 values, high contrast between blood and nonmoving tissues is easier to obtain. This permits more ¯exibility in trading vessel contrast for depth of penetration (®eld of view) than at low ®eld, where T1 values are short. For low-®eld systems, phase contrast angiography becomes an extremely attractive alternative. Figure 6 shows an example. Imaging of chemical shift is also accessible at low ®eld for protons in water, fat, and silicone, by reconstruction of phase images for three-dimensional FT sequences. We have found this method particularly powerful in searching for extracapsular silicone in breast implant patients.12 The chemical shift images have the same resolution and signal-to-noise ratio and do not require additional imaging time (Figure 7).

needed to highlight periventricular lesions (Figure 8). Fat signal gains in relative intensity compared with low ®eld, but inversion recovery and multipoint Dixon techniques make it easier to suppress fat (Figure 9). In gradient echo imaging, water and fat are out of phase at a convenient TE of 10 or 30 ms and in phase at TE of 20 ms. The former results in fast sequences for screening for bone marrow abnormalities (Figure 10). The T1 of blood lengthens enough to make time-of-¯ight MRA techniques attractive (Figure 11). Also, CSF is easier to highlight (Figure 12). Single-shot fast spin echo imaging results in effective imaging of long T2 tissues, principally CSF. Doublecontrast fast spin echo imaging adds to the ¯exibility with which contrast can be varied (Figure 13). Asymmetric spin echo sequences13 can be used to highlight T2* effects (Figure 14). A crossover occurs between the ef®cacy of two- and threedimensional imaging 14 so either is effective. Particularly effective are three-dimensional spin echo techniques, which provide ®ne anatomic detail (Figure 15).

5 SEQUENCES FOR MID-FIELD IMAGING

6

MRI started as a mid-®eld technique. Many of the sequences that are currently considered `routine' were originally developed at mid-®eld, so contrast characteristics are more familiar. The contrast characteristics of mid-®eld MRI are closer to low than to high ®eld. For instance, brain lesions are bright while CSF remains of lower intensity; consequently, ¯uid attenuation inversion recovery techniques (FLAIR) are not

A perception exists that the signal-to-noise (S/N) ratio increases linearly with ®eld strength. The physics of MRI shows that signal increases as the square of the ®eld, and measurements and theoretical considerations (based on certain assumptions about the body) support a linear increase in noise with ®eld, although the latter assumption should be considered

SIGNAL-TO-NOISE RATIO AND FIELD STRENGTH

6 LOW-FIELD WHOLE BODY SYSTEMS (a)

(b)

Figure 8 FSE imaging in a subject with multiple sclerosis. At high ®eld, when a TR is long enough to make lesions bright, CSF is also very bright. Consequently, inversion recovery (IR) techniques are used to suppress CSF (FLAIR). The cost is in the inef®ciency of IR imaging. FLAIR techniques can also be used at mid-®eld (a) (TR = 8 s, TE = 120 ms, T1 = 1.9 s) but are not necessary, as shown in (b) (TR = 2000 ms, TE = 60 ms). Both images were obtained at 0.35 T. The mid-®eld image (a) requires TR = 267 ms per slice, while without the IR pulse (b) TR is 91 ms

with caution. Based on these two factors, the `intrinsic' S/N ratio should increase linearly with ®eld. This simplistic view depends on disregarding all the factors that are important in imaging, all of which result in a much smaller increase than presumed. In addition to the `intrinsic' factor, the S/N ratio depends on the imaging conditions. Of interest is the S/N ratio per unit time for ®xed voxel dimensions and imaged volume, as well as constant number of slices. For a 90 ¯ip angle the relative S/N ratio will be proportional to fexp…ÿTE=T2 †‰1 ÿ exp…TR=T1 †Š

(a)

Figure 9

pp p window length slices= TRg …1†

for two-dimensional FT imaging. The reasons for the Hslices term is that if sequence A yields 10 slices and sequence B yields 20 slices with one excitation, then to cover the same volume sequence A has to be run twice, while sequence B in the same time could be run once with two excitations and a gain of H2 in the S/N ratio. Similarly, the HTR term comes in because, if two sequences provide the same coverage for one excitation, then the sequence with the shorter TR can be run with more excitations and increase the S/N ratio accordingly. For three-dimensional FT there is an additional Hslices term. For a discussion of the relative merits of two- and three-dimensional FT imaging the reader is directed to the paper by Carl-

(b)

Knee imaging in a subject with a bone marrow tumor. (a) Single-shot image for water. (b) Single-shot image of fat.

LOW-FIELD WHOLE BODY SYSTEMS

(a)

7

(b)

Figure 10 Gradient echo images of the knee in a subject with a vascular necrosis. (a) Water and fat in phase (TE = 20 ms at 0.35 T). (b) Water and fat out of phase (TE = 30 ms)

son et al.15 Based on imaging considerations we can analyze how the S/N ratio will behave.

16% to 50%, corresponding to a 1/20 to 1/6 power of ®eld for the loss in S/N ratio due to T2. The loss will be smaller for shorter TE and bigger for longer TE values.

6.1 T1 As ®eld strength increases, so does T1.1,2 A longer T1 means smaller signals. For discussion purposes, consider T1 varying as the square root of ®eld strength. For a short TR sequence the signal will decrease linearly with T1, or as the square root of the ®eld. For a long TR sequence the signal will not change much, but for a ®xed TR the contrast will be lower, or for ®xed contrast TR will be longer (longer imaging time). In this case the equivalent loss in S/N ratio will be smaller, approximately as the fourth root of ®eld strength. Thus, for longer TR values, T1 effects result in a loss in signal that increases as the fourth root to the square root of the ®eld. 6.2 T2

6.3

Window Length

Chemical shift, susceptibility, and inhomogeneity limit window lengths. As discussed above, the artifacts depend linearly on ®eld strength and window length. A shorter window reduces the artifacts, but also decreases the S/N ratio as the square root of window length. The net effect is that at a constant artifactual level there is a square root of ®eld strength term available in the S/N ratio, favoring low ®eld operation. In some sequences

(a)

(b)

There is clear evidence that T2 drops with ®eld strength.2 Between 0.063 and 3 T the T2 of brain drops from about 80 to 58 ms, for muscle from 37 to 30 ms, and for liver from 49 to 25 ms. For a TE of 30 ms, this results in signal drops from

Figure 11 Three-dimensional time-of-¯ight MRA at 0.35 T, acquisition time of 13.5 min

Figure 12 Sagittal C-spine images obtained with gradient echo imaging (TR = 360 ms, TE = 30 ms, ¯ip angle = 15 )

8 LOW-FIELD WHOLE BODY SYSTEMS (a)

(b)

Figure 13 Double-contrast fast spin echo (FSE) imaging in a subject with multiple sclerosis, using half-Fourier techniques. (TR = 2 s, TE = 20/120 ms, with 24 slices in 6.75 min)

this term is not usable, (for instance where TE is very short), and in others it is not fully usable, but in general it provides relative advantages of factors of 2±3 in S/N ratio for low-®eld operation. 6.4 Radiofrequency Power As discussed above, high-®eld systems compromise slice pro®le in the interest of savings in peak and average rf power. At mid- and low-®elds these conditions are not important, and square pro®les can be obtained.16 For a system where a gap of 50% is needed to avoid interslice cross-talk, the loss in the S/N

(a)

ratio for the slice itself may amount to 20%, and the gap can be considered as an unused potential source of 50% more signal.17 Thus, the full S/N ratio is not being achieved if the slice pro®le is poor. The loss in two-dimensional FT imaging is sequence dependent. For three-dimensional FT imaging there is no difference. 6.5

Artifact Reduction

As discussed above, artifact reduction schemes take time away from data collection, either by reducing the echo window length or the number of sections, or by lengthening TE, and

(b)

Figure 14 Asymmetric spin echo sequence (TE = 15ms). An MRI-compatible biopsy needle inserted in beef is imaged with 5 ms offset (a) and zero offset (b). By varying the offset, the visibility of the needle can be altered during an interventional procedure

LOW-FIELD WHOLE BODY SYSTEMS

9

broad based evaluation of ®ve units at 0.3, 1.0, and 1.5 T.18 Another study was limited to head imaging, comparing 0.064 and 1.5 T units,19 and a ROC study of MS and knee imaging was recently published.20 The earliest such study compared ®eld strengths of 0.5, 1 and 1.5 T.21 None of these studies showed a statistically signi®cant difference in ef®cacy. At the end of the 1990s, approximately 50% of the market in the USA is for open (low- and mid-®eld) MRI systems.

8

RELATED ARTICLES High-Field Whole Body Systems.

9

Figure 15 High-resolution head imaging using three-dimensional FSE (3 mm thickness, 0.5 mm resolution, 32 slices in 9.5 min)

may result in penalties in the S/N ratio per unit time. This loss depends on the particular sequence and can be as small as zero where artifact reduction is not needed, and as big as 50% in some extreme cases such as presaturation in multislice imaging.

6.6 Net S/N Ratio Change From the above considerations it can be seen that the S/N ratio changes with ®eld in a complex manner, which depends on the clinical purpose of the sequence, the type of sequence used and tissue imaged, on coverage (including gaps) and on artifacts. For a sequence such as MRA, a major part of the `theoretical' S/N ratio available from increasing ®eld strength can be obtained. For other sequences the gain is small. An approximate rule is that in many situations the S/N ratio increases as the third to fourth root of ®eld strength, so that for a 20-fold increase in ®eld, the S/N ratio is doubled or tripled.

7 DISCUSSION In the excitement of academic work, commercial competition, and a desire to share by publication, we tend to forget that MRI, outside a relatively few research institutions, is nothing more than a diagnostic tool. Its value is based on diagnostic accuracy and noninvasiveness. The lower the cost at which the service can be delivered, the more the people with need can have access to it. Very few studies have rigorously compared diagnostic ef®cacy at different ®eld strengths. One study, commissioned by the Australian Government, was a

REFERENCES

1. S. H. Koenig, R. D. Brown, D. Adams, D. Emerson, and C. G. Harrison, Invest. Radiol., 1984, 19, 76. 2. J.-H. Chen, H. Avram, L. E. Crooks, M. Arakawa, L. Kaufman, and A. C. Brito, Radiology, 1992, 184, 1. 3. D. M. Kramer, A. Li, L. Kaufman, and K. Hake, J. Neuroimaging, 1992, 2, 195. 4. P. A. Rothschild, M. Schulz, D. M. Kramer, and L. Kaufman, J. Neuroimaging, 1991, 1, 79. 5. D. M. Kramer, R. J. Guzman, J. W. Carlson, L. E. Crooks, and L. Kaufman, Radiology, 1989, 173, 541. 6. D. A. Ortendahl, N. M. Hylton, L. Kaufman, J. C. Watts, L. E. Crooks, C. M. Mills, and D. Stark, Radiology, 1984, 153, 479. 7. L. E. Crooks, M. Arakawa, J. C. Hoenninger, J. C. Watts, R. W. McRee, L. Kaufman, P. L. Davis, and A. R. Margulis, Radiology, 1982, 143, 169. 8. M. Brant-Zawadzki, D. Norman, T. H. Newton, W. M. Kelly, B. Kjos, C. M. Mills, W. Dillon, D. Sobel, and L. E. Crooks, Radiology, 1984, 152, 71. 9. A. A. Maudsley, J. Magn. Reson., 1980, 41, 112. 10. D. D. Blatter, D. L. Parker, S. S. Ahn, A. L. Bahr, R. O. Robison, R. B. Schwartz, F. A. Jolesz, and R. S. Boyer, Radiology, 1992, 183, 379. 11. L. Kaufman, L. E. Crooks, P. E. Sheldon, W. Rowan, and T. Miller, Invest. Radiol., 1982, 17, 554. 12. K. A. Derby, S. D. Frankel, L. Kaufman, D. M. Kramer, J. W. Carlson, M. I. Mineyev, K. A. Occhipinti, and R. Friedenthal, Radiology, 1993, 189, 617. 13. W. T. Dixon, Radiology, 1984, 153, 189. 14. D. M. Kramer, L. Kaufman, P. Rothschild, J. Hale, J. Wummer, and K. K. Hake, IEEE Trans. Med. Imag., 1991, 10, 382. 15. J. C. Carlson, L. E. Crooks, D. A. Ortendahl, D. M. Kramer, and L. Kaufman, Radiology, 1988, 166, 266. 16. D. A. Feinberg, L. E. Crooks, J. C. Hoenninger, J. C. Watts, M. Arakawa, H. Cheng, and L. Kaufman, Radiology, 1986, 158, 811. 17. J. B. Kneeland, A. Shimakawa, and F. Wehrli, Radiology, 1986, 158, 819. 18. MRI Technical Committee of the National Health Technology Advisory Panel, `MRI Assessment Program Final Report', Australian Institute of Health, Canberra, August 1990. 19. W. W. Orrison, G. K. Stimac, E. A. Stevens, D. K. LaMasters, M. C. Espinosa, L. Cobb, and F. A. Mettler, Radiology, 1991, 181, 121. 20. B. K. Rutt and D. H. Lee, JMRI, 1996, 6, 57. 21. US Food and Drug Administration, `Summary of Safety and Effectiveness Data, Premarket Approval Application No. P830074', submitted to the US Food and Drug Administration by the General Electric Company, 1989, p. 11.

10 LOW-FIELD WHOLE BODY SYSTEMS Biographical Sketches Leon Kaufman. b 1942. B.S., 1964, (engineering), Ph.D., 1967, (physics), University of California, Berkeley. University of California, San Francisco, 1970±present. Introduced to NMR in a junior modern physics class; started NMR imaging project at UCSF in 1975. Currently, Professor of Physics and Director, UCSF-RIL. Approx. 400 publications. Research interests include image formation and interpretation, and socioeconomic aspects of the technology. David M. Kramer. b 1952. B.A., 1973, (chemistry), Ph.D., 1978, (chemistry), State University of New York at Stony Brook. Worked on development of instrumentation and methods for applications of NMR in medicine. Previously employed by Technicare, 1981±86. Currently, Associate Adjunct Professor of Physics at the University of California at San Francisco, Radiologic Imaging Laboratory and a development scientist for Toshiba America MRI.

Joseph W. Carlson. b 1957. A.B. 1979, (physics) Princeton University, Ph.D. 1985, (physics), University of California, Berkeley. Presently on faculty in radiology, University of California, San Francisco. Research interests are centered on the development of medical MRI technology, with emphasis on the physics of MRI, design and analysis of electromagnetic components of magnetic resonance imagers and data analysis.

Mitsuaki Arakawa. b 1943. B.S., 1966, (physics), Seoul University. Introduced to NMR at Japan Electron Optics Laboratory, 1970, and by Lawrence Crooks and Leon Kaufman. University of California, San Francisco, 1977±present. Approx. 60 publications, including a chapter in the book NMR in Medicine±Its Basic and Clinical Application (in Japanese). Holds 18 US patents. Research and engineering interests include MRI detectors, and their interface and general instrumentation.

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

Magnetic Resonance Imaging: A Historical Overview E. Raymond Andrew University of Florida, Gainesville, FL, USA

1 INTRODUCTION MRI has become a most powerful modality of radiological imaging and is now found in all major hospitals throughout the world. It is a technique which provides the physician with clear pictures of the interior of the human body from any angle without using hazardous radiations. In the future it promises to provide interventional images to guide operations in progress. It is already giving valuable insights into the detailed functioning of the brain. Physicians sometimes express surprise that the basic phenomenon on which MRI is based was discovered 50 years ago. In this historical overview of MRI we therefore begin with a brief account of the origins of NMR and its applications in physics and chemistry and then go on to describe in more detail its applications in biology and medicine, with particular reference to MRI.1,2

2 DISCOVERY The ®rst successful demonstrations of NMR in bulk matter were published in 1946. Two independent groups, unknown to each other, working in physics laboratories on opposite coasts of the USA, discovered NMR almost simultaneously. Bloch, Hansen, and Packard,3 working at Stanford University, and Purcell, Torrey, and Pound,4 working at Harvard University, published their results in consecutive issues of Physical Review. The impact of their work was immediate, and applications of NMR have widened steadily from physics and chemistry to a broad range of disciplines from archeology to medicine. The importance of the discovery was recognized by the joint award of the 1952 Nobel Prize for Physics to the two leaders, Bloch and Purcell. Professor Bloch died in 1983; Professor Purcell died in 1997. These early experiments were described by their discoverers in quite different physical terms, and it was only after discussion between the two groups that it was realized that their discoveries were essentially identical. At Harvard University the emphasis was on transitions of magnetic nuclei between quantized states in a magnetic ®eld and on resonant absorption of radiofrequency energy. At Stanford University the description was of the precession of nuclear magnetization in a magnetic ®eld, inducing an electromotive force (EMF) in the surrounding radiofrequency coil. The ®rst description led to the

1

name `nuclear magnetic resonance' for the phenomenon, and the second to the name `nuclear induction'. These two descriptions, which can be shown to be quantitatively equivalent,5 still provide complementary views of the basic phenomenon, which are mutually illuminating. Today everyone uses the name `nuclear magnetic resonance', but the alternative `nuclear induction' is still often used by physicists and particularly survives in the term `free induction decay' (FID) for the NMR signal following an exciting pulse. Although NMR was ®rst discovered and experimentally demonstrated in ordinary materials in 1945, the subject did have a prehistory. In 1936 Gorter6 looked for the resonance of 7 Li nuclei in crystalline lithium ¯uoride and of protons in crystalline potassium alum, but without success. The failure of these early experiments and of some later attempts7 was attributed to the use of unfavorable materials.8 Meantime, NMR was ®rst demonstrated in molecular beams by Rabi and co-workers9 at Columbia University in the USA in 1939. Although the fundamental principles are closely similar, applications of NMR to atomic and molecular beams represent a separate subject from NMR in ordinary materials with which we are concerned here. Now that hospitals are spending vast sums on NMR clinical equipment it is sobering to recall that the ®rst NMR experiments were done at almost no cost at all. The Harvard University group borrowed a large electromagnet previously used in cosmic ray experiments, which in turn had been converted from a discarded generator from the Boston elevated railway. At Stanford University an electromagnet was borrowed from the physics lecture theater, and only a few hundred dollars were spent, mainly on an oscilloscope.

3

PHYSICS

Many physicists joined this exciting new ®eld of research over the next few years, and NMR was detected for almost all magnetic nuclei in the Periodic Table. NMR was detected in solids, liquids, and gases, in insulators, semiconductors, and metals, in polymers, membranes, and adsorbed and occluded phases, indeed in matter in all its forms. Resonances in liquids and gases were remarkably sharp, those from solids somewhat broader. Exploiting the sharpness of the resonances from ¯uids, extremely precise measurements were made of nuclear magnetic moments; nuclear spins were determined or con®rmed and other nuclear parameters obtained. Since the NMR frequency is directly proportional to the magnetic ®eld strength, NMR presented the physicist with an excellent magnetometer for measurement of unknown magnetic ®elds in the laboratory. The method was readily extended to the Earth's magnetic ®eld, and became useful in civil engineering projects, in archeological surveys for buried magnetic materials, for aerial geomagnetic surveys, for oceanographic surveys, and for measurements in space. The high precision of magnetic ®eld measurement enabled more precise measurements to be made of some of the fundamental constants of physics.5 From the very beginning, the importance of relaxation times in NMR had been realized. At Stanford University the proton NMR was ®rst found in water, and a paramagnetic salt was dissolved in the water in some experiments with a view to shortening the relaxation time.3 At Harvard University the pro-

2 MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW ton NMR was ®rst found in solid paraf®n,4 and great pains were taken to use an extremely low radiofrequency power level to avoid saturation. The relaxation times T1 and T2 were introduced at the outset, and the foundations for understanding these basic quantities were laid in famous pioneering papers by Bloch,10,11 Purcell,12 and their colleagues; to all NMR practitioners interested in relaxation processes this last paper12 is known simply as `BPP' (Bloembergen, Purcell, and Pound). Fundamental studies of relaxation behavior in solids and liquids have given great insight into molecular dynamics in a wide range of materials, including cellular systems and living tissues. In the early years NMR was detected using a steady source of resonant radiofrequency radiation, often referred to as continuous wave (CW) detection. The spectacular discovery of spin echoes by Hahn13 stimulated the widespread use of pulsed NMR. The NMR spectrum could be obtained by Fourier transformation of the free induction decay with tremendous savings of time, and relaxation times could be determined with greater simplicity. In the ®rst decade of its activity, NMR was largely the province of the physicist and the physical chemist. By 1954 some 400 papers had been published, and the ®rst book on NMR5 was able to make reference to all of them. Two later books14,15 devoted to the fundamental physical principles of NMR have become landmarks in the development of the subject.

4 CHEMISTRY Quite early it had been noted by several workers16±18 that the NMR frequency of a given nucleus was to a small degree dependent on the chemical form in which the element was present. The molecular electrons slightly shielded the nucleus and shifted the resonance; the effect was appropriately called the chemical shift. Consequently, a molecule with nuclei in several environments generated a spectrum with several distinct NMR responses, and the subject of NMR spectroscopy had begun.19 Spin multiplets provided ®ner features in the NMR spectrum.20,21 The NMR spectrum was thus a ®ngerprint of the chemical compound and enabled NMR spectroscopy to become one of the chemist's most valuable structural and analytical tools. The NMR spectral lines are narrow and close together; consequently the development of high-resolution NMR spectroscopy called for very uniform magnetic ®elds, uniform to 1 part in 108 or better over a 1 mL sample. Varian Associates in California can be credited fairly with opening up this ®eld to chemists generally in the late 1950s by responding to the challenge posed by small proton chemical shift differences and providing NMR spectrometers with magnets having the requisite uniformity of ®eld. In the quest for improved resolution and sensitivity the proton NMR frequency was steadily advanced from 30 to 60 to 100 MHz, this last frequency corresponding to a magnetic ®eld of 2.3 T. Many other nuclei, especially 13C, 19F, and 31P, were exploited under high-resolution conditions. Pulse methods replaced CW, and a dedicated computer became an integral part of the spectrometer system, controlling the pulse sequences, gathering the data, performing the Fourier transforms, and displaying the spectra.

The pursuit of still higher resolution and sensitivity called for still higher magnetic ®elds, which in turn required a change of magnet technology from the iron electromagnets used hitherto. Using superconducting magnets, proton NMR spectra are now recorded at frequencies up to 1 GHz in ®elds up to 20 T. Designs are in hand at the US National High Magnetic Field Laboratory in Florida for high-resolution proton NMR spectroscopy at 900 MHz and eventually at 1 GHz, which will require a high-resolution superconducting magnet with a ®eld of 23.5 T. The investigation of larger molecules including peptides, proteins, enzymes, and nucleic acids yielded spectra with a forest of ®ne lines dif®cult to resolve and assign. The advent of 2D NMR spectroscopy,22 spreading out the spectra in two dimensions with detailed connectivity, gave great spectral clarity and renewed impetus. This has been extended to three dimensions and beyond. Nowadays no chemistry research laboratory is properly equipped without its high-resolution NMR spectrometer. The widespread use of NMR spectrometers in chemical laboratories has led to the growth of a substantial NMR instrument industry. The principles and practice of NMR are now a feature of chemistry courses at quite elementary levels. Most science students and medical students today will have heard of NMR before going to university. The use of high-resolution NMR spectroscopy has spread from chemistry to biochemistry, microbiology, pharmacy, agricultural chemistry, polymer science, and, as we shall see, biology and medicine. We note in passing that specimens investigated by chemists and physicists are usually homogeneous and rather less than 1 ml in volume. Consequently, the bore of a superconducting NMR magnet for such work is usually about 5 cm in diameter. This provides just suf®cient access for the specimen, its surrounding radiofrequency probe, and other accessories.

5

BIOLOGY

The earliest biological NMR experiments take us back to the inventors of the subject. Soon after the ®rst successful NMR experiments at Stanford University,3 Bloch obtained a strong proton NMR signal when he inserted his ®nger in the radiofrequency coil of his spectrometer. In 1948, Purcell and Ramsey in turn inserted their heads into the 2 T ®eld of the Harvard University cyclotron; around their heads was a coil connected to a powerful radiofrequency generator tuned to the proton NMR frequency. The only sensation recorded was that of EMFs generated in the metal ®llings of their teeth as their heads were moved into and out of the magnet and detected by their tongues. Fifty years later there is no evidence of any damage arising from this magnetic adventure. Occasional NMR experiments were subsequently reported on materials of biological interest.23,24 A pioneering series of investigations was carried out by Odeblad and his colleagues in Stockholm on tissues and ¯uids from animals and humans starting in 1955 and continuing over the next decade. These studies included blood cells,25,26 cervical mucus,27 tissues and ¯uids of the eye28 and the eye lens,29 saliva,30 and muscle.31 However, substantial advances in the application of NMR in biology awaited the development of high-®eld, high-resolution, Fourier transform spectrometers in the late 1960s. As men-

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

tioned in the previous section, these technical advances enabled high-resolution NMR spectra in one and two dimensions to be obtained, and yielded structural and dynamic information of importance in biochemistry and molecular biology. At the same time, high-resolution NMR spectroscopy began to be applied to living systems. First Moon and Richards32 reported high-resolution 31P NMR studies of intact red blood cells in which they could assign lines to individual metabolites. This was followed the next year by recordings of 31P spectra from an intact freshly excised muscle from the leg of a rat by Hoult et al.33 It was found possible to maintain muscles in good physiological condition in the spectrometer and to record the effects of electrical stimulation. The 5 cm diameter bore of a conventional NMR superconducting magnet was nevertheless a severe restriction. Soon magnets with a 10 cm diameter bore were provided, and a wide range of studies on perfused animal hearts, kidneys, livers, and other organs were pursued, using mainly 31P NMR but also 1H and 13C NMR spectroscopy. It was then a natural step to examine whole intact living organisms from bacteria to mice, rats, and rabbits. With animals it was of course important to know from what anatomical part or organ the NMR spectrum originates. In some cases this could be achieved by winding the radiofrequency coil around the part concerned; in other cases a radiofrequency coil was placed on the surface of the animal to record the NMR signal from the region immediately below. From animals it was just a further step to the in vivo NMR spectroscopy of humans and its applications in medicine. Spatial localization continued to be an important challenge, and other more sophisticated methods were introduced such as ISIS and spectroscopic imaging. However, in this article we shall not follow this trail further. Instead we follow the parallel trail of MRI in the next section. Further information on magnetic

Figure 1

The ®rst 2D NMR image (Lauterbur37)

3

resonance spectroscopy (MRS) may be found in books by Gadian,34 Andrew et al.,35 Gillies36 and in other articles in these volumes.

6

MAGNETIC RESONANCE IMAGING

In contrast to the steady onward march of NMR spectroscopy from physics through chemistry and biology to medicine, NMR imaging represents a distinctly different approach which appeared rather suddenly on the scene in 1973. In 1973, Lauterbur37 published the ®rst NMR 2D image of a heterogeneous structured object: two tubes of water. He pointed out the simple fact that if a ®eld gradient is applied to a structured object each nucleus responds with its own NMR frequency determined by its position; the NMR spectrum is the 1D projection of nuclear density along the gradient direction. Applying the gradient in a series of directions and so obtaining a series of projections, he devised an algorithm to generate a 2D proton NMR image of the object from its projections, in just the same way as it is done in X ray computed tomography (CT) scanning. This ®rst image is reproduced in Figure 1. The impact of this ®rst image and of the work that followed has been tremendous. Within a decade of its publication, manufacturers worldwide were producing NMR scanners that generated high quality images of all parts of the human body, challenging traditional modalities of radiological imaging in clinical practice. After two decades all major hospitals are now equipped with NMR whole body scannersÐover 10000 systems worldwideÐand the total expenditure for this equipment overshadows that for all previous uses of NMR. Moreover, it

Nuclear induction apparatus and display

Figure 2 An early schematic diagram for human NMR imaging (Damadian45,46)

Absorption (arb. scale)

A magnetic resonance image is a 2D representation of the NMR signal from the resonant nuclei in a particular thin slice of the object of interest. We may imagine the slice divided up into an array of n  n elements. If we are able to measure the NMR signal from all n2 elements in the slice, we can construct a picture or image consisting of these n  n picture elements (pixels). The larger the value of n, the ®ner is the detail in the

3

3 layers, Gz = 0

r (z) (arb. scale)

(a)

2 1 0 0

3

1 n (unscaled) (kHz)

2

3 layers, Gz = 0.77 G cm–1 (b)

2 1 0 0

r (z) (arb. scale)

has brought NMR to the notice of the man and woman in the street, and NMR has entered our everyday vocabulary. In the interests of simplicity and the happiness of patients the adjective `nuclear' has been dropped in the context of medicine, and the subject is now just `magnetic resonance imaging' or MRI. Other names, such as Zeugmatography, introduced by Lauterbur,37 have now largely dropped out of use. Just as in NMR itself there was a prehistory before the ®nal discovery (see above), so in NMR imaging there was a prehistory in the use of magnetic ®eld gradients. NMR 1D projections in a magnetic ®eld gradient were investigated quite early in 1951±52 with simple glass and liquid structures by Gabillard38±40 in France. The dynamic NMR response is given by the Fourier transform of the distribution function of nuclei in the gradient applied to the structure, and it follows conversely that a 1D projection is obtained from the Fourier transform of the free induction decay. Magnetic ®eld gradients are a basic feature in the study of molecular diffusion in liquids by the NMR spin echo method. The diffusion of molecules from their original locations in a liquid sample in a ®eld gradient reduces their contribution to the echo, and this enables the diffusion constant of the liquid to be determined.13,41 In 1956, Walters and Fairbank42 studied the distribution of 3 He in liquid 3He±4He solutions in three vertically arranged connected containers by applying a ®eld gradient from top to bottom. A separate 3He NMR signal was recorded from each container. In this way phase separations below 0.8 K were determined. In this experiment the ®eld gradient was being used to provide a 1D NMR image of the 3He density in the containers. Magnetic ®eld gradients are a feature of NMR methods of information storage.43,44 In one method,44 information is stored in a 3D cellular structure and refreshed frequently to avoid loss by relaxation. The information is then retrieved by application of gradients which enable the speci®c location of the desired piece of information in the store to be excited. Reading the information from such an information store was in fact an early NMR imaging procedure, which could be used in one, two, or three dimensions. In 1972, Damadian45,46 ®led a prophetic patent in which he proposed without detail a method for scanning the human body (Figure 2) by NMR and detecting pathological deviations, for example cancer, through differences in the relaxation times T1 and T2. This proposal was based on his pioneering observations47 that the relaxation times T1 and T2 were signi®cantly higher in cancerous rat tissue than in the corresponding normal tissue. In 1973, Mans®eld and Grannell48 introduced the concept of NMR diffraction in solids. Multiple pulse line narrowing sequences sharpened the NMR response, and a magnetic ®eld gradient provided spatial resolution. A model 1D lattice of several plates of camphor was used to generate the 1D images shown in Figure 3. While technical dif®culties might prevent application to crystallographic structures, it was suggested that the method would be useful in the study of biophysical systems with regular or approximately regular macroscopic structures such as cell membranes and ®lamentary or ®brous structures. In 1975 in an ampli®cation and extension of this work49 the authors discuss the application of these ideas to more general nonperiodic structures, including microstructures.

,

,

,


4 MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

5

z (mm)

10

15

5 layers, Gz = 0.77 G cm–1

3 2

(c)

1 0 0

5

z (mm)

10

15

Figure 3 1D NMR images of (a) one, (b) three, and (c) ®ve layers of camphor (Mans®eld and Grannell48)

5

, , , ,, , ,,, ,,

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

Sequential point imaging

Figure 4

Line scanning

Planar imaging

3D imaging

A classi®cation of imaging methods35

image. However, if we were to double the value of n in order to improve the resolution, we should have to measure four times as many pixels, each containing only a quarter the number of nuclei from which to generate their NMR signals. In practice, n is usually 256 or 512; computers are happy with numbers that are powers of 2. A ®eld gradient is an essentially 1D probe, and many different methods have been advanced for manipulating the gradient to secure 2D or 3D information. A basic classi®cation of methods is illustrated in Figure 4. First we might arrange to isolate a small volume and collect the NMR signals from it. This elemental volume may then be traversed through a plane in the object of interest, enabling us to build a magnetic resonance image of a plane, point by point. This is the sequential point method, and it is inevitably slow. Next we might arrange to gather the NMR signal from a whole line or column of n elements simultaneously by a Fourier transform procedure, which is n times faster. If n is 256, this represents a substantial gain. This is line scanning. We can then scan the line through the plane of interest to build an image. Then we might gather the NMR signal from a whole plane of n2 pixels simultaneously and process the data in such a way that we get an image of the plane. This is planar imaging. It may not necessarily be faster than line scanning, but since NMR signals are being gathered from the whole plane all the time it will certainly give a much better quality image. Finally, we might gather the NMR signal from the whole 3D object simultaneously and process the data in such a way as to characterize all n  n  n = n3 volume elements (voxels). This is 3D imaging. An image of any desired slice can then be displayed. The ®rst two methods (sequential point and line scanning) were important in the early days of MRI,50±53 but have now been superseded by the third (planar imaging). 3D imaging is available on many instruments but takes longer and gives more information than is generally needed. Planar imaging may be achieved by the projection-reconstruction method37 or more usually by the 2D Fourier imaging method devised by Kumar, Welti, and Ernst,54 which is readily extensible to three dimensions. A careful assessment of the relative performance of the various imaging methods was carried out by Brunner and Ernst.55 In order to do planar imaging it is of course necessary to de®ne the slice to be imaged. This is usually achieved by the selective excitation method of Garroway, Grannell, and Mans®eld.56 Implementation of the 2D Fourier imaging method usually makes use of the spin warp technique developed by Edelstein, Hutchison, Johnson, and Redpath.57 The ®rst MRI experiments were carried out in physics and chemistry laboratories using modi®ed NMR spectrometers with

magnets which had limited access. The imaging subjects were therefore necessarily small. Over the period 1974±78, NMR images were published from the fruit and vegetable kingdom: pecan,58 onion,51 lemon59,60 (see Figure 5), pepper,61 and okra.62 Images were also published from the animal kingdom: clam,58 chicken wing,51 turkey leg,63 mouse,58,64±66 rat,67±69 and rabbit70 (see Figure 6). These pictures showed that geometrically faithful NMR images could be obtained of the interior of biological structures, with contrast arising from differences in tissue density, mobility and relaxation. The stage was set for applications of MRI to the human body.

Figure 5 (a) Thin transverse section proton NMR image of an intact lemon and (b) a photograph of the actual section cut subsequently (Andrew et al.59 and Andrew60)

6 MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

Figure 6 Thin coronal section proton NMR image through the orbits of the head of a rabbit (Hinshaw et al.70)

7 HUMAN MRI The ®rst human NMR image, reported in 1976 by Mans®eld and Maudsley,71,72 was of a live human ®nger (Figure 7). The ®nger could be inserted into the 5 cm gap of a conventional iron NMR electromagnet. Using a magnet with a 12 cm gap, NMR images were obtained of a live human hand,67 wrist73,74 (Figure 8), and forearm75 by Hinshaw, Andrew, Bottomley, and their co-workers. The anatomical detail and tissue contrast

Figure 8 Thin proton NMR image of a live human wrist (bottom). For comparison, a cross section of an actual wrist is also shown (top) (Hinshaw et al.73,74)

Figure 7 Cross-section proton NMR image of a live human ®nger (Mans®eld and Maudsley71,72)

Figure 9 Whole body NMR imaging magnet at Nottingham University, 1978

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

7

Figure 10 Commercial whole body NMR imaging magnet at the University of Florida, 1983 Figure 13 Transverse NMR image of a normal human abdomen (Mans®eld et al.79)

Figure 11 The ®rst superconducting whole body magnet for regular clinical use, installed at Hammersmith Hospital, London, 1981

were good and gave optimism for successful NMR imaging of the whole human body, head, chest, and abdomen. An essential requirement of human MRI was a magnet capable of accepting the whole human body and generating a magnetic ®eld uniform and stable to at least 1 part in 105 over a central sphere about 30 cm in diameter. Early whole body magnets were air cored resistive solenoids producing ®elds up to 0.15 T, with an aperture of about 80 cm in diameter. An

Figure 12 Transverse NMR image of a normal human thorax (Damadian et al.52,77)

example used in the University of Nottingham in 1978 is shown in Figure 9; a commercial version in clinical use in 1983 in the University of Florida is shown in Figure 10. A similar magnet but with a vertical axis was used in Aberdeen.66 For higher ®eld strengths superconducting solenoids are generally used, although at least one system uses a large permanent magnet and another uses an iron electromagnet. The development of high-®eld whole body superconducting magnets has played a crucial role in opening up clinical MRI; Oxford Instruments in the UK in particular were pioneers in this development. The ®rst superconducting whole body magnet for clinical use was installed at Hammersmith Hospital, London in 1981 (Figure 11). Biological tissues are moderate electrical conductors, and we might therefore expect some attenuation of the electromagnetic radiation at the NMR frequency as it penetrates the body. Calculations by Bottomley and Andrew76 suggested that signi®cant attenuation and phase distortion might be encountered above 40 MHz, corresponding to 1 T for proton NMR. Actually the calculations were based on a homogeneous cylinder 20 cm in radius, which was somewhat pessimistic since the human body contains many cavities. In practice little attenuation is seen at 2 T (85 MHz for protons), but has been noticed at 4 T. The ®rst human whole body MR image was an image of the human thorax by Damadian, Goldsmith, and Minkoff52,77 in 1977 (Figure 12). This was closely followed in 1978 by the ®rst magnetic resonance image of the human head, by Clow and Young,78 and the ®rst image of the human abdomen, by Mans®eld, Pykett, Morris, and Coupland79 (Figure 13). In computerized X-ray tomography the slice imaged was usually transverse. Such a restriction does not apply in MRI, and this was clearly demonstrated with the publication in 1980 of the ®rst coronal and sagittal images of the human head by Holland, Hawkes, and Moore80 (Figure 14). These early human images were all of normal healthy volunteers. Soon afterwards in 1980 the ®rst images of human pathology began to appear. Hawkes, Holland, Moore, and Worthington81 demonstrated a wide range of intracranial pathological behavior by MRI. This included examples of

8 MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

Figure 14 Transverse and ®rst coronal and sagittal NMR images of a normal human head (Holland et al.80)

hydrocephalus, a variety of tumors, aneurysms, arteriovenous malformations, and chronic sinus infection. Damadian46 demonstrated a number of cases of carcinoma of the chest. We should notice that when imaging the human body, the proton is by far the most commonly used nucleus. After all, hydrogen is the most abundant chemical element in the body, 1 H is isotopically almost 100% abundant, and it has the highest magnetic moment among stable nuclei. The next most abundant chemical element, oxygen, has no convenient magnetic isotope. For carbon, the isotope 13C is only 1.1% abundant and it has a low magnetic moment. The principal isotope of nitrogen is 14N, which has a low magnetic moment and other complications, while 15N has a low abundance. Phosphorus is an element of considerable interest on account of the important metabolites that contain it. The isotope 31P is 100% abundant and it has a reasonable magnetic moment. However, concentrations of phosphorus are very low in living systems. So summarizing, protons (hydrogen nuclei) are by far the best nuclei for human MRI. Nevertheless there has been some MRI interest in 2D, 7Li, 14N, 19F, 23Na, and 31P. Some early work has also been done on MRI of unpaired electrons.82 Since the ®rst successful human magnetic resonance images of healthy volunteers and patients there has been a rapid devel-

opment of technique and a great improvement in the quality of the images. By 1983 the ®rst commercial MRI systems were being installed in hospitals, yielding images in 4 min with 1 mm resolution. No longer do physicists and chemists and others make their own MRI systems. Images of the author taken in March 1984 using the 0.15 T magnet of Figure 10 are shown in Figure 15, illustrating the performance reached at that time. With the availability of higher ®elds to 2 T images have improved steadily since then. Magnets with ®elds of 3, 4, and 5 T have been built for human imaging but are not yet in general use. Over the past decade, MRI has become an accepted modality of clinical radiology. We may note some of the advantages which recommend it. First it does not use ionizing radiation and is intrinsically safer than modalities using X-rays,

rays, positrons, or heavy ions. In contrast with ultrasound the radiation penetrates bony structures. Besides giving morphological information, MRI gives additional diagnostic insights and improved contrast through relaxation parameters, susceptibility, and diffusion, not available from other modalities. In contrast with CT, MRI provides direct images in transverse, coronal, or sagittal slices, or slices of arbitrary orientation. Flow can be measured. Contrast may be improved by injection of paramagnetic agents. Conventional magnetic resonance images take several minutes to acquire whereas CT images take several seconds only. However, it is possible to acquire simultaneously magnetic resonance images from typically 10 parallel slices and four T2weighted images of each, a total of 40 images in the time of one, making an average acquisition time of only a few seconds. Further, using new techniques such as FLASH83 a single magnetic resonance image can be acquired in a few seconds. Still faster magnetic resonance images may be obtained using a sequence proposed by Mans®eld84 back in 1977, and demonstrated the following year,62 called echo planar imaging. A complete image could be obtained in less than 0.1 s, enabling real time movie magnetic resonance images to be recorded.85

8

CONCLUDING REMARKS

The rapid expansion of interest in NMR of living systems led to the organization of special conferences to discuss these topics. In March 1979 the Royal Society held a Discussion Meeting in London entitled `Nuclear magnetic resonance in intact biological systems', which reviewed achievements to date in both MRI and MRS. The papers given at this meeting were published as a special issue of Philosophical Transactions of the Royal Society, (1980, Vol. 289, no. 1037), and constitute a historic landmark in the development of the subject. In October 1981 another signi®cant meeting devoted solely to MRI was held in Winston-Salem, North Carolina. With the widespread deployment of commercial MRI systems in all major hospitals it was clear that annual meetings were needed, and the Society of Magnetic Resonance in Medicine was formed, holding its ®rst meeting in Boston, Massachusetts, in August 1982 with an attendance of 800 participants. Since then the Society has met annually in August either in North America or in Europe; by 1993 attendance had trebled. In 1983 the Society started a new journal Magnetic Resonance in Medicine, edited by E. Ray-

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW

9

Figure 15 Magnetic resonance images of the author taken in March 1984, using the magnet of Figure 8: (a) head, transverse; (b) head, sagittal; (c) spine, sagittal; (d) abdomen, transverse

mond Andrew, and since 1991 by Felix Wehrli. Other societies and journals began at this time in various parts of the world. This historical overview covers the period up to 1983. The reader can now go on to embrace the multitude of detailed applications of MRI to every branch and corner of medicine, to microimaging, spectroscopic imaging, functional brain imaging, interventional MRI, to the solid state, and much more. Many of these developments are covered in recent books such as that by Morris86 for the mathematically inclined reader or that by Andrew et al.,35 which is addressed particularly to physicians; it has no mathematics and covers both MRI and MRS. More detailed information on very many topics relating to MRI will found elsewhere in these volumes.

9 RELATED ARTICLES Image Formation Methods; MRI in Clinical Medicine; Outcome and Ef®cacy±Analysis of Healthcare Models.

10

REFERENCES

1. E. R. Andrew, Br. Med. Bull., 1984, 40, 115. 2. E. R. Andrew, in `Magnetic Resonance Imaging', ed. C. L. Partain, A. E. James, F. D. Rollo, and R. R. Price, Saunders, Philadelphia, 1988, Chap. 1, p. 3. 3. F. Bloch, W. W. Hansen, and M. E. Packard, Phys. Rev., 1946, 69, 127. 4. E. M. Purcell, H. C. Torrey, and R. V. Pound, Phys. Rev., 1946, 69, 37. 5. E. R. Andrew, `Nuclear Magnetic Resonance', Cambridge University Press, Cambridge, 1955, Chaps. 2 and 4. 6. C. J. Gorter, Physica, 1936, 3, 995. 7. C. J. Gorter and L. F. J. Broer, Physica, 1942, 9, 591. 8. C. J. Gorter, Physica, 1951, 17 169. 9. I. I. Rabi, S. Millman, P. Kusch, and J. R. Zacharias, Phys. Rev., 1939, 55, 526. 10. F. Bloch, Phys. Rev., 1946, 70, 460. 11. F. Bloch, W. W. Hansen, and M. E. Packard, Phys. Rev., 1946, 70, 474. 12. N. Bloembergen, E. M. Purcell, and R. V. Pound, Phys. Rev. 1948, 73, 679.

10 MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW 13. E. L. Hahn, Phys. Rev., 1950, 80, 580. 14. A. Abragam, `The Principles of Nuclear Magnetism', Clarendon Press, Oxford, 1961. 15. C. P. Slichter, `Principles of Magnetic Resonance', 3rd edn., Springer-Verlag, Berlin, 1990. 16. W. D. Knight, Phys. Rev., 1949, 76, 1259. 17. W. C. Dickinson, Phys. Rev., 1950, 77, 736. 18. W. G. Proctor and F. C. Yu, Phys. Rev., 1950, 77, 717. 19. J. T. Arnold, S. S. Dharmatti, and M. E. Packard, J. Chem. Phys., 1951, 19, 507. 20. H. S. Gutowsky, D. W. McCall, and C. P. Slichter, Phys. Rev., 1951, 84, 589. 21. E. L. Hahn and D. E. Maxwell, Phys. Rev., 1951, 84, 1246. 22. W. P. Aue, E. Bartoldi, and R. R. Ernst, J. Chem. Phys., 1976, 64, 2229. 23. T. M. Shaw and R. L. Elsken, J. Chem. Phys., 1950, 18, 1113. 24. B. Jacobsen, W. A. Anderson, and J. T. Arnold, Nature, 1954, 173, 772. 25. E. Odeblad and G. Lindstrom, Acta Radiol., 1955, 43, 469. 26. E. Odeblad, B. N. Bhar, and G. Lindstrom, Acta Biochem. Biophys., 1956, 63, 221. 27. E. Odeblad and V. Bryhu, Acta Radiol., 1957, 47, 315. 28. A. Huggert and E. Odeblad, Acta Radiol., 1959, 51, 385. 29. A. Huggert and E. Odeblad, Acta Opthalmol., 1959, 37, 26. 30. E. Odeblad and E. Soremark, Acta Odontol. Scand., 1962, 20, 23. 31. E. Odeblad and A. Ingelman-Sundberg, Acta Obstet. Gynecol. Scand., 1965, 44, 117. 32. R. B. Moon and J. H. Richards, J. Biol. Chem., 1973, 248, 7276. 33. D. I. Hoult, S. J. W. Busby, D. G. Gadian, G. K. Radda, R. E. Richards, and P. J. Seeley, Nature, 1974, 252, 285. 34. D. G. Gadian, `Nuclear Magnetic Resonance and Its Application to Living Systems', Clarendon Press, Oxford, 1980. 35. E. R. Andrew, G. Bydder, J. Grif®ths, R. Iles, and P. Styles, `Clinical Magnetic Resonance: Imaging and Spectroscopy', Wiley, Chichester, 1990. 36. R. J. Gillies, `NMR in Physiological and Biomedicine', Academic Press, San Diego, 1994. 37. P. C. Lauterbur, Nature, 1973, 242, 190. 38. R. Gabillard, C. R. Acad. Sci. Paris, 1951, 232, 1551. 39. R. Gabillard, Phys. Rev., 1952, 85, 694. 40. E. R. Andrew, `Nuclear Magnetic Resonance', Cambridge University Press, Cambridge, 1955, p. 134. 41. H. Y. Carr and E. M. Purcell, Phys. Rev., 1954, 96, 630. 42. G. K. Walters and W. M. Fairbank, Phys. Rev., 1956, 103, 262. 43. A. G. Anderson, R. L. Garvin, E. L. Hahn, J. W. Horton, G. L. Tucker, and R. M. Walker, Phys. Rev., 1955, 26, 1324. 44. E. R. Andrew, A. Finney, and P. Mans®eld, Royal Radar Establishment Research Report, 1970, PD/24/026/AT. 45. R. Damadian, US Patent 3 789 832 ®led 17 March 1972. 46. R. Damadian, Philos. Trans. R. Soc. London, Ser. B, 1980, 289, 489. 47. R. Damadian, Science, 1971, 171, 1151. 48. P. Mans®eld and P. K. Grannell, J. Phys. C: Solid State Phys., 1973, 6, L422. 49. P. Mans®eld and P. K. Grannell, Phys. Rev. B, 1975, 12, 3618. 50. W. S. Hinshaw, Phys. Lett., 1974, 48A, 87. 51. W. S. Hinshaw, J. Appl. Phys., 1976, 47, 3709. 52. R. Damadian, M. Goldsmith, and L. Minkoff, Physiol. Chem. Phys., 1977, 9, 97. 53. P. Mans®eld, A. A. Maudsley, and T. Baines, J. Phys. E: Sci. Instrum., 1976, 9, 271. 54. A. Kumar, D. Welti, and R. R. Ernst, J. Magn. Reson., 1975, 18, 69. 55. P. Brunner and R. R. Ernst, J. Magn. Reson., 1979, 33, 83. 56. A. N. Garroway, P. K. Grannell, and P. Mans®eld, J. Phys. C: Solid State Phys., 1974, 7, L457.

57. W. A. Edelstein, J. M. S. Hutchison, G. Johnson, and T. W. Redpath, Phys. Med. Biol., 1980, 25, 751. 58. P. C. Lauterbur, Pure Appl. Chem., 1974, 40, 149. 59. E. R. Andrew, P. A. Bottomley, W. S. Hinshaw, G. N. Holland, and W. S. Moore, Nature, 1977, 270, No. 5638 front cover. 60. E. R. Andrew, Philos. Trans. R. Soc. London, Ser. B, 1980, 289, 471. 61. P. C. Lauterbur, in `NMR in Biology', ed. R. A. Dwek, Academic Press, Orlando, 1977. 62. P. Mans®eld and I. L. Pykett, J. Magn. Reson., 1978, 29, 355. 63. P. Mans®eld and A. A. Maudsley, Phys. Med. Biol., 1976, 21, 847. 64. J. M. S. Hutchinson, J. R. Mallard, and C. C. Goll, in `Magnetic Resonance and Related Phenomena', ed. P. S. Allen, E. R. Andrew, and C. A. Bates, North-Holland, Amsterdam, 1975, p. 283. 65. R. Damadian, L. Minkoff, M. Goldsmith, M. Stanford, and J. Koutcher, Science, 1976, 194, 1430. 66. J. Mallard, J. M. S. Hutchison, W. A. Edelstein C. R. Ling, M. A. Foster, and G. Johnson, Philos. Trans. R. Soc. London, Ser. B, 1980, 289, 519. 67. E. R. Andrew, P. A. Bottomley, W. S. Hinshaw, G. N. Holland, W. S. Moore, and C. Simaroj, Phys. Med. Biol., 1977, 22, 971, errata 1291. 68. L. E. Crooks, T. P. Grover, L. Kaufman, and J. R. Singer, Invest. Radiol., 1978, 13, 63. 69. I. L. Pykett and P. Mans®eld, Phys. Med. Biol., 1978, 23, 961. 70. W. S. Hinshaw, E. R. Andrew, P. A. Bottomley, G. N. Holland, and W. S. Moore, Br. J. Radiol., 1978, 51, 273. 71. P. Mans®eld and A. A. Maudsley, Proc. 19th Congress Ampere, Heidelberg, 1976, 247. 72. P. Mans®eld and A. A. Maudsley, Br. J. Radiol., 1977, 50, 188. 73. W. S. Hinshaw, P. A. Bottomley, and G. N. Holland, Nature, 1977, 270, 722. 74. W. S. Hinshaw, E. R. Andrew, P. A. Bottomley, G. N. Holland, W. S. Moore, and B. S. Worthington, Neuroradiology, 1978, 16, 607. 75. W. S. Hinshaw, E. R. Andrew, P. A. Bottomley, G. N. Holland, W. S. Moore, and B. S. Worthington, Br. J. Radiol., 1979, 52, 36. 76. P. A. Bottomley and E. R. Andrew, Phys. Med. Biol., 1978, 23, 630. 77. R. Damadian, L. Minkoff, M. Goldsmith, and J. A. Koutcher, Naturwissenschaften, 1978, 65, 250. 78. H. Clow and I. R. Young, New Sci., 1978, 80, 588. 79. P. Mans®eld, I. L. Pykett, and P. G. Morris, Br. J. Radiol., 1978, 51, 921. 80. G. N. Holland, R. C. Hawkes, and W. S. Moore, J. Comput. Assist. Tomogr., 1980, 4, 429. 81. R. C. Hawkes, G. N. Holland, W. S. Moore, and B. S. Worthington, J. Comput. Assist. Tomogr., 1980, 4, 577. 82. M. J. R. Hoch and A. R. Day, Solid State Commun., 1979, 30, 211. 83. A. Haase, J. Frahm, D. Matthaei, W. Hanicke, and J. Merboldt, J. Magn. Reson., 1986, 67, 258. 84. P. Mans®eld, J. Phys. C: Solid State Phys., 1977, 10, L55. 85. R. Rzedzian, B. Chapman, P. Mans®eld, R. E. Coupland, M. Doyle, A. Chrispin, D. Guilfoyle, and P. Small, Lancet, 1983, ii, 1281. 86. P. G. Morris, `Nuclear Magnetic Resonance Imaging in Medicine and Biology', Clarendon Press, Oxford, 1986.

Biographical Sketch E. Raymond Andrew. b 1921. B.A. (Physics), M.A. Ph.D., Sc.D., Cambridge University, UK, Hon. D.Sc. Turku, Poznan, Leipzig and Wales, FRSE and FRS. Postdoctoral Fellow, Harvard with E. M. Purcell. Lecturer, St. Andrews University, Scotland. Professor of

MAGNETIC RESONANCE IMAGING: A HISTORICAL OVERVIEW Physics, University of Wales, Bangor. Professor and Dean of Science, Nottingham University, England. Currently Research Professor, University of Florida, USA. President, Groupement AMPERE, 1974±80. President, ISMAR, 1983±86. Editor, Magnetic Resonance in Medicine,

11

1983±91. Four books: Nuclear Magnetic Resonance, Magnetic Resonance and Related Phenomena (jtly), Clinical Magnetic Resonance (jtly), NMR in High Magnetic Fields (jtly), 400 publications. Research interests: NMR in condensed matter, NMR imaging, MRI in medicine.

MRI AT MIDFIELD STRENGTH

MRI at Mid®eld Strength Jane M. Hawnaur and Ian Isherwood Department of Diagnostic Radiology, University of Manchester, UK

1 INTRODUCTION Whole body magnetic resonance imaging (MRI) was initially performed in the early 1980s using magnetic resonance (MR) scanners based on resistive electromagnets. In the decade that followed, superconducting MR scanners became the most popular scanners for both research and clinical purposes. The superior signal-to-noise ratio (S/N), ®eld homogeneity and ®eld stability, particularly for high ®eld (1.5 T) systems, proved to be major factors in this selective process. An initial emphasis on brain and spine imaging during the early development of MRI served to reinforce the concept that a high ®eld strength was necessary for good quality images, a view encouraged by some of the systems manufacturers. Since then, manufacturers have expended signi®cant research and development effort on their mid®eld range of systems, many of which now offer the same capabilities (apart from spectroscopy) as higher ®eld MRI systems. As diagnostic applications for MRI have expanded outside the central nervous system into the musculoskeletal and cardiovascular systems, abdomen, and pelvis, some of the disadvantages of high-®eld MRI have become more apparent. Chemical shift, susceptibility effects and movement artifacts cause greater interference during body imaging than cranial or spinal imaging. In addition, the practical and economic dif®culties associated with the installation and running of high-®eld systems in some hospital settings have prompted reassessment of the value of low and mid®eld systems as a necessary compromise between S/N, versatility and cost. Of 719 MRI systems installed in Japan by 1990, 72 were low-®eld (less than 0.2 T), 488 were mid®eld (0.2±1.0 T), and 159 were 1.5 T superconducting systems.1 A similar trend away from high-®eld systems was apparent worldwide; sales of mid®eld MRI scanners worldwide exceed sales of high-®eld systems by a factor of 2± 3, and the single most popular model in Europe is a 1.0 T superconducting MR scanner. 2 TYPES OF SCANNER AT MIDFIELD (0.2±1.0 T) Most of the components of MRI scanners for use at mid and high ®eld are identical apart from the magnet itself. There are three types of magnet design capable of operating in the mid®eld range.

2.1 Superconducting Magnets 2.1.1

Design

MRI scanners using a superconducting magnet offer a wide range of ®eld strengths from 0.2 T upwards and are the most

1

popular choice for clinical whole body systems operating at 0.5 or 1.0 T. The ®eld-generating coil is constructed from a superconducting alloy of niobium/titanium (NbTi) ®laments in a copper matrix through which an electric current is passed. To remove electrical resistance, the system is cooled to 4 K. Once the superconducting state has been achieved, minimal electrical energy is required to maintain the magnetic ®eld. The conducting NbTi wires are suspended in liquid helium, which evaporates as it absorbs heat conducted away from the wires. In older systems, the cooling system required several layers of vacuum insulation, liquid nitrogen and insulating materials around the helium cryostat, and was relatively inef®cient. The helium boil-off rate was of the order of 1 L hÿ1 and the inconvenience and cost of weekly or monthly cryogen re®lls was a major disadvantage of superconducting systems. In modern mid®eld scanners, however, ef®cient refrigeration is achieved without liquid nitrogen, and systems are designed to minimize cryogen costs. Helium boil-off rates vary between manufacturers, but are of the order of 0.04±0.2 L hÿ1. Re®ll rates have consequently fallen, with recently introduced models offering re®ll intervals of up to 7 years. 2.1.2

Field Homogeneity and Shielding

Superconducting magnets produce a stable and homogeneous magnetic ®eld; their solenoidal design produces a main magnetic ®eld B0 oriented along the length of the patient. Homogeneity is of the order of ‹2±5 ppm on a 40 cm diameter spherical volume (DSV). Early designs reduced the fringe ®eld passively by means of bolt-on steel shielding around the magnet, resulting in a combined weight of up to 40 ton. Active shielding, with an integral shielding coil built around the primary coil, is more ef®cient and less costly, reducing the weight of the scanner to around 5 ton for a 1.0 T magnet. Actively shielded magnets are thus more compact and lightweight, but there is a small reduction in magnetic ®eld homogeneity and less protection from variation in external rf ®elds. Typical 5 G contours for a 0.5 T passively shielded superconducting MR system are 3.7 m axial by 2.5 m radial. Many manufacturers offer compact mid®eld systems with integral passive shielding, also resulting in a small fringe ®eld. 2.1.3

Speci®cations

Many mid®eld superconducting magnets have been adapted from high-®eld systems in the same manufacturer's range, and offer similar facilities (Table 1). 2.1.4

Installation and Running Costs

A superconducting MR scanner in the mid®eld range weighs about 5 ton, including the enclosure, helium and gradient coils. This compares with a weight of 5±15 ton for a similar system operating at 1.5 T. Most current mid®eld MR scanners are designed for installation in an existing room within the imaging department. Minimum space requirements are typically 6.0 m5.0 m with a 2.65 m ceiling height. Additional space is required for computers and console area, generally less than 10 m2 each. The capital cost of superconducting MR equipment is relatively high and installation may involve the expense of fringe ®eld shielding, rf shielding, and ventilation to enable effective operation of the system.

2 MRI AT MIDFIELD STRENGTH Table 1

Selected Speci®cations for MR Systems from the Same Manufacturer Operating at 0.5 T and 1.5 T (Phillips) 1998

Magnet homogeneity Weight (similar dimensions) (kg) Fringe ®eld (5 G) Helium consumption (L hÿ1) Gradient: Strength (mT mÿ1) Slew rate (mT msÿ1) Capabilities Spectroscopy Diffusion imaging Functional brain MRI Angiography Receiver coils Surface coils Flexible torso surface coil Multiple phased-array torso coil Endoluminal coil Siting: Magnet room Technical room

0.5 T

1.5 T

‹1 ppm (40 cm DSV) 2600 2.9 m  2.1 m 0.004

‹1 ppm (40 cm DSV) 3100 3.9 m  2.5 m 0.004

15 17

15 (standad), 23 (upgrade) 17 (standard), 53 or 105 (upgrade)

No Yes No Yes

Yes Yes Yes Yes

Yes Yes No Yes

Yes No Yes Yes

6 m5 m 2 m3.2 m

6 m5 m 2 m3.2 m

DSV, diameter spherical volume

2.2 Resistive Magnets 2.2.1

Design

The static ®eld from a resistive magnet is produced by electric current passing through coils or sheets of resistive wire wound around a central iron core. For example, a resistive magnet with a bore of about 1 m and a ®eld strength of 0.15 T requires approximately 1500 turns of wire through which around 200 A of current is passed. Resistive magnets usually have a solenoidal con®guration with B0 oriented along the long axis of the patient, but their relatively light weight enables them to be built with a vertical con®guration which facilitates the use of rf receiver coils to generate a ®eld perpendicular to B0. This design also increases access to the magnet bore for claustrophobic patients, interventional procedures, or for supervision of high-dependency patients. Magnetic ®eld strength is limited by power consumption, the amount of heat produced, and the weight of magnet required to support a large number of turns of wire. Power is dissipated in the coil windings as heat is removed by air/water cooling, necessitating the delivery of large volumes of deionized water. A stable, continuous supply of electricity is required, and power consumption is high compared with permanent or superconducting magnets. A typical power requirement for a resistive magnet is of the order of 50 kW for a 0.15 T ®eld, the power requirement increasing with increasing ®eld strength. The ®elds of commercially available resistive MR systems range from 0.02 to 0.4 T. Air cored designs are also available. 2.2.2

Field Homogeneity and Shielding

A typical main ®eld homogeneity in a shimmed resistive magnet over a 40 cm FOV is 20±200 ppm, i.e. less than for most superconducting magnets. Any problems in maintaining a

constant power supply can result in temporal instability of the magnetic ®eld. Stability will also be affected by temperature variations if cooling and heat exchange systems malfunction. However, the magnetic ®eld can be switched off easily to allow removal of metallic objects attached to the bore of the magnet. 2.2.3

Speci®cations

The weight and installation requirements for a 0.2 T resistive magnet are similar to those of 0.5 T superconducting scanners, and the imaging capabilities are broadly similar (Table 2). 2.2.4

Installation and Running Costs

Resistive MRI scanners are cheaper to buy than superconducting types, and cost less to install. Despite the relatively high electricity costs, running costs overall are less than those of superconducting magnets. 2.3 2.3.1

Permanent Magnets Design

The magnetic ®eld is generated by units of permanently magnetized material arranged to construct a magnet with north and south poles separated by an air gap of suf®cient size for imaging. Permanent magnets usually have a horseshoe or bipolar shape, with the patient lying perpendicular to B0. Containment of the magnetic ¯ux lines to produce a uniform ®eld between the poles of the magnet requires a large mass of iron, with the result that a permanent MRI scanner is heavier than types using superconducting magnets. Modern systems weigh approximately 10 ton but lighter permanent magnets using rare earth alloys such as neodymium alloy are also available. MRI systems based on a permanent magnet have a limited ®eld strength of up to 0.2±0.3 T. The extent of the

MRI AT MIDFIELD STRENGTH Table 2

3

Comparison of Representative Speci®cations of MR Systems available in mid-1990s

Field strength/type

Homogeneity

Installation area (m2)

Fringe ®eld, axial  vertical (m)

0.2 T Permanent

‹20 ppm 1810 cm DSV

25

1.7  2.2

8900

‹5 ppm 40 cm DSV

30

2.6  2.4

12 000

‹5 ppm 40 cm DSV

23

3.7  2.5, passive shield

7000

0.5 T

‹5 ppm 40 cm DSV

25

3.5  2.5, active shield

2800

1.0 T

‹7.5 ppm 50 cm DSV

40

2.4  4.3, active shield

4900

1.5 T

‹5 ppm 50 cm DSV

65

3.1  5.0, active shield

13 100

0.2 T Resistive Superconducting: 0.5 T

Weight (kg)

DSV, diameter spherical volume

stray ®eld is usually less than for wire-wound open-core magnets, but magnetic ®eld intensity often rises sharply near the bore of the magnet and the `missile effect' may be greater than for wire-wound magnets.2 Systems are available with an open con®guration, which reduces claustrophobia and improves access to high-dependency patients or for MR-guided interventional procedures. Solenoidal rf coils can be used to improve the S/N. 2.3.2

Field Homogeneity and Stability

Magnetic ®eld homogeneity is of the order of 40 ppm on a 36 cm DSV, compared with 20 ppm on a 45 cm DSV for a 0.5 T actively shielded superconducting magnet. Variations in the temperature of the iron core will affect ®eld strength, and ®eld homogeneity can be ruined by damage to the magnet surface. 2.3.3

Speci®cations

The limited ®eld strength is associated with an increase in examination time of 5±15% depending on the area being examined. A vertical ®eld orientation is said to produce image quality similar to that of a 0.5 T system, and models are available with a three-dimensional imaging capability. Fast scan and MR angiography packages are available. 2.3.4

Installation and Running Costs

Permanent magnets are becoming more popular as alternatives to superconducting or resistive MR scanners in budgetconscious markets. Installation does not require magnetic shielding, a separate computer room, or special air conditioning. Running costs are about one-tenth of those for a superconducting magnet system, since permanent magnets use little electrical power and no cooling water or cryogens. The latter consideration is particularly important in Japan and Italy, where cryogens are expensive. Operational costs are also

reduced by a low downtime of around 5%. Installation requires a space of 25 m2, similar to that for a computerized tomography (CT) scanner, and can be completed in 10 days. 3

ADVANTAGES OF MIDFIELD MRI

The speci®cation of commercially available MRI scanners in the mid®eld range has improved in recent years, while the purchase price, installation, and running costs have fallen. Manufacturers have concentrated on reducing site costs whilst upgrading capabilities. Ef®cient refrigeration with low cryogen consumption helps to reduce running costs at mid®eld. Operating costs for mid®eld MR systems can be covered with an average daily patient volume of 6±7 patients, while up to 10 patients are required to cover costs on a high-®eld system, assuming similar reimbursement rates. The cost of a maintenance agreement for a representative 1.5 T MR scanner can be 30% greater than for a mid®eld system. The Guidelines published by the Medical Devices Directorate of the UK Department of Health (1993)2 recommend that MR equipment is contained within a designated controlled area, encompassing the 5 G magnetic ®eld contour. Selfshielded, lightweight magnets with tight fringe ®elds enable easy siting without major structural modi®cation to existing rooms. The compact fringe ®eld also enables magnetic-®eld-sensitive devices such as monitoring equipment to be taken into the magnet room, thus allowing MRI of high dependency patients. Compact fringe ®elds also help to limit staff exposure, which should not exceed 0.2 T to the whole body including the hands and limbs as an 8 h time-weighted-average exposure. Intravascular devices exhibiting ferromagnetism display greater torque and magnetic susceptibility artifacts at 1.5 T than at 0.35 T.3 Acoustic noise levels in the magnet bore do not correlate as might be expected with magnetic ®eld strength, averaging 82± 93 dB on a range of MR instruments operating from 0.35 to

4 MRI AT MIDFIELD STRENGTH 1.5 T.4 The level of noise depends more on the ef®ciency of damping of the gradient coil vibration in different manufacturers' systems and will always tend to be greater when using fast gradient recalled echo (GRE) sequences with short TR and TE times and thin sections. Ear protection should be worn when the level of acoustic noise exceeds 85 dB.

4 EFFECT OF FIELD STRENGTH ON IMAGING PARAMETERS AND OPTIONS 4.1 General Speci®cations Advances in hardware and software ®rst introduced for high®eld systems are now available at mid®eld. Software packages for cardiac imaging and angiography, and software for reducing acquisition times are all available for mid®eld systems. The rf components of 0.5, 1.0, and 1.5 T systems available from the same manufacturer are in many cases identical, and 0.5 T systems are now available with 15 mT mÿ1 gradients. While systems with a wide variety of ®eld strengths are capable of producing images of diagnostic quality, high-®eld systems can normally produce esthetically pleasing images more rapidly. The introduction of quadrature coils to mid®eld systems and the use of powerful computers to facilitate multiple system operations and improve patient throughput have reduced this differential. Endoluminal coils are becoming more available on mid®eld systems. 4.2 Signal-to-Noise Ratio The S/N increases as the ®eld strength. This advantage of high ®eld strength MRI has to be weighed against the concomitant potential for greater image artifacts due to increased eddy currents, chemical shift, and susceptibility, as well as the potential risk of excessive rf power deposition or heating (see below). At any ®eld strength, image quality is improved by using a surface receiver coil to improve the S/N. 4.3 Contrast The ability to detect a structure on an MR image and differentiate it from neighboring tissues depends on spatial resolution and signal intensity differences. The latter can be partly manipulated by selection of appropriate pulse sequences but is largely determined by the T1 and T2 relaxation times of the tissues. The resonant frequency and T1 relaxation time of protons increase with increasing magnetic ®eld strength. T1 values increase by approximately 400±500 ms Tÿ1, the values for normal and pathological tissue tending to converge and the T1-dependent contrast to decrease as ®eld strength increases. T2 contrast does not show the same degree of ®eld dependence, most tissues having a T2 value in the range 50±150 ms. Taking the effect of noise on image quality into consideration, however, there is a trend for the tissue contrast-to-noise ratio to increase with ®eld up to 1.5 T.5 Tissue components exhibiting susceptibility effects, e.g. iron in liver or basal ganglia, may result in an overall reduction in T2 relaxation time at higher ®eld strength.

4.4

Fat Suppression

Chemical shift imaging depends on the difference in resonant frequency between water and fat hydrogen protons, which allows selective excitation of one or other species. The amount of chemical shift (3.3 ppm) is independent of ®eld strength, but the difference in frequency is proportional to ®eld strength. Separation of fat and water resonant frequencies is less pronounced at mid®eld (70 Hz at 0.5 T) than at high ®eld (200 Hz at 1.5 T). Magnetic ®eld homogeneity will also tend to be less at mid®eld than at high ®eld, with the variation in resonant frequencies due to ®eld inhomogeneity causing imperfect selective excitation. The effect of ®eld inhomogeneity will be least when using small ®elds of view near the magnet center. When using a binomial pulse to achieve selective saturation, the interpulse spacing must be longer at lower ®eld strengths to achieve the same chemical shift effect as at high ®eld, thus limiting the minimum TE.6 The TE times at which fat and water protons are out of phase are inversely proportional to the ®eld strength, i.e., are longer at mid®eld than at high ®eld. 4.5

Magnetization Transfer Contrast

Magnetization transfer contrast (MTC) using saturation transfer techniques re¯ects cross relaxation between free water and water bound to macromolecules. The reduction of the net available magnetization and of T1 in the `free' pool by application of a saturating off-resonance rf pulse to the `bound' protons is a ®eld dependent phenomenon. Magnetization transfer rate decreases with increasing ®eld strength, for example the magnetization transfer rate constant for white matter is 2.27 sÿ1 at 0.5 T and 1.42 sÿ1 at 1.5 T.7 Although the degree of MTC in a given tissue would be expected to diminish as the T1 relaxation time shortens with decreasing ®eld strength, satisfactory MTC images have been obtained at low and mid®eld strengths. Selective saturation is possible at mid®eld using offresonance rf irradiation of short duration. Power deposition in the subject from the rf dose generated by the preirradiation pulse is also less of a limiting factor at mid®eld strength. The safety guidelines for exposure to rf energy in MRI currently recommend a limit of 1 W kgÿ1 body weight.8 4.6

Fast/High-Resolution Scanning

The ability to achieve sequences with short TR and TE times and low ¯ip angles (fast scanning) and small ®eld of view (FOV)/thin slice imaging (high-resolution) scanning is dependent on gradient strength and rise time rather than ®eld strength. New systems of any ®eld strength typically offer gradient strengths of 15±23 mT mÿ1 and a rise time of less than 1 ms. Image contrast and S/N may be compromised in fast/high resolution sequences, and image quality is more dif®cult to maintain at mid®eld than high ®eld, when the amount of tissue signal at mid®eld is intrinsically less.

4.7

Magnetic Resonance Angiography

Magnetic resonance angiography (MRA) is not usually limited by S/N and does depend on contrast between ¯owing

MRI AT MIDFIELD STRENGTH

blood and background tissues and adequate spatial resolution. Intravoxel phase dispersion, which is an important cause of loss of contrast in MRA, is reduced by minimizing ®eld inhomogeneity and by using the minimum voxel size and shortest TE time possible. Systems with excellent magnetic ®eld homogeneity, high gradient strengths and slew rates, and good S/N characteristics are required, these properties generally being associated with high-®eld MRI systems. However, satisfactory MR angiograms have been obtained on mid®eld superconducting MR scanners and on systems employing a permanent magnet with a ®eld strength of 0.2 T.9,10 Permanent magnets suffer less from eddy current artifacts during gradient switching, thereby preserving ®eld homogeneity and thus improving background suppression in MR angiograms. At low and mid®eld, T1 relaxation times are relatively short, so that with time-of-¯ight (TOF) techniques based on ¯ow-related enhancement, signal from ¯owing blood will tend to be greater at mid®eld than high ®eld because of the reduction in saturation due to the more rapid T1 recovery. Conversely, saturation of stationary tissue is more dif®cult for the same reason, so that relatively larger ¯ip angles and shorter TE times have to be employed at mid®eld than at high ®eld to achieve an adequate degree of background suppression. For example, at 0.5 T the ¯ip angle used for two-dimensional TOF is typically 85 , whilst at 1.5 T a similar sequence would employ a ¯ip angle of about 60 . The degree of artifactual loss of ¯ow signal from local ®eld inhomogeneities caused by metallic surgical clips adjacent to vessels is less at mid®eld than high ®eld.

4.8 Echo Planar Imaging Peripheral nerve stimulation has been experienced by volunteers undergoing echo planar imaging (EPI) on a 1.5 T MR system modi®ed by the addition of high-performance gradient sets. Subjective twitches were felt when EPI was being performed near peak gradient operating levels of around 60 T sÿ1.11 US Food and Drug Administration guidelines recommend that the gradient operating levels are kept below 20 T sÿ1 or one-third of the observed threshold for peripheral nerve stimulation. On 0.5 T EPI scanners (Mans®eld, Nottingham), gradient operating levels are one-sixth of that inducing peripheral nerve stimulation. An inversion±recovery (IR) preparation pulse combined with EPI at 0.5 T produces T1-weighted images in which tissue contrast can be optimized, with potential applications in dynamic contrast-enhanced studies.12 Gradient strengths and switching rates are very similar for snapshot FLASH techniques and EPI.

4.9 MR Spectroscopy 31

P MR spectroscopy (MRS) requires a minimum ®eld strength of 1.5 T, but 1H MR spectroscopy is feasible at lower ®eld strengths. Stimulated echo acquisition mode 1H MRS at 1.0 T demonstrates cerebral metabolites such as N-acetylaspartate and creatinine without signi®cant loss of relative spectral resolution, compared with higher ®elds.13

5 5.1

5

ARTIFACTS Motion

Ghost artifacts due to physiological motion during image acquisition occur in the phase-encoding direction and are most conspicuous when the moving tissue has a bright signal, e.g. fat in the abdominal wall or ¯owing blood on gradient echo sequences. Breathing artifacts are more apparent at high ®eld strength than at low to mid®eld strength, due to the greater S/N (and fewer averages used) and higher fat±muscle contrast on the image.14 5.2

Chemical Shift

Chemical shift misregistration causes displacement of fat± water interfaces along the frequency-encoding direction due to the difference in the resonant frequencies of fat and water protons. This phenomenon is an important cause of error in the evaluation of tissue interfaces, particularly in areas containing abundant fat±soft tissue interfaces such as the orbit, extracranial head and neck, abdominal and pelvic cavities, and the musculoskeletal system. The amount of chemical shift increases with increasing ®eld strength, despite the larger imaging gradients that are often integral to high-®eld systems.15 Many manufacturers of mid®eld systems offer the option of variable bandwidth, narrowing the bandwidth when an increase in S/N is desirable, or using a wider bandwidth to minimize chemical shift. 5.3

Susceptibility

The magnetic susceptibility of a material describes its ability to become magnetized in an external magnetic ®eld. Susceptibility artifacts occur at the interface between materials having different susceptibilities and result in local ®eld inhomogeneities and image distortion in the frequency-encoding direction. Artifactual image degradation is greater at high ®eld than at mid or low ®eld and when using GRE sequences rather than spin echo (SE) sequences, particularly T2*-weighted GRE sequences. 6

6.1

CLINICAL APPLICATIONS OF MIDFIELD VERSUS HIGH-FIELD MRI Diagnostic Accuracy

Images obtained at both 0.5 and 1.5 T using similar protocols have been reviewed in a blind study of patients with various craniospinal, abdominal, and joint conditions. receiveroperated characteristic (ROC) analysis of interpretation by expert radiologists at 0.5 and 1.5 T suggested that the differences in ®eld strength are not signi®cant for diagnostic accuracy.16 The S/N increases linearly with ®eld strength, conferring an advantage on high-®eld systems, particularly when imaging small structures using a high-resolution matrix and thin slices. Advantages of mid®eld systems include superior T1-dependent contrast, fewer problems with chemical shift, susceptibility effects, and certain motion artifacts. Narrowing the

6 MRI AT MIDFIELD STRENGTH Table 3

Imaging of Intracranial Hemorrhage on Mid®eld MRI.18,19

Stage

Compositiona

Signal intensity and optimal sequencesb

Hyperacute (<24 h)

Fresh blood

Hyperintense + on T1 WIs Hyperintense +++ on STIR, > T2* WIs > T2 WIs

Acute (1±3 days)

DeoxyHb in intact RBCs

Hypointense + on T1 WIs, Hypointense +++ on T2 WIs, GRE > SE

Subacute (4±7 days)

DeoxyHb in lysed RBCs

Hyperintense +++ on T1 WIs Hypo- to hyperintense on T2 WIs

Chronic (1±3 weeks)

MetHb

Hyperintense on T1 WIs Hyperintense on T2 WIs, GRE > SE

a

DeoxyHb, deoxyhemoglobin; MetHb, methemoglobin; RBCs, red blood cells. WIs, weighted images; IR, inversion±recovery; SE, spin echo; +, mild; +++, marked.

b

bandwidth at mid®eld increases the S/N and is a strategy used to improve image quality towards that of high-®eld systems where the concomitant increase in chemical shift is tolerable. The theoretical reduction in rf power deposition at mid®eld will become of practical interest with the use of MTC imaging and fast spin echo techniques. 6.2 Craniospinal MRI When systems with hardware and software speci®cations that are virtually identical apart from ®eld strength are compared, image S/N is greater at 1.5 T than at 0.5 T, but there is no signi®cant change in diagnostic accuracy for head and spine imaging between the two ®eld strengths.17 Most conditions, such as cerebrovascular disease and tumors, have similar signal intensity characteristics at mid and high ®eld. A notable exception is the appearance of intracranial hemorrhage, which depends on the stage of evolution of the hematoma as well as the ®eld strength of the MR scanner used (Table 3).18,19 Twelve to 24 h after an intracranial bleed, acute hematoma containing oxyhemoglobin is most readily detected on mid®eld MRI using T2*-weighted GRE sequences or a short  inversion±recovery (STIR) sequence which takes advantage of the combination of high proton density and relatively long T1 relaxation time of fresh blood. Subsequent T2 shortening due to deoxyhemoglobin formation is more easily demonstrated on T2*-weighted GRE than T2-weighted SE sequences at mid®eld strength. As deoxyhemoglobin converts to methemoglobin, paramagnetic T1 shortening results in hyperintensity on T1weighted sequences. Visibility of low signal intensity due to T2 shortening is maximized on T2*-weighted sequences. When hemolysis occurs after about 7 days, hematoma becomes hyperintense on all sequences. Comparative studies have demonstrated that mid®eld MR is more sensitive than CT in detecting both hemorrhagic and nonhemorrhagic lesions resulting from closed head trauma, with the exception of subarachnoid hemorrhage.20 The greater sensitivity of high-®eld MRI to magnetic susceptibility confers a theoretical advantage for demonstrating acute hemorrhage. In practice, selection of appropriate pulse sequences at mid®eld strength, i.e. heavily T2-weighted SE or T2*-weighted GRE sequences to maximize differences in susceptibility, compen-

sates for ®eld strength differences. Similarly, acute hemorrhagic contusion of the spinal cord can be demonstrated at mid®eld, using T2*-weighted GRE sequences to detect the susceptibility effects of deoxyhemoglobin. In a clinical setting, imaging of high dependency patients requiring ventilation or close monitoring is more easily achieved at mid®eld than high ®eld. Speci®c circumstances where high ®eld strength may be advantageous in craniospinal MRI include high resolution imaging, for example of the pituitary gland, and demonstration of brain iron. 6.3

Extracranial Head and Neck MRI

Many studies evaluating MRI for the demonstration and staging of tumors in the maxillofacial region and anterior neck have been performed at mid®eld strength. MRI of small structures such as the parathyroid glands has been successful at both mid and high ®eld. The need to obtain thin slices with high spatial resolution confers some advantage on high ®eld MRI, but images of diagnostic quality are achievable at ®eld strengths of 0.35±0.5 T. The use of surface receiver coils speci®cally designed for anterior neck imaging overcomes the handicap of reduced S/N at mid®eld strength, and image quality is high due to relatively little motion or chemical shift artifacts. 6.4

Cardiovascular MRI

Much of the early experience of cardiovascular MRI was at mid®eld strength, typically employing T1-weighted SE sequences gated to the cardiac cycle to provide anatomic information in congenital and acquired cardiovascular disease. Cardiac gated fast SE sequences are now also available at mid®eld strength; uses include the demonstration of cardiac and paracardiac masses, myocardial disease, and anomalies of the great vessels such as aneurysms and stenosis. Phase contrast sequences providing ¯ow information are also readily available at mid®eld strength, providing semiquantitative ¯ow information in conditions such as valvular heart disease and coarctation of the aorta. High-®eld MRI scanners offer increased S/N in the body coil and very fast sequences, thus effectively freezing cardiac motion and enabling `real-time' stu-

MRI AT MIDFIELD STRENGTH

dies of cardiac motion to be performed. Monitoring equipment may be more dif®cult to use in the proximity of a high ®eld strength magnet however, and cardiac-gated acquisitions may be more dif®cult to set up than at mid®eld strength. Chemical shift artifacts mimicking intraluminal disease, e.g. dissection ¯aps, are more troublesome at high ®eld strength, as are susceptibility artifacts in the postoperative patient. MR angiography of the coronary arteries is more successful at high ®eld strength where the ability to reduce voxel size while maintaining the S/N is crucial. Routine display of larger visceral and peripheral vessels, however, is possible using commercially available software and receiver coils on mid®eld MR scanners. 6.5 Musculoskeletal MRI There is no important advantage of high-®eld MRI over mid®eld MRI when evaluating bone and soft tissue tumors, trauma, or in¯ammation. Indeed, the greater degree of chemical shift and susceptibility effects may be deleterious to image quality. The availability of speci®c surface receiver coils for imaging joints and small peripheral body parts usually overcomes the potential problem of inadequate signal at mid®eld strength. The exception may be high-resolution imaging of small structures; studies have demonstrated a signi®cant increase in accuracy for imaging the medial meniscus at 1.5 T compared with imaging at 0.35 T (p < 0.0002).21 No signi®cant difference was demonstrated between ®eld strengths for imaging the lateral meniscus or cruciate ligaments. More recent studies however, have shown no clinically signi®cant ®eld strength dependent difference between 0.5 T and 1.5 T systems for diagnosis of internal derangements of the knee.22 6.6 Abdominal MRI The convergence of T1 relaxation times for normal liver parenchyma and neoplasm with increasing ®eld strength renders T1-weighted SE imaging at high ®eld strength less sensitive to focal liver pathology than low or mid®eld MRI. Hepatic tumors are also demonstrated using heavily T1-weighted inversion± recovery (IR) sequences or T2-weighted sequences. Using these sequences, there is no signi®cant difference in the ability of 0.5 and 1.5 T MRI systems to detect focal hepatic lesions.23,24 Measured relaxation times for normal and pathological tissues vary between mid- and high-®eld MRI systems. Magnetic susceptibility effects tend to increase as the square of the ®eld strength, resulting in relative shortening of the observed T2 relaxation time at high ®eld. Variability will also be introduced by individual and lesional differences, the speci®c MRI system and pulse sequences used, and the method of calculation. However, statistically signi®cant differences in calculated T2 values have been demonstrated between normal liver and tumor, and between hepatocellular carcinoma and cavernous hemangioma imaged at both 0.35 and 1.5 T.25 No signi®cant differences in focal liver lesion detection can be demonstrated between 0.2 and 1.5 T systems when using T2-weighted sequences enhanced by intravenous superparamagnetic iron oxide contrast medium.26 In abdominal MRI, conventional SE sequences are often degraded by movement artifacts. Very fast GRE sequences allowing breath-hold imaging of abdominal structures, which

7

effectively freeze respiratory motion, are available on mid®eld systems. Although motion artifact is reduced, contrast on the fast sequences is relatively poor, and such dynamic MRI sequences are usually acquired in conjunction with a bolus of intravenous paramagnetic contrast agent. Maximum intensity pixel (MIP) reconstructions from this type of acquisition produce an elegant display of portal venous anatomy. The combination of very fast GRE sequences with steady state precession allows MIP projections of the main bile ducts and peripheral biliary tree. In pancreatic imaging the lack of a reliable gastrointestinal contrast agent limits the ability of MRI to delineate the margins of the gland distinctly from adjacent small bowel. Contrast between pancreas and retroperitoneal fat may be poor, particularly when fatty involution of the gland is present. Solid pancreatic masses can be very similar in signal intensity to normal parenchyma on both T1- and T2-weighted sequences, and the inability of MRI to identify calci®cation is a disadvantage when attempting to characterize pancreatic masses. Spatial resolution is also currently less than is possible with CT. MR image degradation from motion artifacts due to respiration, peristalsis, and vascular pulsation are less of a problem at mid®eld than high ®eld. Breath-hold sequences can overcome some motion artifacts, but the intrinsically poor soft tissue contrast of the GRE sequence on which they are based does not improve detection of small pancreatic masses, unless gadolinium enhancement with fat suppression is used. The results of studies assessing the ability of mid®eld MRI to identify pancreatic tumors and in¯ammation have been encouraging. Solid pancreatic carcinomas are demonstrated as an area of low signal intensity on T1-weighted and medium signal intensity on T2-weighted SE sequences at 0.5 T, whereas a very high signal intensity is seen on T2-weighted sequences of cystadenocarcinomas.27 Islet cell tumors less than 2.5 cm in diameter can be successfully localized using T1- and T2-weighted sequences.28 In severe acute pancreatitis, contrast-enhanced fast GRE MRI at 1.0 T is equivalent to CT in determining pancreatic viability.29 MRI has a limited role in demonstrating renal masses, and the potential for characterizing renal tumors and differentiating between small benign and malignant lesions has not yet been realized at mid®eld strength. Results have been more encouraging using higher ®eld strength systems, breath-hold imaging, dynamic contrast enhancement, and fat suppression. Studies of the adrenal glands performed at mid®eld strength (0.35±0.5 T) have analyzed signal intensity ratios of adrenal masses, and claim the ability to separate a proportion of benign and malignant adrenal masses on the basis of their signal intensities or calculated relaxation times. No consistent difference in the signal intensity of adrenal adenomas and malignant adrenal tumors has been demonstrated with T1- or T2-weighted sequences at 1.5 T (using fat or liver tissue as the internal reference). Calculated T2 values may be more helpful, a calculated T2 value of 65 ms or greater being associated with malignancy.30 6.7

Pelvic MRI

Staging of pelvic malignancy is useful in carcinoma of the body of the uterus, the uterine cervix, bladder, and rectum. It may also help to differentiate between recurrent pelvic tumor

8 MRI AT MIDFIELD STRENGTH and the effects of previous surgery or radiotherapy. Good results have been achieved using MRI, both at mid®eld and high ®eld strengths. The S/N at all ®eld strengths is improved by the use of a pelvic surface coil, e.g. of Helmholtz design, or by using multiple phased array coils. Motion artifacts due to respiration are less of a problem in the pelvis than in the upper abdomen, but the effects of peristalsis, vascular pulsation, chemical shift, and susceptibility are important causes of reduced image quality, their effects being greater at high ®eld than at mid®eld. Endoluminal MRI is available at the end of the 1990s at both mid®eld and high ®eld and offers the prospect of improved accuracy of prostate cancer staging over body coil imaging.

7 SUMMARY There is little objective evidence of a substantial difference in diagnostic accuracy between MR images obtained at mid®eld or high ®eld. Image quality depends more on using a modern scanner which is maintained to speci®cation by means of a routine quality assurance program, and the use of appropriate pulse sequences by experienced personnel. Studies of machines of different ®eld strength using test objects have shown wide variability in performance between machines of the same ®eld strength and of individual systems at different times due to variable quality assurance. For certain diagnostic applications, the relatively good tissue contrast and easier operation of mid®eld systems may be preferable to the higher S/N and spatial resolution of high-®eld systems. Very fast GRE sequences obtained at 1.5 T are available at 1.0 T but do not necessarily reduce the overall examination time, which includes the time necessary to prepare the patient for scanning.

8 RELATED ARTICLES Cryogenic Magnets for Whole Body Magnetic Resonance Systems; High-Field Whole Body Systems; Low-Field Whole Body Systems; MRI in Clinical Medicine; Resistive and Permanent Magnets for Whole Body MRI.

9 REFERENCES 1. Japan Municipal Hospital Association data provided by Hitachi for Diagnostic Imaging International 1991, Jan./Feb., 32±38. 2. Department of Health Medical Devices Directorate, `Guidelines for Magnetic Resonance Diagnostic Equipment in Clinical Use', HMSO, London, 1993. 3. G. P. Teitelbaum, W. G. Bradley, Jr., and B. D. Klein, Radiology, 1988, 166, 657. 4. R. Hurwitz, S. R. Lane, R. A. Bell, and M. N. Brant-Zawadski, Radiology, 1989, 173, 545. 5. H. R. Hart, Jr., P. A. Bottomley, W. A. Edelstein, S. G. Karr, W. M. Leue, O. Mueller, R. W. Redington, J. F. Schenck, L. S. Smith, and D. Vatis, Am. J. Roentgenol., 1983, 141, 1195. 6. C. J. Baudouin, D. J. Bryant, and I. R. Young, Br. J. Radiol., 1992, 65, 132.

7. R. P. Gullapalli, P. Margosian, and L. Kasuboski, in Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, Vol. 3, p. 1292. 8. National Radiation Protection Board, Br. J. Radiol., 1983, 56, 974. 9. P. Pavone, L. Marsili, C. Catalano, G. A. Petroni, E. Aytan, G. P. Cardone, and R. Passariello, Radiology, 1992, 184, 401. 10. C. B. Grandin, P. Mathurin, T. Duprez, G. Stroobandt, F. Hammer, P. Goffette, and G. Cosnard, Am. J. Neuroradiol., 1998, 19, 245. 11. M. S. Cohen, R. M. Weisskoff, R. R. Rzedzian, and H. L. Kantor. Magn. Reson. Med., 1990, 14, 409. 12. M. K. Stehling, R. J. Ordidge, R. Coxon, and P. Mans®eld, Magn. Reson. Med., 1990, 13, 514. 13. R. Sauter, W. Loef¯er, H. Bruhn, and J. Frahm, Radiology, 1990, 176, 221. 14. M. L. Wood and R. M. Henkelman, Magn. Reson. Imag., 1986, 4, 387. 15. R. Lufkin, M. Anselmo, J. Crues, W. Smoker, and W. Hanafee, Comput. Med. Imaging Graphics, 1988, 12, 89. 16. D. Vellet, D. Lee, M. Eliasziw, P. Munk, W. Romano, C. Romano, R. Farb, R. Sevick, and L. Vidito, in Proc. XIIth Ann Mtg. Soc. Magn. Reson. Med., New York, 1993, Vol. 3, p. 1480. 17. C. R. Jack, Jr., T. H. Berquist, G. M. Miller, G. S. Forbes, J. E. Gray, R. L. Morin, and D. M. Ilstrup, J. Comput. Assist. Tomogr., 1990, 14, 505. 18. R. D. Zimmerman, L. A. Heier, R. B. Snow, D. P. C. Liu, A. B. Kelly, and M. D. F. Deck, Am. J. Roentgenol., 1988, 150, 651. 19. R. A. Brooks, G. Di Chiro, and N. Patronas, J. Comput. Assist. Tomogr., 1989, 13, 194. 20. L. R. Gentry, J. C. Godersky, B. Thompson, and V. D. Dunn, Am. J. Roentgenol., 1988, 150, 673. 21. S. P. Fischer, J. M. Fox, W. Del Pizzo, M. J. Friedman, S. J. Snyder, and R. D. Ferkel, J. Bone Joint Surg., 1991, 73A, 2. 22. M. J. Barnett. Am. J. Roentgenol., 1993, 161, 115. 23. J. W. Reinig, A. J. Dwyer, D. L. Miller, J. A. Frank, G. W. Adams, and A. E. Chang, Radiology, 1989, 170, 149. 24. H. V. Steinberg, J. J. Alarcon, and M. E. Bernardino, Radiology, 1990, 174, 153. 25. K. Ohtomo, Y. Itai, K. Yoshikawa, T. Kokubo, and M. Iio, Radiology, 1988, 168, 621. 26. F. Deckers, B. Corthouts, Y. Nackaerts, O. Ozsarlak, P. M. Parizel, and A. M. de Schepper, Eur. Radiol., 1997, 7, 887. 27. P. Pavone, R. Occhiato, O. Michelini, S. Guiliani, G. Cardone, G. A. Petroni, N. De Stafano, E. Aytan, and R. Passariello, Eur. Radiol., 1991, 1, 124. 28. P. Pavone, D. G. Mitchell, F. Leonetti, M. Di Girolamo, G. Cardone, C. Catalano, G. Tamburrano, and R. Passariello, J. Comput. Assist. Tomogr., 1993, 17, 403. 29. A. Saifuddin, J. Ward, J. Ridgway, and A. G. Chalmers, Clin. Radiol., 1993, 48, 111. 30. R. Kier and S. McCarthy, Radiology, 1989, 171, 671.

Biographical Sketches Jane M. Hawnaur. b 1957. MBChB, 1981, MRCP, 1984, DRMD, 1987, FRCR, 1987. Senior lecturer in Radiology, University of Manchester, 1991±present, and Honorary Consultant Radiologist, Central Manchester Healthcare Trust. Clinical radiologist, with research interest in MRI in oncology. Ian Isherwood. b 1931. MBChB, 1954, DMRD, 1957, FFR, 1960, FRCR, 1975, FFR RCSI (Hon.), 1980, FRCP, 1986, Honorary MD, University of Zaragoza, 1986, CBE 1996. Emeritus Professor of Radiology, University of Manchester.

MRI IN CLINICAL MEDICINE

MRI in Clinical Medicine Fergus V. Coakley and Alexander R. Margulis University of California, San Francisco, CA, USA

Magnetic resonance imaging (MRI) is the third and most recently developed of the major cross-sectional imaging techniques. MRI entered clinical usage in 1980, well after ultrasonography and computed X-ray tomography. MRI has revolutionized radiology by greatly extending the soft tissue contrast resolution of conventional imaging techniques. Clinical MRI was introduced in Nottingham and Aberdeen, and this was followed in 1981 by high-quality images from the Hammersmith Hospital, the Cleveland Clinic, the Massachusetts General Hospital, and the University of California, San Francisco.1±10 At that time, ultrasound was the most common and most readily available cross-sectional modality, while computed tomography (CT) was regarded as the pre-eminent imaging technique: CT sections of the body were obtainable with an acquisition time of 5±6 s each. Despite these established alternative imaging modalities, the diagnostic advantages of MRI were apparent from the very beginning. Starting in 1981, all meetings where MRI reports were presented attracted over¯owing audiences. What particularly captured the interest of the physicians in these audiences was the soft tissue contrast resolution of MRI, the ability to image in any plane, the wide ®eld of view, the rapidly improving quality of the images, the ability to visualize blood vessels because of the effects of ¯ow, and, above all, the dependence of the method on many variable physical parameters offering a never-ending number and variation of sequences and coil improvements. All of these advantages provided for better lesion detection and characterization and greater speed of imaging. In addition, sequences could be manipulated to suppress many artifacts, such as those owing to respiratory motion, cardiac and arterial pulsatility, eddy currents and chemical shift.11±14 An important later advance was the development of intravenously administered MR contrast media which are even safer than the iodinated intravascular contrast media used in conventional radiography and CT. The versatility of the method is manifested by the vast array of instruments that have been available since the early stages of clinical MRI. Images of acceptable quality can be generated from permanent, resistive or superconducting magnets, with ®eld strengths varying from 0.03 to 1.5 T. Higher ®eld strength scanners of up to 3±4 T are available for investigational and animal studies. Because of its many advantages, MRI rapidly became a popular and widely accepted imaging tool in developed countries, particularly the United States and Japan, and later in western Europe. The number of MRI scanners in the United States in 1997 was 3670, or 13.85 per million population. Differences in the availability of MRI scanners is largely a consequence of government and state control over the proliferation of high-cost procedures and technology and does not relate to the desires of either physicians or patients. For example, the number of MRI scanners in California in 1997

1

was 55 per million population, while in Massachusetts, a state that requires certi®cates of need, the number was 13 per million population. In the United States, most of the MRI scanners have a ®eld strength of 1.5 T. In the rest of the industrialized world, notably Japan and western Europe, 0.5 to 1 T instruments predominate. In 1997, MRI had a 21% share of all global medical imaging expenditure. By comparison, conventional rediography had a 34% share. CT and nuclear medicine had a comparatively minor share of the global imaging expenditures.15±18 Until the late 1980s, the main clinical applications of MRI were in the brain and spine. In these anatomic regions, MRI was superior to other imaging modalities mainly because of the relative lack of motion. As the number of MRI scanners increased, and with the development of motion suppression, faster sequences, and fat suppression, other anatomic regions started to be imaged routinely. A further contributory factor to the growth in interest in non-neurological applications of MRI was the emergence of a cadre of body-imaging radiologists who were adequately trained in the performance and interpretation of MRI. These factors resulted in a spectacular increase in the application of MRI to areas other than the brain and spine in the United States. For example, between 1991 and 1997, the total increase in the use of MRI was a startling 81%. When examined by system, chest imaging rose 88%, brain imaging rose 78%, spinal imaging rose 61%, and general abdominal imaging rose 57%. However, pelvic imaging decreased by 19%. Despite the disproportionately greater rise in non-neurological applications, the absolute number of studies remained higher for neurological examinations. For example, in 1997, the number of brain and spinal MRI studies performed in the United States was 8 032 000, compared with 199 000 abdominal studies, and 109 000 chest studies. The number of pelvic MRI was a relatively modest 130 000. By the 1990s, MR had become the primary imaging technique for diseases of the brain, head and neck, major blood vessels, spine, musculoskeletal system, and pelvis, including the male and female reproductive organs. MRI with intravenously administered Gd-DTPA has essentially replaced contrast-enhanced CT in the diagnostic evaluation of brain tumors, stroke, and multiple sclerosis (Figures 1 and 2).19±21 MRI has not yet been able to demonstrate speci®c diagnostic ®ndings in Alzheimer's disease, and it is often unable to identify accurately the source of focal epileptic seizures. In the head and neck,22 MRI is outstanding in showing parathyroid and thyroid neoplasms, tumors of salivary glands, enlarged lymph nodes, and tumors of the upper air passages.13,14 It is also valuable in diagnosing brachial plexus injuries. In the musculoskeletal system, MRI is the optimal modality for studying degenerative, in¯ammatory and trauma induced changes in the spine22±24 and major joints: the knee, hip, shoulder.25±27 As experience increases, smaller joints in hand and foot are being examined with MRI and the ®ndings guide therapy. MRI is also the best method for the evaluation of bone and soft tissue tumors. With newer techniques, it is very promising in the evaluation of different types of arthritis (Figure 3). In the male pelvis, MRI is used predominantly for staging cancer of the prostate (Figure 4) and the evaluation of major trauma prior to de®nitive surgical repair. MRI of the prostate is

2 MRI IN CLINICAL MEDICINE

Figure 1 Spin-echo T1-weighted coronal MR image of the brain, showing marked enhancement of a left lateral geniculate glioma

generally not used for cancer diagnosis but rather for local and regional staging of biopsy-proven cancer. Assessment of extracapsular extension, seminal vesicle invasion, and lymphadenopathy are critical observations in staging prostate cancer and can all be assessed with relatively high accuracy by MRI. The addition of magnetic resonance spectroscopy to routine clinical MRI of the prostate can also help in tumor evaluation.28±31 In the female pelvis, MRI is used for the staging of tumors of the endometrium (Figure 5), cervix uteri, ovaries, and

Figure 2 Spin-echo T1-weighted axial MR image of the posterior fossa, before (left) and after (right) intravenous administration of Gd± DTPA. An enhancing small left acoustic neuroma (arrow) is evident. The use of Gd±DTPA helps in lesion detection

Figure 3 Saturation-transfer subtraction axial MRI section of the knee of a patient with nonspeci®c synovitis. The image was generated by subtracting T2-weighted steady-state gradient-echo images without an on-resonance pulsed saturation transfer from the same sequence obtained with a pulsed saturation transfer. Patellar cartilage can be clearly distinguished from adjacent joint ¯uid and subchondral bone. Hypertropic synovial tissue is sharply delineated, without the use of intravenous gadolinium-containing contrast media. The patient subsequently underwent surgical synovectomy.

vagina. Intravenous Gd-DTPA is helpful in staging most gynecologic malignancies, with the exception of cervical cancer.31 The use of Gd-DTPA is usually only helpful in cervical cancer when there is suspected invasion of adjacent organs, tumor necrosis, or post-radiation change. In the heart, MRI can diagnose and evaluate various types of congenital and valvular heart disease.32±35 However, cardiac applications of MRI remain relatively limited because of a lack of enthusiasm for the technique among most cardiologists and

Figure 4 Fast spin-echo T2-weighted axial MR image of the prostate, obtained using an endorectal coil. Invasion of the right seminal vesicle (arrow) by carcinoma of the prostate is seen

MRI IN CLINICAL MEDICINE

Figure 5 Fast spin-echo T2-weighted sagittal MR image of the uterus, showing a large cervical carcinoma (arrow) with deep cervical stromal invasion and extension above the internal os into the uterine body

because of the relatively small number of cardiac radiologistinvestigators. Major breakthroughs in the area of cardiac MRI are likely in the coming years, as ultrafast sequences and strategies to overcome image degradation by cardiac motion continue to evolve. Although CT is still the principal imaging method for demonstrating renal abnormalities, MR with Gd±DTPA is becoming at least as accurate.36 MRI may be preferable in patients with an elevated creatinine or a history of previous reactions to iodinated contrast. MRI may also be helpful in assessing vascular invasion in renal cell carcinoma. MRI is extremely useful in the detection and characterization of focal liver lesions (Figure 6). Initially, dynamic Gd-

(a)

3

DTPA-enhanced MRI appeared inferior to CT arterioportography, but more recent results with improved technology suggest that the two techniques are of comparable overall accuracy. MRI may be slightly less sensitive but more speci®c than CT arterioportography, when intraoperative exploration and ultrasound are used as the standard of reference. Newer speci®c contrast media such as Mn-DPDP and superparamagnetic iron oxide particles have recently become available. These agents may help in lesion detection and characterization, although a de®nite advantage over the standard use Gd-DTPA has not been demonstrated.37±41 In the future, optimized use of such newer contrast agents is likely to be more rewarding. A relatively recent application of MRI in the liver is the use of ultrafast T2-weighted sequences to image the ¯uid in the biliary and pancreatic ducts. This technique of MR cholangiopancreatography appears set to replace diagnostic endoscopic retrograde cholangiopancreatography in the next few years (Figure 7). MRI has established at least an equality with other imaging techniques in the examination of the heart, mediastinum, retroperitoneum, and for focal lesions of the spleen.11±14 MRI is still inferior to CT in the evaluation of the gastrointestinal tract and its mesentery.42,43 This may change very shortly with shorter sequences and the use of antiperistaltic agents such as glucagon.44 Interesting advances are also being made in the examination of the lung, particularly in the evaluation of pulmonary edema, embolism, and ®brosis.45±47 More recently, there has been considerable interest in the use of inhaled hyperpolarized 3He or 129Xe to produce images of lung ventilation.48 Currently, MRI is as limited as CT in the evaluation of tumor involvement of lymph nodes, although it is very likely that with new contrast media this dif®culty may be overcome. Recently, MRI of the breast combined with the intravenous use of Gd±DTPA has shown great promise by being more sensitive in discovering cancer in dense breasts than conventional X-ray mammography. Dramatic progress has been made since MR mammography was ®rst described in the early 1980s, although many questions remain unanswered. The development of better and more speci®c contrast agents is likely to be the

(b)

Figure 6 In phase (a) and out of phase (b) fast spoiled gradient-echo T1-weighted axial MR image of the liver, showing a large irregular lesion in the right hepatic lobe. The lesion has markedly reduced signal intensity on the out of phase image (b), con®rming the presence of fat and establishing the diagnosis of focal fatty in®ltration. A prior CT image had been interpreted as suggestive of malignancy

4 MRI IN CLINICAL MEDICINE

Figure 7 Single-shot fast spin-echo T2-weighted coronal MR cholangiopancreatography in a patient with pancreas divisum. The main pancreatic duct is seen draining to the minor papilla (arrow) at a level superior to the entry of the common bile duct to the major papilla. The image was generated as a maximum-intensity projection of a slab of multiple contiguous thin coronal sections through the pancreaticobiliary tree

Figure 8 Gradient-echo T1-weighted sagittal MR image of the breast after the intravenous administration of Gd-DTPA, with fat suppression. This MR mammogram shows a small enhancing carcinoma (arrow). The lesion was not detected using conventional radiographic mammography

next major advance. A practical system of MRI-guided breast biopsy is also essential, especially for those lesions that are seen only on MRI and not evident on conventional radiographic mammography or ultrasound (Figure 8). Better coils and low-cost dedicated equipment are also required.49±52 MRI of the fetus has been slow to develop, because of image degradation by fetal motion and the need for high spatial resolution. The recent development of ultrafast spin-echo based T2-weighted sequences has been a major advance, and fetal imaging is emerging as one of the more exciting new clinical applications of MRI. The use of fetal MRI remains con®ned to those cases where antenatal ultrasound is indeterminate or inconclusive, since in most cases ultrasound is an adequate study. In dif®cult situations, MRI can provide signi®cant incremental information (Figure 9) and impacts on management in a substantial proportion of patients.53,54 The ®eld of MRI has also advanced with the construction and application of a multitude of surface coils, single or assembled in phased arrays, sometimes in conjunction with

Figure 9 Single-shot fast spin-echo T2-weighted MR image of a fetus with suspected laryngeal atresia from a prior sonograph, in an oblique longitudinal plane relative to the fetus. The constellation of diaphragmatic eversion (white arrows), a large volume of fetal ascites (asterisk), marked subcutaneous edema (black arrow), and a blindending ¯uid-®lled upper airway (curved white arrow) is diagnostic of laryngeal atresia. The fetus was successfully treated by in utero tracheostomy

MRI IN CLINICAL MEDICINE

5

onary MR angiography is the `holy grail' of clinical MRI, but despite the technical problems is likely to become clinically available in the ®rst decade of the 21st century. Techniques for the evaluation of the proximal segments of the major coronary arteries already exist. Clinical acceptance requires the procedure to produce diagnostic images in a cost-effective manner. Improved contrast agents may be needed.35 Interventional MRI units are being feverishly developed. These interventional units vary greatly in design and cost. Permanent magnets with wide separation allowing manipulation, a resistive magnet with four ports to provide good ®eld homogeneity and allow ample access, and a unit with two separated tunnels (the separation to permit space for the surgeon to work), are all at various stages of completion. These units will signi®cantly differ in cost. The quality of the image, the speed of acquiring the image, the size and weight of the units, and shielding requirements will all play a signi®cant role in clinical acceptance. Interventional MRI requires MR-compatible instruments and anesthesia equipment. Better techniques for MRI tracking of guidewires and catheters are also needed. Initially, interventional MRI is likely to be most useful in the evaluation of interstitial therapies such as laser treatment, radiofrequency ablation, and cryosurgery.59 It remains to be seen how the roles

Figure 10 Gradient-echo ¯ow-sensitive axial MR time-of-¯ight angiogram in a patient with a large arteriovenous malformation. The image was generated as a maximum-intensity projection of a slab of multiple contiguous thin axial sections through the circle of Willis

endoluminal coils such as with the endorectal coil used for examination of the prostate (Figure 4).30 MR angiography has rapidly advanced since it was ®rst introduced in the late 1980s. Initially, MR angiography was limited by relatively long acquisition times and low spatial resolution. These problems have been largely overcome by improvements in hardware, pulse sequence design, and sequencing strategies. MR angiography is increasingly used in the evaluation of arterial and venous disease throughout the body, especially in patients who have contraindications to conventional angiography, such as renal impairment, iodinated contrast allergy, and coagulopathy. MR angiographic techniques may be based on time of ¯ight (Figure 10), phase contrast, black blood, or contrast enhancement. Magnetization transfer can be used to suppress signal from stationary tissue approaches. Recently, ®rst-pass gadolinium-enhanced MR angiography has received considerable attention. In the future, the emergence of blood pool contrast agents is likely to contribute further to the development of MR angiography. This technique, with its noninvasive nature and ease of performance, is already a clinical procedure in the evaluation of major vessels in the brain, neck, body, and extremities.55±58 Imaging of the coronary arteries with a resolution equivalent to conventional angiography remains a crucial goal that has not yet been achieved. Coronary MR angiography poses unique challenges because of rapid motion and the small size of the vessels. Cor-

Figure 11 Diffusion-weighted axial MR image in a patient with an acute right hemiparesis. Increased signal (arrow) in the distribution of the left middle cerebral artery territory con®rms the diagnosis of cerebral infarction

6 MRI IN CLINICAL MEDICINE

Citrate Choline

Creatine

Figure 12 MR spectroscopy performed in a patient with prostate cancer. The T2-weighted fast spin-echo axial image of the prostate shows an area of reduced signal intensity in the right peripheral zone. The corresponding MR spectra demonstrate elevation of choline and reduction of citrate in this area, con®rming the presence of malignancy. Normal data, obtained from the healthy part of the gland, are shown for comparison and demonstrate a high level of citrate and no detectable choline

of radiologists, surgeons, anesthesiologists, and other health care professionals will develop in this complex ®eld. Advances in the basic sciences and continued improvements in hardware and software have resulted in an increasing array of MRI techniques capable of evaluating physiological rather than anatomical information. Such techniques include diffusion MRI, perfusion MRI, functional MRI, MR spectroscopy, MR thermography, and MR elastography.11±14 Diffusion MRI has been used mainly in the assessment of brain disease, because disruption in the regular three-dimensional con®guration of nerve ®bers results in marked alterations in the local pattern of molecular diffusion. Diffusion MRI has become clinically useful in the diagnosis of stroke, because diffusion abnormalities can be detected in the earliest stages of a cerebral infarct (Figure 11) at a time when other imaging techniques are often nondiagnostic and when thrombolytic therapy still has the potential to improve outcome.60 Diffusion MRI may also be

helpful in locating epileptogenic foci prior to neurosurgical treatment.61 Perfusion MRI utilizes exogenous or endogenous contrast to measure blood ¯ow at the capillary level and in the future may have clinical applications in the brain, heart, and kidneys.62 Functional MRI utilizes focal changes in blood volume, ¯ow, or oxygenation to produce images that re¯ect local organ function. Functional MRI has been used predominantly to map the spatial and temporal distribution of neuronal activity in the brain during speci®c cognitive, motor, or sensory tasks.63 MR spectroscopy utilizes the slight differences in Larmor frequency between protons in different molecules to assess the level of different proton-containing tissue metabolites. Clinical applications are emerging in tumors of the breast, prostate (Figure 12), and brain.30,64,65 Many of the current and future developments in MRI will be accompanied and augmented by the introduction of new MR contrast media. Such new contrast media are likely to

MRI IN CLINICAL MEDICINE

7

scanners are available to purchasers who do not have the resources to acquire an expensive high-®eld-strength multipurpose scanner. This wide choice helps make MRI more widely available. Almost all scanners have the potential to be upgraded. The size of scanners and extent of fringe ®elds is being reduced, making siting less expensive. As governments realize that MRI is cost effective, the clinical use of MRI will be expanded and encouraged. However, strict regulation of reimbursement for MRI studies will undoubtedly become universal.18 Such regulations already exist in many parts of the industrialized world. MRI offers an unparalleled choice of techniques for the imaging of normal and abnormal anatomy and pathology. Current and future developments in MR contrast media are likely to expand the clinical utility of MRI still further. The complexity of MRI and the necessity to be familiar with anatomy in all planes and all body parts makes MRI a more dif®cult diagnostic modality to master than any other imaging technique. As a result, for optimal MRI performance and interpretation, clinical expertise and technical pro®ciency are essential. The radiological community must continue to maintain a strong in¯uence over clinical standards in MRI.

1

RELATED ARTICLES

Contrast Agents in Whole Body Magnetic Resonance: An Overview; High-Field Whole Body Systems; Low-Field Whole Body Systems; MRI at Mid®eld Strength; Outcome and Ef®cacy±Analysis of Healthcare Models. Figure 13 A resistive 0.5-T MR imager transported through a narrow doorway and later assembled in room, making siting much more economical

evolve along the following lines.66±70 (a) High-molecularweight contrast media will be useful for perfusion and bloodpool MRI techniques, since these remain in the circulation. (b) Tissue-speci®c contrast agents, which enhance only a single organ or tissue type, will increase the ability to detect and characterize lesions. Such contrast media are already available, for example the manganese-based contrast media that are picked up and excreted by hepatocytes and the ferumoxides that are taken up by reticuloendothelial cells. (c) Contrast media that exit the capillaries and enter the lymphatic system to be picked up by lymph nodes will increase the ability to detect and characterize lymph node abnormalities. (d) Contrast media enveloped in liposomes can be designed for uptake by different targeted organ systems, for example the bone marrow. (e) Contrast media incorporated with monoclonal antibodies may provide the ultimate `magic bullet' for imaging of targeted abnormal cells, for example cancer cells. Cost remains a major hurdle to the greater availability and use of clinical MRI in much of the world. As competition between companies grows and the marketplace becomes increasingly global, a vast array of scanner types is emerging. Scanner technology now includes permanent magnets, resistive magnets (Figure 13), and various types of superconductive magnet. These different machines allow considerable choice in the degree of sophistication and cost, so that small dedicated

2

REFERENCES

1. R. C. Hawkes, G. N. Holland, and W. S. Moore, J. Comput. Assist. Tomogr., 1980, 4, 429. 2. R. C. Hawkes, G. N. Holland, W. S. Moore, E. J. Roebuck, and B. S. Worthington, J. Comput. Assist. Tomogr., 1981, 5, 605. 3. R. C. Hawkes, G. N. Holland, W. S. Moore, E. J. Roebuck, and B. S. Worthington, J. Comput. Assist. Tomogr., 1981, 5, 613. 4. F. H. Doyle, J. C. Gore, J. M. Pennock, G. M. Bydder, J. S. Orr, R. E. Steiner, M. Burl, M. Clow, D. J. Goldisdale, D. R. Bailes and P. E. Walton, Lancet, 1981, 2, 53. 5. G. M. Bydder, R. E. Steiner, I. R. Young, A. S. Hall, D. J. Thomas, J. Marshall, C. A. Pallis, and N. J. Legg, Am. J. Nucl. Reson., 1982, 3, 459. 6. R. J. Al®di, J. R. Haaga, S. J. El Yousef, P. J. Bryan, B. D. Fletcher, J. P. Li Puma, S. C. Morrison, B. Kaufman, J. B. Richey, W. S. Hinshaw, D. M. Kramer, H. N. Yeung, A. M. Cohen, H. E. Butler, A. E. Ament, and J. M. Lieberman, Radiology, 1982, 143, 175. 7. I. R. Young, D. R. Bailes, M. Burl, A. G. Collins, D. T. Smith, M. J. McDonnell, J. S. Orr, L. M. Banks, G. M. Bydder, R. H. Greenspan, and R. E. Steiner, J. Comput. Assist. Tomogr., 1982, 6, 1. 8. L. Crooks, M. Arakawa, J. Hoenninger, J. Watts, R. McRee, L. Kaufman, P. L. Davis, A. R. Margulis, and J. DeGroot, Radiology, 1982, 143, 169. 9. T. J. Brady, M. C. Gebhardt, I. L. Pykett, F. S. Buonanno, J. H. Newhouse, C. T. Burt, R. J. Smith, H. J. Mankin, J.P, Kistler, M. R. Goldman, W. S. Hinshaw, and G. M. Pohost, Radiology, 1982, 144, 549.

8 MRI IN CLINICAL MEDICINE 10. A. R. Margulis, C. B. Higgins, L. Kaufman, and L. E. Crooks, `Clinical Magnetic Resonance Imaging', Radiology Research and Education Foundation, San Francisco, CA, 1983. 11. R. R. Edelman and J. R. Hesselink, `Clinical Magnetic Resonance Imaging', Saunders, Philadelphia, PA, 1990. 12. V. M. Runge, `Clinical Magnetic Resonance Imaging', Lippincott, Philadelphia, PA, 1990. 13. C. B. Higgins, H. Hricak, and C. A. Helms, `Magnetic Resonance Imaging of the Body', 3rd edn., Raven Press, New York, 1997. 14. D. Stark and W. G. Bradley, Jr., `Magnetic Resonance Imaging', 3rd edn., Mosby Year Book, St Louis, MO, 1998. 15. G. J. Schieber, J. P. Poullier, and L. M. Greenwald, Health Care Financ. Rev., 1992, 1, 87. 16. G. E. Medical Systems data, 1999. 17. American College of Radiology data, 1999. 18. R. L. Arenson, C. A. Gooding, and A. R. Margulis, Invest. Radiol., 1993, 28, Suppl. 3. 19. J. A. Barkovich, `Pediatric Neuroimaging', 2nd edn., Raven Press, New York, 1995. 20. J. Kucharczyk, M. Moseley, and J. A. Barkovich, `Magnetic Resonance Neuroimaging', CRC Press, Boca Raton, FL, 1994. 21. S. W. Atlas, `Magnetic Resonance Imaging of the Brain and Spine', 2nd edn, Lippincott-Raven Press, New York, 1996. 22. R. B. Lufkin and W. N. Hanafee, `Pocket Atlas of Head and Neck MRI Anatomy', Raven Press, New York, 1989. 23. R. Kricun and M. E. Kricun, `MRI and CT of the Spine', Raven Press, New York, 1994. 24. M. T. Modic, T. J. Masaryk, and J. S. Ross, `Magnetic Resonance Imaging of the Spine', Mosby Year Book, St Louis, MO, 1993. 25. D. Resnick, `Bone and Joint Imaging', Saunders, Philadelphia, PA, 1989. 26. W. P. Chan, P. Lang, and H. K. Genant, `MRI of the Musculoskeletal System', Saunders, Philadelphia, PA, 1994. 27. D. W. Stoller, `Magnetic Resonance Imaging in Orthopaedics and Sports Medicine', Lippincott, Philadelphia, PA, 1993. 28. C. Langlotz, M. Schnall, and H. Pollack, Radiology, 1995, 194, 645. 29. E. K. Outwater, R. O. Petersen, E. S. Siegelman, L. G. Gomella, C. E. Chernesky, and D. G. Mitchell, Radiology, 1994, 193, 333. 30. J. Kurhanewicz, D. B. Vigneron, H. Hricak, P. Naraan, P. R. Carroll, and S. J. Nelson, Radiology, 1996, 198, 795. 31. H. Hricak and B. M. Carrington, `MRI of the Pelvis: A Text Atlas', Martin Dunitz, London, 1991. 32. A. M. Hubbard, K. E. Fellows, P. M. Weinberg, and M. A. Fogel, Semin. Roent., 1998, 23, 218. 33. J. L. Boxerman, T. J. Mosher, E. R. McVeigh, E. Atalar, J. A. C. Limar, and D. A. Bluemke, RadioGraphics, 1998, 18, 543. 34. C. B. Higgins, `Essentials of Cardiac Radiology and Imaging', Lippincott, Philadelphia, PA, 1992. 35. T. F. Budinger, A. Berson, E. R. McVeigh, R. J. Pettigrew, G. M. Pohost, J. T. Watson, and S. A. Wickline, Radiology, 1998, 208, 573 36. R. A. Huch Boni, J. F. Debatin, and G. P. Prestin, MRI Clinics North Am., 1996, 4, 101. 37. P. Soyer, S. C. deGivry, C. Gueye, S. Lenormand, E. Somveille, and A. Scherrer, Am. J. Roentgenol., 1996, 166, 1115. 38. A. D. P. Lang, and R. N. Gibson, JMRI, 1998, 8, 337. 39. J. P. Heiken, P. J. Weyman, J. K. T. Lee, et al., Radiology, 1989, 171, 47. 40. P. Soyer, M. Levesque, C. Caudron, D. Elias, G. Zeitoun, and A. Roche, J. Comput. Assist. Tomogr., 1993, 17, 67. 41. R. C. Semelka, R. F. Schlund, P. L. Molina, et al., JMRI, 1996, 1, 39. 42. R. M. Gore, I. Laufer, and M. Levine, `Gastrointestinal Radiology', Saunders, Philadelphia, PA, 1994. 43. P. C. Freeny and G. W. Stevenson, `Alimentary Tract Radiology', Mosby Year Book, St Louis, MO, 1994.

44. J. K. T. Lee, H. B. Marcos, and R. C. Semelco, Am. J. Roentgenol., 1998, 170, 1457. 45. E. A. Zerhouni, `CT and MRI of the Thorax', Churchill Livingstone, New York, 1990 46. D. P. Naidich, E. A. Zerhouni, S. S. Siegelman, and J. P. Kuhn, `Computed Tomography and Magnetic Resonance of the Thorax', Raven Press, New York, 1991. 47. J. F. Meaney, J. G. Wegg, T. L. Chenevert, D. Stafford-Johnson, B. H. Hamilton, and M. R. Prince, N. Engl. J. Med., 1997, 336, 1422. 48. J. P. Mugler III, B. Driehuys, J. R. Brookeman, et al., Magn. Reson. Med., 1997, 37, 809. 49. M. B. Williams, E. D. Pisano, M. D. Schnall, and L. L. Fajardo, Radiology, 1998, 206, 297. 50. W. A. Kaiser, `MR Mammography (MRM)', Springer-Verlag, Berlin, 1993. 51. S. G. Orel, M. D. Schnall, R. N. Newman, C. M. Powell, M. H. Torosian, and E. F. Rosato, Radiology, 1994, 193, 97. 52. C. K. Kuhl, A. Elevett, C. C. Leutner, J. Gieseke, E. Pakos, and H. H. Schild, Radiology, 1997, 204, 667. 53. D. Levine, P. D. Barnes, S. Sher, R. C. Semelka, W. Li, C. R. McCardle, S. Worawattanakul, and R. R. Edelman, Radiology, 1998, 206, 549. 54. F. V. Coakley, H. Hricak, R. A. Filly, A. J. Barkovich, and M. R. Harrison, Radiology, 1998, 209(P), 189. 55. D. B. Stafford-Johnson, M. R. Prince, and T. L. Chenevert, Acad. Radiol., 1998, 5, 289. 56. M. R. Prince, MRI Clin. North Am., 1998, 6, 257. 57. E. A. Knopp, Neuroimag. Clin. North Am., 1996, 6, 769. 58. Y. Anzai, M. R. Prince, T. L. Chenevert, J. H. Maki, F. Londy, M. London, and S. J. McLachlan, JMRI, 1997, 7, 209. 59. K. D. Hagspiel, K. Kandarpa, and F. A. Jolesz, JVIR, 1997, 8, 745. 60. N. J. Beauchamp Jr, A. M. Ulug, T. J. Passe, and P. C. M. van Zijl, RadioGraphics, 1998, 18, 1269. 61. K. L. Weiss, R. E. Figueroa, and J. Allison, MRI Clin. North Am., 1998, 6, 95. 62. P. Jezzard, Radiology, 1998, 208, 296. 63. B. R. Buchbinder and G. R. Cosgrove, MRI Clin. North Am., 1998, 6, 67. 64. J. R. Roebuck, K. M. Cecil, M. D. Schnall, and R. E. Lenkinski, Radiology, 1998, 209, 269. 65. M. Castillo, L. Kwock, and S. K. Mukherji, Am. J. Neuroradiol, 1996, 17, 1. 66. W. Schima, A. Mukerjee, and S. Saini, Clin. Radiol., 1996, 51, 235. 67. H. Niendorf, J. Dinger, J. Haustein, I. Cornelius, A. Alhassan, and W. Clauss, Eur. J. Radiol., 1991, 13, 15. 68. R. C. Brasch, Radiology, 1992, 183, 1. 69. V. Runge, `Contrast Media in Magnetic Resonance Imaging: A Clinical Approach', Lippincott, New York, 1992. 70. R. Brasch, `MRI Contrast Enhancement in the Central Nervous System: A Case Study Approach', Raven Press, New York, 1993.

Biographical Sketches Fergus V. Coakley. b 1964. M.B. B.Ch., University College, Cork, Ireland, 1988. Intern, Mercy Hospital, Cork, 1988±89. Resident, Medicine, Mater and St Vincent's Hospitals, Dublin, 1989±91. M.R.C.P.Irel., 1990. Resident, Radiology, Leicester Royal In®rmary, Leicester, 1991±96. F.R.C.R., 1994. Fellow, Body Imaging, Memorial Sloan-Kettering Cancer Center, New York, 1996±97. Visiting Assistant Professor, University of California, San Francisco, 1997±98. Visiting Associate Professor, 1998±present. Over 50 publications. Research specialties: abdominopelvic MRI, including MR cholangiopancreatography, prostate MRI, and fetal MRI.

MRI IN CLINICAL MEDICINE Alexander R. Margulis. b 1921. M.D., Harvard Medical School, 1951, residency in radiology, University of Michigan, Ann Arbor, American Board of Radiology, 1951±54. Assistant Professor, University of Minnesota, Minneapolis, 1959±1963; Assistant, Associate, and Full Professor, Washington University, St. Louis, MO, 1963; Professor & Chairman, Department of Radiology, University of California, San

9

Francisco, 1963±89; Professor of Radiology, Associate Chancellor of Special Projects & Director, Magnetic Resonance Center, 1989±93; Special Consultant to Vice Chancellor & Professor of Radiology, 1993±present. Over 250 publications. Research specialties: gastrointestinal radiology, magnetic resonance imaging in magnetic resonance spectroscopy, and radiologic and health policy issues.

SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS

Sensitivity of Whole Body MRI Experiments David I. Hoult Institute for Biodiagnostics, National Research Council Canada, Winnipeg, MB, Canada

1 INTRODUCTION Surprisingly, the signal-to-noise ratio of a routine, clinical proton NMR image is governed by parameters that are somewhat subjective, and therefore dif®cult to assess. Concomitantly, optimizing the amount of diagnostic information that can be extracted by NMR imaging in a given time is fraught with dif®culty and uncertainty. The subjectivity arises from the juggling of a number of often ill-de®ned variables, and from the fact that the end product of the experiment is not a spectrum, as in conventional NMR spectroscopy, but a picture. Both the variables and the picture are manipulated and interpreted in the light of substantial experience and with all the extra information that such experience brings to bear. Thus, for example, imaging parameters that produce `perfectly acceptable' images from one patient may produce `images with artifacts' from other, and the removal of these artifacts, at the whim (in the eyes of an untutored observer) of the viewer, may entail a reduction of image signal-to-noise ratio, but an increase of information. In the light of dif®culties such as these, and the huge impact that magnetic resonance has had upon the discipline of radiology, the subject of imaging sensitivity is here treated separately from the more general topic of Sensitivity of the NMR Experiment. However, the present discussion builds upon the foundations laid in that article, with its relatively `®rm' conclusions, which the reader is therefore encouraged to read before becoming immersed in the complexities of the present analysisÐnotwithstanding its being only a brief overview aimed at routine imaging with 90 and 180 pulses at ®eld strengths of 1.5 T or less. To summarize the conclusions of Sensitivity of the NMR Experiment, it was shown that, following a 90 rf pulse, the voltage induced in a receiving coil by a small volume's precessing nuclear magnetic moment could easily be calculated, provided the magnetic ®eld that would be created at that volume by unit receiving coil current was known. Thus the signal (the free induction decay or FID) from a volume element (a `voxel') in an imaging experiment can usually be found without dif®culty. However, it was also shown that the random voltages (`noise') present in the receiving coil were somewhat more dif®cult to calculate, being associated with the resistance of the coil conductor and the electrical conductivity of the sample. As human tissues exhibit a range of conductivities, we shall be particularly interested in the latter type of noise, for the Brownian motion of electrolytes in the body induces random voltages in the receiving coil that usually predominate over the voltages associated with Brownian motion of the coil's electrons. Finally, it was also shown that once an NMR protocol was run repeatedly, the accumulated signal became a

1

function of the longitudinal relaxation time T1 of the sample. Thus, our aim in the present article is to extend these results to estimate the sensitivity of the whole body imaging experiment, particularly as a function of magnetic ®eld strength, while introducing those additional factors, such as spatial resolution and power deposition, that are unique to the imaging experiment. Underlying all the arguments to be presented is an analytical technique based upon the principle of superposition. The end result of most imaging experiments is the production of one or more large matrices of data (typically 256  256), where the row and column positions correspond to grid coordinates (u,v) within the sample plane or `slice'. Now such a matrix is most fruitfully examined (at least, by a human) not by scrutiny of its individual numbers (which are functions of amount of hydrogen, relaxation times, etc. at the corresponding spatial position), but rather by causing the numbers to modulate, according to a given law, the brightness of a video display. Once again, the matrix row and column positions correspond to grid coordinatesÐin this instance on the video screen. Thus, via the intermediary of the data matrix, there is a direct correspondence between position on the screen and position in the sampleÐthere is an image, which is much easier to appreciate than a large matrix. Now the content of each matrix element is a measure of the signal emanating from the corresponding spatial position, and, ideally, each signal would come just from a cuboidal volume, the voxel, associated with that position. (In practice, the voxel is not so well de®nedÐit is fuzzy round the edges, and overlaps its neighbors a little.) The process of spatial selectivity that allows the multitude of signals from different parts of a person to be sorted into their volumes of origin, is, of course, the essence of the imaging experiment (see Image Formation Methods). However, because the latter is a linear process, we may take the stunning liberty, right from the beginning of our analysis, of ignoring the vast majority of the signals and concentrating our attention upon the signal from a particular voxel of interest! We may then follow that signal, through all the indignities to which the imaging experiment subjects it, to ®nd the ®nal matrix element associated with the voxel. We may also ®nd how the indignities affect the noise, and thus arrive for our voxel at a measure of its signalto-noise ratio in the image. This then will be our course, with various digressions and comments as necessary.

2

FREE INDUCTION DECAY SIGNAL AND NOISE

With the above philosophy in mind, consider ®rst a free induction decay signal (maybe in the form of an echo) received by a coil wrapped around a portion of the human body, for example, the head.1,2 The signal is from a voxel of volume dV subject to a 90 pulse, at a spatial position de®ned by Cartesian coordinates (u,v,w). Let the proton magnetic moment in the voxel be initially dm = M0 dV z where z is the unit vector in the z direction, and M0 is the spatially averaged equilibrium magnetization hM0(u,v,w)i of the voxel. Using the principle of reciprocity, it may be shown [see equation (5) of Sensitivity of the NMR Experiment] that, following the pulse, the precessing magnetic moment dm induces in the receiving coil a voltage d given by

2 SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS ~ 1xy M0 dV† exp…ÿt=T2 † cos…20 t ‡ † d ˆ 20 …B

…1†

where  0 is the Larmor frequency, 1xy is the ®eld at the voxel, parallel to the (x,y) plane, that would be produced if a current of 1 A at the Larmor frequency were passed through the receiving coil, T2 is the transverse relaxation time of the elementary volume, and  is a phase determined by the imaging protocol. As M0 is proportional to the main ®eld strength B0 [see equation (1) of Sensitivity of the NMR Experiment], the amplitude of the FID is proportional to the square of B0, via the Larmor relationship 2 0 = B0. At the risk of stating the obvious, we emphasize that the signal from a voxel is directly proportional to the volume of the voxel, and therefore proportional to the cube of voxel side, or spatial resolution. This strong dependence is easily forgotten when one is deep in the intricacies of imaging, but can return to haunt one. That little factor-of-2 improvement in spatial resolution can lead to an order-of-magnitude loss of sensitivity! As was shown in Sensitivity of the NMR Experiment, the noise induced in the receiving coil by a conducting sample may also be estimated with the aid of the principle of reciprocity.1,3 Experience shows that with a well-designed head coil, this noise dominates the noise associated with the coil's resistance at frequencies above about 8 MHz, while with a good body coil, the same is true above approximately 4 MHz.4 (Hence, there is little point at the frequencies usually used in imaging, 20±70 MHz, to attempt to cool head or body coils, for the improvement in performance will be negligible.) Some idea of the magnitude of the induced noise may be obtained by ®nding the power that would be deposited in a sphere of the tissue under consideration, by a homogeneous, on-resonance, radiofrequency transmitting ®eld parallel to the (x,y) plane. Assuming this B1 ®eld is produced by an RMS current I in the receiving coil, the power deposited when the wavelength within the sphere is considerally greater than the latter's radius is1,3 8 3 2 2 ~2 W ˆ I 2 Rm ˆ 15  0 I B1xy b5 …0 † 

645 02 12 b5 …0 † 15 2

…2†

where Rm is the equivalent resistance in series with the receiving coil, b is the sphere's radius, is the gyromagnetic ratio,  1 is the nutational frequency of the rf ®eld, and ( 0) is the tissue conductivity. (Note the explicit inclusion of the conductivity's frequency dependence;  is usually of the order of 0.4 S mÿ1 in the NMR frequency range, but can vary considerably.5) The noise in bandwidth
 is now easily found [see equation (9) of Sensitivity of the NMR Experiment] from the effective resistance Rm, and when we use it to ®nd the signalto-noise ratio (S/N) of the FID, we obtain from equations (1) and (2) (

d FID



15 ˆ M0 …0 † dV 8b5 …0 †kB T
   t cos…20 t ‡ †  exp ÿ T2

1=2 )

(radius b) and composition [( 0) and M0( 0)]. Thus, while in no way can a sphere of tissue be considered a good approximation to the human head or torso, we would expect the S/N associated with a voxel in the head to be considerably better than that for a voxel in the torso (say, by a factor of approxi5 mately 1.52 = 2.75), while a body scan of a pregnant patient would have relatively poor sensitivity. We might also expect the direction of the BÄ1 ®eld to in¯uence the S/N when the sample is asymmetric, for the effective value of b then depends on orientation. Further, the advantages and limitations of using a surface coil, with its very small effective value of b, have been discussed in Sensitivity of the NMR Experiment. Thus, in a set of experiments aimed at determining the validity of equation (3) and the magnitude of the term in braces,4 it has been shown that for water in a human head, with a typical receiving coil and a 90 pulse, over the range 5±80 MHz, the intrinsic S/N ratio is " d FIDhead  440

0 dV

#

 10% 1=2

…
†   t  exp ÿ cos…20 t ‡ † T2

while for the torso, with the B1 ®eld from a typical coil running laterally, " d FIDtorso  2100

00:8 dV

#

 60% …
†1=2   t  exp ÿ cos…20 t ‡ † T2

where kB is Boltzmann's constant and T is the sample temperature (310 K). This is a remarkable equation in that all dependence upon the ®eld BÄ1xy has vanished. The sample determines its own `intrinsic' signal-to-noise ratio by its effective size

…5†

Note that  0 is in hertz and dV in cubic meters. On the other hand, with an anterior±posterior B1 ®eld, it was found that the amplitude of equation (5) was roughly halved, bearing out the assertions concerning the shape of the sample. At ®rst sight, it might appear odd that the coef®cient in equation (4) is less than that in equation (5)Ðuntil it is remembered that the equations are only valid in the range 5±80 MHz. However, the interesting facet of equations (4) and (5) is their differing frequency dependences, attributable directly to the conductivity frequency dependences of the various tissues subject to the probe BÄ1 ®eld. Note too the large percentage range allowed in the values: not error in the measurements, but rather the range of values likely to be encountered, given the great variety of the human form. An offshoot of these measurements was an estimate of the power W [equation (2)] needed to produce a given transmitting B1 ®eld for the pulses. The power is deposited overwhelmingly in the patient in the form of heat from rf eddy currents, and general agreement with equation (2) was found. Thus, for the head, W  4:8  10ÿ20 02 12

…3†

…4†

…6†

while for the torso with an anterior±posterior ®eld, W  3  10ÿ21 02:4 12

…7†

We shall encounter the rf power deposited again later in the article, but the reader is correct in assuming that this power is potentially dangerous. It is the only known hazard associated

SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS

with routine imaging. (See Health and Safety Aspects of Human MR Studies).

follows that, since we require the error du <<
u, the gradient must obey the inequality
uGu  H B0 ˆ 3:5  10ÿ6 B0

3 THE FOURIER TRANSFORM The expectation aroused by the con®rmation of theory contained in equations (4) and (5) is that the S/N of a brain image should improve linearly with ®eld strength. Concomitantly,3 we expect the power absorbed in the head of a patient to increase as the square of the ®eld strength for a given nutational frequency (i.e., pulse width and strength). For the torso, we might expect the sensitivity to increase somewhat more modestly, and the power deposited to increase more dramatically. An image's information content, however, is dependent not just upon intrinsic S/N, but on other factors tooÐprimarily the resolution, the chemical shift range H, and the relaxation times T1 and T2, which are functions of frequency. We must therefore consider how FIDs [and, in particular, the FID from our chosen voxel at position (u, v, w)] are utilized in an imaging experiment involving these factors, and it is at this juncture that uncertainty clouds the analysis. With most imaging techniques, each FID or echo in an imaging protocol suffers Fourier transformation as described in Sensitivity of the NMR Experiment, as the ®rst stage in image formation. Now the essence of any imaging experiment (see Image Formation Methods) is a linear correspondence between Larmor frequency  0 and spatial position u during data acquisition: 20 ˆ …B0 ‡ Gu u†

…8†

This correspondence is imposed by the user by impressing a known ®eld gradient Gu (the `read gradient') on the main (homogeneous) ®eld B0. From the value of  0, as determined from the `spectrum' (the Fourier transform of the FID), the position u of our voxel is easily determined. However, suppose there is a small, additional ®eld dB present, caused, for example, by nuclear shielding (a `chemical shift'). From equation (8) we then have uˆ

20 = ÿ B0 ÿ dB Gu

…9†

and there is a positional error of du = ÿdB/Gu in the determined value of u and hence in the position of our signal in the ®nal matrix of data, the image. If that error is more than a fraction of the image's `spatial resolution' in u (i.e., the distance
u equivalent to a shift of one picture element, a voxel side, or a jump of one matrix columnÐthey are all approximately equivalent), it is potentially serious, for it represents a visible and possibly misleading misregistration of a portion of the image. However, to assess the potential for damage, we need to know both the range of any chemical shifts in the portion of the human body being imaged, and also the desired spatial resolution. Now the hydrogen NMR spectrum of a region within a human, though complicated, is sometimes utterly dominated by only a single lineÐthat of water, an ideal situation. Sometimes, however, it is dominated by a spectral structure, associated with both fat and water, covering a chemical shift range of roughly H = 3.5 ppm. (Thus, dB = HB0 & 3.5  10ÿ6 B0.) It

3

…10†

In other words, if both water and fat are present in the portion of the body being imaged, the stronger the static ®eld B0, the greater the read gradient Gu must be if we wish to render `chemical shift artifacts' negligible. If this condition is disobeyed, water and fat signals from the same spatial location will appear at different locations in the image, while water and fat from different spatial locations may appear in the same image location. Thus, voids and overlaps are seen, with the potential for confusion and misdiagnosis, particularly if the viewer is inexperienced. If the only consequence of obeying equation (10) were an increase in engineering dif®culty, it would be worthy of only passing consideration: we should note that extra expense was incurred, while a cynic might suppose that a salesman would therefore make sure his sale pitch would show only images that resulted from water alone. However, the employed gradient strength bears a direct relation to the `acquisition window' (the time for which the FID is collected), to the resolution, and to the available S/N. To understand this statement, recall that in Section 4 of Sensitivity of the NMR Experiment, the S/N following Fourier transformation was optimized by collecting the FID for an optimum time 1.26T2, or, even better, by imposition of an exponential ®lter of time constant T2 prior to transformation. Other ®lters, beyond the scope of this article, may also be advantageously imposed in imaging experiments, but no matter what method is employed to shape and de®ne the data collection period, there is an `uncertainty relation' linking the characteristic duration
t of the timedomain data with the width
 of the ensuing spectral `line'. (`Protuberance' might be a better appellation than `line', since the latter implies smoothness and symmetry.) This relation is of the form

t  1

…11†

(A well-known example, wherein the duration
t is naturally limited, is
 = 1/T2 for a Lorentzian line.) Now the frequency width of the spectral `line' that results from transformation of the FID of our voxel of interest determines the spectral resolutionÐhow close in frequency two `lines' can be while still being distinguishable. (This is the Rayleigh criterion from the theory of optics.) However, because of the synonymy of frequency and distance, courtesy of the imposed gradient Gu, spectral resolution is directly linked to spatial resolution. Thus, 2
 ˆ
uGu

…12†

and 1 2 ˆ

uGu

…13†

2 1 2:86  105 ˆ ˆ

H B0 H 0 0

…14†


t 

Hence, from equation (10),
t 

4 SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS In other words, if chemical shift artifacts are to be avoided, not only must the read gradient increase with ®eld strength, but the effective data acquisition window
t must also become shorter and shorter. Once
t becomes much less than T2, the S/N of the Fourier transform data begins to diminish rapidlyÐwe are throwing away most of the FID signal and just using its beginning. With a typical value of T2 for water in tissue of 50 ms, this transition occurs above approximately 6 MHz. Thus, for example, when imaging with a 1.5 T machine ( 0 = 64 MHz), equation (14) states that the data acquisition window
t should be much less than 4.5 ms. Not surprisingly, the interpretation of `much less than' can range from the pedantic to the ¯agrantly false, depending on the importanceÐcommercial, diagnostic or scienti®cÐplaced upon the image sensitivity and chemical shift artifacts. Concomitantly,
t is a variable of some importance that is frequently overlooked. However, once
t << T2, from equation (3) and (14), and from equation (15) and (16) of Sensitivity of the NMR Experiment, the S/N of the frequency-domain signal becomes 

1=2 15
t 8b5 …0 †kB T  1=2 15 << M0 …0 † dV 8H 0 b5 …0 †kB T

d SP ˆ M0 …0 † dV

…15†

while the resolution in the image becomes independent of variations in T2: a redeeming feature of the condition. To summarize, we may conclude that for a head experiment, following a single 90 pulse, if chemical shift artifacts are to be avoided, the S/N ratio following Fourier transformation is d SPhead ˆ 4400 dV…
t†1=2  10%  1=2 0 10% << 440 dV H 1=2

ˆ 2:4  105 dV 0

 10%

…16†

Thus, the price paid for restricting the acquisition window to render negligible the chemical shift artifact, is that the sensitivity increases only as the square root of ®eld strength. For the torso, the rate of increase is even less, being approximately as  00.4. To recoup this loss and ful®l our initial expectations, Hahn echo formation, using a number of 180 pulses, is needed. Images formed from each echo of a Carr±Purcell sequence may then be coadded [with weighting as exp (ÿt/T2)] to recover the lost sensitivity, and, as an added bonus, T2 information may also be obtained. However, as we shall see, the power deposition in the patient is then substantially increased, and, further, the use of the sequence in a multislice experiment is not trivial. Other possible strategies include the use of chemical shift imaging techniques to give separate fat and water images (see Chemical Shift Imaging), or presaturation of either water or fat magnetization to remove its in¯uence. However, chemical shift methods at least double the time taken to get an image, while presaturation demands excellent B0 ®eld homogeneity.

4 4.1

FROM TRANSFORM TO IMAGE Signal-to-Noise Ratio

One FID does not an image make, and, with the exception of certain fast imaging methods (see Echo-Planar Imaging), data must usually be garnered from a prescribed number of FID's or echoes. Typically, that number is related to one of the dimensions of the image matrix (e.g., 256), but we need not inquire too closely into the detailsÐthe important point for our analysis is that a minimum number of repetitions is required. (The protocol may then be repeated, and images effectively coadded if better S/N is wanted.) This subject had been examined in some detail in Sensitivity of the NMR Experiment, and Figure 5 of that article shows a contour plot of relative sensitivity versus ¯ip angle and repetition rate. (For convenience, the plot is included in Figure 1 of the present article.) However, 90 pulses are mostly used in imaging protocols, and it may be seen that with a repetition period of 1.26T1, only about a 10% loss in sensitivity is suffered thereby. However, if T1 changes, the length of time taken to collect the 256 FIDs must also change if we wish to maintain sensitivity. In short, the T1 of the region of interest in the patient is the gold standard by which the repetition period TR must be judged. It follows that if we are to further our examination of signal-to-noise ratio, we must know how T1 varies with tissue type and frequency. There are two reviews of the pertinent literature,6,7 and each shows that in the frequency range 1±100 MHz, T1 for most tissues can be modeled in the form T1  0

…17†

where the constants  and  are contained in brief summary in Table 1. An inescapable conclusion, then, is that as we increase the ®eld strength, the minimum time taken to build an image increases as per equation (17) unless the imaging protocol is changed for some reason. The gains reaped as this increase in imaging time is suffered are that (i) the expected increase in S/N as  10| 2 (or as  0 if chemical shift artifacts are ignored or multiple echoes are used) is realized, and (ii) as the ratio T1/
t increases with increasing ®eld, so multiple slice methods gain in their ability to produce more cuts. (This improved ability is usually only of consequence at the lowest ®elds, and we do not consider it further.) However, in all fairness, we must ask how much the S/N of a low-®eld image could be improved if we were to allow for its acquisition the same time taken, perforce, at high ®elds. For example, if the minimum time Tmin we are forced to take to acquire a scan at Table 1 Tissue Longitudinal Relaxation Times T1 (&   0 )a Tissue Adipose Kidney Gray matter White matter Heart muscle Liver Skeletal Muscle a





0.012 0.007 3 0.003 6 0.001 5 0.001 3 0.000 52 0.000 45

0.17 0.25 0.31 0.35 0.36 0.38 0.42

Derived from Bottomley et al.;6 T1 in s and  0 in Hz.

SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS

5

10 0.1

8 7 6 5 4

0.2

0.3

0.4

0.5 0.6 –0.1

3

0.8

–0.4

2 0.9 TR/T1

–0.2

–0.6 –0.8

1 0.8 0.7 0.6 0.5 0.4

–0.95

0.8

–0.9

–0.8

0.3 0.6 0.2

–0.7

0.1 0°

20°

40°

60°

80°

100°

Flip angle q

Figure 1 Normalized signal-to-noise ratio and relaxational contrast-to-noise ratio q /q T1 in a repeated FID image as functions of effective ¯ip angle  and repetition period TR. The effects of any pulse used to produce an echo are neglected, and the plot is for a constant imaging time rather than a constant number of pulses. The C/N contours are the black lines, with values in bold print; the S/N contours are white lines, with values in plain print. The topologies of the two functions are similar, and it is possible to obtain simultaneous high absolute values. However, the plots are for constant hydrogen concentration, and therefore must be employed with care when there is a correlation between concentration (water content) and relaxation time T1: q /q M0 and q /q T1 have opposite signs and tend to cancel

80 MHz is four times longer (thanks to the frequency dependence of T1) than that taken to acquire an equivalent image at 5 MHz then, by coadding four 5 MHz images (thereby also taking time Tmin), the S/N of the 5 MHz picture could be doubled. In short, a factor of T11| 2 has to be included in any assessment of image signal-to-noise ratios. Thus in a simple, S/N-optimized, 90 pulse, repeated FID imaging experiment (sometimes inaccurately termed a `saturation±recovery' experiment) that avoids chemical shift artifacts and does not employ a Carr±Purcell multiple echo sequence, the head image S/N varies for ®xed imaging time T as  d Ihead  3500 dV

T
t T1

1=2

10%

…1ÿ†

T0 << 350 dV H 

!1=2 10%

…18†

Thus, taking gray matter with 80% water as an example, d Igray << 2:5  106 dV 00:35 T 1=2  10%

…19†

and in 4 min at 64 MHz, an image with an S/N of 40 : 1 can have a voxel size of about 5 l, or in other words a resolution of about 1 mm in a 5 mm thick slice. Summarizing, we have established the credentials of NMR imagingÐresolutions of the order of 1 mm in 5 mm thick slices in times of the order of a minute or two. However, lest other

credentials be sought (`Why not 1 m in 30 s?'), the relationships in equations (18) and (19) are worthy of note. 1. We remarked earlier that the linear resolution
u in an image, if all three dimensions are similarly scaled, is proportional to the cube root of the volume. Thus the time taken to attain a given S/N (whatever may be neededÐnormally in the range 16 :1±64 : 1) is proportional to the inverse sixth power of the linear resolution
u. It follows that if the S/N in an image is marginal, any idea of increasing resolution by imaging for longer should be promptly abandoned: the return is almost worthless. Keeping the resolution constant and imaging for longer (effectively coadding images) to improve the sensitivity is a reasonable course (improvement as the square root of time), but the overwhelming temptation must be to suffer the chemical shift artifacts and increase the data acquisition window! 2. The relationship between spatial resolution and ®eld strength for constant S/N and imaging time is also weak, since
u !  0( ÿ 1)/6. For example, for gray matter, a 10-fold increase in ®eld strength affords only a 25% improvement in linear resolution, unless one can be absolutely certain that only water is being observed. 3. Looking at the relationship between ®eld and time for constant S/N and resolution, we see that T !  0( ÿ 1), and as the ®eld is increased, so the time needed to gather and coadd multiple images decreases fairly rapidlyÐfewer additions are needed. However, once only a few additions are

6 SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS needed, the use of imaging time T in the above arguments and equations is, in truth, a little arti®cial, since T must be granularÐcorresponding to say 128, 256, 2048, etc. FIDs, depending on the desired resolution and number of repetitions. Thus, as the ®eld strength is increased and T diminishes, there will come a point at which one run of the imaging sequence, with its ®xed number of FIDs, suf®ces. Thereafter, further increase in ®eld strength serves to increase the time to collect that ®xed number (thanks to the frequency dependence of T1) while admittedly improving S/N. Perhaps the overriding message of equations (18) and (19) is that the richest dividends come from innovative and minute attention to every detail of rf design. For example, a change of preampli®er from one with a noise ®gure of 4 dB to one with a noise ®gure of 1 dB, coupled with a switch to quadrature receiving coils, can improve linear resolution by 25%. It is perhaps for this reason that the literature is full of special coil designs for every conceivable part of the human body, even though there are, for the most part, no new design principles enunciated. Finally, there are those who would image other nuclei. The overwhelming obstacle to this desire is the paucity of elements other than hydrogen in the body. (Carbon-12 and oxygen-16 have no nuclear magnetic moments.) For example, the magnetization of metabolic phosphorus is less than 10ÿ4 that of the body's hydrogen, leading very roughly from equation (19), to a voxel size in the head of the order of 30 ml at 64 MHz (3.7 T). Such a size is incompatible with the formation of a recognizable image. 4.2 Contrast-to-Noise Ratio A common aspect of imaging is the desire to impart to the data a strong dependence on either relaxation time T1 or T2, so that tissues with similar water content but different relaxation behavior can easily be distinguished. In this context, we draw a sharp distinction between absolute and relative contrast. The former relates to the difference in signal strength between two regions of differing relaxation, and is here our main concern. One's ability to distinguish that difference depends on its size relative to that of the noise in the image. However, it may be dif®cult to view the difference adequately in the context of the full image, since careful `windowing' may be neededÐthe image brightness must go from dark black to full white over a small range of signal strengths, thereby blacking- or whitingout detail in the rest of the picture. The relative contrast is simply the ratio of two signals, without consideration of the noise. If the relative contrast is large and the S/N is reasonable, there are no windowing problems, but only gross changes in relaxation behavior can be viewed. We begin by considering absolute T1 contrast, a measure of which for repeated FID imaging can be obtained by differentiation of equation (18) with respect to T1. Figure 1 is a normalized contour plot of absolute T1 contrast-to-noise ratio (C/N) q /qT1 versus ¯ip angle and repetition period for a ®xed imaging time T. Note that the contrast is negative, since a longer T1 implies a greater degree of saturation and therefore a reduction in signal. It may be seen that for a 90 pulse, the point of maximum contrast (at TR = 0.50 T1) is somewhat removed from that of maximum sensitivity at TR = 1.26 T1. The relative values of sensitivity and C/N at these two rep-

Table 2 Normalized Maximum and Compromise Signal-to-Noise and Contrast-to-Noise Ratios in a Repeated FID Imagea Experiment

TR/T1

S/N

C/N

b

0.5 0.81 1.26

0.78 0.87 0.90*

ÿ0.92* ÿ0.87 ÿ0.50

`High-®eld'c

1 1.75 ?

0.63 0.83 1*

ÿ1* ÿ0.83 0

`Low-®eld'

a

For a 90 pulse. `Low-®eld' values are normalized to the maxima of the absolute values obtainable with a variable ¯ip angle pulse. Contrast is negative because increasing T1 results in decreasing signal. b Imaging for a constant time, by adjustment of the number of repetitions of the imaging protocol. Usually applicable at ®elds of about 0.1 T. c Imaging for a constant number of pulses, regardless of the time taken. Usually applicable at ®elds of about 1.5 T. *Maxima of absolute values.

etition periods are shown in Table 2, and it may be seen that at the compromise value of TR = 0.81T1, 87% of both the maximum possible S/N and C/N are obtained. This percentage is only a few points less then the maximum possible with a 90 pulse. As when considering sensitivity, the overall best T1 contrast is obtained when repeating as fast as possible, albeit with a small ¯ip angle  obeying the equations exp…TR=T1 † ˆ

2 ‡ cos  ; 1 ‡ 2 cos 

cos  ˆ

2 exp…ÿTR=T1 † ÿ 1 2 ÿ exp…ÿTR=T1 †

…20†

While normal imaging methods employ 90 and 180 pulses, fast imaging techniques can take advantage of such low-¯ipangle pulses. When the C/N in an image is so good that multiple runs of the imaging protocol are not required, the optimum conditions are somewhat different. With a 90 pulse, the best T1 contrast, as shown in Table 2, is obtained with TR = T1, but then only 63% of the maximum sensitivity is garnered. A reasonable compromise is TR = 1.75T1. An unfortunate aspect of imparting T1 contrast by partial saturation is that there is often a correlation between tissue water content and T1: the more water, the longer is T1. The two variables then work in opposition, the greater saturation of the longer lived species tending to reduce its larger signal, thereby reducing contrast by the tendency to equate signal strengths. In these circumstances, an `inversion±recovery' pulse sequence (see Image Formation Methods) is often employed, but, as shown in Figure 2, a considerable price is paid in both C/N and S/N. The lure of the inversion±recovery method is that the relative contrast (the ratio of the signal strengths of two species) can be very large, and thus the image is easily viewed. However, since both positive and negative signals can exist, a magnitude image can often not be used, and this can be both a considerable inconvenience and a trap for the unwary. If we wish to determine how the contrast varies with ®eld strength, we must have excellent knowledge of the frequency dependence of each T1 of interest, and such data may be unavailable. Using the values for brain gray and white matter in

SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS 10 8 7 6 5

–0.3

–0.4

–0.2

7

–0.1 0

–0.5 –0.6

4

–0.7

3

0.1

–0.8

TR/T1

2

0.2 1 0.8 0.7 0.6 0.5

0.3 0.4

0.4

0.5

0.8 0.6

0.3 0.7 0.2

0.7 0.6 0.5

0

0.4

0.3

0.2

0.1

–0.2

0.03

0.04

0.06

0.08 0.1 Inversion time/T1

–0.4

–0.6

–0.8

0.1 0.01

0.02

0.2

0.3

0.4

0.6

0.8

1

Figure 2 Normalized signal-to-noise ratio and relaxational contrast-to-noise ratio q /q T1 in an inversion±recovery image as a function of inversion time (the time between the inverting 180 pulse and the following, signal-generating 90 pulse) and of recovery time TR (the time between the 90 pulse and the next 180 inverting pulse). The functions are normalized to the maximum values attainable under the conditions pertinent to Figure 1, the effects of any pulse used to produce an echo are neglected, and the plot is for a constant imaging time rather than a constant number of pulses. The C/N contours are the black lines, with values in bold print; the S/N contours are white lines, with values in plain print. The area with the lighter shading is where a positive correlation between hydrogen content and longitudinal relaxation time enhances T1 contrast. It may be noted that the maximum absolute contrast in this area is obtained where the signal is weakest. Thus, the relative contrast of the inversion±recovery method can be very large. However, there is a considerable price to be paid, as may be seen from the values of the contours

Table 1, the absolute C/N is easily calculated at any frequency, but it is found that in going from 21 to 64 MHz (0.5 to 1.5 T), the increase in absolute contrast is only 12%. Again, it is stressed that this result is for the same imaging time at the two ®eld strengths, and that chemical shift artifacts have been suppressed by an appropriate choice of data acquisition window. Turning to T2 contrast, dependence is easily imparted by generating an echo from the FID; indeed, most modern imaging methods use an echo anyway, since it helps to deal with a number of phasing problems. If we assume, as is usually the case, that T1 >> T2, the repetition rate has almost no bearing on T2 contrast. It may then be shown that to excellent accuracy for
t 4 1.5 T2, the maximum absolute contrast is obtained when the acquisition window
t is centered on echo time TE = T2. However, this positioning can entail a loss of up to 63% in image sensitivity when
t << T2. On the other hand, the maximum relative contrast is obtained when the echo time is as long as possible, and the S/N in the image is then very poor indeed. The frequency dependence of T2 (which is normally quite small) has little effect on the calculation once
t << T2. However, if multiple echoes are accumulated with an optimum ®lter, or if chemical shift artifacts are neglected, what little dependence there is, is re¯ected directly in the image sensitivity and contrast.

5

POWER DEPOSITION

In the preceding paragraphs, the discussion of the frequency dependence of an image's sensitivity and contrast has been dominated by the possible effects of chemical shift artifacts and the changing longitudinal relaxation time. However, there is an additional aspect to imaging that we dare not ignore, and that is the power deposited in the patient by the rf pulses. While a healthy individual can safely absorb 1 kW of infrared radiation for many minutes, our concern must be with a patient who may have cardiovascular or thermoregulatory problems. Further, it is obvious that the eddy currents induced in the body by the transmitter's B1 ®eld may be forced, at certain parts of the body (e.g., the shoulder blades), to ¯ow through relatively thin layers of tissue, thereby generating so-called `hot spots'. How the body handles such insults is not clear, but, in the context of this article, it is clear that there has to be a limitÐand indeed, government agencies advise on the issue.8,9 To attempt an assessment of the issue's seriousness, we must look at the factors that govern the rf pulses used in imaging. It is obvious that they must have a duration that is much less than T2, but they must also be powerful enough to cover the required bandwidth. While there may be problems with

8 SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS main ®eld inhomogeneity that require an increase in pulse bandwidth, such dif®culties can usually be solved by the application of moneyÐno amount of money can change the fundamental limit that is determined by the chemical shift range H. If a composite image of water and fat is to be obtained, the exciting ®eld must obey the inequality 1  H 0

…21†

where  1 is once again the nutational frequency and  0 is the Larmor frequency. Needless to say, the meaning of `very much greater than' is open to a wide variety of interpretations, and is also dependent on the type of pulse used. However, substitution into equations (6) and (7) to ®nd the instantaneous power needed to produce the necessary ®eld in the human body gives 2 4 Whead  4:8  10ÿ20 H 0 ;

2 4:4 Wbody  3  10ÿ21 H 0

…22†

As usual, it is helpful to insert numbers, since with them we obtain the power speci®cations of the transmitter. Setting H = 3.5  10ÿ6, we have at 10 MHz, Whead >> 6 mW and Wbody >> 230 mW, whereas at 64 MHz, Whead >> 10 W and Wbody >> 820 W. The very rapid increase with Larmor frequency demonstrated in these numbers, and in equation (22), shows that rf engineering problems become immense and expensive at ®elds not much larger than 1.5 T if composite imaging of water and fat is still to be entertained. Indeed, a typical speci®cation for a transmitter at 64 MHz is already 10 kW. Turning to the energy absorbed by the human body, we have already mentioned that 180 pulses are commonly used in imaging protocols. It is usually these pulses, rather than 90 pulses, that deposit the most energy in the patient. Now, with a rectangular envelope, the duration of such a pulse is (2 1)ÿ1, and so the energy absorbed from each pulse is Ehead ˆ 2:4  10ÿ20 "H 03 ;

Ebody ˆ 1:5  10ÿ21 "H 03:4

…23†

where " (>> 1) is a factor that takes account of the inequality in equation (21). In practice, pulses are often selective, which increases the energy a little, but this minor uncertainty is lost amid the greater uncertainty surrounding the factor ", the size of the patient, etc. Now the recommendations of most governments concerning absorbed power P are related, among other factors, to the resting human metabolic rate (of order 100 W). (There are also recommendations concerning localized power depositionÐthe so-called `speci®c absorption rate', SAR.) It follows from equation (23) that the number of 180 pulses per second that may be safely applied is R180head ˆ

4:2  1019 P ; "H 03

R180body ˆ

6:7  1020 P "H 03:4

…24†

For example, consider imaging at 80 MHz on the torso. Let P = 30 W, " = 3, and H = 3.5  10ÿ6. Then R180body = 2.6, which is highly restrictive. Such a value would prohibit, for example, a single echo, multislice experiment. We may manipulate the factor " in equation (24) or justify to ourselves a little more power, but the overwhelming dependence in the equation is upon Larmor frequency. A small lowering of frequency makes a large difference to the ¯exibility and versatility of the equipment. Turning this statement about, even if our estimates

of power deposition are considerably in error or we manage to invent a pulse where " can be honestly as low as 1, it will make little difference to the maximum frequency to which we can aspire. Put plainly, `high-®eld' imaging with large-angle pulses presents serious problems for any method that aims to excite the whole chemical shift range encountered in the human body.

6

CONCLUSIONS

In any attempt to assess the imaging experiment, and in particular, the composite imaging of fat and water, which is the mainstay of clinical imaging practice, some objective criterion of usefulness is neededÐa commodity hard to come by, for signal-to-noise ratio alone does not suf®ce. One possible approach to this dilemma is to realize that, with each 180 pulse used for echo formation, there is, in general, an associated and useful image. Thus, the `information' potentially available to the radiologist can, to some degree, be measured by the number of 180 pulses that can be gainfully applied. We have already encountered an upper limit on that number as ®eld strength increases, equation (24), caused by power deposition problems. However, there is another limit, as ®eld decreases, caused by the fact that we presumably do not wish to apply 180 pulses more frequently than one every data acquisition period
t. Thus, from equation (14), the maximum average rate at which they are likely to be applied is R180 max ˆ H 0

…25†

where the factor  (< 1) allows for slower running, to accommodate the phase-encoding period, for example. To gather data at the maximum rate, a multislice, multiecho experiment would almost certainly be required. However, we note that the limits implied by equation (24) and (25) intersect at a frequency given by  P 0:25 2 "H   P 0:227 ˆ 5:4  104 2 "H

0head ˆ 8:1  104 0body



…26†

Below these frequencies, the rate of application of 180 pulses is limited by the long data acquisition window; and above, by the allowable power deposition. For example, with P = 30 W, " = 3, and H = 3.5  10ÿ6, as previously, and with  = 0.5, we have for the torso an intersection frequency of 32 MHz, allowing about ®fty-six 180 pulses per second. These pulses could be used, for example, to obtain eight-slice, seven-echo images (if a slice repetition rate of one per second were used), thereby giving full T2 information for each slice, and by weighted, coaddition of images, the full S/N unhampered by chemical shift artifacts. On the other hand, at both lower and higher frequencies, the maximum rate of application would be less. The crux of the argument being presented here is that as ®eld strength is increased in a single slice experiment, so a point slowly arrives at which resolution and S/N are those desired (without any need for repetition of the imaging protocol), and further ®eld increases give little in return while

SENSITIVITY OF WHOLE BODY MRI EXPERIMENTS

consuming some extra time. However, for a multislice experiment (and even more so for multislice, multiecho techniques), there comes a point at which power dissipation problems rapidly impose severe constraints, and versatility and ¯exibility are quickly hampered. It is a little like taking a walk over a gentle hill and suddenly ®nding a precipice on the other side! The reason for this behavior is clearly contained in equation (26): the frequency at which it occurs can be in¯uenced but little by changing the allowable power deposition P, or by fudging one's notions of what are acceptable chemical shift artifacts, for these factors are raised to a small powerÐabout 0.25. The numerical constants in the equations totally dominate, and they in their turn are dominated by the effective radius and conductivity of the sample. Thus, for the head, it may well be that a frequency of 70 MHz might maximize the information available; while for the torso, about 30 MHz might be best. As the novelty of NMR imaging has worn off, and it has become a routine clinical tool, it has become clear that the bulk of the market has, in fact, settled for machines in the frequency range mentioned. Neuroradiologists tend to prefer the higher ®eld strengths, for example, 1.5 T (64 MHz); many average radiology departments have settled for a strength of 0.5 T (21 MHz). The technology is reasonably mature, and its limits of resolution and sensitivity are understood, and have been, for the most part, reached. Having said this, the application of the technique is, at the time of writing, still expanding. However, it is salutary to think that if the amount of electrolytes in the body were considerably greater (thereby greatly increasing conductivity), or if hydrogen, like oxygen and carbon, had no nuclear spin, NMR imaging would have remained a curiosity.

7 RELATED ARTICLES Chemical Shift Imaging; Image Formation Methods; EchoPlanar Imaging; Health and Safety Aspects of Human MR Studies; Sensitivity of the NMR Experiment.

8

9

REFERENCES

1. C.-N. Chen and D. I. Hoult, `Biomedical Magnetic Resonance Technology', Adam Hilger, Bristol, 1989. 2. D. I. Hoult and R. E. Richards, J. Magn. Reson., 1976, 24, 71. 3. D. I. Hoult and P. C. Lauterbur, J. Magn. Reson., 1979, 34, 425. 4. C-N. Chen, V. J. Sank, S. M. Cohen, and D. I. Hoult, Magn. Reson. Med., 1986, 3, 722. 5. C. H. Durney, C. C. Johnson, P. W. Barber, H. Massoudi, M. F. Iskander, J. L. Lords, O. K. Ryser, S. J. Allen, and J. C. Mitchell, `Radiofrequency Radiation Dosimetry Handbook', 2nd edn., USAF School of Aerospace Medicine, Brooks Air Force Base, TX, 1978. 6. P. A. Bottomley, T. H. Foster, R. E. Argersinger, and L. M. Pfeifer, Med. Phys., 1984, 11, 425. 7. P. T. Beall, S. R. Amtey, and S. R. Kasturi, `NMR Handbook for Biomedical Applications', Pergamon Press, New York, 1984. 8. National Radiological Protection Board, Br. J. Radiol., 1983, 56, 974. 9. Of®ce of Radiological Health, `Guidelines for Evaluating Electromagnetic Risk for Trials of Clinical NMR Systems', US Food and Drug Administration, Rockville, MD, Communications of February 25 and December 28, 1982.

Acknowledgements The author gratefully acknowledges the hospitality of the In Vivo NMR Center, University of Utrecht.

Biographical Sketch David I. Hoult b. 1945; B.A. (Physics) 1968. D.Phil. (supervisor Sir Rex Richards F.R.S.) 1974, both University of Oxford, UK. Biochemistry Department, Oxford University, 1974±1977. National Institutes of Health, Bethesda, MD, 1977±1982. Head, NMR Instrumentation Group, NIH, Bethesda, MD, 1982±1991. University of Utrecht, The Netherlands, 1991±1994. Institute for Biodiagnostics, NRC, Winnipeg, 1994±present. Approx. 70 publications, and with C-N Chen, the book `Biomedical Magnetic Resonance Technology'. 7 awards including Gold Medal, Society of Magnetic Resonance in Medicine, 1985. Research interests: all aspects of instrumentation (mainly NMR), mostly in the biological and medical ®elds.

Whole Body MRI: Strategies Designed to Improve Patient Throughput

unavoidable penalty in signal-to-noise ratio (S/N). Since the minimum tolerable S/N is usually dictated by diagnostic criteria such as lesion detectability, increases in temporal resolution will usually have to be traded against spatial resolution. This article briefly reviews the concepts underlying the most important classes of rapid imaging techniques, each of which is discussed in more detail in other articles in this Encyclopedia, referenced later. For a simple introduction to rapid MRI imaging techniques, the reader is referred to Wehrli.7

Felix W. Wehrli University of Pennsylvania Medical School, Philadelphia, PA, USA

2 k -SPACE FORMALISM

1 2 3 4 5 6 7 8

Introduction k -Space Formalism Classification of Imaging Techniques Various Embodiments of Spin Warp Imaging Multiline Scanning Techniques Other Methods of Scanning k -Space Related Articles References

1 1 2 2 5 7 7 8

We shall in the following resort to the classical k-space formalism8 (see also Image Formation Methods). In brief, data sampling occurs in reciprocal space, referred to as ‘kspace’ or the ‘spatial frequency domain’. For an object of spin density ρ(r), ignoring relaxation, the spatial frequency signal ρ
(k) is given by ρ
(k) =




r2

ρ(r)e−ik(t) · r d3 r

(1)

r1

Here k(t) is the spatial frequency vector, expressed in units of rad cm−1 : 1

INTRODUCTION




t

k(t) = γ

G(t
) dt


(2)

0

Since the inception of clinical MRI at the beginning of the 1980s, the rate at which a subject can be scanned—often denoted ‘patient throughput’—has been reduced by nearly an order of magnitude. Some of the improved scan efficiency is unrelated to a shortening of the actual data acquisition time, and can be accounted for by improvements in some of the ancillary procedures such as patient set-up, rf coil placement, scan parameter entry, prescan (MRI jargon for rf carrier selection and adjustment of transmit and receive gain), and advances in digital system architecture (shortening in reconstruction times, concurrence of data acquisition and reconstruction, etc.). The focus of this article, however, will be on techniques for reducing scan time, since it is this that determines patient tolerance. Scan times in excess of 5–10 min typically lead to image degradation from subject motion. Another strong motivation for shortening scan time is physiologic motion such as cardiac pulsation, respiration, and peristalsis. Clearly, if the data acquisition time could be lowered below the period of physiologic motion, the effect of motion unsharpness in the image data could effectively be suppressed. Many of the new applications, such as the assessment of organ function by monitoring the passage of a bolus of contrast agent1 or the study of brain function by recording the response to stimuli,2 – 4 demand temporal resolution on the order of 0.1–1 s (see also Contrast Agents in Whole Body Magnetic Resonance: An Overview). While many of the ideas for improved scan efficiency have been in existence for a decade or longer, they have become practical only recently, owing to engineering advances such as digital transceiver electronics (see also Whole Body Magnetic Resonance Spectrometers: All-Digital Transmit/Receive Systems) and gradient technology (see Gradient Coil Systems). A case in point is echo planar imaging (EPI), which was first conceived in 1977,5 but has only recently become practical.6 Finally, an increased data sampling rate typically exacts an

where G(t
) is the time-dependent spatial encoding gradient in vector format. Pictorially, the spatial frequency may be regarded as the phase rotation per unit length of the object experienced by the magnetization after being exposed to a gradient G(t
) for some period t. We further recognize from equation (1) that ρ
(k) and ρ(r) form a Fourier transform pair, and so
ρ(r) =

k2

ρ
(k)eik(t)·r d3 k

(3)

k1

The path of the spatial frequency vector during execution of the pulse sequence is determined by the time-dependent gradient waveforms. Since k-space is sampled discretely, sampling needs to satisfy the Nyquist criterion 1 2 Li ki,max

= 12 Ni π

(4)

where k i,max is the highest spatial frequency, Li and N i represent the field-of-view and number of data samples, respectively, and i = x , y, z . We can then write for the sampling interval k i ki =

ki max 1 2 Ni

=

2π 2π = Li Ni Li

(5)

with Li representing the pixel size. The factor of 2 in equation (5) arises because we collect N i samples from −k max to +k max or 12 N i samples from 0 to k max . Since the data sampling time increases with the number of data samples, one obvious approach toward shortening the sampling time is to reduce N i . It is readily seen from equation (5) that this can be achieved in two different ways. If we wish to maintain the pixel size, this will result in a reduction in the

2 WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT field-of-view and thus an increase in k i . Conversely, we may want to maintain the field-of-view, in which case k i remains invariant but k max is lowered, and thus pixel size is increased. In the latter case, we trade spatial resolution for a reduction in sampling time. Further, since k ≈ γ t G, the sampling interval t scales inversely with the amplitude of the spatial encoding gradients. Therefore, the duration of the spatial encoding process is governed significantly by the achievable gradient amplitudes. At a given resolution, the latter essentially determine what fraction of k-space can be covered following a single excitation (since the signal decays with time constant T 2 ). Therefore, the properties of the gradients, notably their amplitudes, and for many imaging sequences also their slew rates, are paramount in determining ultimate imaging speed.

3

CLASSIFICATION OF IMAGING TECHNIQUES

Most of the currently used imaging techniques use rectilinear k-space sampling, and as such may be regarded as descendants of the spin warp technique of Edelstein et al.,9 which itself is an outgrowth of Fourier zeugmatography, pioneered by Kumar, Welti, and Ernst.10 Since the latter is an embodiment of two-dimensional NMR spectroscopy,11 conceptionally all rectilinear sampling techniques may be regarded as originating from 2D NMR. By contrast, projection–reconstruction imaging12 – 14 or imaging with nonperiodic time-varying gradients15 such as spiral scanning16,17 are not rectilinear sampling techniques, since the k-space path is non-Cartesian. Common to all rectilinear k-space sampling techniques are two distinct phases of spatial encoding, a first period during which a gradient G y is applied whose time integral determines k y and during which no sampling occurs, followed by a second period during which a gradient G x is applied while the free precession signal is sampled, resulting in a line k x . The former is typically called ‘phase encoding’ and the latter ‘frequency encoding’. In this manner, a 2D k-space grid is sampled by incrementally stepping the amplitude of the phase-encoding gradient. The concept can be extended to a third dimension by phase encoding along a second spatial coordinate to afford k z , in which case k-space is three-dimensional. In both backprojection and spiral scanning, two or all three spatial encoding gradients are simultaneously active while data are sampled. A generic 3D spin warp pulse sequence and k-space map are shown in Figure 1.

4

4.1

VARIOUS EMBODIMENTS OF SPIN WARP IMAGING Scan Time, Signal-to-Noise Ratio and Spatial Resolution

Except in those situations where k-space can be scanned in a period on the order of T 2 , such as in EPI5 or multiple pulse techniques,18 resulting in a series of consecutive echoes that can be individually encoded, only a fraction of k-space is typically scanned upon creating transverse magnetization, and the process repeated several times, each cycle affording another k-space segment. Longitudinal magnetization builds

t rf

Gy

Gz

Gx

(a)

kz

ky

kx k(t)

(b)

Figure 1 (a) Generic 3D spin warp imaging with phase encoding on the x and y axes and frequency encoding on z . Hatched areas indicate the spatial frequency vector at time t (b). Amplitudes of phaseencoding gradients are arbitrary. During readout, a line k x is scanned corresponding to intersection between the (k x ,k y ) and (k x ,k z ) planes (shaded areas)

up in a recurrent fashion between cycles of period TR (also called ‘pulse repetition time’). The data acquisition time T ac thus scales with TR and the number of pulse sequence cycles, the latter being a function of the number of spatial encoding steps or k-space samples. It is customary to denote the frequency-encoding axis by x and the phase-encoding axis by y (or y and z , respectively, if phase encoding in an additional dimension is effected), with N i (i = x , y, z ) representing the number of data samples. Assuming further that during a frequency-encoding period a single line k x,i=1,...Nx is sampled, T ac to sample k-space is given by Tac = Ny Nz nT R

(6)

where n represents the number of sequence repetitions (also called ‘number of excitations’ in imaging jargon) at given values of k. At typical sampling rates, the sampling time for frequency encoding is on the order of 10 ms or less, and so does not appear in equation (6). It will, however, become critical in high-speed gradient echo imaging19,20 and EPI and its derivatives.21 – 23

WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT

The scan time scales with the number of phase encodings (N y in 2D, and N y and N z in 3D spin warp imaging), and thus a reduction in the number of phase-encoding samples entails a concomitant reduction in scan time. As previously pointed out, this can be achieved in two different ways: either by lowering k max or decreasing sampling density 1/k . The former lowers resolution, the latter leads to reduced fieldof-view while preserving resolution [cf. equation (4) and (5)]. Thus, the term ‘matrix size’ is ambiguous, though in common parlance a change in matrix implicitly implies fixed field-of-view. The two different situations, each lowering data acquisition time of a 3D image acquisition by a factor of 4, are illustrated in Figure 2. Let us now examine the S/N implications of the two different operations. The signal amplitude S scales with voxel size, i.e., S ∝ x y z =

Lx Ly Lz Nx Ny Nz

3

where the product x y z expresses the voxel volume, the remaining parameters have previously been defined. Further, the noise-reducing effect of sampling leads to S/N ∝



Nx Ny Nz n

(8)

√ Lx Ly Lz n S/N ∝  Nx Ny Nz

(9)

and thus

From equation (9), we conclude that halving the number of data samples N y and N z by halving k y,max and k z,max , increases S/N by a factor of 2 (Figure 2a). Skipping alternate data samples, on the other hand, penalizes S/N by a factor of 8 since this operation halves the field of view for both Ly and Lz [equation (5); Figure 2b].

(7)

4.2 Slice Multiplexing and Multiecho Acquisition kz (a)

ky

kz (b)

In most embodiments of 2D spin warp imaging, the period comprising slice selection, phase encoding, and frequency encoding accounts for only a fraction of the TR interval. Therefore, minimizing the dead time by shortening the pulse recycle time seems to be a straightforward means of optimizing scan time. However, TR is typically dictated by contrast requirements. Shortening of TR leads to progressive saturation, and thus, in spin echo imaging at least, is incompatible with proton density and T 2 -weighted contrast A solution to this problem was first conceived by Crooks et al.,24 who introduced what is commonly called ‘2D multislice imaging’. By making the rf pulses slice-selective, only the spins in the slice of interest are stimulated. Adjacent slices are then excited in sequence without perturbing the steady state. The gain in efficiency achieved in this manner depends on TR, the echo time TE , and the time required for applying the spatial encoding gradients, and typically ranges from about 10 to 50. The second important innovation was the incorporation of Carr–Purcell echoes into spin warp imaging. In this manner, images with different contrast25 can be derived from a single acquisition. It is obvious that adding echoes reduces the number of slices that can be interrogated in a given TR period. Both 2D multislice and multiecho spin warp imaging, perhaps appearing trivial in hindsight, are, in historic perspective, among the most significant milestones in the quest for imaging efficiency.

ky

4.3 Conjugate Synthesis

A salient property of k-space with implications for imaging speed is its conjugate symmetry, i.e., ρ(k) = ρ(−k)

Figure 2 Halving the number of phase and slice encodings (N y , N z ) in a 3D spin warp acquisition can be achieved by reducing either the k-space area (a) or the sampling density (b), leading to lower resolution (a) or reduced field-of-view (b). Open circles represent skipped data samples

(10)

This implies that two data samples corresponding to spatial frequencies k and − k are identical. In principle, therefore, it suffices to acquire only half of k-space. In reality, however, the conjugate symmetry is often not perfect. Causes of deviations are phase shifts from magnetic field inhomogeneity and motion, which require appropriate correction of the raw

4 WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT data before Fourier reconstruction26 (see also Partial Fourier Acquisition in MRI). As already pointed out, sampling in the k y and k z directions is much slower than sampling in the k x direction. Therefore, by exploiting the redundancy in phase-encoding lines, scan time is halved. Halving imaging time by conjugation was described by Margosian et al.27 and also by Feinberg et al.28 Because the phase correction requires sampling of a small number of data lines on the conjugate half, the time saving is usually somewhat less than 50%.

In contrast to FLASH, the latter techniques provide true spin echo images in which the signal is attenuated as exp (−TE /T 2 ) rather than exp (−TE /T 2 *) as in FLASH. The FLASH signal can readily be calculated from the Bloch equations to yield S(T R, α, T E) ∝

1 − exp(−T R/T1 ) sin α 1 − cos α exp(−T R/T1 )

(12)

× exp(−T E/T2∗ )

For α = α opt , equation (12) simplifies to 4.4

Variable Flip Angle Imaging

Sαopt (T E) ∝ tan( 12 α) exp(−T E/T2∗ )

In their seminal work on pulse Fourier transform spectroscopy Ernst and Anderson29 showed that for a train of equidistant rf pulses, the extent of saturation can be controlled by adjusting the rf pulse flip angle α and further that the signal peaks for the condition αopt = cos−1 exp(−T R/T1 )

(11)

where α opt is the optimum pulse flip angle (also called the ‘Ernst angle’), which is valid provided that TR  T 2 . It is readily recognized that flip angles α < 90◦ are not compatible with the simple Hahn or Carr–Purcell imaging pulse sequence, since inversion of the residual longitudinal magnetization by the subsequent phase reversal pulse would lead to a steady state of very low magnetization. Haase et al.30 modified the original spin warp technique in which a gradient echo is sampled and showed that in this manner TR could be reduced at least 10-fold relative to corresponding spin echo implementations, which may have prompted the authors to the acronym ‘FLASH’ (f ast l ow angle shot). Other solutions to the problem involve the idea to restore the longitudinal magnetization with a second 180◦ pulse.31 An alternative solution is to select a flip angle so that 90◦ < α < 180◦ , followed by the usual phase-reversal 180◦ pulse.32,33

(13)

In Figure 3(a), the relative S/N is plotted as a function of TR/T 1 for α = 90◦ and α = α E , equation (12). In either case, a shortened data acquisition time exacts an S/N penalty, which, however, is less severe if the flip angle is optimized. At typical imaging field strengths (1 ± 0.5 T), mammalian tissue T 1 relaxation times span a range of about one order of magnitude (from about 0.3 s for fat to about 3–4 s for extracellular fluids). Hence, a compromise setting will have to be chosen for α. Figure 3(b) shows a series of images for which the flip angle was gradually increased, demonstrating the effect of increasing saturation and the progression from proton density to T 1 -weighting. Low flip angles emphasize structures with long T 1 , such as cerebrospinal fluid in the ventricles. It is noteworthy, however, that the flip angle for optimum contrast between two tissues of differing T 1 is greater than α opt .34 The assumption implicit in equation (12), i.e., TR  T 2 between successive rf pulses, may be invalid at high pulse repetition rates. As first described by Carr,35 a steadystate situation arises for spins subjected to a train of rf pulses in inhomogeneous magnetic fields, with magnetization building up in a recurrent fashion from cycle to cycle. A similar situation applies in imaging, at least as long as the

1

(a)

90 68 53 S/N (arbitrary units)

39 25 18 0.1

13

0.01 0.01

0.1 TR/T1

1

Figure 3 (a) Signal amplitude as a function of TR/T 1 for repetitive pulses after attaining steady state: α=α opt (upper curve, with optimum flip angle indicated), and α = 90◦ (lower curve), assuming negligible residual transverse magnetization. (b) Effect of increasing pulse flip angle in 2D gradient echo images (TR = 0.3 s), showing effect of gradual differential saturation

WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT

gradient moments are constant during successive cycles. In spin warp imaging, however, this latter condition is not satisfied, since the phase-encoding gradient is stepped in an incremental fashion. The ensuing cycle-dependent phase shifts cause intensity artifacts that have a positional dependence, since the phase depends on the spins’ location on the phaseencoding axis.36 If, on the other hand, the phase imparted by the phase-encoding gradient is unwound by a gradient of opposite moment prior to the next-following rf pulse,36,37 the spatial dependence of the steady state vanishes. Common acronyms for this modification to the generic FLASH pulse sequence are ‘GRASS’ (gradient recalled acquisition in the steady state) and ‘FISP’ (f ast i maging with steady state precession). From a tissue contrast point of view, it is often desirable to suppress transverse coherences. Application of spoiler gradients of varying amplitude has been found to be relatively ineffective for this purpose,38 though schemes were later reported that provide some spoiling efficiency.39 Nevertheless, a more effective means of eliminating the carry-over of transverse magnetization from previous cycles is to step the rf transmit and receive phase in constant increments from cycle to cycle. Although predicted by Crawley et al.38 rf-spoiled gradient echo imaging was developed by Matt Bernstein and his colleagues in 1989, but, unfortunately, never published. The contrast characteristics of the two embodiments of FLASH are substantially different. In the high-flip-angle regime, GRASS affords images in which the signal scales as T 2 /(T 2 + T 1 ), hence favoring fluids, where T 2 ≈ T 1 (as opposed to tissues, where T 1  T 2 ). By contrast, with spoiling, the signal obeys equation (12). 4.5

High-Speed Gradient Echo Techniques

Combination of increased sampling frequency bandwidth and fractional echo sampling with partial Fourier processing,40 reduction in the number of frequency-encoding samples, and other strategies such as truncated rf pulses permit shortening of pulse repetition times to 5 ms or less.19,20 The impact of reduced pulse repetition time on S/N is illustrated in Figure 3(a). These methods therefore need to be assessed in terms of the benefits and trade-offs they entail. Besides S/N, contrast is perhaps the single most important criterion for clinical utility. The typically unsatisfactory contrast inherent in highspeed steady-state techniques can be enhanced by means of magnetization preparation,19 discussed in more detail in Whole Body Magnetic Resonance: Fast Low-Angle Acquisition Methods. The underlying idea is to create a nonequilibrium situation by preceding the train of low angle rf pulses by an inversion pulse for T 1 -dependent contrast41 – 43 or drivenequilibrium pulses to achieve T 2 -weighted contrast.44 Since, in this case, the magnetization evolves during scanning of kspace, effectively filtering the data in a manner determined by the spin dynamics, the usual sequential scanning from −k max to +k max is not optimal. Remembering that the signal components pertaining to k ≈ 0 contribute most to contrast, it is necessary to acquire these data at the desired time point during magnetization evolution (e.g., when the signal from a particular tissue is nulled). These requirements can, for example, be reconciled with a k-space ordering scheme that has been called ‘centric phase encoding’,42 in which data collection starts at k-space center and proceeds outward.

5

ECG

Sampling ky ky

kx

Figure 4 Principle of segmented k-space acquisition in cardiac cine imaging. The RR interval is divided into time periods (‘segments’) of short duration, during which several lines k y are encoded. The process is repeated during subsequent cycles until all lines are scanned

Finally, real-time implementations of fast gradient echo scanning are possible by confining images to a single spatial dimension. In this manner, projections can be displayed in real time at a frame rate of 100 s−1 as a scrolling bar, analogous to M-mode ultrasound, with possible applications in cardiology.45 Another interesting idea based on fast gradient echo imaging is a method designed to track interventional devices. It consists of the consecutive generation of three orthogonal projections from a small pointlike object, followed by computation and display of its coordinates in a real-time mode.46 A comprehensive review of short-TR imaging techniques has been given by Haacke et al.47 4.6 Cardiac Gated Techniques

Except when the entire k-space can be covered in a small fraction of the cardiac cycle, such as in EPI,48 data acquisition in cardiac imaging needs to be synchronized to the cardiac cycle. Therefore, it appears that the minimum scan time in 2D spin warp imaging of the heart is given as N y T c , with N y and T c being the number of phase-encoding samples and cardiac period, respectively. The inherent speed of the gradient echo then can be exploited to subdividing the cardiac period by collecting data at multiple cardiac phases, resulting in a series of temporally resolved images that can be played back in a movie loop, a technique that has been called ‘cardiac cine’ imaging.49,50 The minimum scan time can be considerably shortened (albeit at some cost in temporal resolution) if, instead of a single line k y , multiple lines are scanned per cardiac phase, as illustrated in Figure 4. This modification, also termed ‘segmented k-space acquisition’43 permits shortening of the data acquisition time in cardiac cine imaging to 16–24 heartbeats,51 which is within the range of a breath-hold period, thus considerably improving cardiac image quality.

5 MULTILINE SCANNING TECHNIQUES

All previously discussed embodiments of spin warp imaging have in common that within one pulse sequence cycle, a single

6 WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT data line is sampled. Additional echoes are merely a replication of the procedure affording images of increasing T 2 -weighting. An inherently more efficient approach makes use of the idea to exploit a plurality of echoes in such a manner that each echo affords a separate k y line.

5.1

Echo Planar Imaging

In 1977, Mansfield first proposed what he termed ‘echo planar imaging’ (EPI), in which the entire 2D encoding process could be completed during a single FID. The fundamental principle underlying the technique is to oscillate the read gradient G x in the presence of a constant gradient G y .5 It is readily seen that in this manner, half of k-space is traversed in a zig–zag trajectory. The difficulties in reconstructing images from only half the data and nonorthogonal sampling grids were remedied with the introduction of pre-encoding gradients and discontinuous phase encoding.22,23 In this manner, entire images of 64 × 128 data samples could be obtained in as little as 50 ms at 2 T field strength.23,48 For an overview of the subject, the reader is referred to a review by Cohen and Weisskoff6 (see also Echo-Planar Imaging). An interesting corollary of single shot EPI is that the pulse repetition time loses its meaning, since the data are collected in single sequence cycle. However, concatenation of individual acquisitions for the purpose of generating movies, for example, underlies the same restrictions as the conventional FLASH method, demanding adjustment of the rf flip angle to control saturation. Cohen and Weisskoff6 showed 16-frame cardiac movies obtained by acquiring complete images every 67 ms within a single heart beat. Unlike gated techniques, where each image data set, or fraction thereof, is collected during consecutive heart beats, the method is insensitive to variations in heart rate. In contrast to the high-speed FLASH-type methods, however, echo planar imaging puts far more stringent demands on instrumentation, in particular to amplitudes and slew rates of the gradients, which explains its currently limited diffusion. Some of these conditions are relaxed if only a fraction of k-space is scanned from a single echo train. For a detailed discussion of the subject, see Echo-Planar Imaging. Finally, one wonders what S/N penalty this extraordinary scan speed exacts. At a given resolution, three parameters contribute to S/N: √ 1. sampling time (S/N ∝ ts ); 2. transverse relaxation losses (S/N ∝ exp (−TE /T 2 )); 3. the fraction of magnetization available for signal creation (determined by saturation) (S/N ∝ f (TR, T 1 , α). The following comparison may be instructive. Suppose a typical tissue such as neural white matter is scanned at 1.5 T field strength (T 1 ≈ 700 ms, T 2 ≈ 75 ms). Then consider a high-speed gradient echo acquisition with 128 phase encodings and 2 ms readout time, TR = 5 ms, TE ≈ 0, operated at the optimum flip angle, which is 6.8◦ , and compare it with a single shot EPI with 50 ms total sampling time, performed such that k y = 0 is sampled at TE = 30 ms. The results are summarized in Table 1. They imply a net advantage of about a factor of 5 in favor of EPI, defeating the common notion that shortened scan time inevitably exacts a S/N penalty. Similar arguments in favor of EPI have been put forward by Cohen and Weisskoff.6 Such a comparison, however, does not withstand more rigorous scrutiny. Suppose we were to

Table 1 Comparison of S/N in Single Shot EPI and High-Speed FLASH; For Assumptions, See Text

Sampling time (ms) Magnetization exp (-TE /T 2 ) Net gain

EPI

FLASH

(S/N)EPI /(S/N)FLASH

50 1 0.67

256 0.06 1

0.44 16.7 0.67 4.9

repeat the EPI scan every 640 ms, i.e., the equivalent of the FLASH data acquisition time. This would mean that the longitudinal magnetization has only incompletely recovered, therefore requiring operation at the Ernst angle, which is 65◦ . Doing so reduces the transverse magnetization following the rf pulse to 0.64 [from equation (13)], thus lowering the relative S/N advantage of EPI to (a still substantial) factor of 3. 5.2 RARE

In 1986, Hennig introduced the spin echo counterpart of EPI,52,53 initiating a development with significant impact on clinical efficiency. Surprisingly, this work did initially not elicit broad interest. The perception that the images emphasized structures with long T 2 , thereby producing significantly T 2 -weighted images, prompted Hennig to dub his technique ‘RARE’, short for rapid acquisition with relaxation enhancement. While obviously highly efficient, the extreme T 2 weighting appeared to limit the method to specialized applications such as imaging of cerebrospinal fluid. Besides scan time reduction by up to an order of magnitude, the current widespread clinical use of the method owes its popularity to its spin echo nature, i.e., being insensitive to magnetic field inhomogeneities. Equally important, however, is the almost unlimited latitude in image contrast RARE provides if implemented in a hybrid mode with only a fraction of k-space covered per echo train. It is interesting that the possibilities for contrast manipulation of the RARE sequence were already recognized by Hennig in his 1986 paper.52 By k-space weighting the echoes appropriately, Melki et al.54 showed that virtually any arbitrary contrast is possible. For this purpose, they devised an algorithm for phase-encoding ordering while minimizing adverse effects from discontinuous sampling of k y -space. For example, assigning early echoes to low spatial frequencies de-emphasizes T 2 weighting. The convolution of the data with the relaxation function, of course, may result in point spread function blurring, since high-k signals are attenuated.55 Since each of the n echoes generates n lines k y , the minimum data acquisition time is lowered n-fold. Nevertheless, when assessing the method’s efficiency in terms of scan time per image, one has to consider that the prolonged echo train reduces the number of slice locations that can be interleaved during the dead time following data collection. For example, at an echo train length of 250 ms consisting of 16 echoes, a TR = 3 s acquisition provides for fewer than 12 slice locations, rather than about 30 as for a conventional dual echo scan with a second echo time of 100 ms. Finally, Hennig showed that the per-unit-time S/N can be up to a factor of 100 higher in single shot RARE, which again emphasizes what we have seen for EPI, i.e., that increased scan speed does not inevitably penalize S/N. Therefore, the greatly improved efficiency can be traded

WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT

Figure 5

Three-dimensional RARE images of the lumbar spine: (a) axial section; (b) oblique reformation showing nerve roots

for improved resolution (by a factor of four in linear resolution) without adversely affecting S/N when compared with single line spin echo imaging. Further, it enables 3D imaging with all its benefits such as retrospective data rearrangement with image display in secondary planes, which was hitherto impractical in spin echo imaging because of excessive scan time. Examples of 3D RARE images are given in Figure 5. Finally, it has also been shown that RARE and EPI principles can be combined in such a manner that gradient echoes are interweaved into a spin echo train,56,57 and thus the efficiency is further augmented. (See also Multi Echo Acquisition Techniques Using Inverting Radiofrequency Pulses in MRI.)

6

7

OTHER METHODS OF SCANNING k -SPACE

An alternative k-space scanning technique that, like EPI, uses oscillating gradients is spiral scanning. As implied by the term, the k-space trajectory in this case is a spiral originating at the k-space center. Ahn et al.,16 who published the first spiral scan images, used a back-projection algorithm for reconstruction. Most of the more recent work is based on constant-velocity interleaved k-space spirals (as opposed to constant angular frequency), an embodiment that is credited to Macovski’s laboratory17 (see Spiral Scanning Imaging Techniques). This scanning mode has the advantage that kspace is covered more uniformly and that the amplitude of the oscillating gradients is constant. Compared with EPI, the gradient amplitude and slew rate requirements are less demanding, and the technique has been found to be tolerant to flow-related dephasing, an observation that was attributed to the low gradient moments near the k-space center and the periodic simultaneous return to zero of all moments.17 Spiral scans require different reconstruction algorithms. Fourier transformation along the radial coordinate affords a series of projections, from which the image is obtained by filtered back-projection.16 Alternatively, the data can

be rearranged into a rectilinear grid by interpolation, thus allowing Fourier reconstruction17 (see also Spiral Scanning Imaging Techniques). Another nonrectilinear sampling technique suited for highspeed imaging is projection imaging, where k-space is sampled radially in two or three dimensions by simultaneous application of two or three constant-amplitude gradients that are proportional to the sine and cosine of the projection angle. In this manner, a radial path is traced, starting at the center of k-space.13,14 k-space is covered by stepping the gradient amplitudes by incrementing the projection angle in successive sequence cycles. By selecting shallow rf pulses, the pulse sequence can be run at rates comparable to gradient echo spin warp imaging. Since sampling can start almost immediately following excitation, the technique is particularly suited for imaging short-T 2 tissues such as lung parenchyma58 or imaging of nuclei with inherently short T 2 such as 11 B.59 (See also Projection–Reconstruction in MRI.) In summary, during the past two decades since its inception, MRI has revealed a wide spectrum of technical approaches for covering data space. Scan speed is a continuum ranging from milliseconds to minutes, with trade-offs between temporal resolution, S/N and spatial resolution. The rate at which k-space can be scanned is determined significantly by the imager’s hardware, in particular the achievable amplitudes of the spatial encoding gradients and their switching rates, and the receiver’s bandwidth. Besides intrinsic S/N, the ultimate scan speed may be limited by adverse bioeffects, notably peripheral nerve stimulation caused at increased dB /dt of the switched gradients.

7 RELATED ARTICLES

Echo-Planar Imaging; Gradient Coil Systems; Image Formation Methods; Multi Echo Acquisition Techniques Using Inverting Radiofrequency Pulses in MRI; Partial Fourier

8 WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT Acquisition in MRI; Projection–Reconstruction in MRI; Spiral Scanning Imaging Techniques; Whole Body Magnetic Resonance: Fast Low-Angle Acquisition Methods; Whole Body Magnetic Resonance Spectrometers: All-Digital Transmit/Receive Systems.

8

REFERENCES

1. A. Villringer, B. R. Rosen, J. W. Belliveau, J. L. Ackerman, R. B. Lauffer, R. B. Buxton, Y. S. Chao, V. J. Wedeen, and T. J. Brady, Magn. Reson. Med., 1988, 6, 164. 2. J. W. Belliveau, D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen, Science (Washington, DC), 1991, 716. 3. K. K. Kwong, J. W. Belliveau, D. A. Chesler, I. E. Goldberg, R. M. Weisskoff, B. Poncelet, D. N. Kennedy, B. E. Hoppel, M. S. Cohen, R. Turner, H. Cheng, T. Brady, and B. Rosen, Proc. Natl. Acad. Sci. USA, 1992, 89, 5675. 4. S. Ogawa, D. W. Tank, R. Menon, J. M. Ellerman, S. G. Kim, H. Merkle, and K. Ugurbil, Proc. Natl. Acad. Sci. USA, 1992, 89, 5951. 5. P. Mansfield, J. Phys. C , 1977, 10, L55. 6. M. S. Cohen and R. M. Weisskoff, Magn. Reson. Imaging, 1991, 9, 1. 7. F. W. Wehrli, Fast-Scan Magnetic Resonance: Principles and Applications, Raven Press, New York, 1991, p. 1. 8. S. Ljunggren, J. Magn. Reson., 1983, 54, 338. 9. W. A. Edelstein, J. M. S. Hutchison, G. Johnson, and T. Redpath, Phys. Med. Biol., 1980, 25, 751. 10. A. Kumar, D. Welti, and R. Ernst, J. Magn. Reson., 1975, 18, 69. 11. W. Aue, E. Bartholdi, and R. Ernst, J. Chem. Phys., 1976, 64, 2229. 12. P. C. Lauterbur, Nature (London), 1973, 243, 190. 13. C.-M. Lai and P. C. Lauterbur, Phys. Med. Biol., 1981, 26, 851. 14. I. R. Young, D. R. Bailes, M. Burl, A. G. Collins, D. T. Smith, M. J. McDonnell, J. S. Orr, M. L. Banks, G. M. Bydder, R. H. Greenspan, and R. E. Steiner, J. Comput. Assist. Tomogr., 1982, 6, 1. 15. A. Macovski, Magn. Reson. Med., 1985, 2, 29. 16. C. Ahn, J. Kim, and Z. Cho, IEEE Trans. Med. Imaging, 1986, 5, 2. 17. C. H. Meyer, B. S. Hu, D. G. Nishimura, and A. Macovski, Magn. Reson. Med., 1992, 28, 202. 18. O. Heid, M. Deimling, and W. Huk, Magn. Reson. Med., 1993, 29, 280. 19. A. Haase, Magn. Reson. Med., 1990, 13, 77. 20. J. Frahm, K. D. Merboldt, H. Bruhn, M. L. Gyngell, W. H¨anicke, and D. Chien, Magn. Reson. Med., 1990, 13, 150. 21. R. Rzedzian, B. Chapman, P. Mansfield, R. E. Coupland, M. Doyle, A. Crispin, D. Guilfoyle, and P. Small, Lancet, 1983, 2, 1281. 22. B. Chapman, R. Turner, R. Ordidge, M. Doyle, M. Cawley, R. Coxon, P. Glover, and P. Mansfield, Magn. Reson. Med., 1987, 5, 246. 23. I. L. Pykett and R. R. Rzedzian, Magn. Reson. Med., 1987, 5, 563. 24. L. Crooks, M. Arakawa, J. Hoenninger, J. Watts, R. McRee, L. Kaufman, P. L. Davis, A. R. Margulis, and J. DeGroot, Radiology, 1982, 143, 169. 25. L. E. Crooks, D. A. Ortendahl, L. Kaufman, J. Hoenninger, M. Arakawa, J. Watts, E. R. Cannon, M. Brant-Zawadzki, P. L. Davis, and A. R. Margulis, Radiology, 1983, 146, 123.

26. J. R. MacFall, N. J. Pelc, and R. M. Vavrek, Magn. Reson. Imaging, 1988, 6, 143. 27. P. Margosian, F. Schmitt, and D. Purdy, Health Care Instrum., 1986, 1, 195. 28. D. A. Feinberg, J. D. Hale, J. C. Watts, L. Kaufman, and A. Mark, Radiology, 1986, 161, 527. 29. R. R. Ernst and W. A. Anderson, Rev. Sci. Instrum., 1966, 37, 93. 30. A. Haase, J. Frahm, and D. Matthaei, J. Magn. Reson., 1986, 67, 258. 31. J. A. Tkach and E. M. Haacke, Magn. Reson. Imaging, 1988, 6, 373. 32. A. R. Bogdan and P. M. Joseph, Magn. Reson. Imaging, 1990, 8, 13. 33. H. Jara, F. W. Wehrli, H. Chung, and J. C. Ford, Magn. Reson. Med., 1993, 29, 528. 34. N. J. Pelc, Magn. Reson. Med., 1993, 29, 695. 35. H. Carr, Phys. Rev., 1958, 112, 1693. 36. J. Frahm, K. D. Merboldt, and W. Hanicke, J. Magn. Reson., 1987, 27, 307. 37. K. Sekihara, IEEE Trans. Med. Imaging, 1987, 6, 157. 38. A. P. Crawley, M. L. Wood, and R. M. Henkelman, Magn. Reson. Med., 1988, 8, 248. 39. H. Z. Wang and S. J. Riederer, Magn. Reson. Med., 1990, 15, 175. 40. T. K. Foo, F. G. Shellock, C. E. Hayes, J. F. Schenck, and B. E. Slayman, Radiology, 1992, 183, 277. 41. J. P. Mugler III and J. R. Brookeman, Magn. Reson. Med., 1990, 15, 152. 42. A. E. Holsinger and S. J. Riederer, Magn. Reson. Med., 1990, 16, 481. 43. R. R. Edelman, B. Wallner, A. Singer, D. J. Atkinson, and S. Saini, Radiology, 1990, 177, 515. 44. J. P. Mugler, III, T. A. Spraggins, and J. R. Brookeman, J. Magn. Reson. Imaging, 1991, 1, 731. 45. J. D. Pearlman, C. J. Hardy, and H. E. Cline, Radiology, 1990, 175, 369. 46. C. L. Dumoulin, S. P. Souza, and R. D. Darrow, Magn. Reson. Med., 1993, 29, 411. 47. E. M. Haacke, P. A. Wielopolski, and J. A. Tkach, Rev. Magn. Reson. Med., 1991, 3, 53. 48. R. R. Rzedzian and I. L. Pykett, Am. J. Roentgenol., 1987, 149, 245. 49. G. Nayler, D. N. Firmin, and D. B. Longmore, J. Comput. Assist. Tomogr., 1986, 10, 715. 50. G. H. Glover and N. J. Pelc, in Magnetic Resonance Annual , ed. H. Y. Kressel, Raven Press, New York, 1988, p. 299. 51. R. R. Edelman, W. J. Manning, D. Burstein, and S. Paulin, Radiology, 1991, 181, 641. 52. J. Hennig, A. Nauerth, and H. Friedburg, Magn. Reson. Med., 1986, 3, 823. 53. J. Hennig, H. Friedburg, and D. Ott, Magn. Reson. Med., 1987, 5, 380. 54. P. S. Melki, R. V. Mulkern, L. P. Panych, and F. A. Jolesz, Magn. Reson. Imaging, 1991, 1, 319. 55. P. S. Melki, F. A. Jolesz, and R. V. Mulkern, Magn. Reson. Med., 1992, 26, 328. 56. K. Oshio and D. A. Feinberg, Magn. Reson. Med., 1991, 20, 344. 57. D. A. Feinberg and K. Oshio, Radiology, 1991, 181, 597. 58. C. J. Bergin, G. H. Glover, and J. M. Pauly, Radiology, 1991, 180, 845. 59. G. H. Glover, J. M. Pauly, and K. M. Bradshaw, J. Magn. Reson. Imaging, 1992, 2, 47.

WHOLE BODY MRI: STRATEGIES DESIGNED TO IMPROVE PATIENT THROUGHPUT

Biographical Sketch Felix W. Wehrli. b 1941. M.S., 1967, Ph.D., 1970, chemistry, Swiss Federal Institute of Technology, Switzerland. NMR application scientist, Varian AG, 1970–79; Executive Vice President, Bruker

9

Instruments, Billerica, 1979–82; NMR Application Manager, General Electric Medical Systems, Milwaukee, 1982–88. Currently Professor of Radiologic Science and Biophysics, University of Pennsylvania Medical School. Approximately 95 publications. Current research specialty: NMR imaging of biomaterials, specifically trabecular bone.

WHOLE BODY STUDIES: IMPACT OF MRS

Whole Body Studies: Impact of MRS

1

goals in MRS research is the study of the biochemical basis of human disease. Many human diseases have no useful animal models; the response to pharmacological interventions is often species-speci®c, and human morphology and anatomy has a unique in¯uence on the expression and development of disease.

George K. Radda University of Oxford and The John Radcliffe Hospital, Oxford, UK

1.2

1 INTRODUCTION It is common to refer to MRS studies of humans as `whole body' MRS, or, as indicated in the title of this article, `whole body studies'. This is a misnomer, since much effort is expended in obtaining MR spectra from a de®ned organ or even a de®ned part of the given organ and not from the whole body. The term arose because generally, though not exclusively, this kind of measurement is taken in magnets that can accommodate the human body. In the study of limbs and the head, this is not always necessary, and much human MRS has been done using magnets that only accommodate part of the body. This article is therefore about the impact of MRS in human investigations, but excludes discussion of work in which MRS is used to study samples from biopsies or body ¯uids of humans. Much of the human data is inevitably backed up, extended, or elucidated by parallel animal investigations. Although such studies have played an essential role in the development and evaluation of human MRS, they are not included in this article for organizational rather than conceptual reasons.

1.1 The Varying Aims of Human MRS Following from the success of MRI as a radiological tool in clinical diagnosis, it is not uncommon to view MRS as just another extension of the technique for the characterization and description of pathology. Thus one aim is to ®nd clinical applications that can contribute to diagnosis in a routine examination. It will be shown below that this aim is far from having been achieved, and it will be argued that, given the nature and cost of the technique, the major justi®cation for using MRS rests elsewhere. The second main aim of MRS is to investigate the dynamics of normal human biochemical processes. It is necessary to ask, however, why it is important to study fundamental biochemical processes in humans when the same cellular mechanisms are likely to be operative in all mammalian species. Signi®cant special aspects of human biology, among others, include the wide range of polymorphism that results in the biochemical heterogeneity, the nature of the immune system, the advanced central nervous system, and the ease by which the effects of training, adaptation and environmental factors can be studied in humans. In addition, developmental processes at the structural and biochemical level require direct human investigations. The most important of the

Approaches to Clinical MRS

A successful clinical study by MRS has several basic requirements: (a) to obtain well-resolved spectra from de®ned regions of human organs and tissues in a time acceptable to patients; (b) to obtain as much information as possible, and to assign and interpret the biochemical meaning of the spectra; (c) to use the biochemical information derived to advance clinical understanding or management. (a) Localization strategies fall into two major categories: single volume selection and spectroscopic images obtained in the form of one-dimensional slices or with lower frequency as two- or three-dimensional data acquisition. The choice of the method used is often limited by the instrumentation provided by the manufacturer (different manufacturers emphasize their own preferred options). It would be desirable if the localization could be tailored to the nature of the clinical condition to be studied. This is likely to become increasingly the case. Since the majority of current human spectrometers are based primarily on advanced MRI systems, image guided localization is the common practice. (b) There are instances where a single spectroscopic measurement provides valuable clinical data. In most instances, however, the information content of the study is considerably increased by the measurement of reaction rates either directly, or more generally, in response to some intervention or stress test. For example, almost all the investigations of muscle metabolism and energetics in disease rely on following changes in the concentrations of high- and low-energy phosphates and in the pH values in response to exercise (dynamic or isometric), and during the recovery phase after exercise. In organs such as the liver, stress might be induced by the infusion or oral administration of substrates (glucose, fructose, etc.), while the intervention of most interest in tumors could well be the therapeutic method that is being used. In the human brain, sensory stimulation may be used in some measurements (e.g., glucose uptake1) to provide regional activation. It is also possible that metabolic changes may be detected in responses to the administration of drugs that interfere with neurotransmitter release or function. Cardiac work can be increased by exercise (dynamic leg exercise or isometric hand grip) or by infusion of dobutamine. The assignment of peaks in the spectra observed in vivo and their biochemical interpretation are largely based on extensive studies of cells, isolated organs, and animals. Spectral resolution and quanti®cation often remain a problem, but progress is being made to replace measurements of metabolite ratios with derivation of absolute amounts or intracellular concentrations. (c) Increasingly, the understanding of mechanisms of human disease should lead to the design and evaluation of new therapies. MRS, in the ®rst instance, contributes to this process in a general sense. There are, at present, only a handful of speci®c For References see p. 9

2 WHOLE BODY STUDIES: IMPACT OF MRS cases where a spectroscopic evaluation altered the management of a particular patient. Indeed, the view has been advanced by the present author that the success of MRS will be measured by how readily the expensive and technically demanding MRS study can be replaced by a relatively simple but perhaps less informative and more empirical test for following the clinical status of a patient. This view is based on the belief that if we can de®ne the nature and extent of the disease from MRS studies, it should be possible to design tests that can be correlated. 1.3 What Nuclei? The nuclei that have been used so far in human investigations are 31P, 1H, 13C, 19F, 23Na, and 7Li. This choice is often governed by the technology, i.e., relative sensitivities, ease of detection, resolution, and quanti®cation. Since we now know that high-resolution spectra can be obtained from all these nuclei in human organs (with the exception of 23Na, where only total, largely extracellular, Na+ is measured), it is essential that their choice should be governed by the biomedical problems we wish to solve. 1.3.1

31

P MRS

This is widely used to investigate cellular bioenergetics in vivo.2 Essentially, the following parameters that relate to the energetics are measured: the relative and, more seldom, absolute concentrations of ATP, PCr, Pi, and hexose phosphate, and the value of intracellular pH, which is often equated with the production of lactic acid. The concentration of ADP is often calculated assuming that creatine kinase is at equilibrium. Following changes in these values with stress (e.g., exercise in skeletal muscle and heart), substrate infusion, or administration (e.g., fructose, galactose, and glucose) yields direct or indirect measures of speci®c enzyme activities or of rates of transport of substrates or ions (e.g., Na+/H+ exchange rates). Additionally, magnetization transfer has been used to measure the activities of creatine kinase in human muscle,3 brain,4 and heart,5 and in principle the activity of ATP synthase, so far only achieved in animal studies in vivo.6 In addition to the study of energetics, 31P MRS can be used to follow pathways associated with the synthesis and degradation of phospholipids. Key molecules often detected include phosphoethanolamine and phosphocholine, which are intermediates in the Kennedy pathway of phospholipid biosynthesis, but may also be generated by speci®c phospholipase C catalyzed degradation of membrane phospholipids, perhaps involved in signaling pathways during cell proliferation. Other degradation products observed are glycerophosphorylcholine and glycerophosphorylethanolamine. In vivo, the resonances of these small metabolites may be obscured by the much larger and broader phosphodiester (PDE) signal arising from the phospholipids in the bilayers of membranes.7 1.3.2

13

C MRS

This provides an extremely powerful approach, since, through the use of 13C-enriched substrates, it can lead to measurements of individual reaction rates or ¯uxes through speci®c pathways.8 The formation and breakdown of muscle For list of General Abbreviations see end-papers

and liver glycogen can be studied in this way.9 The sensitivity of 13C detection is considerably enhanced by the proton observe carbon edit (POCE) method. This allows detection of the 13C label in the 1H spectrum with heteronuclear editing, thus obtaining the information in the 13C spectra with the sensitivity of 1H MRS. This technique, as well as direct 13C measurements, has been used to observe the ¯ow from [1-13C]glucose into the C-4 of glutamate via the tricarboxylic acid cycle in human brain.10 1.3.3

1

H MRS

This is now widely used in clinical and physiological investigations, particularly of the human brain. Special techniques are required to suppress the water resonance, and in addition it is sometimes necessary to edit the spectra if speci®c components (e.g., lactate) are to be observed. The major metabolites that can be detected include N-acetylaspartate, glutamate and glutamine, creatine + phosphocreatine, and, under some conditions, lactate.11 Additional signals are obtained from the so-called `choline-containing compounds' (likely to represent a mixture of several substances), taurine and inositol phosphates. Unlike the metabolites that are detected in the 31P MR spectra, the components seen in 1H NMR are not directly linked through major pathways, or in several instances have no established functional role. The lactate signal, however, represents a sensitive handle on anaerobic glycolysis. Knowledge of total creatine (from creatine and phosphocreatine) provides useful data when combined with measurements of phosphocreatine by 31 P MRS. N-acetylaspartate has been used empirically as a `neuronal marker', though this remains to be established. 1.3.4

19

F MRS

This is used for the measurement of the tissue concentration and metabolism of ¯uorinated drugs (e.g., the antidepressant ¯uoxetine in the brain)12 or ¯uorinated agents like halothane. 1.3.5

7

Li MRS

The unique use of lithium therapy in psychiatric patients is beginning to be studied by this technique, allowing the determination of brain concentrations and pharmacokinetics.13 In the discussion below, studies will not be classi®ed according to the nuclei observed, but rather will be presented in relation to the investigation of physiological and pathological problems. 2

NORMAL HUMAN BIOCHEMISTRY

The distinguished Harvard physiologist Walter B. Cannon (1871±1945) coined the term `homeostasis' to describe the coordinated physiological processes that maintain most of the steady state in an organism. The structural and functional features of the components of the organism are optimized to provide adjustments to the required performance of the system (input/output balance). Ultimately, cellular biochemical regulatory processes perform this function. In addition to short-term molecular control, the long-term response to chronically increased needs involves adjustments in the design properties of the system. Weibel14 introduced the principle of symmorphosis

WHOLE BODY STUDIES: IMPACT OF MRS

and de®ned it `as a state of structural design commensurate to functional needs resulting from regulated morphogenesis'. MRS in vivo provides a unique way of studying control, development and adaptation, since it provides quantitative data about `molecular integrative physiology' through the measurement of dynamics, structure and interactions within any part of the whole organism. While traditionally medicine is `organoriented' and MRS is used to examine organs, the exciting aspect of the method is that it is particularly suited for the observation of coordinated bodily functions at the molecular level. 2.1 Control of Bioenergetics In Vivo 2.1.1

Skeletal Muscle

One of the fundamental questions in the understanding of bioenergetics in vivo is how the demand for ATP utilization is linked to increased supply of energy via oxidative metabolism and therefore to increased rate of oxygen delivery to the tissue. Oxygen delivery to the mitochondrion within the intact cells has four components: 1. the oxygen-carrying capacity of blood, i.e., the nature and amount of hemoglobin present; 2. the rate of ¯ow of blood through major blood vessels; 3. capillary density and diffusion from the capillary to the mitochondrion; 4. the ability of oxyhemoglobin to release its oxygen, i.e., the dissociation of curve for oxyhemoglobin. 1 H NMR has been used to observe the ratio of oxy/deoxymyoglobin, since the chemical shifts of the proximal histidine resonance of the two forms are different.15 Indirectly, blood oxygenation can be measured through the effect of paramagnetic deoxyhemoglobin on the T2 relaxation time of blood,16 although this effect has not yet been explored in studies of human muscle. It is believed to be partly responsible for the changes seen in the brain during sensory stimulation.1 It is possible to link NMR studies to measurement of tissue oxygenation using infrared detection of oxy/deoxymyoglobin and hemoglobin. During exercise, the decrease in the oxy forms can be related to oxygenation, and the rate of recovery is a measure of the oxidative capacity of the system.17 While the various pathways involved in the energetics of skeletal and cardiac muscle are well mapped out, the way in which they are controlled in the intact cell is less well understood. NMR studies in vivo have contributed a great deal to our understanding of some of the control functions, particularly in skeletal and cardiac muscle. In skeletal muscle, mitochondrial oxidation in vivo was shown to depend, in a hyperbolic manner, on the concentration of ADP, with a Km of about 30 M. This value is very close to that reported by Chance and Williams for isolated mitochondria. Three types of experiments have been performed to arrive at this conclusion. 1. Chance and his colleagues have shown that in a graded exercise protocol in which the work is measured with an ergometer, during performance going from rest (state 4) to exercise (state 3), the transfer function, as de®ned in Chance et al.,18 approximates a rectangular hyperbola (i.e., the plot of work against Pi/PCr is hyperbolic), giving a Km for ADP of 28 M. This follows from the fact that if there are no pH changes, as is the case in the relatively submaximal protocol

3

used, the Pi/PCr ratio is proportional to the concentration of ADP.18 2. It was shown that the ADP concentration reached at the end of exercise and the initial rate of phosphocreatine resynthesis during recovery in human arm muscle also follow the same relationship, giving a Km value of 27 mM and a Vmax for the ATP synthesis rate of 43 mM minÿ1.19 3. Measurements of the ¯ux between ATP and Pi during steady state isometric muscle contraction using magnetization transfer in animal muscle give a direct indication of the relationship between ADP and mitochondrial oxidative phosphorylation.6 Recovery from exercise is a function of mitochondrial ATP synthesis.20 It can therefore be thought of as an aerobic work jump, in which the rate constant of PCr recovery is a function of Vmax for ATP synthesis, with the complication that the pH may also be recovering from a low value. As no work is done, the absolute rate of PCr resynthesis is an estimate of the oxidative ATP synthesis rate minus a small component of basal ATP turnover. The hyperbolic relationship between oxidation rate and [ADP] can be used to estimate apparent Vmax and mitochondrial Km in the same way as in oxidative exercise. The apparent Vmax is a function of the density and capacity of working mitochondria and the supply of substrate and oxygen and is independent of muscle mass. A preliminary calculation for human forearm muscle suggests that the Vmax calculated from PCr recovery kinetics is 60±70% of that predicted from the intrinsic activity and muscle content of mitochondria. A reduction in apparent Vmax can be caused by an intrinsic mitochondrial defect or a reduction in mitochondrial density (e.g., heart failure) in arterial oxygen carrying capacity or arterial pO2 or in muscle blood ¯ow.21 The quantitative analysis of the energetic processes during exercise is more complex, since both oxidative and anaerobic glycolytic processes contribute to ATP synthesis while the ATP used against the work done depends on muscle mass and other parameters. In a detailed analysis of exercise performed at three levels of mechanical power output under ischemic and aerobic conditions, it was suggested that oxidative ATP synthesis was negligible during the ®rst half-minute of aerobic exercise, and increased with a half-time of around 0.75 min, although the mitochondrial controller [ADP] increased much more rapidly.22 Initial PCr resynthesis after exercise appeared to be an adequate estimate of oxidative synthesis at the end of aerobic exercise. The pH dependence of proton ef¯ux inferred from analysis of recovery could be used to estimate ATP synthesis by glycogenolysis/ glycolysis to lactic acid during aerobic exercise. Measurement of lactic acid production by 1H NMR, which has been demonstrated as a possibility,23 might be used to con®rm these conclusions. The quantitative interpretation of 31P MRS data from human skeletal muscle, which has many applications to the study of physiology and pathophysiology, and of muscle bioenergetics and proton handling, has been analyzed in detail.20 31 P NMR is particularly suited for the examination of the relationship between work output and energetics in vivo. This has been achieved in animal models using, preferentially, sciatic nerve stimulation inducing tetani at different frequencies and with different time intervals. In humans, a variety of experimental protocols have been worked out. Some studies have used the steady state or graded work rate from a constant load For References see p. 9

4 WHOLE BODY STUDIES: IMPACT OF MRS or a Cybex ergometer, while others have used force produced in an isometric contraction. Chance and co-workers have examined the relationship between reaction velocities (e.g., ATP utilization) and tissue work rates to the concentration of regulatory molecules in skeletal muscle.18 They examined the forearm muscle and measured values of Pi/PCr and the work by an ergometer in a graded metabolic load situation. They expressed the relationship between work and velocities as a `transfer function', showing the hyperbolic relationship between velocity and Pi/PCr. The form of this transfer function varies between normal individuals and well-trained athletes, and can be used as an indication of exercise performance. Boska studied the ATP cost of force production in the human gastrocnemius muscle using 31 P NMR and an exercise protocol of isometric maximum voluntary contraction, and estimated the contributions to ATP production from three different processes.24 The rate of change of PCr was used to estimate ATP production rates from creatine kinase rates during exercise and from oxidative phosphorylation during the ®rst 10 s of recovery. The anaerobic glycolysis was estimated from the rate of change of pH, from an assumed buffering capacity and hydrogen ion production stoichiometry. The results showed that by the end of 30 s of exercise, the total ATP production and ATP cost of force production had stabilized and remained constant until the end of a 2 min period. It was also shown that the ATP cost of force production is lower in the ®rst second than at any time, or possibly that ATP production is underestimated at later time points. 2.1.2

13

C NMR and Glycogen

A major new approach to the study of substrate utilization in bioenergetics is provided by the use of 13C NMR and in particular in the use of 13C-enriched substrates, which can lead to measurements of individual reaction rates or ¯uxes through speci®c pathways (for a recent review of 13C NMR, see Cerdan and Seelig25). An important new observation for muscle investigations was that, in spite of its high molecular weight, glycogen is fully visible by 13C NMR in organs and tissues.26 Thus the synthesis and breakdown of glycogen during muscle exercise can now be followed for the ®rst time noninvasively in human muscles. There are many opportunities to study fundamental questions of the regulation of carbohydrate metabolism under a variety of conditions. One example of this was the pioneering study in which the rate of human muscle glycogen formation was measured in normal and diabetic subjects, using infused, isotopically labeled [1-13C]glucose and 13C NMR.27 The important conclusions from these measurements were that (i) synthesis of muscle glycogen accounted for most of the total body glucose uptake and all of the nonoxidative glucose metabolism; (ii) in subjects with non-insulin-dependent diabetes, the rate of glycogen formation was decreased by 60% compared with the rate in normal controls. 2.1.3

Human Heart

Many of the same parameters can, in principle, be measured in heart muscle, and have indeed been studied in isolated and in vivo animal hearts. So far, only the PCr/ATP ratio as an index of energetics has been quantitatively measured in the human heart, using special localization or For list of General Abbreviations see end-papers

spectroscopic imaging techniques (for a review see Conway and Radda28). In normal subjects, the PCr/ATP ratio remained constant when the heart was stressed, either by an isometric hand grip exercise29 or by a dynamic leg exercise performed in the prone position.28 Thus, in the normal situation, the heart can adequately regulate its phosphates over a range of workloads. The results show that free ADP is not the primary regulator of increased ATP synthesis in the normal heart in these situations. 2.1.4

Brain Metabolism

Glucose is the main, and generally the sole, energy source for the brain. Thus measurement of the regional cerebral rate of glucose consumption together with that of oxygen consumption and regional blood ¯ow has been used to describe brain energetics. Using radiolabeled compounds, in humans [2-11F]¯uorodeoxyglucose accumulation and 17O uptake are followed by positron emission tomography. Recently, largely as a result of work by Shulman and colleagues (for a review, see Shulman et al.1) NMR spectroscopic methods have been developed for measuring regional metabolic parameters. The kinetics of glucose transport in the human brain has been determined by measuring brain glucose concentration by 13 C NMR at steady state as a function of plasma glucose concentration. Using a model for facilitated transport, values of Km = 4.9 mM and Vmax = 1.1 mol gÿ1 minÿ1 were derived from the transporter. The maximal glucose in¯ux was 3.6 times the normal glycolytic rate, showing that increases in energy demand can be accommodated by glucose transport. In addition, the concentration of glucose in the brain (about 1.2 mM) is well above the Km for hexokinase, so that changes in plasma glucose will not change the hexokinase ¯ux except at very low concentrations. The turnover rate of the brain glutamate pool can be measured by following the incorporation of 13 C isotope from [1-13C]glucose into C-4 of glutamate on the ®rst timing of the tricarboxylic acid (TCA) cycle. The isotopic ¯ux has been measured directly by 13C NMR and by the proton observe carbon edit method. By ®tting the results to a metabolic model, the rates of the TCA cycle can be determined, which gives a measure for the O2 consumption rates. For the human occipital lobe/visual cortex, this was found to be 1.5 mol O2 gÿ1 minÿ1, a value close to that reported from PET studies.1 Visual stimulation results in observable metabolic changes in the visual cortex that include increased lactate production (observed by 1H NMR),1 decreased brain glucose concentration,30 and a change in the PCr/ATP ratio31 as seen from 31 P NMR. 2.1.5

Magnetization Transfer and Reaction Fluxes

One of the many advantages of NMR in cellular studies is that it can provide dynamic information about intracellular events, and give a measure of reaction ¯uxes catalyzed by enzymes, whether in the steady state or at equilibrium. The role of creatine kinase in skeletal muscle, heart, and brain has been debated for some years, and has been studied in these organs (in animal preparations) by magnetization transfer techniques.2 In humans, creatine kinase ¯uxes have been measured in skeletal muscle, where it was shown that the PCr ! ATP

WHOLE BODY STUDIES: IMPACT OF MRS

¯ux decreases during exercise,3 and in the brain, where the activity was shown to be twice as high in gray as in white matter.4 These kinds of measurements as well as studies of ATP-synthase activities have not yet been explored further in human investigations, largely because of the time required for obtaining reliable data.

3 HUMAN DISEASE In broad terms, human diseases that have been studied using MRS can be conveniently divided into four major categories (Sections 3.1±3.4 below). 3.1 Impaired Oxygen and Substrate Delivery Oxygen delivery to tissues and organs depends on three factors: blood ¯ow, hemoglobin concentration in the blood, and oxyhemoglobin dissociation characteristics. Alterations in any one of these parameters may impair oxidative phosphorylation, with a concomitant increase in anaerobic glycolysis leading to the production of lactic acid. 3.1.1

Muscle

Stenosis in limb arteries produces cramp and pain in patients, with intermittent claudication. 31P MRS has been used by several groups to investigate the metabolic consequences and severity of peripheral vascular disease. While at rest changes in metabolites and intracellular pH (pHi) were observed in severe claudicants (the pH being signi®cantly alkali), the most characteristic and speci®c changes were detected during exercise and during the following recovery phase. During exercise, PCr utilization and intracellular acidosis were enhanced, indicating a greater dependence on glycolytic mechanisms for energy production. The reduced rate

Table 1

5

of PCr resynthesis is consistent with abnormal mitochondrial oxidation, and the slow pH recovery is indicative of substantially reduced blood ¯ow and hence a decreased oxygen delivery (for a review, see Radda et al.32). The hyperbolic relationship between cytosolic [ADP] and rate of PCr resynthesis after exercise has been used to estimate the apparent maximum rate of oxidative ATP synthesis (Qmax) in several human diseases in which mitochondrial oxidation may be impaired (see Table 1).21 Muscle responds to impaired oxidation by stimulating anaerobic ATP synthesis and/or by increasing [ADP], the stimulus to the mitochondrion. However, these responses interact: [ADP] depends on pH and [PCr], and lactic acid production tends to lower [ADP] (by lowering pH), while proton ef¯ux has the opposite effect. Four patterns were identi®ed: (d) in mitochondrial myopathy, apparent Qmax is reduced and [ADP] is appropriately increased, because increased proton ef¯ux reduces the pH change in exercise despite increased lactic acid production; (e) in some conditions (e.g., cyanotic congenital heart disease), apparent Qmax is reduced, but there is no compensatory rise in [ADP], probably because anaerobic ATP synthesis during exercise is increased without increase in proton ef¯ux; (f) in other conditions (e.g., myotonic dystrophy) [ADP] is increased during exercise, but apparent Qmax is normal, suggesting either an increase in proton ef¯ux and/or decrease in anaerobic ATP synthesis during exercise; (g) there are also conditions (e.g., respiratory failure) where, despite impaired oxygen supply, both apparent Qmax and end-exercise [ADP] are normal. Patients with heart failure might have been expected to have reduced muscle blood ¯ow and, perhaps, reduced oxygen delivery due to arterial hypoxemia. Several groups, however, found no blood ¯ow changes,32 but demonstrated reduced mitochondrial content and activity and increased glycolytic ca-

Conditions with Probable or Possible Defects of Mitochondrial ATP Synthesis

Mechanism

Examples analyzed

Intrinsic mitochondrial defects

Mitochondrial myopathy Iron de®ciency Cardiac failure Uremia

Impaired blood ¯ow

Peripheral vascular disease

Reduced blood oxygen tension

Chronic respiratory failure Cyanotic congenital heart disease

Reduced blood oxygen content

Iron-de®cient anemia Myelodysplastic anemia

Increased Hb oxygen af®nity

Treatment with 12C79a

Other conditions with increased [ADP] in exercise

Hypertension Duchenne dystrophy carriers Myotonic dystrophy

a

12C79 is an experimental `left-shifter' drug, 5-(2-formyl-3-hydroxyphenoxy)pentanoic acid.

For References see p. 9

6 WHOLE BODY STUDIES: IMPACT OF MRS pacity in the muscles of such patients, the pattern of response falling largely in group (b) above. 3.1.2

Heart

The elucidation of the metabolic abnormalities in ischemic heart disease has been a major objective of MR spectroscopists for some time. Recently Weiss et al.29 have demonstrated a decrease in the PCr-to-ATP ratio during hand grip exercise in patients with coronary heart disease and ischemia, re¯ecting a transient imbalance between O2 supply and demand in the myocardium with compromised blood ¯ow. Exercise testing in ®ve of these patients after revascularization showed no changes in high-energy phosphates, as was also observed in normal subjects and in nine patients with nonischemic heart disease. The possibility that an exercise-linked spectroscopic test may markedly improve the recognition and evaluation of myocardial ischemia is certainly of great interest. 3.1.3

Brain

Several important clinical problems arising out of impaired O2 delivery have been investigated by both 31P and 1H MRS. Perinatal asphyxia is the commonest cause of neurological handicap in full-term infants. Reynolds and colleagues33 have studied infants with birth asphyxia from birth onwards. 31P NMR spectra from neonatal brain differ from those from the adult brain in that a large monoester peak, identi®ed as phosphoethanolamine, is present and the ratios PCr/ATP are lower. These workers found that birth asphyxia caused signi®cant changes in the ratio PCr/Pi, the magnitude of which was helpful in prognosis. These changes occurred early, and the ratios returned to normal within a few days. Several 1H MRS studies have reported that the N-acetylaspartate (NAA) signal in neurologically abnormal infants was lower than in normal controls (for a review, see Howe et al.34). Cerebrovascular disease and stroke are among the major causes of death in the adult population in the Western world. Numerous 1H MRS studies in patients with stroke have suggested that NAA may be a marker for neuronal loss. Serial measurements of NAA following stroke demonstrated recovery of the NAA peak with neurological improvement.34 Elevation of brain lactate during acute ischemia could be detected, but surprisingly lactate was also found within the infarcted area months after the event. The metabolic turnover of this lactate pool was demonstrated by the appearance of [13C]lactate, following infusion of [1-13C]glucose.35 It has been suggested that brain macrophages, which begin to appear three days after infarction and gradually disappear over several months, could be a major source of elevated lactate signals that persists for months after stroke.36 Levine et al. investigated early human focal ischemia with 31P MRS to characterize the temporal evolution and relationship of brain pH and phosphate energy metabolism.37 Serial ischemic brain pH levels indicated a progression from early acidosis to subacute alkalosis. When acidosis was present, there was a signi®cant elevation in the relative signal intensity of inorganic phosphate and signi®cant reductions in signal intensities of ATP compared with those of control subjects. Ischemic brain pH values correlated directly with the relative signal intensity of PCr and the PCr index, and correlated inversely with the signal intensity of Pi. There was a For list of General Abbreviations see end-papers

general lack of correlation between either ischemic brain pH or phosphate energy metabolism and the initial clinical stroke severity. The data suggested a link between high-energy phosphate metabolism and brain pH, especially during the period of ischemic brain acidosis, and the authors proposed that effective acute stroke therapy should be instituted during this period. Subarachnoid hemorrhage may be complicated by cerebral ischemia, which, though reversible initially, can progress to an irreversible neurological de®cit. 31P MRS studies of 10 patients on 30 occasions at various times after hemorrhage showed in some of the patients (n = 5) focal areas of intracellular acidosis (pH < 6.8), which in most cases returned towards normal.38 The recovery of pHi to normal paralleled an improvement in clinical conditions in each case. These areas of acidosis probably affect ischemia, which is potentially reversible. The MRS measurement provides an opportunity for assessing methods of treatment.

3.2

Genetic and Metabolic Diseases

MRS is particularly valuable in the study of metabolic diseases and genetic conditions. 3.2.1

Mitochondrial Disease

Abnormal mitochondria are increasingly recognized as being responsible for muscle and brain disorders. In primary mitochondrial diseases, there are mutations in the nuclear or mitochondrial genes necessary for the synthesis of the components of the electron transport chain. The functional bioenergetic consequences of these abnormalities can be studied by 31P MRS.39 At rest, in both skeletal muscle and the brain, PCr/Pi and PCr/ATP are decreased, while the Pi/ATP ratio is often increased. This means that the phosphorylation potential, a measure of the available cellular energy, is decreased. In skeletal muscle, oxidative ATP synthesis is impaired, and this often results in slow PCr resynthesis during recovery. As already pointed out (see Table 1), the reduction in mitochondrial Qmax is partly compensated for by an increase in free ADP during exercise and recovery, which stimulates the ATP synthesis rate. This arises from an apparent adaptation in the muscle to chronic mitochondrial insuf®ciency in which the rate of H+ removal from the cell by the Na+/H+ exchange mechanism is enhanced. This provides a protection to the cell against low pH that would arise out of excess lactate production, and also results in an increase in ADP through the creatine kinase equilibrium. 3.2.2

Diseases in Carbohydrate Metabolism

Defects of glycogenolysis and glycolysis are inborn errors of metabolism mainly of muscle or liver. As in the mitochondrial myopathies, the biochemical defects result in impaired production of ATP and in reduced (or absent) amounts of pyruvate for mitochondrial oxidation. Not only are the MRS ®ndings in this group of disorders quite distinct from the oxidative disorders, but glycogenolytic defects are distinguishable from glycolytic defects.32 Among several metabolic liver disorders, abnormal metabolism of fructose was studied in livers of patients with

WHOLE BODY STUDIES: IMPACT OF MRS

hereditary fructose intolerance.40 In this autosomal recessive disease, aldolase B activity is greatly reduced in liver, kidney, and small intestine. Homozygotes were easily detectable by MRS. Heterozygotes for this disorder cannot be identi®ed by conventional tolerance tests, but in the MRS investigations they showed impaired tolerance to fructose compared with control subjects. In the heterozygotes, the rise in plasma urate after fructose (attributed to the activation of AMP deaminase by the low Pi concentration) correlated signi®cantly with the decrease in liver Pi. The drop in liver Pi after fructose was very marked in those heterozygotes who also suffered from gout, suggesting that some patients suffering from gouty attacks might bene®t from a low-fructose diet. Indeed, subsequent studies on a group of patients with a family history of gout show that in some families this resulted from being carriers of hereditary fructose intolerance, and a fructose-free diet improved their condition.41 1 H MRS has been used to detect speci®c signals in metabolic brain diseases. For example, high lactate concentrations were found in the brains of patients with Leigh's disease, and N-acetylaspartate was consistently increased in Canavan's disease, an inborn error of N-acetylaspartate metabolism.34 3.2.3

Dystrophy

31

P MRS studies of skeletal muscle bioenergetics in patients with muscular dystrophies have given new clues as to the functional consequences of the genetic lesion. For example, in patients with Duchenne (DMD) and Becker (BMD) muscular dystrophy, there is an increase in intracellular Pi and decrease in PCr. Intracellular pH was raised substantially in DMD, moderately in BMD, and slightly but signi®cantly in carriers of the disease.2,42 Investigations on the mouse equivalent of DMD (the mdx mouse) have shown that the raised pHi is linked to increased intracellular Na+ and Ca2+, suggesting that the absence of dystrophin in the muscle leads to a series of ionic abnormalities. Measurements of these biochemical parameters in vivo may provide a way of following the success of therapies such as might be achieved by gene replacement. Dystrophin is also known to be absent in the brain (neuronal cells) of patients with DMD, and it is known that about 30% of such patients are mentally retarded. It is of interest, therefore, that abnormal Pi/PCr ratios are also observed in the brains of children with DMD and in the brain of the mdx mouse. Neuropsychological studies on such children have shown a relationship between full-scale IQ and the Pi/PCr in the brain.43

3.3 Abnormal Control and Disease The control and coordination of metabolic processes require an intricate network of input functions. Some of these have been emphasized in Section 2, while others include the role of speci®c ions, general ionic environment, potentials across intraand extracellular membranes, concentrations of substrates, and hormonal input. We may consider diseases arising out of a `control lesion' as being primary intracellular or extracellular, i.e., systemic in origin.

3.3.1

7

Intracellular Control

There are several examples where MRS has led to the identi®cation of defects that affect intracellular control. These include a patient with abnormal intracellular substrate transport in skeletal muscle (malate aspartate shuttle defect) and the recognition of a sarcoplasmic reticulum Ca2+-ATPase defect.32 As already mentioned, perturbations in the ionic balance occur in Duchenne and Becker muscular dystrophy. There are other conditions where the rates of ion-exchange processes (Na+/H+ antiport/Na+/K+-ATPase) are signi®cantly increased. For example, in essential hypertension, skeletal and heart muscle as well as vascular smooth muscle are affected in this way, as are other peripheral cells (white cells and erythrocytes). It is not known in this case to what extent these changes are secondary or whether they are related to the fundamental mechanism of the disease.2 3.3.2

Extracellular Control

The relationships between the extracellular environment (blood biochemistry) and intracellular processes is central to our understanding of physiological regulation. 3.3.2.1 Acid/Base Regulation In Vivo. This is an important physiological problem, with profound clinical implications. Intracellular pH regulation in individual organs and its dependence on the physiological acid/base balance can be examined very well by 31P MRS,44 and many studies have been carried out on this problem in animal models and humans. Some illustrative examples are given below. Hood et al.45 examined the effect of systemic pH on pHi using 31P MRS by measuring lactic acid generation in the forearm muscle of a group of volunteers who performed exhaustive arm exercise with arterial blood ¯ow occluded. On ingestion of NH4Cl (acidosis), blood pH decreased to 7.30 from 7.39 (control), whereas muscle pHi did not change. During alkalosis induced by NaHCO3 ingestion, blood pH increased, while pHi was again well controlled. Systemic acidosis inhibited and alkalosis stimulated lactic acid output (during exercise), which suggests that systemic pH regulates cellular acid production, protecting muscle pH at the expense of energy availability. In contrast, in severe respiratory acidosis produced by CO2 breathing, muscle pH is signi®cantly reduced.45 The adult mammalian brain is protected from oscillations in arterial blood pH by a different mechanism, namely the blood±brain barrier, but it is not protected from changes in arterial partial pressure of CO2 that have been shown to produce changes in the pH of cerebrospinal ¯uid. 31P MRS was used to measure intracellular pH in different parts of the human brain while the subject was breathing air, and during hypercapnea. The white matter responded with a greater decrease in pHi than gray matter.46 The response of liver to acidosis was examined in rats.47 Varying severity of acidosis was induced by HCl infusion, NH4Cl ingestion, or induction of experimental diabetic ketoacidosis. Whereas at the more severe degrees of systemic acidosis a marked decrease in hepatic pHi was observed, there was little change in ketoacidosis. This explained why, in the latter condition, gluconeogenesis from lactate is not inhibited, despite evidence in vitro of inhibition of glyconeogenesis by systemic acidosis. 3.3.2.2 Cellular Inorganic Phosphate (Pi) Concentration and For References see p. 9

8 WHOLE BODY STUDIES: IMPACT OF MRS Total Intracellular Phosphate. These are dependent on blood Pi homeostasis, which in turn is linked to renal Pi clearance and whole body phosphate incorporation and regulation in bones. The relationship between extracellular and intracellular concentrations of Pi was studied in the muscles and erythrocytes of patients with vitamin-D-resistant rickets and patients with Paget's disease of bone before and after they had been made hyperphosphatemic by treatment with the drug ethylidene-1hydroxy-1,1-biphosphonate. Even though the plasma Pi concentration in these patients spanned a fourfold range (0.5± 2.0 mmol Lÿ1), the corresponding intramuscular Pi concentration increased by only 70%.32 Patients in renal failure also have elevated blood phosphate, and again this is accompanied by a proportional but smaller increase in muscle Pi. In spite of this, in this condition, the resting phosphorylation potential remains unaltered, implying that cellular levels of ADP and ATP are adjusted to compensate for increase in Pi. 3.3.2.3 Hormonal Regulation of Metabolism. This has long been recognized, and MRS has been used in many investigations of patients with abnormal hormonal levels on hormonal control. The study of diabetes using 13C MRS has already been mentioned in Section 2.1.2. An extension of this work to insulin resistance is of some interest, since the latter may precede the diabetic condition. Thyroid hormones affect metabolism in many tissues. Skeletal muscle metabolism has been examined in patients with both hypo- and hyperthyroid conditions.2 The abnormalities found are totally reversible with treatment, and MRS is therefore particularly useful in the evaluation of the relationship between clinical symptoms (e.g., decreased exercise capacity in hypothyroidism) and the underlying metabolic abnormalities. 3.3.3

Tumor Biochemistry

Uncontrolled cellular growth and proliferation represent an extreme case of disease where normal molecular homeostasis is not observed. The study of tumor biochemistry by MRS represents a major area of clinical research, and advances in the ®eld have been regularly reviewed since the 1980s and more recently examined in some detail.34,48 In the majority of studies, 31P or 1H MRS have been used but there are reports using 13C, 19F, and 23Na as the nuclei of interest. Several speci®c developments have arisen out of the 31 P MRS measurements. 1. The fact that the majority of human tumors were shown to have alkaline intracellular pH valuesÐand not acidic as was expectedÐfocused attention on the role of ionic control in relation to cellular proliferation. 2. The observation that many tumors had prominent signals in the phosphomonoester (PME) region, which were mostly associated with increased phosphoethanolamine and phosphocholine, brought into the forefront research on phospholipid metabolism in tumors. While it is still not known whether such changes are the result of alterations in the biosynthetic pathways or are associated with phospholipid degeneration and cellular signaling, the possibility that they can be used as an index of cellular proliferation rates has generated much interest. 3. Some researchers hoped that the bioenergetic status of the tumour as re¯ected in the 31P MR spectrum might be characteristic of tumor types. Others argued32 that such For list of General Abbreviations see end-papers

measures of the metabolic state are more likely to provide an indication of the physiological state of a particular tumor in its speci®c location. The latter property could then be used to guide the choice of therapies that would utilize the state of oxygenation and activity of a given tumor at a de®ned stage of its development. 4. There is considerable evidence that response to therapy, particularly chemotherapy, might well be measurable by quanti®cation of the phospholipid components.48 Recently, there has been a rapid expansion in the use of 1H MRS in clinical investigations of tumors.34 Given that in most 1 H MRS experiments, only four peaks are measured, namely those assigned to N-acetylaspartate, total creatine, choline-containing molecules, and lactate, much emphasis has been placed on empirical correlations with other measurements (histology, PET, etc.) or on pattern recognition methods.34 It is perceived that multicenter clinical trials might clarify the applicability of such approaches to the grading and possible diagnosis of human tumors. 3.4

Disease and Foreign Invasion

Biochemical changes associated with the administration of toxic substances, viral infections, in¯ammation, and autoimmune disorders are readily observed by MRS. Alcoholic liver disease, the hepatic consequences of paracetamol poisoning, and viral hepatitis are just some of the conditions that have been studied by 31P MRS.49 Both the extent of damage and the liver regeneration process can be assessed from the MR measurements. Acute liver failure can result in hepatic encephalopathy, possibly through the action of toxic substances like ammonia, mercaptans, and amino acids. Several 1H MRS studies demonstrated an elevation of glutamine levels and the parallel decrease in brain myoinositol.34,50 Liver transplantation completely reversed the 1H MRS changes. Other forms of treatment (e.g., by neomycin or lactulose) were also examined by Kreis et al.50

4

CONCLUSIONS AND FUTURE PROSPECTS

This article has been written from the point of view of the author with emphasis on the biochemical aspects of human MRS. There can be no doubt that we have already learned a great deal from MRS about biochemical aspects of human diseases and control and ¯uxes in normal situations. The method is thus established as a truly new approach in clinical research. In some areas, such as muscle investigations and possibly the empirical uses of brain 1H spectroscopy, it could be considered as part of clinical examination aimed at evaluation of the nature and severity of the disease. As the capability of obtaining spectra in a relatively routine manner becomes available on standard diagnostic imaging instruments, such examinations will become more common, though probably restricted to specialist centers. The author's bias remains, however, that, being a quantitative biochemical technique, the strength of MRS will remain in bringing new understanding about disease mechanisms. Once such understanding is derived from careful and focused studies

WHOLE BODY STUDIES: IMPACT OF MRS

on speci®c conditions, we should attempt to devise simpler, cheaper, and generally more readily available tests, probably with a less precise information content, that can be adopted in daily clinical use.

5 RELATED ARTICLES Applications of 19F-NMR to Oncology; Brain Infection and Degenerative Disease Studied by Proton MRS; Brain MRS of Human Subjects; Brain Neoplasms in Humans Studied by Phosphorus-31 NMR Spectroscopy; Fluorine-19 MRS: General Overview and Anesthesia; In Vivo Hepatic MRS of Humans; Localization and Registration Issues Important for Serial MRS Studies of Focal Brain Lesions; NMR Spectroscopy of the Human Heart; Peripheral Muscle Metabolism Studied by MRS.

6 REFERENCES 1. R. G. Shulman, A. M. Blamire, D. L. Rothman, and G. McCarthy, Proc. Natl. Acad. Sci. USA, 1993, 90, 3127. 2. G. K. Radda and D. J. Taylor, in `Molecular Mechanisms in Bioenergetics', ed. L. Ernster, Elsevier, Amsterdam, 1992, Chap. 19, p. 463. 3. D. Rees, M. B. Smith, J. Harley, and G. K. Radda, Magn. Reson. Med., 1988, 9, 39. 4. T. A. D. Cadoux-Hudson, M. J. Blackledge, and G. K. Radda, FASEB J., 1989, 3, 2660. 5. P. A. Bottomley and C. Hardy, J. Magn. Reson., 1992, 99, 443. 6. K. M. Brindle, M. J. Blackledge, R. A. J. Challis, and G. K. Radda, Biochemistry, 1989, 28, 4887. 7. P. M. Kilby, J. L. Allis and G. K. Radda, FEBS Lett., 1990, 272, 163. 8. R. G. Shulman, Ann. NY Acad. Sci., 1987, 508, 10. 9. T. Jue, D. L. Rothman, B. A. Tavitian, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1989, 86, 1439. 10. D. L. Rothman, E. J. Novotny, G. I. Shulman, A. M. Howseman, O. A. C. Petroff, G. Mason, T. Nixon, C. C. Hanstock, J. W. Pritchard, and R. G. Shulman, Proc. Natl. Acad. Sci. USA, 1992, 89, 9603. 11. J. Frahm, T. Michaelis, K. D. Merboldt, W. Hanicke, M. L. Gyngell, D. Chien, and H. Bruhn, NMR Biomed., 1989, 2, 188. 12. P. F. Renshaw, A. R. Guimares, M. Fava, J. F. Rosenbaum, J. D. Pearlman, J. G. Flood, P. R. Puopolo, K. Clancy, and R. G. Gonzale, Am. J. Psychiatry, 1992, 149, 1592. 13. R. A. Komorski, J. E. Newton, J. R. Sprigg, D. Cardwell, P. Mohanakrishnan, and C. N. Karson, Psychiatry Res., 1993, 50, 67. 14. E. R. Weibel, `The Pathway of Oxygen', Harvard University Press, Cambridge, MA, 1984. 15. Z. Wang, E. A. Noyszewski, and J. S. Leigh Jr., Magn. Reson. Med., 1990, 14, 562. 16. K. R. Thulborn, J. C. Waterton, P. M. Matthews, and G. K. Radda, Biochim. Biophys. Acta, 1982, 714, 265. 17. W. J. Bank, G. Tino and B. Chance, Neurology, 1992, 42, 146. 18. B. Chance, B. J. Clark, S. Nioka, H. Subramanian, J. M. Maris, Z. Argov, and H. Bodle, Circulation, 1985, 72, 103. 19. G. K. Radda, Phil. Trans. R. Soc. Lond. Ser. A., 1990, 333, 515. 20. G. J. Kemp and G. K. Radda, Magn. Reson. Quart., 1994, 10, 43. 21. G. J. Kemp, D. J. Taylor, C. H. Thompson, L. J. Hands, B. Rajagopalan, P. Styles, and G. K. Radda, NMR Biomed., 1993, 6, 302. 22. G. J. Kemp, C. H. Thompson, P. R. J. Barnes, and G. K. Radda, Magn. Reson. Med. 1994, 31, 248.

9

23. H. P. Hetherington, J. R. Hamm, J. W. Pan, D. L. Rothman, and R. G. Shulman, J. Magn. Reson., 1989, 82, 86. 24. M. Boska, NMR Biomed., 1991, 4, 173. 25. S. Cerdan and J. Seelig, Annu. Rev. Biophys. Chem., 1990, 19, 43. 26. J. R. Alger, L. O. Sillerud, K. L. Behar, R. J. Gillies, R. G. Shulman, R. E. Gordon, D. Shaw, and P. E. Hanley, Science, 1981, 214, 660. 27. G. I. Shulman, D. L. Rothman, T. Jue, P. Stein, R. A. De Fronzo, and R. G. Shulman N. Engl. J. Med., 1990, 322, 223. 28. M. A. Conway and G. K. Radda, Trends Cardiovasc. Med., 1991, 1, 300. 29. R. G. Weiss, P. A. Bottomley, C. J. Hardy, and G. Gerstenblith, N. Engl. J. Med., 1990, 323, 1593. 30. K.-D. Merboldt, H. Bruhn, W. Hanicke, T. Michaelis, and J. Frahm, Magn. Reson. Med., 1992, 25, 187. 31. D. Sappey-Marinier, G. Calabrese, G. Fein, J. W. Hugg, C. Biggins, and M. W. Weiner J. Cereb. Blood Flow Metab., 1992, 12, 584. 32. G. K. Radda, B. Rajagopalan, and D. J. Taylor, Magn. Reson. Quart., 1989, 5, 122. 33. P. L. Hope, A. M. Costello, E. B. Cady, D. T. Delpy, P. S. Tofts, A. Chu, P. A. Hamilton, E. O. R. Reynolds, and D. R. Wilkie, Lancet, 1984, 2, 366. 34. F. A. Howe, R. J. Maxwell, D. E. Saunders, M. M. Brown, and J. R. Grif®ths, Magn. Reson. Quart., 1993, 9, 31. 35. D. L. Rothman, A. M. Howseman, G. D. Graham, O. A. C. Petroff, G. Lantos, P. B. Fayad, L. M. Brass, G. I. Shulman, R. G. Shulman, and J. W. Pritchard, Magn. Reson. Med., 1991, 21, 302. 36. O. A. Petroff, G. D. Graham, A. M. Blamire, M. al-Rayess, D. L. Rothman, P. B. Fayad, L. M. Brass, R. G. Shulman, and J. W. Pritchard, Neurology, 1992, 42, 1349. 37. S. R. Levine, J. A. Helpern, K. M. Welch, A. M. Vande-Linde, K. L. Sawaya, E. E. Brown, N. M. Ramadan, R. K. Deveshwar, and R. J. Ordidge, Radiology, 1992, 185, 537. 38. N. S. R. Brooke, R. Ouwerkerk, C. B. T. Adams, G. K. Radda, J. G. G. Ledingham and B. Rajagopalan, Proc. Natl. Acad. Sci. USA, 1994, 91, 1903. 39. D. J. Taylor and G. K. Radda, in `Current Topics in Bioenergetics', ed. C. P. Lee, Academic Press, San Diego, 1994, Vol. 17, p 99. 40. R. D. Oberhaensli, B. Rajagopalan, D. J. Taylor, G. K. Radda, J. E. Collins, J. V. Leonard, H. Schwarz, and N. Herschkowitz, Lancet, 1987, 24, 931. 41. J. E. Seegmiller, R. M. Dixon, G. J. Kemp. P. W. Angus, T. E. McAlindon, P. Dieppe, B. Rajagopalan, and G. K. Radda, Proc. Natl. Acad. Sci. USA, 1990, 87, 8326. 42. G. J. Kemp, D. J. Taylor, J. F. Dunn, S. P. Frostick and G. K. Radda, J. Neurol. Sci., 1993, 116, 201. 43. I. Tracey, R. Scott, C. H. Thompson, J. F. Dunn, P. R. J. Barnes, P. Styles, G. J. Kemp, C. D. Rae, M. Pike, and G. K. Radda, in Proc. IInd Ann Mtg. (Int.) Soc. Magn. Reson. Med., San Francisco, 1994, p. 345. 44. G. K. Radda, FASEB J., 1992, 6, 3032. 45. V. L. Hood, C. Schubert, U. Keller, and S. Muller, Am. J. Physiol., 1988, 255, 479. 46. T. A. D. Cadoux-Hudson, B. Rajagopalan, J. G. G. Ledingham, and G. K. Radda, Clin. Sci., 1990, 79, 1. 47. J. S. Beech, S. R. Williams, R. D. Cohen, and R. A. Iles, Biochem. J., 1989, 263, 737. 48. W. Negendank, NMR Biomed., 1992, 5, 303. 49. G. K. Radda, R. M. Dixon, P. W. Angus, and B. Rajagopalan, in `Regulation of Hepatic Function (Alfred Benzon Symposium 30)', eds. N. Grunnet and B. Quistorff, Munksgaard, Copenhagen, 1990, p. 433. 50. R. Kreis, N. Farrow, and B. D. Ross, Lancet, 1990, 336, 635.

For References see p. 9

10 WHOLE BODY STUDIES: IMPACT OF MRS Biographical Sketch George K. Radda. b 1936. B.A. 1960, chemistry, M.A., D. Phil., 1962, University of Oxford, UK. Introduced to NMR by Rex

For list of General Abbreviations see end-papers

Richards. Successively junior research fellow, lecturer (biochemistry), and Professor (Molecular Cardiology), University of Oxford, 1962± present. Approx. 650 publications. Research specialties: in vivo NMR, clinical spectroscopy, control of bioenergetics.

GASTROSCOPY AND COLONOSCOPY

Gastroscopy and Colonoscopy Nandita M. deSouza Imperial College School of Medicine, London, UK

Glyn A. Coutts and David J. Larkman Marconi Medical Systems, UK

and David J. Gilderdale Engineering Consultant, UK

1 INTRODUCTION Since the 1960s, endoscopy has revolutionized the diagnosis and treatment of both upper and lower gastrointestinal (GI) disease and has become the gold standard in management. However, detection of disease is limited to visualization of the surface mucosa. The extent of invasion beneath the surface may be shown with biopsy, but lesions with no super®cial component or those with in®ltration remote from the biopsy site may be missed.1 Ultrasound probes have been used in conjunction with endoscopy to address this problem, and transducers that can be passed down ¯exible gastroscopes2 or bronchoscopes3 have been used in order to locate lesions and demonstrate their vascularity. However, low soft-tissue contrast limits the application of this technique. MRI provides a valuable method for visualizing soft tissue; however, with conventional external whole-body or phased array receiver coils, image resolution of small areas of interest is generally low. Dedicated surface receiver coils can greatly improve image resolution of adjacent structures, and intracavitary coils have been used in the rectum and vagina for imaging the adjacent prostate and cervix.4,5 It is possible to use coils of this type in conjunction with a nonferromagnetic endoscope for MRI of lesions of the upper and lower GI tract.

2 MR-COMPATIBLE ENDOSCOPE DESIGN Ferromagnetic materials are normally used in the construction of endoscopes. These not only produce an unacceptable degree of artefact but may be subject to movement forces when placed within the magnetic ®eld. It is now possible to produce endoscopes from MR-compatible components. We have developed a MR-compatible gastroscope and colonoscope that contains no ferromagnetic materials (Endoscan Ltd, Burgess Hill, UK). The general handling attributes are similar to those of conventional instruments. An alloy of chrome is used as a substitute for the normal stainless steel guide wires. The conventional steel shroud, which provides integral strength, is replaced by copper braid. Each of the substitute materials is tested to ensure there is no signi®cant MR artefact. The whole

1

assembly is encased in a heat-shrink sleeve that seals the instrument in an acceptable fashion. The tip of the instrument is associated with the MR receiver coil and must be modi®ed accordingly. In early versions of the endoscope, the coil was separated from the endoscope but used in conjunction with it. The coil could, therefore, be advanced away from the tip itself so that artefacts associated with components in the tip were not critical. The tip endpiece in this instrument is con®gured from an amalgam for the viewing lens, irrigation, and biopsy channels. Although polymers would be preferable, forming an impervious seal to the optics proved impossible and alloys of gold-plated copper were used instead. A newer version of the endoscope uses a coil design integral to the tip itself, which means that the tip must be artefact free and the tip design has to be appropriately modi®ed to accommodate the coil circuitry. The ®nished instrument incorporates conventional steering controls, an eyepiece, a light guide with an illumination source (Olympus CLE-3 with 150 W dichroic bulb), and a biopsy channel 2.8 mm in internal diameter (Figure 1). The length of the upper GI endoscope from eyepiece to tip is 1 m with a diameter of 12.5 mm, which equals that of a therapeutic gastroscope. The colonoscope is 1.7 m in length and 14 mm in diameter.

3

RECEIVER COIL DESIGN

The simplest design uses a solid coil that is separate from the endoscope itself but can be used in conjunction with it. The outer diameter of a rigid coil cannot be larger than that of the endoscope itself. Such a coil may be advanced some distance from the endoscopic viewing lens or pulled back up to it. It is maintained in the latter position during intubation of the pharynx. Advantages of its rigidity are not only ease of maintenance but also that it does not require tuning on every occasion. We have used a single turn saddle geometry receiver coil 25 mm in length and 10 mm in diameter wound on an acetyl homopolymer (Delrin) former. The conductor is of 18 swg varnish insulated copper and the on-board electronics contain no ferro- or paramagnetic materials. The receiver coil is connected to the external electronics by a specially constructed 50 ohm coaxial cable 2.1 mm in external diameter in a lowfriction polytetra¯uoroethylene (PTFE) jacket purpose built with MR-compatible components. The cable is accommodated within the biopsy channel of the endoscope. The addition of an immobilization device to the coil, such as an in¯atable balloon or suction cup, would greatly improve image quality. In¯atable balloon devices are currently in use with gastroscopic2 and bronchoscopic3 transducers. However, in¯atable designs require individual tuning and matching and are subject to distortion while the endoscope is being maneuvered. Using the receiver soil separately from the tip itself has several disadvantages: it reduces maneuverability of the instrument, compromises optical ®eld-of-view, and decreases suction (because its cable is fed through the biopsy channel). The incorporation of the receiver coil into the wall of the endoscope itself improves maneuverability and optical viewing. However, it is more challenging to construct and has important implications for housing and repair to the electronics. In particular,

2 GASTROSCOPY AND COLONOSCOPY

Figure 1 MR-compatible endoscope and receiver coil (arrow). The appearance and handling of the endoscope are identical to a standard instrument. The receiver coil is the same diameter as the endoscope itself

immersion of the electronics during sterilization of the endoscope must be considered. In addition, the tip of the endoscope must be completely MR compatible in order to avoid unacceptable artefact on the images. In our design, the coil incorporates an electronic switch that renders the coil electrically invisible during the transmit phase of the MR sequence. Computer controlled switching of rf is used to achieve this. Diodes are selected for their optimal rf switching characteristics while avoiding the ferromagnetic materials that are routinely found in conventional diodes. The small inductors used to form part of the rf switch have minimal detrimental effect on coil performance. A further circuit is provided to transform the impedance of the coil to match that of the transmission cable, thereby minimizing signal degradation within the cable. The capacitors used for both coil tuning and impedance transformation are selected for their high quality factor (Q), while avoiding the ferromagnetic nickel barrier terminations that are often preferred for capacitors in a massproduction environment. The board required to accommodate the additional coil control circuitry is typically 11.5 mm10 mm and is placed immediately behind the coil. The signal-to-noise ratio (SNR) ®gures for coil performance are often misleading. When compared with the standard pelvic phased array coil, the gastroscope coil performs signi®cantly better at distances up to 2 cm from the coil surface. A clinically acceptable ®eld-of-view for diagnostic purposes is obtained up to 2.5 cm from the coil surface. However, coil orientation relative to the ®xed magnetic ¯ux density ®eld, B0, also has a signi®cant effect on the local ®eld-of-view. For a single coil, the regions in which the ¯ux is parallel to B0 pro-

duce no useful signal. Employing two or more coils each with a different orientation such that the blind spot of one is compensated for by the other may overcome this problem. Such coil arrays are crucial in imaging a tortuous structure such as the gut, where coil orientation is extremely variable.

4

SCANNER DESIGN

Modern short-bore magnet designs can be usefully adapted to perform MR endoscopy. Our experience has been with a short-bore design Picker Asset, where the distance from the magnet face to center is 60 cm and an operator can reach at arm's length into the center of the bore (Figure 2). Traditional MR scanner designs in which the patient is at a distance down the long magnet bore do not provide suf®cient patient access and are unsuitable. The newer `interventional' MR scanners (C-arm, double doughnut) speci®cally address the issues of patient accessibility, and their open-plan features are designed to cater for endoscopic procedures. 4.1

Pulse Sequences

Imaging strategies for use with the gastroscopic receiver coil seek a balance between two major considerations. On the one hand the coil gives a large gain in sensitivity over a limited region that can be exploited to reduce patient scan times as well as to provide high-resolution images of immediately adjacent tissue. In particular, image wrap around is not an issue

GASTROSCOPY AND COLONOSCOPY

Figure 2

MR endoscopic procedure being performed

when choosing small ®elds-of-view. On the other hand, it is important to avoid image degradation associated with motion. The exact sources of motion artefact depend on the positioning of the coil. Major contributions arise from the pulsatile motion of the heart and great vessels in the thorax and from respiratory motion, as well as from patient movement if scan times are too long. If no immobilization is provided for the receiver coil, the sensitivity pro®le is such that smearing of the very high signal from regions immediately adjacent to the coil occurs and can swamp signal from regions of interest a few centimeters from the coil. There are many well-known strategies for dealing with motion artefact in MR imaging. In the upper GI tract, cardiac gating is useful. Typically a standard gated spin echo sequence will have a repetition time (TR) of 700±800 ms, so T1-weighted images, which provide useful anatomic landmarks, may be obtained. These sequences do not have a high gradient demand; consequently, with normal imaging gradients (10 mT mÿ1 with a rise time of 1 ms), 3 mm slice thicknesses and 10 cm ®elds-of-view can be easily achieved. A small reduction in the achievable resolution and the number of slices that can be scanned (although still more than enough to cover the extent of

3

the coil) allows the use of gated fast-spin echo sequences, which signi®cantly reduce the scan times. With longer pseudoecho times, these sequences give more T2-weighted images. Although gating greatly improves image quality, images collected in this way are still subject to extensive motion artefact. In these sedated patient examinations, breath-hold techniques are impractical while the requirement for high resolution means that single-shot techniques that have proved useful for body imaging, such as HASTE (half-Fourier acquisition single-shot turbo spin echo),6 require gradient performances well above those that are generally available. Single-slice, fastscanning techniques may be used to further reduce artefact by `freezing' motion. An alternative technique is to use gated fast gradient echo sequences with segmented k-space acquisition.7,8 At a TR of 15 ms, 32 lines or more of data can be collected per R±R interval and scan times of a few seconds are possible. Collection of the full image data can be divided over a few heartbeats. Optimization of the SNR depends on further re®nements of the sequences, such as varying the ¯ip angle through each cycle of data acquisition.9 Preparation pulses can be used before the data collection in each cardiac cycle in order to achieve the required image contrast. In combination with this, centrally ordered phase encoding seeks to collect views from the center of k-space early in each cycle of data collection. In the colon, cardiac motion is not an issue; however residual bowel air reduces contact between the receiver coil and the bowel wall and facilitates movement of the coil within a distended lumen. Modi®ed gradient optimized gradient echo sequences with a TR of 8±15 ms and TE of 3.7±6.1 ms can achieve a 1±25 s acquisition. In addition, versions of this sequence may be used with an inversion prepulse of 300±700 ms to maximize contrast between bowel wall layers. Finally navigator echoes for line-by-line quanti®cation of motion, together with postcorrection or phase encode re-ordering techniques, have been used in, for instance, cardiac imaging.10,11 For the application here, the limited ®eld-of-view of the imaging coil means that navigator echoes are impractical. We have investigated exploiting similar techniques by adding an additional small receiver coil (internal diameter ~3 mm) within the tip of the endoscope, attached to a separate receiver channel and containing a small MR-visible ®ducial marker (~1 mm diameter).12 Motion of the endoscope tip can then be monitored by application of four gradient projections,13 acquired in 60 ms, before each line of data acquisition, from which the current position of the ®ducial can be measured. In situations where fast scan times are not necessary for patient tolerance, this method allows full exploitation of the coil sensitivity for high-resolution imaging within the gradient performance of clinical scanners, as well as permitting the full range of diagnostic image contrast.

5

EX VIVO IMAGING

From inner to outer, four layers are seen in normal bowel (Figure 3). First, a layer of high signal intensity on both T1 and T2-weighted images that represents the mucosa. Second, there is a layer of low signal intensity on T1-weighted imaging that represents the muscularis mucosa. This layer is intermediate in signal intensity on T2-weighted imaging and, therefore, indistinguishable from the adjacent outer layer. Third, a layer of

4 GASTROSCOPY AND COLONOSCOPY

Figure 3 The normal colon ex vivo showing normal colonic layers. (a) Transverse spin echo T1-weighted image [repetition time (TR) 355 ms; echo time (TE) 20 ms]; (b) T2-weighted image (TR 2500 ms; TE 80 ms); (c) histologic section. The innermost ring of high signal intensity represents mucosal epithelium (arrowhead). The next thin layer of low signal intensity represents the muscularis mucosa (short arrow). The submucosa is seen as a band of high signal intensity in (a) and as an inner ring of intermediate signal intensity and an outer ring of low signal intensity in (b) (curved arrow). The muscularis propria is a thick band of low signal intensity in both (a) and (b) (long arrows). This correlates with the appearance on histologic section

high signal intensity on T1-weighted imaging is seen that has two counterparts on the T2-weighted images: an inner layer of intermediate signal intensity and an outer rim of lower signal intensity. These correspond to the submucosa. Finally, a homogeneous thick outer layer has low signal intensity on both T1and T2-weighted imaging; this represents the muscularis propria. 6 IN VIVO STUDIES 6.1 Upper Gastrointestinal Tract The procedure initially follows standard endoscopy practices. Following local anesthesia to the back of the pharynx, the patient's esophagus is intubated with the coil in advance of the tip of the endoscope. After visual inspection of the upper GI tract, the coil is placed adjacent to a lesion or region of interest under direct vision and the patient positioned with the region to be imaged in the center of the magnetic ®eld.14 Use of a short-bore scanner (0.5 T, 60 cm from magnet face to center) allows access to the patient's mouth. Imaging strategies described in Section 4 are then used to study normal

anatomy (Figure 4), the extent of tumors (Figure 5), varices not detectable on visual inspection (Figure 6), and post-surgical appearances at the gastroesophageal junction. 6.2

Lower Gastrointestinal Tract

Patient preparation is identical to that used for a standard colonoscopy. During visual inspection of the colon, note is taken of areas where imaging may be of interest. On completion of inspection, the receiver coil is placed adjacent to these areas and images obtained. As in the ex vivo studies, three layers of bowel wall can be identi®ed: mucosa, submucosa, and muscularis propria (Figure 7).

7 7.1

SAFETY ISSUES Technical Factors

An alternating magnetic ®eld gives rise to an electric ®eld in a plane normal to the magnetic ®eld. This statement of Faraday's law may be made more concisely as

GASTROSCOPY AND COLONOSCOPY

5

Figure 4 The normal mid-esophagus. Spin echo T1-weighted transverse image (repetition time 660 ms; echo time 20 ms) showing normal esophagus (arrowhead), ascending and descending aorta (open arrows), and azygos vein (short arrow)

Figure 5 Carcinoma of the esophagus. Spin echo T1-weighted transverse image (repetition time 660 ms; echo time 20 ms) through the midesophagus showing a large mass of intermediate signal intensity in the mediastinum (arrowheads) surrounding the esophagus. A pulmonary vessel is seen passing through it (arrow)

6 GASTROSCOPY AND COLONOSCOPY

Figure 6 Gastric varices. Electrocardiograph-gated spin echo T1-weighted transverse image (repetition time 660 ms; echo time 20 ms) at the level of the gastric fundus (open arrow) and coeliac axis (arrowhead). Numerous signal voids (arrows) at the splenic hilum represent recurrence of varices in the left gastric territory

‡ Edl ˆ

… ÿd BdA dt

where E is the induced electric ®eld strength along a path of length dl, t is time, and B is the inducing magnetic ¯ux density over an area dA, all variables being vector qualities. If we take the relatively simple case of a closed circuit path of radius r †in free space in a plane normal to the magnetic ¯ux, then Edl ˆ E:2r. The corresponding rate of ¯ux change is: … ÿ

dB dA dt

which will have magnitude dB 2 r : dt Figure 7 Normal colon. In vivo spin echo T1-weighted transverse image (repetition time 660 ms; echo time 20 ms) through the sigmoid colon showing a trilaminar wall structure of mucosa (intermediate signal intensity; short arrow), submucosa (high signal intensity; curved arrow), and muscularis propria (low signal intensity; long arrow)

Then E:2r ˆ

dB 2 r dt

GASTROSCOPY AND COLONOSCOPY

and Eˆ

dB r dt 2

Consequently, the induced electric ®eld increases linearly with the enclosed loop radius. If part of this circular path is replaced by a conducting medium, a copper wire for example, the resulting electric ®eld within the wire will be zero and the electric ®eld around this loop will be concentrated into the lower conductivity portion of the loop outside the wire. Since, for a given diameter, the total loop electromotive force (E.2r) is ®xed, increasing the length of the wire portion of the loop to create a smaller gap will increase the electric ®eld in the gap proportionately. Should this gap be now ®lled with a nonmetallic medium capable of supporting electric currents (e.g., human tissue), this relatively high electric ®eld could give rise to currents suf®ciently high to produce signi®cant tissue heating and a potential safety hazard. Such a situation can exist within a MR scanner if suitable precautions are not taken. The B1 magnetic ®eld, alternating at the Larmor frequency, typically irradiates the whole body. Under normal operating conditions, the induced electric ®elds and their associated currents are small enough that the resultant tissue heating is well within safety guidelines. However, the introduction of a conductor, for example an electrocardiograph lead positioned so as to form a large loop, can result in a large electric ®eld between the ends of the loop. If this high-®eld region is adjacent to the body surface, the local electric current can produce tissue heating that far exceeds the safety guidelines, resulting, in extreme cases, in localized tissue burning. This conducting-loop problem could equally apply to a colonoscope, for example as it performs loops within the abdomen. This hazard is minimized by providing a suf®ciently thick layer of insulating material of high dielectric strength between conductive material within the colonoscope and the external tissue, so ensuring that electric ®elds are con®ned largely within these layers and that their levels are kept within reasonable limits. The coil is coated in an epoxy-based adhesive commonly used in medical devices. It is biologically inert and nontoxic, meeting US pharmacopeia biocompatibility standards, thus avoiding allergic reactions and long-term toxic effects. In addition, it is chemical and water resistant, which allows sterilization of the coil by soaking it in a solution of glutaraldehyde. 7.2 Clinical Factors Upper GI endoscopy carries a signi®cant morbidity and mortality.15 In a large prospective audit across 36 hospitals to include over 14 000 examinations, the death rate was 1 in 2000 and the morbidity rate 1 in 200; cardiorespiratory problems were most prominent and there was a strong relationship between the lack of monitoring and the use of high-dose benzodiazepines and adverse outcomes. A lack of recovery area and poor staf®ng exacerbates this problem. No complications have been experienced in our unit so far, but as the procedure is considerably longer than a standard upper GI endoscopy and often requires an additional dose of sedative, an

7

increase in the morbidity and mortality must be anticipated. Also although the patient is accessible, this is less than achieved in a standard procedure and all monitoring requires increased vigilance. Continuous oxygen and close observation of the sedated patient during the entire period of imaging must be maintained. Performing endoscopy in the MR environment, therefore, requires additional considerations. The MR suite must be equipped with full resuscitation facilities nearby. This includes oxygen available to the patient in the scanner and a MR-compatible trolley to transfer a patient in an emergency to a space outside the scanner room where normal resuscitation procedure may be carried out. Finally, all staff must be trained in metal safety and in resuscitation practice peculiar to the MR environment.

8

FUTURE APPLICATIONS

MR endoscopy will provide a useful means of staging neoplastic masses in the upper and lower GI tract. It may also be used as a way of monitoring or guiding therapeutic interventions administered endoscopically. For example, injection or banding esophageal varices may bene®t from being performed with simultaneous cross-sectional images so that the completeness and ef®cacy of the procedure can be monitored in realtime. Laser treatments for the palliative recanalization of esophageal carcinoma may also be monitored in this way, thus reducing the risks of perforation or inadequate treatment and optimizing treatment outcome.

9

CONCLUDING REMARKS

MR imaging using a surface coil placed within the GI tract during an endoscopic procedure enables visualization of mural and extramural lesion extension and may provide a useful adjunct to diagnostic upper and lower GI endoscopy. Formal evaluation of this technique and comparison with ultrasound, computed tomography, and surgical ®ndings is warranted.

10

RELATED ARTICLES

Coils for Insertion into the Human Body; Design and Use of Internal Receiver Coils for MRI.

11

REFERENCES

1. T. W. Rice, G. A. Boyce, and M. V. Sivak, J. Thorac. Cardiovasc. Surg., 1991, 101, 536. 2. M. D. Rifkin, S. J. Gordon, and B. B. Goldberg, Radiology, 1984, 151, 175. 3. B. B. Goldberg, R. M. Steiner, J. B. Liu, D. A. Merton, G. Articolo, J. R. Cohn, J. Gottlieb, B. L. McCombe, and P. W. Spirn, Radiology, 1994, 190, 233. 4. B. N. Milestone, M. D. Schnall, R. E. Lenkinski, and H. Y. Kressel, Radiology, 1991, 180, 91.

8 GASTROSCOPY AND COLONOSCOPY 5. N. M. deSouza, I. C. Hawley, J. E. Schwieso, D. J. Gilderdale, and W. P. Soutter, Am. J. Roentgenol., 1994, 163, 607. 6. T. Miyazaki, Y. Yamashita, T. Tsuchigame, H. Yamamoto, J. Urata, and M. Takahashi, Am. J. Roentgenol., 1996, 166, 1297. 7. A. Haase, J. Frahm, and K. D. Matthaei, J. Magn. Reson., 1986, 67, 258. 8. A. Haase, Magn. Reson. Med., 1990, 12, 77. 9. M. K. Stehling, Magn. Reson. Imag., 1992, 10, 165. 10. Y. Wang, R. C. Grimm, J. P. Felmlee, S. J. Riederer, and R. L. Ehman, Magn. Reson. Med., 1996, 36, 117. 11. A. M. Taylor, P. Jhooti, D. N. Firmin, and D. J. Pennell, JMRI, 1999, 9, 395. 12. G. A. Coutts, D. J. Gilderdale, K. M. Chui, L. Kasuboski, and N. M. deSouza, Magn. Reson. Med., 1998, 40, 908. 13. C. L. Dumoulin, S. P. Souza, and R. D. Darrow, Magn. Reson. Med., 1993, 29, 411. 14. N. M. deSouza, A. H. Gibbons, G. A. Coutts, A. S. Hall, R. Puni, J. Calam, and I. R. Young, Minim. Invas. Ther., 1995, 4, 277. 15. A. M. Thompson, K. G. Park, F. Kerr, and A. Munro, Br. J. Surg., 1992, 79, 1046.

Acknowledgements We gratefully acknowledge the help we received in undertaking the work described here from Picker International Inc., Cleveland, OH, and Endoscan Ltd, Burgess Hill, UK. The work was supported by MedLINK grant P37.

Biographical Sketches Nandita M. deSouza. b 1957. B.Sc. Hons (Physiology) 1978, MBBS, 1983, University of Newcastle upon Tyne, UK, M.R.C.P. 1986, F.R.C.P. 1991, M.D. 1996. House Surgeon and Physician and Senior House Physician, Newcastle upon Tyne 1983±85. Registrar in internal medicine 1986±87. Registrar and Senior Registrar in Diagnostic Radiology, Guy's Hospital, London 1987±92. Senior Research Fellow in MR Imaging, Royal Postgraduate Medical School, Hammersmith Hospital, 1993±97. Senior Lecturer 1997±present. Research Interests: internal receiver coils, interventional MRI. Glyn A. Coutts. b 1959. B.A. (Physics) University of Oxford, UK, 1980; D.Phil, University of Oxford 1985. Senior Research Associate at Picker Research UK, 1987±99. Marconi Medical Systems, UK, 1999± present. Approximately 30 publications in the ®eld of MRI and MRS. Current interest: interventional MRI. David J. Larkman. b 1969. 1993, Ph.D. Physics, Manchester University, 1997, BSc (Hons) Physics, Liverpool John Moores University, Research Scientist at the Robert Steiner MRI Unit, Hammersmith Hospital, London, UK, 1997±99. Marconi Medical Systems, UK, 1999± present. Publications as conference and journal papers on small-coil MR, fast imaging using multicoil techniques, and artefact control in MR images. David J. Gilderdale. b 1947. B.Sc. telecommunications, Ph.D. computer simulation as an aid to lighting design. Research Assistant, Plymouth Polytechnic, 1974±78. Electronic Engineer, Central Research Laboratories, Thorn-EMI Ltd, 1978±81. Senior Scientist, GEC Hirst Research Centre, 1981±84. Senior Lecturer, Department of Electrical Engineering, Plymouth Polytechnic, 1981±88. Engineering Consultant, 1988±present.

Interventional MRI: Specialist Facilities and Techniques Glynn A. Coutts and Nandita M. deSouza Royal Postgraduate Medical School, Hammersmith Hospital, London, UK

1 2 3 4 5 6

Introduction MR-Guided Diagnostic Procedures MR-Guided Therapeutic Procedures Conclusions Related Article References

1

INTRODUCTION

1 1 7 9 9 9

The excellent anatomical detail and high level of soft tissue contrast provided by magnetic resonance imaging (MRI) are potentially of enormous value in guiding both diagnostic and therapeutic procedures. However, the use of MRI has been limited because of the poor access to the patient provided by most scanners and the practical difficulties of working within a magnetic field. Both these problems are now being addressed. Newer more open-plan magnet configurations provide greatly improved access to the patient, and manufacturers are now producing a wide range of MR-compatible equipment. The indications for MR-guided diagnostic procedures include lesions visualized only with MRI, e.g. focal lesions in the breast where MR-guided biopsy may be carried out using MR-compatible needles. The technical aspects of such procedures may also be simpler under MR guidance when particular access routes are preferred because of the oblique planes and 3D acquisition facility offered by MRI. In addition, MR-guided diagnostic procedures may be performed via endoscopes and facilitated by the use of specially designed insertable coils. These dedicated surface coils inserted through body orifices or into blood vessels and placed adjacent to localized areas of interest, e.g. cervix, prostate, oesophagus, and inner ear, provide remarkably high resolution images. MR-guided therapeutic procedures are largely performed through flexible fiber-optic endoscopes, and may embrace a wide clinical spectrum of applications including treatment of very common conditions such as benign prostatic hypertrophy and prolapsed intervertebral disks.

2 2.1

MR-GUIDED DIAGNOSTIC PROCEDURES Breast Biopsy

Over a decade has elapsed since the first clinical trials predicted an important role for MRI in breast cancer diagnosis.1 – 3 Use of gadopentate dimeglumine as a contrast agent in the

1980s showed that malignant breast tumors enhanced reliably and predictably,4 – 6 and thus lack of enhancement excludes malignancy. However, other lesions such as atypical hyperplasia and areas of fibrocystic change also enhance, so that the specificity of MRI for malignancy is low.7 More detailed studies demonstrated that cancers enhanced early, and were best discriminated from benign lesions in the first 2 min after injection of contrast agent.5 The rise time for the enhancement of benign lesions, such as sclerosing adenosis and fibroadenomas, is generally slower than for neoplastic lesions,7 but there is overlap in individual cases, which makes it difficult to exclude a definitive diagnosis of malignancy with the necessary high level of certainty. These difficulties in interpretation are compounded following treatment with surgery and/or radiation therapy. Posttreatment changes mimic malignancy clinically, mammographically, and on ultrasound.8 – 11 Mammographic evaluation of such breasts is frequently impaired by decreased compressibility, generalized or locally increased density, architectural distortion, skin thickening, mass-like areas of fibrosis or fat necrosis, and indeterminate microcalcification. Furthermore, different radiation protocols as well as individual differences in tissue response may affect the timing of this enhancement seen on MRI. As a result, histology is frequently necessary not only for the diagnosis of the primary lesion but also for the diagnosis of disease recurrence in treated patients. MR-guided biopsy can therefore provide histological diagnosis of primary lesions not visible on mammography, secondary foci and nonspecific areas of contrast enhancement, and lesions visible in mammographically uninterpretable cases, e.g. dense and treated breasts. It provides a means for identifying multifocal disease where mastectomy rather than lumpectomy may be indicated. MRI may also be used for evaluating and biopsying focal clinical abnormalities during pregnancy when mammography and radiation exposure are undesirable. 2.1.1 Techniques

Several schemes for MR-guided breast biopsy have so far emerged. One method uses the patient in the prone position with the breast under gentle compression in a multicoil array. Following the diagnostic study the lesion is reimaged using an interventional device consisting of a perforated lateral plate incorporating a single coil and a medial plate incorporating two coils.12 Intravenous gadolinium contrast agent is administered for lesion identification, and fiducials are used to calculate the coordinates of the entry site. Stereotactic MR-compatible biopsy equipment is now becoming commercially available, and may be used to place needles with 1 mm accuracy. The first step is the acquisition of the 3D data set, which is equivalent to obtaining the scout views for stereotactic mammographic biopsy. The coordinates of the lesion are then calculated from the reconstructed images, and entered into the unit which aligns the gun and needle along the proper trajectory. In comparison with conventional stereotactic mammography, use of the 3D data set reduces the errors associated with estimation of tip position and makes it possible to visualize the lesion and the needle trajectory in three planes. In patients who have had surgery and radiotherapy to the axilla, elevation of the arm is difficult and sometimes painful, making the prone position unacceptable. In these patients the

2 INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

Figure 1 Breast biopsy device comprising baseplate with integral surface coil and (a) upper compression plate with holes for needle access and (b) shaped thermoplastic material. This is immobilized on three sides to the modified baseplate using Plexiglas blocks

breast may be immobilized in the lateral decubitus position. This provides the added advantage of an easy medial or lateral needle approach that automatically runs parallel to the chest wall. This is essential in the distorted scarred breast, where the lesion is often retracted toward the chest wall. Also, in the lateral decubitus position, good access is provided to lateral lesions of the uppermost breast and medial lesions of the lowermost breast. Our biopsy device consists of a rectangular receiver coil coupled to a mechanism for immobilization (Figure 1). The receiver coil is a 180 × 80 mm rectangle of copper foil. The coil is mounted at a distance of 2 mm from a double split radiofrequency screen to minimize electric field effects.13 The whole assembly is mounted 10 mm behind the breast support plate. Immobilization is achieved by using two flat compression plates with one edge contoured to the chest wall. The lower plate incorporates the rectangular receiver coil mounted on its outer surface. The upper plate contains a 250 × 140 mm area of holes approximately 5 mm apart. The plates are connected by two posts: the position of the upper plate can be adjusted, and it is locked in place when breast compression is adequate (Figure 1(a)). Adequate immobilization with a greater degree of comfort may be achieved using a nonshrinking thermoplastic that is commercially available for use as a radiation therapy mask (Figure 1(b)). It is a thermoplastic that is rendered deformable by heating in a water bath at 70–80 ◦ C for 30–60 s. It can then be stretched and molded around the breast to be biopsied. Hardening occurs in 60 s, after which it forms a rigid exoskeleton. It is fixed to the baseplate on three sides and fastened around the chest wall using tapes. Two MR-visible scales provide the X - and Y -coordinates of the lesion to be biopsied, and the depth is calibrated from the images in a perpendicular plane using an electronic caliper. Another approach is to use MRI guidance without having to perform the biopsy within the bore of the scanner. This can be achieved by using frameless stereotaxy with an ultrasonic localizer. It is based on a localizing tip with a handle onto which two ultrasound emitters are mounted (the ‘wand’). Ultrasound detectors on a board placed nearby can locate the position of the tip in three dimensions to an accuracy of 2 mm (Figure 2). A 3D data set is then acquired with the breast

immobilized, and the data are transferred to a workstation. MR-visible fiducials placed on the breast and imaged at the same time are then registered in relation to the images of the breast using the ‘wand’. The image of the lesion and the position of the biopsy needle mounted on the localizing ‘wand’ can be displayed simultaneously on the screen so that the entry point and the needle trajectory can be planned. 2.1.2 Needles

A variety of MR-compatible hollow needles suitable for fine-needle aspiration cytology of the breast are currently commercially available in 22G, 20G, and 18G diameters and in 5, 10, and 15 cm lengths. The tip designs include straight short bevels (Lufkin needle, E-Z-M, New York, U.S.A.; Chiba needle, William Cook, Europe), a serrated edge (Franseen needle, William Cook, Europe), and a spiral cutting edge (Tip Cut needle, William Cook, Europe). With the serrated or spiral bevel needles, a 360◦ quick rotation of the needle prior to withdrawal may yield small fragments of tissue for histology. However, fine-needle aspiration has a failure rate due to insufficient tissue as high as 37%, and reported falsenegative rates as high as 31%.14 The technical failure rate is higher in the treated breast, where sampling from an enhancing scar can produce a very acellular aspirate. In this situation, a two-piece cutting-edge core needle is usually necessary. Even with MR-compatible alloys a small susceptibility effect may be seen around the needle. This may arise from the metal itself or the air within. Seen in cross-section, this artifact appears as a triangular central signal void with bright edges (Figure 3). It varies with magnetic field strength, alloy, and diameter of needle. Operators should be familiar with the type and amount of artifacts produced by the needles they are using at their particular operating frequencies in order to avoid errors in locating the needle. 2.1.3 Image Localization and Marking of Lesions Prior to Surgery

Needle marking for localization of breast lesions based on imaging immediately prior to surgery was developed

INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

3

Figure 2 Targeting a breast lesion using a 3D data set. The needle trajectory is projected in three planes in relation to the lesion and the surface entry point

to reduce the amount of tissue excised during biopsy.15 – 18 The original technique involved the freehand placement of a needle in the breast as a marker to guide the surgeon.15 Although needle localization is more accurate than the external localization alone, there are significant problems due to needle displacement before or during surgery. Localization wires with hooked ends have been developed to secure the tip of the marker wire in place within the breast,19 – 21 but with mammography these may become displaced because patient and breast movement may occur when altering the position of the compression plates in order to obtain an image in a perpendicular plane. MRI not only provides 2D imaging in multiple planes without moving the patient, but a volume acquisition allows reconstruction in multiple planes. MR-compatible marker wires are now commercially available (Kopans wire, William Cook, Europe), and, using a technique similar to that for MRguided breast biopsy, it is possible to place a hooked wire in a lesion and confirm its positioning using a 3D imaging data set. With faster scanning methods it will also be possible to use real-time guidance to advance and steer the wire into the center of the lesion. 2.2

Upper Gastrointestinal Endoscopy

Upper gastrointestinal endoscopy is limited to visualizing lesions on the surface mucosa, and though the extent of invasion beneath the surface may be shown with biopsy,

lesions with no surface component or those with infiltration remote from the biopsy site remain a problem in diagnosis. MRI provides a means of visualizing the full submucosal and intramural extent of a lesion deep to the surface, and thus complements existing techniques. Endoscopes require detailed reengineering to be acceptable for use in an MR scanner. Because many of the component parts of a standard endoscope such as the steel tension steering wires have a detrimental effect on the uniformity of the static magnetic field (B 0 ), it is necessary to replace them with MR-compatible alloys. In other respects the appearance and function of such an instrument [Figure 4(a)] are identical to those of a standard endoscope.22 A small coil positioned just ahead of the endoscope and inserted with it may be placed in the upper gastrointestinal tract adjacent to surface lesions or suspected sites of pathology. A coil 50 mm in length with a diameter identical to that of the endoscope is optimal for ease of insertion whilst retaining a useful field of view for imaging. We have developed such a coil which incorporates a hybrid circuit to satisfy tune and match as well as decoupling requirements [Figure 4(b)]. The cable to the coil is fed through the biopsy channel of the endoscope. The procedure is performed under sedation in the center of the magnetic field. This can easily be managed in an openplan or even a short-bore magnet such as the Picker Asset (70 cm magnet face to center plane) whilst maintaining full monitoring with pulse oximetry and electrocardiography, as well as providing pharyngeal suction as necessary.

4 INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES endoanal use to image the anal sphincter, and for use in the external auditory meatus to image the cochlea. 2.3.1 Cervix

Figure 3 T 1 -weighted SE (660/20) (a) sagittal and (b) axial scans with an MR-compatible needle in situ during a biopsy procedure. The needle tip lies at the periphery of enhancing scar tissue (arrow)

Imaging in the esophagus with the coil in position necessitates the use of cardiac gating or fast imaging techniques to reduce motion artifacts from the heart and great vessels. High-resolution images of the esophagus and adjacent mediastinum (Figure 5) may be obtained.22 Initial experience with this technique is promising for visualizing intramural changes (for example, mural thickening may be seen at the level of a stricture in reflux esophagitis). 2.3

Insertable Coils

These have been developed for endovaginal use in order to image the cervix, for endorectal use to image the prostate, for

In early cervical neoplasia, conventional MRI is of limited value in detecting invasion of the cervix when this is only of the order of a few millimeters.23 There are also difficulties in demonstrating extension of disease into the urinary bladder,24 as well as involvement of adjacent parametrial lymph nodes. Endorectal coils have therefore been used to increase the resolution of cervical images,25 but while this helps to visualize the posterior lip of the cervix, the drop off of signal anteriorly makes demonstration of the anterior lip of the cervix and the fascial planes between the cervix and bladder difficult. We have developed a solenoidal receiver coil mounted on an acetal homopolymer (Delrin) former (internal diameter, 37 mm) for endovaginal placement around the uterine cervix which provides high-resolution images of this structure.26 The coil is attached to a 20 cm tubular Delrin handle (Figure 6). Although designs with an angle of 45◦ between the coil and handle have been used, in most cases a design with no angle between the coil and handle suffices. A 10–15 cm field of view is optimal, as good parametrial detail is seen up to 6 cm from the center of the coil. With this technique (pixel size, 0.6 mm2 ), details of the endocervical mucosa are easily distinguished, as well as the inner and outer zones of cervical stroma [Figure 7(a)]. The mucosa shows a smooth and regular outline in the nulliparous cervix and a more irregular and indented outline in the parous cervix. In addition, dilated glands filled with secretions are sometimes seen in the latter group. This detail is rarely provided by standard body coil or even phased-array coil images. The stromal zones are readily distinguished, and no obvious differences are seen in women taking the oral contraceptive pill or between the follicular and luteal phases of the menstrual cycle. On administration of 0.1 mmol kg−1 of gadopenetate dimeglumine, brisk mucosal enhancement is seen at 30 s, which peaks at about 120 s after injection. The inner and outer stroma enhance more slowly, with the outer zone showing more prominent enhancement, and zonal differentiation being maximal at about 90 s.26 In Stage I carcinoma, use of an internal coil provides accurate delineation of neoplastic lesions [Figure 7(b)], including the precise position and extent of stromal invasion as well as of extension of the tumor into the parametrium, bladder base, rectum, and locoregional lymph nodes.27 2.3.2 Endorectal Coil

Carcinoma of the prostate is the second leading cause of cancer death in males. For younger patients, surgical excision offers the best chance of survival. MRI using an endorectal coil helps distinguish whether the disease is confined to the capsule and is therefore suitable for treatment by radical prostatectomy. The first prototype endorectal coil was developed in 1988, and a 3.5 × 8 cm inflatable version is now commercially available with a tune and match box (Medrad, Pittsburgh, USA). It provides a two- to fourfold increase in the signal-tonoise ratio, and allows thin slices (3.0 mm) and small fields of view (10 cm) to be used. It is flexible, and may be inserted with a minimum of discomfort. As the device requires inflation of

INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

5

Figure 4 (a) MR compatible gastroscope. The internal receiver coil cable is passed through the biopsy channel. (b) Internal receiver coil with integral electronics for an endoscope. The diameter of the coil is the same as that of the endoscope

Figure 5 T 1 -weighted SE (600/20) transverse image through the midoesophagus of a normal volunteer showing the trachea (short arrow), aorta (long arrow), and signal void from a coil in the oesophagus (arrowhead)

a balloon in the rectum, potential contraindications to its use are radiation strictures, rectal fistulas, and obstructing rectal masses. Maximal contrast between the central and peripheral zones of the prostate is obtained on the T 2 -weighted images, where the central zone appears of low signal intensity and the peripheral zone, which is more prominent posteriorly, is of intermediate signal intensity [see Figure 12(a)]. With increasing age the

Figure 6 Intravaginal coil with a clamp stand for immobilizing it in situ. The height, angle, and tilt can be adjusted at both universal joints (arrows)

central zone enlarges and may become adenomatous (benign prostatic hypertrophy). Nodular components within such an adenoma are often mucus-containing, and are seen as high signals on the T 2 -weighted scans (Figure 8). Carcinomas tend to occur in the peripheral zone often adjacent to the capsule, and are usually seen as low signal intensity nodules. The use of fat suppression techniques either with an inversion recovery approach28 or with chemical saturation using either frequency selective chemical saturation

6 INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

Figure 8 T 2 -weighted SE (1000/130) transverse image through the mid-prostate in benign prostatic hypertrophy. The central zone of the gland is nodular and massively enlarged. Mucus-containing nodules with long T 2 components (arrow) are prominent

images.32 However, low-signal areas in the seminal vesicles may also result from calculi or adenomatous nodules of benign prostatic hypertrophy protruding posteriorly. Understaging of prostatic carcinoma despite the use of an endorectal coil is common, and may occur because sheets of infiltrating neoplastic cells grow around glandular structures without causing displacement or distortion. Moreover, microscopic extension into the walls of the seminal vesicles is frequent and often cannot be demonstrated even with the increased resolution afforded by the endorectal coil technique. 2.3.3 Anal Coil

Figure 7 T 2 -weighted SE (2500/80) transverse images obtained with a cervical coil in the normal nulliparous cervix show (a) a regular mucosal outline (small arrows) and a low signal intensity inner stromal zone (arrowheads). Nabothian follicles are clearly seen (long arrows). (b) In Stage I cancer the inner stromal zone is displaced by the central tumor (arrowheads)

or a three-point Dixon technique29 has been reported to improve the detection of the capsular penetration. Experience with gadopentate dimeglumine in the preoperative assessment of carcinoma of the prostate has proved it to be of marginal use, with no improvement in the detection of malignant nodules over the T 2 -weighted images.30 Use of an endorectal coil improves the overall staging efficacy of MR over body coil images.31 Minor extracapsular invasion may be identified as an irregular bulge or asymmetry of the neurovascular bundle due to retraction. Tumor infiltration of the normally fluid-filled seminal vesicles is seen as low signal intensity foci within them on the T 2 -weighted

Disruption of the function of the anal sphincter is a common and distressing problem in which preoperative imaging is of considerable value. At present, transrectal sonography is the ‘gold standard’ for this purpose, but differentiation of the muscle layers (particularly anteriorly in females) is not always possible.33 MRI of the anal sphincter with a dedicated coil provides very high-resolution images of the muscle layers. We have developed a coil consisting of a simple saddle geometry on a 9 mm diameter Delrin former.34 The length of the coil is 50 mm, and it is designed to image the anal sphincter from the subcutaneous external component to the lower rectum. The coil is encased in an outer sheath, and molded in shape so that it is anatomically acceptable. It incorporates a hybrid circuit to satisfy tune and match as well as B1 decoupling requirements. To reduce coil motion during imaging, we immobilize it in an external clamp which is mounted on a baseplate under the patient’s thighs. Universal bossheads allow adjustment of the angle and tilt of the coil (Figure 9). Good differentiation is obtained between the external sphincter, longitudinal muscle of the rectum, and the internal sphincter34 (Figure 10). The internal sphincter is of high signal intensity on both T 1 - and T 2 -weighting, and enhances very briskly on administration of intravenous gadopentate dimeglumine. 2.3.4 Ear Coil

An 8 × 10 mm coil angled at 55◦ to mimic anatomically the plane of the tympanic membrane may be inserted under direct

INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

7

Figure 9 Insertable anal coil with integral electronics for imaging the sphincter. The coil may be immobilized in a clamp stand during imaging

Figure 11 T 2 -weighted SE (2500/80) coronal images through the petrous temporal bone in a cadaver specimen using a dedicated ear coil in the external auditory canal. High-resolution images of the cochlea (arrow) and semicircular canals (arrowhead) are obtained

Figure 10 T 1 -weighted SE (720/20) transverse image through the normal anal sphincter showing the internal sphincter (arrowhead), the longitudinal muscle of the rectum (short arrow), and the external sphincter (long arrow)

vision into the external auditory meatus in order to obtain high-resolution images of the cochlea and semicircular canals (Figure 11). Placement of such a coil is best performed with an operating microscope, and may also require local anesthesia of the tympanic membrane.

3 3.1

MR-GUIDED THERAPEUTIC PROCEDURES Laser Thermotherapy Techniques

Tissue damage can be produced for therapeutic purposes using lasers. The extent depends on the distribution of the laser light, the optical properties of the target tissue and the wavelength of the light used. The most important effects are

thermal, as a result of the laser energy being dissipated in the target tissue as heat. The neodymium:yttrium aluminum garnet (Nd:YAG) laser with its near-infrared beam penetrates tissues for several millimeters, and may be transmitted via flexible optical fibers. These fibers can be passed down hollow needles into solid organs in order to carbonize tissue at low power (2–5 W),35 or may be passed down endoscopes into hollow organs and be used at much higher powers (50–70 W) to recannalize them, e.g. in the esophagus, and urethra.36 Such treatments achieve in situ necrosis of tissue, but the dose of laser energy is often empirical, and the depth of penetration and extent of tissue damage is frequently unknown. The use of MR to image tissue changes during the delivery of laser energy helps visualize the degree and extent of the changes, so that the dose of laser energy may be matched to the tissue response. Previous animal studies have correlated laser-induced acute tissue damage with pathological findings.37 – 39 During application of laser energy to rabbit muscle with a direct contact fiber at 2000 J, three layers were seen with T 2 -weighted MR scans: a central cavity with a high signal rim corresponding to carbonized tissue, a broad low signal intensity layer corresponding to denatured tissue with a zone of coagulative necrosis, and a high signal zone corresponding to a layer of edema.39 Similar therapy in patients with liver tumors using a diffusing fiber tip implanted directly in the lesion and 4320 J of laser energy demonstrated an area of hyperintensity on T 1 weighted scans at 2 min.40 This may have been due to the formation of methemoglobin in the heated tissue, as has been shown in vitro.41 The noncontact mode and a side-firing technique at 40–60 W may be used in hollow organs such as the prostate to achieve maximal coagulation and minimal evaporation of tissue.42 With this technique, improvement in symptoms may occur due to sloughing of tissue over a period of months. However, this may also cause obstructive and/or irritative symptoms for variable periods of time.43,44 The technique

8 INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES was perfected on a canine prostate model,43 which, unlike the human prostate, is very acinar and has little stroma.45 It therefore undergoes more dramatic sloughing of tissue when laser coagulation is performed. 3.1.1 Laser Prostatectomy

Endoscopic laser ablation (ELAP) is challenging the position of transurethral prostatectomy (TURP) in the management of benign prostatic hyperplasia because it offers major advantages over TURP such as reduced blood loss and greatly reduced hospital stay.36 However, unlike TURP, the subjective and objective benefits associated with ELAP are usually not apparent for several weeks after treatment, and the resulting cavitation in the prostate is much less. As a result there may be much less long-term benefit from the procedure.46 MR-monitored ELAP provides a method of imaging during treatment, allowing the operator to match tissue changes to the size and shape of the prostate in order to improve the long-term clinical outcome.

The procedure may be performed in the MR suite under general or regional (spinal or epidural) anesthesia. After establishing anesthesia the patient is transferred to the scanning table, and preoperative images are obtained using an endorectal receiver coil. A suprapubic catheter is inserted to maintain adequate irrigation throughout the laser therapy. In a shortbore design magnet, the patient’s head lies outside the magnet whilst imaging the pelvis, thus facilitating continuous anesthetic monitoring.47 Using a Nd:YAG laser (Cardiolase 4000 Triamedyne Inc., USA) with a side-firing fiber (Bard, USA), a standard fourquadrant ablation of 40 W in each quadrant for 90 s (14 400 J) in our group of patients produced a striking increase in glandular size (approximately 34%) on the immediate postoperative images.47 This swelling was accompanied by a loss of the nodular architecture of the gland. An area of T 1 shortening was seen in some cases. On the T 2 W scans, development of a periurethral band of low signal was seen in the immediate postoperative period [Figure 12(a) and (b)].

Figure 12 T 2 -weighted SE (2500/80) transverse images through the prostate (a) before, (b) immediately after, and (c) 3 months following laser prostatectomy. In addition to glandular swelling and loss of nodular architecture, in image (b) there is a high-signal central region immediately adjacent to the catheter and probably related to it. Outside of this area there is a diffuse low-signal region within the prostate (arrows). (c) At 3 months there is a well-demarcated low-signal ring (arrowheads) several millimeters from the urethra (open arrow), but no evidence of cavitation

INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES

Glandular swelling was markedly reduced 1 week later, and resolved 3 months postoperatively. At this time a welldemarcated low-signal ring was seen several millimeters away from the urethra [Figure 12(c)]. The appearance of the gland was otherwise unchanged from the preoperative appearance, and there was no significant reduction in prostatic size or evidence of periurethral cavity formation. 3.1.2 Laser Discectomy

It has been possible to place hollow needles within the nerve root foraminae using real-time MRI, and pass microendoscopes down these in order to visualize fragments of prolapsed disk material and the adjacent nerve root prior to laser discectomy under direct vision.48 3.2

Cryoablation, Radiofrequency Ablation, and Focused Ultrasound

These modalities produce useful tissue ablation and have all been used to a limited extent under MR guidance.49 – 52 The equipment necessary is cumbersome, and not only must it be rendered MR compatible but it must be easy to manipulate if access to the patient is limited. The issue of radiofrequency shielding of such equipment whilst in use during scanning also requires consideration. Cryotherapy has the particular advantage of freezing the mobile protons and creating a ‘black hole’ effect in the treated area, thus sharply demarcating treated from normal tissue.51 Technical advances and the trend toward image-guided minimally invasive intervention are making these techniques more popular.

4

CONCLUSIONS

MRI offers unique advantages of good soft tissue contrast as well as multiplanar cross-sectional imaging. Its potential for monitoring and guiding radiological and surgical interventional procedures is now beginning to be expanded. It should enable many procedures to be optimized whilst reducing their complications, thus making minimally invasive therapy safer and more cost-effective.

5

RELATED ARTICLES

Coils for Insertion into the Human Body.

6

REFERENCES 1. S. J. El Yousef, R. J. Alfidi, R. H. Duchesneau, C. A. Hubay, J. R. Haaga, P. J. Bryan, J. P. LiPuma, and A. E. Ament, J. Comput. Assist. Tomogr., 1983, 7, 215. 2. S. J. El Yousef, R. H. Duchesneau, R. J. Alfidi, J. R. Haaga, P. J. Bryan, and J. P. LiPuma, etc. Radiology, 1984, 150, 761. 3. C. B. Sterling, P. C. Wang, A. Lieber, S. S. Mattingley, W. O. Griffen, and D. E. Powell, Radiology, 1985, 154, 457. 4. W. A. Kaiser and E. Zeitler, Radiology, 1989, 170, 681.

9

5. S. H. Heywang, A. Wolf, E. Pruss, T. Hilbertz, W. Eiermann, and W. Permanetter, Radiology, 1989, 171, 95. 6. J. P. Stack, O. M. Redmond, M. B. Codd, P. A. Dervan, and J. T. Ennis, Radiology, 1990, 174, 491. 7. S. E. Harms, D. P. Flamig, K. L. Hesley, M. D. Meiches, R. A. Jensen, W. P. Evans, D. A. Savino, and R. V. Wells, Radiology, 1994, 187, 493. 8. P. C. Stomper, A. Recht, A. L. Berenberg, M. S. Jochelson, and J. R. Harris, Am. J. Radiol., 1987, 148, 39. 9. M. Rebner, D. R. Pennes, D. D. Adler, M. A. Helvie, and A. S. Lichter, Radiology, 1989, 170, 691. 10. C. Balu-Maestro, J. N. Bruneton, A. Geoffray, C. Chauvel, A. Rogopoulus, and O. Bittman, J. Ultrasound Med., 1991, 10, 1. 11. E. B. Mendelson, Semin. Ultrasound, CT, MR, 1989, 10, 154. 12. M. D. Schnall, R. Newman, and S. Orel, Abstracts, Society of Magnetic Resonance, 1993, p. 163. 13. N. M. deSouza, G. A. Coutts, J. E. Schweiso, A. S. Hall, M. Burl, T. Krausz, and C. Vernon, Abstracts, Society of Magnetic Resonance, 1994, p. 1582. 14. G. S. Grant, J. R. Goellner, J. S. Welch, and J. K. Martin, Mayo Clin. Proc., 1986, 61, 377. 15. G. D. Dodd, K. Fry, and W. Delaney, In Management of Patients with Cancer, ed. T. F. Neaton, Saunders, Philadelphia, 1965, pp. 88–113. 16. S. A. Feig, Radiol. Clin. North Am., 1983, 21, 155. 17. H. A. Frank, F. M. Hall, and M. L. Steer, N. Engl. J. Med., 1976, 295, 259. 18. B. Threatt, H. Apelman, R. Dow, and T. O’Rourke, Am. J. Roentgenol. Radium Ther. Nucl. Med., 1974, 121, 839. 19. J. E. Myer, D. B. Kopans, P. C. Stomper, and K. K. Lindfors, Radiology, 1984, 150, 335. 20. E. J. Urrutia, M. C. Hawkins, B. G. Steinbach, M. A. Meacham, K. I. Bland, E. M. Copeland, and I. F. Hawkins, Radiology, 1988, 169, 845. 21. M. J. Homer, Radiology, 1985, 157, 259. 22. A. S. Hall, D. J. Bryant, M. Burl, and N. M. deSouza, Abstracts, Society of Magnetic Resonance, 1994, p. 1581. 23. D. Rubeno, J. R. Thornbury, C. Angel, M. H. Stoler, S. L. Weiss, R. M. Lerner, and J. Beecham, Am. J. Radiol., 1988, 150, 135. 24. H. Hricak, C. G. Lacey, L. G. Sandles, Y. C. Chang, M. L. Winkler, and J. L. Stern, Radiology, 1988, 166, 623. 25. B. N. Milestone, M. D. Schnall, R. E. Lenkinski, and H. Y. Kressel, Radiology, 1991, 180, 91. 26. N. M. deSouza, I. C. Hawley, J. E. Schwieso, D. J. Gilderdale, and W. P. Soutter, Am. J. Radiol., 1994, 163, 607. 27. N. M. deSouza, D. Scoones, J. Schwieso, D. J. Gilderdale, and W. P. Soutter, Abstracts, Society of Magnetic Resonance, 1994, p. 1542. 28. F. Parivar, V. Rajanavagam, V. Waluch, R. T. Eto, L. W. Jones, and B. D. Ross, J. Magn. Reson. Imaging, 1991, 1, 657. 29. B. Tamlet, F. G. Sommer, G. H. Glover, and E. Schneider, Radiology, 1991, 179, 43. 30. S. A. Mirowitz, J. J. Brown, and J. P. Heiken, Radiology, 1993, 186, 153. 31. M. D. Schnall, Y. Imai, J. Tomaszewshi, H. M. Pollack, R. E. Lenkinski, and H. Y. Kressel, Radiology, 1991, 178, 797. 32. M. L. Schiebler, J. E. Tomaszewski, M. Bezzi, H. M. Pollack, H. Y. Kressel, E. K. Cohen, H. G. Altman, W. B. Gefter, A. J. Wein, and L. Axel, Radiology, 1989, 172, 131. 33. A. H. Sultan, M. A. Kamm, C. N. Hudson, J. R. Nicholls, and C. I. Bartram, Clin. Rad., 1994, 49, 368.

10 INTERVENTIONAL MRI: SPECIALIST FACILITIES AND TECHNIQUES 34. R. Puni, A. S. Hall, G. A. Coutts, and N. deSouza, Proceedings of Progress in Magnetic Resonance, British Institute of Radiology, London, 1994. 35. A. Masters, A. C. Steger, W. R. Lees, K. M. Walmsley, and S. G. Bown, Br. J. Cancer, 1992, 66, 518. 36. A. J. Costello, W. G. Bowsher, D. M. Bolton, K. G. Braslis, and J. Burt, Br. J. Urol., 1992, 69, 603. 37. Y. Anzai, R. B. Lufkin, S. Hirschowitz, K. Farahani, and D. J. Castro, J. Magn. Reson. Imaging, 1992, 2, 671. 38. K. Anson, K. Seenivasagam, R. Miller, and G. Watson., Br. J. Urol., 1994, 73, 225. 39. Y. Anzai, R. B. Lufkin, D. J. Castro, K. Farahani, B. A. Jabour, L. J. Layfield, R. Udkoff, and W. N. Hanafee, J. Magn. Reson. Imaging, 1991, 1, 553. 40. B. Gewiese, J. Benthan, F. Fobbe, D. Stiller, G. Muller, J. BoseLandgraf, K. Wolf, and M. Deimling, Invest. Radiol., 1994, 29, 345. 41. K. Farahani, H. C. Yoon, J. Kumiyoshi, R. E. Saxton, S. Sinha, Y. Anzai, A. A. F. DeSalles, K. L. Black, and R. B. Lufkin, Abstracts, Society of Magnetic Resonance, 1994, p. 576. 42. J. N. Kabalin, J. Urol., 1993, 150, 95. 43. D. G. Assimos, D. L. McCullough, R. D. Woodruff, L. H. Harrison, L. J. Hart, and W. J. Li, J. Endourol., 1991, 5, 145. 44. D. L. McCullough, J. Urol., 1991, 146, 1126. 45. G. Bartch, G. Daxenbichler, and H. P. Rohr, J. Steroid Biochem., 1983, 19, 147. 46. P. Narayan, G. Fournier, R. Indukara, R. Leidich, K. Shinohara, and A. Ingerman, Urology, 1994, 43(6), 813. 47. N. M. deSouza, R. Flynn, G. A. Coutts, D. J. Gilderdale, A. S. Hall, R. Puni, M. Chui, D. N. F. Harris, and E. A. Kiely, Am. J. Radiol., 1995, in press.

48. R. Seibel and D. Groenmeyer, Abstracts, Symposium on Interventional MRI, Los Angeles, 1994 , pp. 135–136. 49. K. Farahani, Abstracts, Symposium on Interventional MRI, Los Angeles, 1994 , p. 39. 50. K. L. Black, Abstract, Symposium on Interventional MRI, Los Angeles, 1994 , pp. 72–74. 51. J. Gilbert, Abstracts, Symposium on Interventional MRI, Los Angeles, 1994 , pp. 20–30. 52. K. Hynyen, Abstracts, Symposium on Interventional MRI, Los Angeles, 1994 , pp. 40–44.

Biographical Sketches G. A. Coutts. b 1959. B.A. Physics, 1980, D.Phil. 1985, University of Oxford, UK. Research associate at GEC Hirst Research Centre in laboratory headed by Ian R. Young, 1987–present. Approx. 30 publications in the field of MRI and MRS. Current research specialty: interventional MRI. Nandita M. deSouza. b 1957. B.Sc. (Hons.) (Physiology), 1978, MBBS, 1983, University of Newcastle upon Tyne, UK. M.R.C.P., 1986, F.R.C.R., 1991, House Surgeon and Physician and Senior House Physician, Newcastle upon Tyne, 1983–85. Registrar in Internal Medicine, 1986–87. Registrar and Senior Registrar in Diagnostic Radiology, Guys Hospital, UK, 1987–1992. Senior research fellow in MR Imaging, Royal Postgraduate Medical School, Hammersmith Hospital, UK, 1993–present. Research specialty: interventional MRI.

MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

MR-Guided Biopsy, Aspiration, and Cyst Drainage Jonathan S. Lewin University Hospitals of Cleveland and Case Western Reserve University, Cleveland, OH, USA

1 INTRODUCTION As the practice of radiology has evolved since the late 1980s, there has been increasing emphasis on intervention. Recent trends in minimally invasive methods for diagnosis and treatments have created a burgeoning interest in the use of MRI for guidance of radiological procedures. Historically, the relatively long imaging times and the dif®culty in patient access resulting from closed-bore superconducting cylindrical system designs have combined to make MR an unlikely guidance modality for radiological procedures. More recently, many of these disadvantages have been overcome with the advent of system hardware and pulse sequence improvements that have allowed the development of rapid imaging on open systems. This chapter discusses the hardware and software improvements that have made interventional MRI a reality and the use of this technology for guidance of biopsy, aspirations, and cyst drainage.

1

cedures has required a compromise between these opposing forces; this balance has been achieved through a number of different concepts and solutions. Some of the ®rst interventional procedures under MRI guidance were performed on conventional cylindrical systems, simplifying imaging but signi®cantly compromising the access to the patient that is necessary for performance of the procedure and monitoring of patient discomfort and safety. Biopsy and aspiration procedures have been and can be performed using these cylindrical superconducting systems, but the patient must be withdrawn from the magnet in order to reposition the needle between scans.2 This results in relatively long procedure times and contributed to the sparse use of MRI for biopsy and aspiration during the late 1980s. In particular, when the lesion in question was visible and accessible for biopsy by computed tomography (CT) or ultrasound, these more conventional methods for procedure guidance were clearly better suited modalities. These disadvantages were overcome, in part, through the access to the patient provided by permanent or resistive `open' magnet imaging systems.3±6 Unfortunately, the open low-®eld imaging systems available in the late 1980s often required signi®cant imaging time to obtain a suf®ciently high SNR to allow needle insertion under MRI guidance. However, the access to the patient provided by these systems led to an increased number of reports of its use, primarily for biopsy in the head and neck region.7±9 Since then, many magnet, gradient, and receiver chain design improvements have occurred and have led to a number of approaches for interventional imaging. A brief description of the most commonly used current designs for intervention follows. Each system has bene®ts and limitations for interventional procedure guidance, as each has chosen a different balance between the spatial constraints necessary for highquality imaging and the freedom necessary for surgical or radiological intervention.

2 IMAGING SYSTEM DEVELOPMENT Recent advances in magnet and system design have accelerated progress in MR-guided intervention. Many different MR system con®gurations have been used to guide percutaneous procedures. Each of these systems has advantages and disadvantages, with a constant tradeoff between signal-to-noise ratio, (SNR) access to the patient, useable ®eld of view, and expense. The use of MRI for interventional procedure guidance includes its use by radiologists during manipulation of needles, electrodes, catheters, or laser ®bers. This form of intervention requires a signi®cant departure from conventional diagnostic concepts and traditional imaging systems. The factors contributing to high image quality in diagnostic MR include system ®eld strength and the homogeneity and stability of the static and gradient magnetic ®elds. These factors are best obtainable by decreasing the `openness' of an imaging system. The optimal design of a magnet with regard to ®eld homogeneity would be a complete sphere, without opening.1 In contrast, the environment best suited to radiological intervention is one with maximum access to the patient, allowing complete freedom in interventional approach and maintaining close proximity of monitoring and therapeutic devices. These attributes are in direct opposition to those facilitating image quality.1 The use of MRI for guidance of interventional pro-

2.1

Cylindrical Superconducting Systems

From an image quality viewpoint, cylindrical superconducting systems enjoy many signi®cant advantages relative to static magnetic ®eld strength and homogeneity, and the adoption of this system design as the standard for image quality in the diagnostic arena is not a chance occurrence. The limited patient access for procedure performance and direct visual observation is a compromise that still allows certain procedures. One more recent approach to the use of cylindrical superconducting systems for interventional procedures has been the siting of a superconducting system in an interventional or surgical environment. With the rapid fall-off of the static magnetic ®eld achieved by actively shielded magnet designs, it is possible to site a superconducting system in near proximity to a surgical work space. This design has been adopted in a system constructed by Philips Medical systems (Eindhoven, the Netherlands) in which a procedure suite including a ¯uoroscopy system has been constructed just outside of the fringe ®eld of a superconducting short-bore 1.5 T imaging system. Fluoroscopic or angiographic procedures can be performed in a conventional manner. However, when desired, intermittent images can be obtained by sliding the patient into the cylindrical imaging system. The imaging system allows all of the standard capabilities of a high-®eld system, including functional imaging, with high

2 MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE image quality. Although there are very few compromises to either the ¯uoroscopic or MRI portions of this environment, this con®guration does not allow interactive MR guidance of the intervention, as access to the patient is markedly limited. Therefore, needle placement under MR guidance must still be performed in an `in and out' manner, which can be time consuming for dif®cult needle placements. Once a needle is placed, the access to ¯uoroscopy can greatly facilitate guidewire and catheter placement during drainage procedures. The use of cylindrical magnets has also been successfully applied for guidance of stereotactic biopsy for breast lesions.10 This type of stereotactic procedure lends itself well to the immobilized breast but is less applicable to most other areas in which motion is more of an issue. 2.2 `Double Donut' Con®guration A novel approach to obtaining access to a patient in a cylindrical system is exempli®ed in the `double donut' con®guration of the General Electric Medical Systems Signa SP system (Milwaukee, WI), designed speci®cally for interventional applications.11±13 This design has taken the central segment out of a cylindrical system, allowing access to the patient from the sides and top at the isocenter of the imaging system. This product design has been directed toward siting in a surgical environment and allows enough room in the gaps between the halves (54 cm) for two physicians or nurses, one on either side of the patient.1 This system provides a marked improvement in access to the patient relative to a closed-bore cylindrical system, although the technology is expensive. This concept represents the ®rst system designed and constructed speci®cally for interventional guidance. This magnet design has been used both to guide and to monitor a large number of surgical procedures, in addition to an early published series of biopsies and aspirations.12,14,15 However, the marked improvement in access to the patient is achieved at the expense of a decrease in ®eld strength at the imaging isocenter (0.5 T imaging ®eld strength) compared with the ®eld generated by each individual superconducting half, and it compromises magnetic ®eld homogeneity.1 The unobstructed side and vertical access to the patient also complicates the engineering of the radiofrequency coil, requiring a local transmit±receive coil with a somewhat limited ®eld of view. 2.3 Biplanar Magnet Designs Another approach that has found widespread application for diagnostic MRI in the burgeoning `open MRI' market has been the use of a biplanar magnet design. With these magnets, the patient is positioned between ¯at magnetic poles, leaving access from a range of side approaches. These systems typically utilize lower-®eld permanent or resistive magnets, with ®eld strengths ranging from 0.064 to 0.3 T, although superconducting mid-®eld models of this design have also been introduced. The degree of access around the circumference of the biplanar magnet depends upon the number and position of supports separating the two magnetic poles. A large degree of access around the circumference is afforded by the `C-arm' design, produced by Siemens Medical Systems (Erlangen, Germany) and Picker International (Highland Heights, OH). In this design, a single column on one side

Figure 1 The C-arm system for percutaneous intervention. Intervention MR suite set up for radiological intervention demonstrates two-camera video sensor array (curved black arrow), which detects the location and orientation of a hand-held probe or needle guide (straight white arrow). The system automatically acquires continuous MR images based on the probe position and automatically updates display of four images on a shielded LCD monitor adjacent to the scanner (straight black arrow). A computer mouse on the LCD console (curved white arrow) and foot pedals (not shown) allow the scanner to be fully operated by the radiologist throughout the procedure

of the patient supports the upper pole, allowing access from the contralateral side, head, and foot end of the magnet (Figure 1). Several manufacturers produce biplanar systems with two supporting posts, resulting in slightly more restricted access to the patient. These include systems produced by Fonar Corporation (Melville, New York), Hitachi Medical Corporation of America (Twinsburg, OH), and General Electric Medical Systems. There are also designs with four support posts from Toshiba America Medical Systems (San Francisco, CA) and Fonar Corporation. The biplanar concept has the advantage of a fairly homogeneous static magnetic ®eld but is limited to lower ®eld strengths than available with the cylindrical superconducting designs. The side access provided by these systems is analogous to that of C-arm ¯uoroscopy and is amenable to needle or catheter-directed procedures. However, anterior or posterior interventional approaches require decubitus or oblique positioning of the patient, which may not be possible with large patients because of relatively limited space between the poles of the biplane magnet. A true direct vertical approach to the patient is only possible when the patient table is brought outside of the magnet, essentially excluding simultaneous vertical intervention and imaging. 2.4

Other Systems

Other concepts for intervention have also been evaluated, although these systems are still in earlier stages of development. A biplanar design with a very large interpole distance approaching a `¯at bed' scanner was also recently described by Fonar Corporation, but this is still in development. Most likely,

MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

different system designs will be advocated for different interventional applications. The choice of interventional system will depend upon each institution's optimal balance between the types of procedure to be performed, required image quality, and ®nancial constraints.

3

performed with the operator sitting next to the patient, with no need to remove the operator's hand from the interventional device at any time (Figure 1). This manner of intervention, analogous to an angiographic or sonographically guided procedure, is well-suited to the skill set developed by radiologists during more conventional types of image-guided intervention.

2.5 Supplemental Technical Developments There are several recent technical developments that are common to many of the magnet and system designs discussed above and that have facilitated the guidance phase of interventional procedures in a time-ef®cient manner. First, the construction of higher-quality low-noise receiver chains has allowed lower ®eld systems to be constructed, providing relatively high SNR images and greatly improving image quality compared with earlier attempts at low-®eld MRI. These systems now provide image quality suf®cient for the procedureguidance phases of minimally invasive procedures. Image acquisition times of 1±3 s per image, or less, provide the necessary temporal resolution. The second development that has impacted on the use of MR methods for procedure guidance has been the modi®cation and development of rapid gradient echo pulse sequences for use in interactive guidance during device placement. Rapid gradient echo sequences have been developed to allow a wide range of tissue contrast in a time frame suf®cient for device tracking (0.3±7 s per image) even at low ®eld strength and with the suboptimal coil position sometimes required to access the puncture site (Figure 2).16±19 Because the area of interest to the interventionalist is often limited to a small portion of the entire image acquired, a number of strategies have been proposed to decrease the imaging time for these methods further. These strategies include alternative reconstruction techniques such as Keyhole imaging, singular value decomposition, and wavelet encoding.20±32 The use of these rapid gradient echo techniques for guidance of interventional procedures also takes advantage of the reduced needle artifact at lower ®elds compared with that at 1.5 T, allowing a wider choice of materials for device construction and still allowing high spatial accuracy despite the increased metallic artifact encountered with gradient echo techniques.33,34 The third innovation of signi®cant bene®t for MR-guided intervention has been the development of an interface between an optically-linked 3D digitizer and the measurement control software of the MR imager. Previously, frameless stereotactic systems had been used primarily for surgical guidance based on historical imaging data sets. The ability to acquire MR images rapidly and interactively in an arbitrary plane determined by a hand-held, sterilizable probe, or other interventional device, allows rapid planning and con®rmation of complex trajectories for rigid instruments. This type of interactive image-guidance system was ®rst described as a component of the `double donut' system and subsequently has been applied for guidance of image acquisition on C-arm systems (Figure 1).12,18 An additional technical factor facilitating near-real-time procedure guidance has been the ability to view images in the magnetic ®eld through the development of in-room high-resolution rf-shielded monitors.12,18,35 In combination with the patient access allowed by open-imaging systems, the ability to view images at scanner side allows the entire procedure to be

3

MR-SPECIFIC ISSUES FOR INSTRUMENT GUIDANCE AND VISUALIZATION

Unlike guidance by X-ray-based techniques such as ¯uoroscopy and CT, there are a number of user-de®ned imaging parameters and needle trajectory decisions that can markedly alter needle visibility and, therefore, accuracy and safety of MR-guided procedures.33 The primary factors affecting needle visibility in clinical application can most simply be divided into those dependent upon the pulse sequence and those dependent upon the needle. The most important pulse-sequence issue with regard to needle visibility is the sensitivity of the sequence to magnetic susceptibility effects.33 At our institution, as in other recent reports,12,18,35±37 near-real-time imaging for primary needle guidance is performed with gradient echo pulse sequences, taking advantage of the rapid scan time and high SNR of this sequence design. Use of these sequences has the primary advantage of rapid temporal resolution and is extremely useful for guiding needles into lesions 1 cm or greater in size that are not immediately adjacent to major neurovascular structures. At 0.2 T, the apparent width of the biopsy needle under gradient echo image guidance generally ranges from approximately 4 mm for smaller needles to 9 mm for larger needles.33 Although this degree of artifactual widening is acceptable for larger lesions in areas of low neurovascular density, such as the abdomen and extremities, it is clearly unacceptable for head and neck lesions adjacent to major vessels. The degree of artifactual widening can be reduced by approximately a factor of two through use of a turbo spin echo pulse sequence or higher sampling bandwidth.33 The use of turbo spin echo pulse sequences should be routinely performed prior to cutting needle biopsy to exclude smaller vessels at the level of the cutting edge of the needle; it will be rarely used as the primary mode of image guidance for lesions immediately adjacent to the carotid or vertebral artery, or for spinal lesions for which precise needle visualization is mandatory (Figures 2±4). Spin echo pulse sequences can also be applied to reduce the artifactual needle widening by a similar factor,33 but are less useful during procedures because of the longer imaging time necessary to obtain a suf®ciently high SNR. Therefore, the use of turbo spin echo imaging for position con®rmation or primary guidance should be strongly considered when needle placement within 5 mm of major neurovascular structures is contemplated (Figure 3), rather than relying on the more rapid gradient echo sequences for guidance of the entire procedure.18 The second sequence-related parameter that can be easily varied to adjust the degree of artifactual widening is less intuitively obvious but has had signi®cant clinical utility in our experience. This variable is the direction of frequency encoding relative to the needle shaft. The presence of a needle within tissue creates a local nonuniform spatial distortion in the magnetic ®eld and, therefore, a variation in the local resonant

4 MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

Figure 2 An 89-year-old woman with Warthin's tumor of the parotid gland. The inferior approach to the parotid tail lesion in this patient allowed this 18-gauge side-notch cutting needle to be placed in the lesion without traversing normal parotid gland and without risk to the facial nerve and its branches. (a,b) Two images from true fast imaging with steady-state precession (FISP) sequence (11.7/5.9/1/90 TR/TE/number of signal averages/ ¯ip angle) frames obtained at approximately 1 image per second during advancement of the inner stylet of this core biopsy needle (arrows). (c) A higher resolution turbo spin echo T1-weighted image (500/24/1 TR/TE/number of signal averages, 70 s scan time) was obtained to con®rm the needle (arrow) position prior to obtaining the core biopsy. (Reproduced with permission from Lewin et al.18)

MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

Figure 3 A 50-year-old subject with recurrent squamous cell carcinoma of the oropharynx. Turbo spin echo T1-weighted image (500/24/1 TR/TE/number of signal averages) clearly documents needle position (arrow) and cutting notch location, particularly important when biopsy is performed near the internal carotid artery (arrowhead) or the larger branches of the external carotid artery. The external carotid artery in this patient had been previously occluded with Gianturco coils to control oropharyngeal bleeding. The coils did not result in signi®cant image distortion and did not preclude biopsy

frequency. For imaging, a spatially linear gradient is used to encode location in the frequency-encoded direction, imparting a linear one-to-one mapping between frequency and position. The local ®eld distortion resulting from the needle disrupts this linear relationship and leads to image artifact along this axis.

Figure 4 Needle position con®rmation in a 43-year-old man with leukemic in®ltration (chloroma) of scalene muscle. Coronal turbo spin echo T1-weighted image (680/24/1 TR/TE/number of signal averages) demonstrates 18-gauge side-notch cutting needle (arrowheads) with stylet extended into the patient's right thoracic inlet paraspinal mass. (Reproduced with permission from Lewin et al.18)

5

When frequency encoding is perpendicular to the needle, this results in artifactual widening. When parallel to the needle, this results in less obvious artifact at the tip and hub of the needle. Depending upon needle composition and orientation, swapping the frequency and phase-encoding axes relative to the needle shaft can reduce or increase the apparent width of the needle by a factor of 0.33 to 2.5 for turbo spin echo sequences.33 This effect can be used to decrease the apparent needle width when the needle artifact obscures adjacent anatomic structures by setting the frequency-encoding axis of the image parallel to the needle shaft.18 In the presence of even mild respiratory motion, the needle tip for thinner (e.g., 22 gauge) needles can be dif®cult to visualize in tissue with certainty. Therefore, this effect is more commonly useful, in our experience, to identify the needle tip location more con®dently on turbo spin echo images by maximizing the needle artifact through frequency encoding the image in a direction perpendicular to the needle shaft. This ability to alter the appearance of the needle signi®cantly is used on a daily basis in a variety of clinical settings. Several other important factors are related more to the needle and its trajectory rather than to the pulse sequence and imaging parameters. The dependence of visualization for safe and accurate biopsy on needle composition is obvious and has been well described in many reports since the inception of MR-guided procedures.2,3,33 The optimal material for needle fabrication will vary with MR system ®eld strength. Relatively less-expensive materials such as high-nickel, high-chromium stainless steel may be adequate at 0.2 T but may give rise to unacceptable artifact at 1.5 T.33,34 Conversely, small-caliber needles constructed from low-artifact materials such as titanium may be dif®cult to identify in certain clinical settings at low ®eld. Another issue that may have signi®cant impact on needle localization may be somewhat less obvious: the angle between the needle shaft and the static magnetic ®eld of the MR imager. The apparent needle diameter diminishes markedly from a decrease in artifact as the needle shaft approaches the axis of the static magnetic ®eld.2,33,34 Artifacts resulting from ®eld distortion arise most signi®cantly where the ®eld enters and exits objects of differing magnetic susceptibility, such as the needle and surrounding tissue. When the needle is parallel to the ®eld, distortion of the ®eld (and, therefore, image artifact) occurs mostly at the tip and hub, and to a lesser extent along the shaft. The ®eld is increased within the needle shaft. However, because there are no protons to image within the needle, no image distortion occurs. The ®eld is also distorted slightly adjacent to the shaft. At low ®elds, the distortion adjacent to the shaft is much less than the effect of the applied imaging gradients, and, therefore, little mismapping occurs, with artifact along the shaft primarily related to mild signal loss caused by decreased T2* relaxation. When the needle is perpendicular to the main magnetic ®eld, the ®eld enters and exits throughout the length of the shaft. Local ®eld distortion and, therefore, more prominent artifact can be observed along the entire needle. Although the apparent needle width decreases as the needle is positioned parallel to the static magnetic ®eld (vertically for most biplanar magnets and along the long axis of cylindrical or `double donut' systems), artifact at the device tip blooms and obscures the true tip position. Steep needle trajectories may be more familiar or appear advantageous in some anatomic locations

6 MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE but may be prohibited because of this poor needle conspicuity and loss of tip position information. Safe clinical application of MR-guided techniques requires careful consideration of these factors during procedure planning and execution.

4 CLINICAL APPLICATION FOR BIOPSY AND ASPIRATION Biopsy, aspiration, and cyst drainage represent some of the most straightforward interventional applications for a cross-sectional imaging modality. The tissue contrast, spatial resolution, and multiplanar capabilities of MR have obvious bene®ts for guidance of biopsy and aspiration applications, and this application was the ®rst reported use of MRI to guide intervention. As noted above, a number of investigators have described the use of MRI for guidance of biopsy and aspiration since the late 1980s.3,7±9,18,35,38 MR guidance for head and neck biopsies has represented a large part of this early literature, owing at least in part to the striking advantages that MR guidance provides for interventional procedures performed in areas of complex anatomy. The bene®ts of multiplanar image acquisition were quickly discovered by Lufkin and colleagues, who ®rst reported the advantages derived from going beyond the single imaging plane, usually axial, available in CT-guided procedures.7 The elimination of beam-hardening artifact, inherent in CT, has also supported the use of MRI for guidance of skull base procedures.9,18 Furthermore, the high vascular conspicuity resulting from ¯ow-related enhancement effects inherent in 2D gradient echo imaging is an additional bene®t of MR guidance for lesions in regions of complex vascular anatomy. In our experience, the use of MRI for guidance of biopsy and aspiration has been of most advantage in sampling lesions of the suprahyoid neck, including high cervical spine lesions (Figure 5). The major bene®t of MR guidance in this region is the ability to visualize the internal carotid, vertebral, and major branches of the external carotid artery continuously during the entire needle insertion (Figure 6). Masses at the base of the neck, in particular those encroaching upon the brachial plexus, have also presented an excellent application, exploiting the multiplanar capabilities, tissue contrast, and spatial resolution of MRI to allow con®dent tissue sampling adjacent to the neurovascular structures of the thoracic inlet (Figure 4).18,39 The ability to guide needle insertion with continuous imaging has also been helpful for biopsies of laryngeal and pharyngeal submucosal lesions, which are often dif®cult to de®ne on CT and may be hidden by normal overlying mucosa on endoscopic examination (Figure 7). The T2-weighted techniques can also be used to increase diagnostic tissue yield by allowing sampling of non-necrotic regions of complex masses. Rapid gradient echo T2-weighted images are also helpful for biopsies and drainages within the spinal canal, in which a clear distinction between cerebrospinal ¯uid or cyst contents and the adjacent spinal cord is essential (Figure 8). Other lesions that are particularly well suited to MR guidance are musculoskeletal tumors, for which CT and ultrasound often lack suf®cient tissue contrast between pathological and normal muscle. MR-guided needle insertion is also time ef®cient for contrast injection prior to MR arthrograms.40

The use of MR guidance may be less useful for biopsy of abdominal lesions where ultrasound and CT are well suited. In our practice, we restrict the use of MR for abdominal procedures to lesions that are high under the hemidiaphragm,35 liver tumors in which ultrasound and CT fail to provide adequate tissue contrast, and for patients in whom prior ultrasound- or CT-guided procedures have been unsuccessful (Figure 9). Depending upon the radiologist's preference, biopsies under MR guidance can be performed using free-hand methods, needle holders, or with real-time guidance from optically linked systems, as described above.7,12,18,35 Currently, most biopsies at our institution are performed with free-hand techniques, with optically linked frameless guidance reserved for particularly complex procedures. Using the free-hand technique,18 most biopsies can be performed in less than 1 h of MR table time. For many of these procedures, the rapid frame-rate and high visibility of nearby vessels provided by 2D gradient echo MR techniques allow needle placement in less time than required with CT guidance (less than 8 min per needle pass for both aspiration and core cutting-needle biopsy).18

Figure 5 A 77-year-old woman with neck pain showing a C2 vertebral body lesion on MRI. The capability to guide biopsy with oblique imaging plane also allows the needle to be angled around the internal carotid and vertebral arteries for biopsy of high cervical spine lesions. Turbo spin echo T2-weighted image (2000/105/3/17 TR/TE/ number of signal averages echo train length, 51 s scan time) image obtained prior to sampling with 18-gauge Menghini-type needle. Frequency encoding must be performed perpendicular to the needle shaft with turbo spin echo or spin echo images to maximize needle visualization. Histological diagnosis was plasmacytoma, and the patient subsequently developed multiple myeloma. (Reproduced with permission from Lewin39)

MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

7

Figure 6 Images from a continuous series obtained at 7 s image with fast imaging with steady-state precession (FISP) sequence (18/7/4/90 TR/ TE/number of signal averages/¯ip angle) obtained during guidance of needle insertion in a 68-year-old man with C1±2 vertebral and prevertebral mass. Previous attempt at surgical transoral biopsy had been unsuccessful. (a) Image early during insertion demonstrates needle tip passing through the left parotid space. An ill-de®ned mass can be seen in the prevertebral space. High vascular conspicuity resulting from 2D Fourier transform technique allows ready visualization of ¯ow-related enhancement within the internal carotid (curved arrow) and vertebral (straight arrow) arteries. The needle tip (arrowheads) can be interactively directed to avoid these major vascular structures. (b) Mid-insertion, the needle (arrowheads) is interactively redirected more anteriorly once safely beyond the internal carotid artery. Histology demonstrated chronic osteomyelitis and cellulitis and the offending organism was successfully isolated

An additional use of the interventional MRI techniques that form the basis for biopsy guidance is their application for the guidance and monitoring of direct intralesional drug injection, including injection for sclerotherapy of vascular malformations. The same rapid image updates used for interactive needle placement can be used to monitor the injection of sclerosing agents for the treatment of low-¯ow vascular malformations.41 The multiplanar cross-sectional images obtained with MR allow the injection of alcohol or other sclerosing agents to be monitored during administration, to ensure ®lling of the entire targeted portion of the malformation and to exclude extravasation or dissipation of the agent through venous egress. 5 UTILIZATION AND ECONOMIC CONSIDERATIONS In addition to the many engineering challenges related to MR procedure guidance, the development of appropriate utilization guidelines to optimize cost-effectiveness of these interventions also represents a dif®cult task. When compared

with current ultrasound- and CT-guided procedures, MR-guided procedures are slightly more expensive (charges for an MRguided procedure are currently approximately 10±15% more than for a similar procedure under CT guidance at our institution). Therefore, when less-expensive CT or ultrasound guidance is suitable, their use is warranted as the most costeffective choice. There are a number of procedures in which other imaging modalities lack the tissue contrast or vascular conspicuity to provide safe routes for biopsy or aspiration.18 It is the procedures for which other imaging modalities are limited or open surgical biopsy is the primary alternative that provide the best area for application of MR-guided techniques. The most formidable barrier to widespread dissemination of these techniques has been the economic issues related to interventional MRI. The ®nancial resources necessary for an institution to acquire interventional MRI capabilities are often initially considered an insurmountable problem outside of the academic setting. The cost of an interventional system and associated siting depends on several factors, including ®eld strength and system fringe ®eld; the more novel the interventional design, the more expensive it is to acquire. The least-

8 MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

Figure 7 Chondrosarcoma of cricoid cartilage in a 65-year-old man. (a) A T2-weighted image obtained before biopsy shows large cricoid cartilage mass (arrow), stretching the cricopharyngeous muscle posteriorly. (b) Fast imaging with steady-state precession (FISP) (18/7/1/90 TR/TE/ number of signal averages/¯ip angle) shows coaxial placement of 22-gauge aspiration needle (arrowheads) through 18-gauge introducer needle. MR guidance allowed avoidance of both large and small external carotid artery branches. (Reproduced with permission from Lewin et al.18)

expensive approach for percutaneous procedures is to add the necessary supplemental equipment for MR-guided procedures to an existing diagnostic MR system. Several vendors offer interventional accessories at prices below US $160 000, making the variable cost required to perform percutaneous MRguided procedures relatively low for those sites with pre-existing open MR scanners. Purchase and siting of a low-®eld system speci®cally out®tted for intervention, a 1.5 T superconducting short-bore system or the `double donut' system are each progressively more expensive. As the speed at which procedures can be performed further increases with improvements in MR-guidance techniques, the contribution to ®xed cost compared with that of CT-guided techniques should continue to decrease. Finally, even with the current level of technology and expense, appropriate utilization of interventional MRI can potentially result in a tremendous reduction in the costs associated with treating certain patients. In particular, there has been a marked decrease in cost at our institution for patients for whom the application of MR-guided biopsy or sclerotherapy has avoided an open surgical procedure. This has resulted from a decrease in cost of the procedure itself, relative to surgery, along with equal or more pronounced savings through avoidance of intensive-care unit and routine care hospitalization. This is in addition to the bene®t to the patient that results from the reduction in morbidity, mortality, postprocedure discomfort, and time required for recovery when open surgical procedures can be avoided. These potential advantages of interventional MRI have provided the rationale for an increased interest in these techniques, with a growing number of centers and radiology practices developing interventional or intraoperative MRI capabilities. The overall success for this developing ®eld within interventional radiology will require the careful application of these techniques to ensure appropriate utilization. Intraoperative and minimally invasive MR-guided techniques hold tremen-

dous potential, but much effort remains necessary to de®ne their appropriate role in the delivery of cost-effective medical care. 6

RELATED ARTICLES

Design and Use of Internal Receiver Coils for MRI; Gastroscopy and Colonoscopy; Head and Neck Investigations by MRI; Thermal Therapies in the Body Monitored by MRI. 7

REFERENCES

1. R. S. Hinks, M. J. Bronskill, W. Kucharczyk, M. Bernstein, B. D. Collick, and R. M. Henkelman, JMRI, 1998, 8, 19. 2. P. R. Mueller, D. D. Stark, J. F. Simeone, S. Saini, R. J. Butch, R. R. Edelman, J. Wittenberg, and J. T. J. Ferrucci, Radiology, 1986, 161, 605. 3. R. Lufkin, L. Teresi, and W. Hanafee, Am. J. Roentgenol., 1987, 149, 380. 4. R. Lufkin, G. Duckwiler, E. Spickler, L. Teresi, M. Chang, and G. Onik, J. Comput. Assist. Tomogr., 1988, 12, 1088. 5. L. Kaufman, M. Arakawa, J. Hale, P. Rothschild, J. Carlson, K. Hake, D. Kramer, W. Lu, and H. J. Van, Magn. Reson. Q., 1989, 5, 283. 6. D. H. Gronemeyer, L. Kaufman, P. Rothschild, and R. M. Seibel, Radiol. Diagn. (Berlin), 1989, 30, 519. 7. R. Lufkin, L. Teresi, L. Chiu, and W. Hanafee, Am. J. Roentgenol., 1988, 151, 193. 8. G. Duckwiler, R. B. Lufkin, L. Teresi, E. Spickler, J. Dion, F. Vinuela, J. Bentson, and W. Hanafee, Radiology, 1989, 170, 519. 9. R. Wenokur, J. C. Andrews, E. Abemayor, J. Bailet, L. Lay®eld, R. F. Canalis, B. Jabour, and R. Lufkin, Skull Base Surgery, 1992, 2, 167. 10. S. G. Orel, M. D. Schnall, R. W. Newman, C. M. Powell, M. H. Torosian, and E. F. Rosato, Radiology, 1994, 193, 97.

MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

9

Figure 8 A 36-year-old woman with paraparesis and pain after a motor vehicle accident with recent worsening of symptoms. (a) A sagittal image from true fast imaging with steady-state precession (FISP) sequence (11.7/5.9 TR/TE) during posterior interlaminar 22-gauge needle placement (arrowhead) into large post-traumatic subarachnoid cyst (white arrows). A posterior approach was used with the patient in the decubitus position. The procedure was performed to see if cyst decompression would be bene®cial prior to de®nitive open surgical repair. (b) True FISP (11.7/5.9 TR/ TE) sagittal image after needle placement and partial aspiration of cyst (arrows). A smaller cyst can also be seen immediately superior as the larger cyst is decompressed. (c) True FISP (11.7/5.9/3/90 TR/TE/number of signal averages/¯ip angle) sagittal image at the end of the aspiration procedure demonstrates small residual cyst (arrows) with re-expansion of the thoracic spinal cord. (Reproduced with permission from Lewin et al.18)

10 MR-GUIDED BIOPSY, ASPIRATION, AND CYST DRAINAGE

Figure 9 A 38-year-old man with pancreatic carcinoma and liver mass with a previous unsuccessful CT-guided biopsy attempt. Liver masses, in particular, are typically very difficult to visualize without T2-weighted images. Guidance of a 22-gauge needle (arrow) was provided in this case with a true fast imaging with steady-state precession imaging sequence (11.7/5.9/1/90 TR/TE/number of signal averages/flip angle, 1 s per frame scan), with a series of 9±12 images obtained with continuous update during each breath-hold. Pathological diagnosis was simple hepatic cyst. (Reproduced with permission from Lewin et al.18)

11. J. F. Schenck, F. A. Jolesz, P. B. Roemer, H. E. Cline, W. E. Lorensen, R. Kikinis, S. G. Silverman, C. J. Hardy, W. D. Barber, and E. T. Laskaris, Radiology, 1995, 195, 805. 12. S. G. Silverman, B. D. Collick, M. R. Figueira, R. Khorasani, D. F. Adams, R. W. Newman, G. P. Topulos, and F. A. Jolesz, Radiology, 1995, 197, 175. 13. S. G. Silverman, F. A. Jolesz, R. W. Newman, P. R. Morrison, A. R. Kanan, R. Kikinis, R. B. Schwartz, L. Hsu, S. J. Koran, and G. P. Topulos, Am. J. Roentgenol., 1997, 168, 1465. 14. P. M. Black, T. Moriarty, E. Alexander, P. Stieg, E. J. Woodard, P. L. Gleason, C. H. Martin, R. Kikinis, R. B. Schwartz, and F. A. Jolesz, Neurosurgery, 1997, 41, 831. 15. E. Alexander, T. M. Moriarty, R. Kikinis, P. Black, and F. M. Jolesz, Stereotact. Funct. Neurosurg., 1997, 68, 10. 16. A. E. Mahfouz, A. Rahmouni, C. Zylbersztejn, and D. Mathieu, Am. J. Roentgenol., 1996, 167, 167. 17. J. L. Duerk, J. S. Lewin, M. Wendt, and C. Petersilge, JMRI, 1998, 8, 203. 18. J. S. Lewin, C. A. Petersilge, S. F. Hatem, J. L. Duerk, G. Lenz, M. E. Clampitt, M. L. Williams, K. R. Kaczynski, C. F. Lanzieri, A. L. Wise, and J. R. Haaga, Am. J. Roentgenol., 1998, 170, 1593. 19. Y. C. Chung, E. M. Merkle, J. S. Lewin, J. R. Shonk, and J. L. Duerk, Magn. Reson. Med., 1999, 42, 335. 20. M. Busch, A. Bornstedt, M. Wendt, J. L. Duerk, J. S. Lewin, and D. Groenemeyer, JMRI, 1998, 8, 944. 21. J. L. Duerk, J. S. Lewin, and D. H. Wu, JMRI, 1996, 6, 918. 22. L. P. Panych, C. Oesterle, G. P. Zientara, and J. Hennig, Magn. Reson. Med., 1996, 35, 554. 23. Y. Cao and D. N. Levin, Magn. Reson. Med., 1995, 33, 140. 24. J. L. Duerk, D. H. Wu, Y. C. Chung, Z. P. Liang, and J. S. Lewin, JMRI, 1996, 6, 957.

25. G. P. Zientara, L. P. Panych, and F. A. Jolesz, Magn. Reson. Med., 1994, 32, 268. 26. N. Gelman and M. L. Wood, Magn. Reson. Med., 1998, 39, 383. 27. L. P. Panych, P. D. Jakab, and F. A. Jolesz, JMRI, 1993, 3, 649. 28. L. P. Panych and F. A. Jolesz, Magn. Reson. Med., 1994, 32, 738. 29. L. P. Panych, R. V. Mulkern, P. Saiviroonporn, G. P. Zientara, and F. A. Jolesz, Magn. Reson. Med., 1997, 38, 964. 30. R. D. Peters and M. L. Wood, JMRI, 1996, 6, 529. 31. J. B. Weaver, Y. Xu, D. M. Healy, and J. R. Driscoll, Magn. Reson. Med., 1992, 24, 275. 32. M. Wendt, M. Busch, G. Lenz, J. L. Duerk, J. S. Lewin, R. M. Seibel, and D. Groenemeyer, IEEE Trans. Med. Imag., 1998, 17, 803. 33. J. S. Lewin, J. L. Duerk, V. R. Jain, C. A. Petersilge, C. P. Chao, and J. R. Haaga, Am. J. Roentgenol., 1996, 166, 1337. 34. C. Frahm, H. B. Gehl, U. H. Melchert, and H. D. Weiss, Cardiovasc. Intervent. Radiol., 1996, 19, 335. 35. H. B. Gehl, C. Frahm, H. Schimmelpenning, and H. D. Weiss, Rofo. Fortschr. Geb. Rontgenstr. Neuen. Bildgeb. Verfahr., 1996, 165, 70. 36. C. Frahm, H. B. Gehl, H. D. Weiss, and W. A. Rossberg, Rofo. Fortschr. Geb. Rontgenstr. Neuen. Bildgeb. Verfahr., 1996, 164, 62. 37. D. S. Lu, H. Lee, K. Farahani, S. Sinha, and R. Lufkin, Am. J. Roentgenol., 1997, 168, 737. 38. G. Hathout, R. B. Lufkin, B. Jabour, J. Andrews, and D. Castro, JMRI, 1992, 2, 93. 39. J. S. Lewin, Am. J. Neuroradiol., 1999, 20, 211. 40. C. A. Petersilge, J. S. Lewin, J. L. Duerk, and S. F. Hatem, Am. J. Roentgenol., 1997, 169, 1453. 41. J. S. Lewin, E. M. Merkle, J. L. Duerk, R. W. Tarr, Radiology, 1999, 211, 566.

Acknowledgements The author would like to acknowledge the members of the Interventional MR Imaging Research Program at University Hospitals of Cleveland and Case Western Reserve University for their ongoing commitment to the development of interventional MRI techniques; in particular, Drs Andrik Aschoff, Elmar Merkle, Michael Wendt, and Jeffrey Duerk of the Department of Radiology. In addition, I would like to thank Elena DuPont and Rayna Lipscomb for their invaluable help with manuscript preparation. The University Hospitals of Cleveland/Case Western Reserve University Interventional MR Program is supported in part through research collaborations with Siemens Medical Systems, Radionics, and Minrad, Inc. This project was also supported through grants from the Whitaker Foundation, American Cancer Society, Mary Ann S. Swetland Fund, and the M. E. and F. J. Callahan Foundation.

Biographical Sketch Jonathan S. Lewin. b 1959. B.A. 1981 Brown University, MD 1985 Yale. Residency in Diagnostic Radiology University Hospitals Cleveland 1985±89, MRI Research Fellowship Siemens Erlangen, Germany 1989±90, Fellowships in Neuroradiology at Cleveland Clinic Foundation 1991±93. Associate Professor and Vice Chairman for Research and Academics in Department of Radiology, Case Western Reserve University and Director MRI at University Hospitals, Cleveland 1993± present. Approx. 105 publications. Current interests: basic science and clinical aspects of interventional MRI, functional MRI, and the imaging of acute stroke.

MR-GUIDED THERAPY IN THE BRAIN

MR-Guided Therapy in the Brain Volker Tronnier University of Heidelberg, Germany

Antonio A. F. De Salles University of California at Los Angeles, CA, USA

Yoshimi Anzai University of Michigan, USA

Keith L. Black University of California at Los Angeles, CA, USA

and Robert B. Lufkin University of California at Los Angeles, CA, USA

1 INTRODUCTION Through the technological and clinical advances of the 1990s, MRI has become one of the most powerful noninvasive diagnostic tools in the evaluation of the brain. The superior soft-tissue contrast, the oblique and multiplanar imaging capability, and the absence of ionizing radiation have made MRI the imaging modality of choice for most diagnostic applications of the brain. These advantages also make MRI an ideal guide for invasive and interventional procedures in the brain. The high sensitivity of MRI to ¯ow enables the delineation of blood vessels without the use of contrast agents. The absence of beam-hardening artifacts from bone and metal devices allows complex instrumentation approaches to anatomic regions that may be dif®cult or impossible to reach under conventional computed tomography (CT) guidance. Nearly a decade of experience with MR-guided stereotaxis has proven MRI to be an effective and reliable means to visualize and localize neurosurgical targets and a valuable aid in optimizing surgical approaches.1±4 Recently, MRI has been applied to treatment monitoring procedures in the brain, such as interstitial therapy and even open surgery.5 New developments in MR imaging techniques and MR-compatible instrumentation and the availability of open magnet designs are revolutionalizing the role of MRI in guiding and monitoring interventional procedures.

2 MR-GUIDED STEREOTAXIS Since the mid 1980s, MRI stereotaxis has been successfully employed in neurosurgery to guide biopsies, in tumor resection, deep electroencephalographic (EEG) electrode placement, functional target development and resection, and other

1

interventional procedures.6±14 The noninvasive visualization and localization provided by MRI greatly reduce the surgical invasion necessary to access the target and signi®cantly lower patient morbidity. Several stereotactic systems with localizer frames designed for MRI localization are commercially available: the BRW and CRW (Radionics, Burlington, MA), Leksell (Elekta Instruments Inc., Atlanta, GA), Laitinen (SandstroÈm Trade & Technology Inc., Welland, Ontario, Canada), and others. Since MRI is more prone to geometric distortion than CT, the accuracy of MRI stereotaxis must be evaluated. The maximum localization error reported in the literature ranges from 1 to 5 mm.15±18 The variation in the maximum error may be accounted for partly by the use of different stereotactic systems and different reference imaging modalities (CT, ventriculography, etc.) with which MRI is compared. The spatial accuracy of MRI depends on the linearity and calibration of magnetic ®eld gradients and on magnetic susceptibility artifacts, which manifest as spatial distortions. Meticulous quality control should decrease errors in magnetic ®eld linearity and calibration. MRI sequences should use high bandwidth signal acquisition to reduce spatial distortion from susceptibility effects. Several researchers have proposed practical correction algorithms to improve the spatial accuracy of MRI further.19,20 Although accurate and reliable, conventional stereotactic systems require the ®xation of a frame to the patient's skull. Recent developments in frameless stereotactic systems will improve patient comfort, localization ef®ciency, and surgical access in open procedures.21±23 Projects are underway to develop MR-compatible navigational systems that use an armbased three-dimensional digitizer or light-emitting diode (LED)-based optical digitizers registered to the coordinate system of the interventional MR scanner for interactive scan plane control and near-realtime MRI monitoring of the interventional procedure (Figure 1).

3

MR-GUIDED BIOPSY

One of the earliest applications of intracranial interventional MR was to obtain biopsy specimens for tissue diagnosis. As with biopsies elsewhere in the body, the approach utilized for accessing an intracranial lesion is chosen to limit the risk of injury to surrounding vital structures. Either ®ne needle aspirates or core biopsies can be obtained by utilizing various gauge MR-compatible needles. Also, either frame-based or frameless stereotaxis may be used to guide the approach, depending on the location and size of the target lesion. The primary goal of an intervention within the MR scanner, however, will be a frameless procedure with on-line imaging control of the needle trajectory. The point of entry is determined with an optical tracking device, a gadolinium-®lled wand, or simply with the surgeon's ®nger. Fast image sequences (e.g., twodimensional fast imaging, with steady-state precession, FISP) allow a near realtime image while advancing the biopsy needle to the desired target. An obvious advantage compared with intraoperative CT is the possibility of taking specimens from different tumor areas best displayed by MRI under direct visual control (Figure 2).

2 MR-GUIDED THERAPY IN THE BRAIN

Figure 1 Mechanical articulated arm systems (Operating Arm System Radionics Software Applications, Inc., Burlington, MA) provides intraoperative neuronavigation using a previously acquired three-dimensional CT or MR data set

4 MR-GUIDED ELECTRODE PLACEMENT Another application of interventional MR is to assist in the placement and in post-placement evaluation of intracerebral depth electrodes.24±27 Intracerebral depth electrodes are used to localize the seizure focus in those patients who are refractory to medical therapy. Prior to the development of this technique, such electrodes were placed by inferring their position relative to previously acquired imaging studies. Postplacement examination of electrode placement could not be accurately performed because of the extensive beam-hardening artifact produced in CT by the stainless steel composition of the electrodes. Previously, MR imaging was not feasible because of the ferromagnetic properties of stainless steel and concern for heat and torque production, which could result in local brain injury. With the manufacture of electrodes with different metal compositions, such as platinum alloys, MR imaging guidance can be employed during electrode placement to con®rm position without risk of brain injury. Suf®cient anatomic detail is obtained to evaluate electrode placement adequately and to avoid possible procedural complications (Figure 3). This should result in improved anatomic±physiologic correlation and surgical outcome for these patients. Intraoperative documentation of longterm implanted depth electrode placed for treatment of chronic pain or movement disorders is another indication in which the anatomical position can be correlated with intraoperative neurophysiological target determination.

5

MR-GUIDED CLOSED MINIMALLY INVASIVE THERAPEUTIC PROCEDURES

In the effort to reduce patient morbidity, hospitalization length, and recovery time associated with conventional craniotomy, several researchers are pioneering new, minimally invasive therapeutic techniques that may achieve results comparable to surgical resection or serve as effective palliative measures. In these minimally invasive procedures, it is particularly appealing to employ MRI beyond mere stereotactic localization and begin to use MRI to monitor treatment administration directly. The sensitivity of MRI to temperature change and other acute tissue effects (e.g. edema) enables the physician to adjust treatment parameters interactively to maximize the destruction of pathologic tissue and minimize undesirable injury to normal tissue.28±30 5.1

Cryoblation

Freezing has been explored as a minimally invasive technique for tissue ablation in the brain. Postoperative MRI of patients who underwent cryothalamotomy demonstrated excellent depiction of the lesions created (Figure 4).31 Intraoperative MRI guidance and monitoring of the cryoablation procedure awaits the development of practical MR-compatible freezing devices.

MR-GUIDED THERAPY IN THE BRAIN

Figure 2 MR-guided biopsy. (a) Direct imaging allows rapid localization of a glioma. (b) MR-guided approach for a deeper lesion

3

4 MR-GUIDED THERAPY IN THE BRAIN

Figure 3 Deep electroencephalographic electrode implantation for patients with medically refractory partial complex seizures who are being considered for possible temporal lobectomy. Coronal MRI was obtained with MR-compatible electrodes in satisfactory position in the hippocampal region

5.2 Ablation by Interstitial Laser Photocoagulation Traditionally, the use of lasers in neurosurgery has been limited to applications in which direct visualization was possible. Now, MRI guidance allows stereotactic placement of an interstitial laser probe into a deep tumor through a small burr hole in the skull. Laser energy travels through the optic ®ber directly into the tumor and produces heat-induced protein denaturation and coagulation necrosis. However, since the penetration and absorption of laser energy are complex functions of tissue characteristics, it is dif®cult to predict a priori the extent of tissue damage as a function of dose. Therefore, a monitoring technique is required to follow the destructive process in tissue as laser energy is applied. Although CT can show laser-induced tissue changes, these are relatively late ®ndings. Early tissue reaction to laser is missed by CT because of its low soft-tissue contrast.32 Fortunately, many researchers have demonstrated the ability of MRI to detect acute tissue changes secondary to the application of laser energy.33±37

Animal studies have characterized the acute and chronic tissue effects of the Nd:YAG (neodymium:yttrium±aluminum± garnet) laser and correlated the histopathologic ®ndings with their MRI appearance.33,35,38 It has been shown that T2weighted images are particularly sensitive to tissue changes and can clearly demonstrate concentric rings of central cavitation, coagulation necrosis, and edema around the point of laser application. Higuchi et al. detected similar patterns in T2weighted images 5 min after laser irradiation in the live rabbit brain.33 Since the primary tissue effect of Nd:YAG laser irradiation is thermal damage,39,40 the thermal sensitivity of MRI can also be exploited to monitor the progress of the laser therapy. The dependence of T1-time, diffusion coef®cient, water proton chemical shift, and other MR parameters on temperature is well known and has been utilized for noninvasive thermal mapping during the application of laser and other thermal ablation techniques.28±30,41±43 Knowledge of the temperature distribution in the treatment area will allow the surgeon to con®rm

MR-GUIDED THERAPY IN THE BRAIN

5

Figure 4 A T1-weighted image obtained several days after stereotactic cryothalamotomy shows the lesion well circumscribed by a hyperintense ring (arrowhead)

the destruction of the target unambiguously while avoiding irreversible thermal damage to vital structures. Interstitial laser therapy is particularly well suited for intraoperative MRI guidance because the laser-delivering optic ®ber is intrinsically MR compatible and because the application of laser energy does not interfere with MR signal acquisition. Initial clinical experience with MR-guided interstitial laser therapy has yielded promising results in the treatment of brain tumors.24 This procedure can be executed under local anesthesia, which allows the neurologic status of the patient to be monitored continuously. MR guidance enables the physician to minimize damage to normal brain tissue, while the absence of systemic toxicity permits repetition of the procedure as necessary. The Dusseldorf group performed a clinical pilot study of MR-guided Nd:YAG laser ablation on 31 patients with brain tumors. Three-dimensional turbo fast low angle shot (FLASH) sequences were used to determine the position of light guide. The time of irradiation varied from 10 to 20 min. Phase-sensitive two-dimensional FLASH and echo-shifted turbo FLASH were used to generate a color-coded heat map. The result is summarized in Table 1. The study shows that MRI is well suited to monitor laser-induced thermotherapy. A typical laser lesion has a central and peripheral zone that make up the total lesion, and it is circumscribed by an enhancing rim demarcating the outer border of the irreversible damaged lesion (Figure 5).

5.3

Radiofrequency-induced Thermal Ablation

Thermal ablation induced by rf is a technique of tissue destruction by resistive heating using rf (~500 kHz) electrical current. A probe with a conductive tip is inserted into the target, and rf current is then delivered from the tip to cause focal heating; this results in protein denaturation and coagulation necrosis. Over three decades of extensive neurosurgical experience with rf has established the ef®cacy and safety of this ablation technique.26,27,44±52 Ablation with rf has been used primarily for functional neurosurgical procedures including thalamotomy and pallidotomy for parkinsonism and other

Table 1 Results of a Clinical Pilot Study of MR-guided Laser Ablation of Brain Tumors: the Dusseldorf Study No. Study population Neurologic deterioration during ablation Transient de®cit after ablation (vasogenic edema) Persistent de®cit after ablation Patients with astrocytoma WHO II Neurologic status unchanged after ablation Neurologic status improved after ablation

31 2 4 1 24 6 18

6 MR-GUIDED THERAPY IN THE BRAIN

Figure 5 Post-processed phase-sensitive two-dimensional fast low angle shot images acquired and displayed during and after laser-induced thermotherapy (LITT). The calculated temperatures are color-coded (bottom right), gradually increasing the heat-affected zone with increasing maximum temperatures. Images acquired during treatment; (a) 1; (b) 4; (c); 8; and (d) 10 min after starting the laser therapy. (Courtesy of T. Kahn. In R. Lufkin: `Interventional MRI', St Louis, 1999, Mosby)

movement disorders, leucotomy for intractable pain, and rhizotomy for trigeminal neuralgia. Postoperative MRI has demonstrated excellent depiction of lesion details in patients who underwent rf thalamotomy.44,53,54 One group has recently completed an initial study of MRguided rf ablation of primary and metastatic brain tumors.5,55 The procedure used MR-guided stereotaxis to place an rf probe (Radionics, Burlington, MA) into the tumor through a 2 mm twist-drill hole in the skull (Figure 6). The rf power was then applied to achieve an intratumoral temperature of 80  C for 1 min. Depending on the shape and size of the tumor, the probe position was adjusted and additional doses were administered. The entire procedure was performed under light sedation and local anesthesia, which allowed the surgeon to monitor the neurologic status of the patient throughout. Pretreatment T1- and T2-weighted MR images provided clear depiction of tumor morphology and anatomic relationships with surrounding vital structures and proved critical in planning the best trajectory for the rf probe. MR angiography with twodimensional time-of-¯ight sequences also proved useful to visualize vascular structures surrounding the planned trajectory of the rf probe. The rf lesion was detected in post-treatment MRI as a region of low signal on T1-weighted images and as a region of high signal on T2-weighted images within 20 to 30

min of the rf ablation. Gadolinium-enhanced T1-weighted images demonstrate a rim of enhancement that outlines the zone of tissue necrosis. A total of 15 tumors were treated in 11 patients. During the 2-year follow-up period, seven tumors were controlled by treatment while eight tumors recurred. Five tumors had a durable response to treatment (11.9±23.8 months of follow up; median 15.1 month) (Figure 7). Neurologic complications were few and transient except for one patient, who still had mild expressive speech dif®culty at 10 weeks after treatment. MR-guided rf thermal ablation promises to be a safe, minimally invasive treatment of brain tumors. This technique may become an effective palliative measure and may serve as an adjuvant to radiation therapy. However, unlike radiation therapy, which is contraindicated by accumulative toxicity, rf ablation can be repeated as many times as necessary to treat recurrences. Greater ef®cacy is anticipated as improved MRcompatible rf generator and therapy probes are developed to allow real-time artifact-free MR imaging of the ablation procedure. As with interstitial laser therapy, the heating effects of the rf ablation can be monitored by temperature-mapping MRI sequences during the application of rf power to con®rm the destruction of target tissue before physiologic changes become apparent.

MR-GUIDED THERAPY IN THE BRAIN

7

Figure 6 MR-guided rf ablation. (a) The therapy probe is placed into the patient's brain through a 2 mm twist-drill hole using MR-guided stereotaxis. The entire procedure is done under light sedation and local anesthesia in the MR suite. (b,c) Images from two other patients treated in the same way

8 MR-GUIDED THERAPY IN THE BRAIN

Figure 7 Contrast-enhanced MR images of a patient who underwent MR-guided rf ablation of a metastatic adenocarcinoma. (a) Image before treatment and (b) immediately after rf ablation. Further images were obtained at (c) 1 week, (d) 5 months, and (e) 1 year after treatment

5.4 High-intensity Focused Ultrasound Ablation High-intensity focused ultrasound (HIFU) is a new technique of thermal ablation by remote heating. An extracorporeal ultrasound transducer focuses vibrational energy to a point inside the patient's body so that only the tissue within a small focal zone is heated to lethal temperatures (60  C or above for thermal coagulation and protein denaturation) while the surrounding tissue experiences only a slight temperature increase. HIFU ablation under ultrasound imaging guidance has been used in clinical trials to treat glaucoma, benign prostatic hyperplasia, tumors of the brain, the breast, the liver, the kidney, and the bladder.56±60 The greatest advantage of HIFU over other thermal ablation techniques is that no surgical invasion is required to access the target provided there is a sonic path clear of bone and air. However, noninvasive methods to measure the temperature of the focal heating zone must be developed to provide adequate control. By combining the high soft-tissue contrast and the temperature sensitivity of MRI with HIFU ablation, a powerful minimally invasive therapy technique can be developed. Several researchers have demonstrated the feasibility of using MRI to guide and monitor HIFU ablation procedures.41,43,61,62 Fast T1-weighted gradient echo sequences can be used to track in realtime the high-temperature focal zone, which appears as a hypointense region, to guide the positioning of the ultrasound transducer. Tissue changes secondary to HIFU heating can be detected easily on T2-weighted images, as with interstitial laser therapy. Although practical application of HIFU to the intracranial region necessitates the removal of a skull bone ¯ap to expose the dura, surgical invasion of the brain parenchyma can be avoided and the potential for infection and other complications greatly reduced. Clinical application of HIFU in

neurosurgery awaits the further development of MRI-compatible HIFU therapy systems and MRI temperature-mapping techniques. 6

CONCLUSION

MRI has traditionally been criticized as an expensive diagnostic technology. However, when compared with more costly open surgical procedures and the associated higher patient morbidity, hospitalization costs, and lost time from work, MRIguided minimally invasive interventions can potentially lower the ®nal cost of medical care in selected patients. More work is clearly warranted in this exciting ®eld to determine its ultimate areas of application in neurosurgery.

7

RELATED ARTICLES

MR-Guided Biopsy, Aspiration, and Cyst Drainage; Temperature Measurement Using In Vivo NMR; Thermal Therapies in the Body Monitored by MRI.

8

REFERENCES

1. P. Mueller, D. Stark, J. Simeone, et al. Radiology, 1986, 161, 605. 2. R. Lufkin, L. Teresi, and W. Hanafee, Am. J. Roentgenol., 1987, 149, 380. 3. R. Lufkin, and L. Lay®eld, J. Comput. Assist. Tomogr., 1998, 13, 1105. 4. S. To, R. Lufkin, and L. Chiu, Comput. Med. Imaging Graph., 1989, 13, 469.

MR-GUIDED THERAPY IN THE BRAIN 5. Y. Anzai, K. Black, A. DeSalles, et al. Am. J. Neuroradiol., 1999, in press. 6. G. Kratimenos, and D. Thomas, Br. J. Neurosurg., 1993, 7, 155. 7. D. Wen, W. Hall, D. Miller, et al. Neurosurgery, 1993, 32, 407. 8. P. Kelly, Adv. Tech. Stand. Neurosurg., 1990, 17, 77. 9. R. Lufkin, S. Jordan, P. Lylyck, et al. Am. J. Neuroradiol., 1988, 9, 953. 10. L. Lunsford, Neurosurgery, 1988, 23, 363. 11. P. Kelly, F. Sharbrough, K. Ball, et al. Mayo. Clin. Proc., 1987, 62, 103. 12. S. Uematsu, A. Rosenbaum, M. Delong, et al. Acta Neurochir. Suppl., 1987, 39, 21. 13. L. Lunsford, A. Martinez, and R. Latchaw, J. Neurosurg., 1986, 64, 872. 14. L. Leksell, D. Leksell, and J. Schwebel, J. Neurol. Neurosurg. Psychiatry, 1985, 48, 14. 15. K. Burchiel, Clin. Neurosurg., 1992, 39, 314. 16. D. Kondziolka, P. Dempsey, L. Lunsford, et al. Neurosurgery, 1992, 30, 402. 17. C. Derosier, G. Delegue, T. Munier, et al. J. Radiol., 1991, 72, 349. 18. J. Villemure, E. Marchand, T. Peters, et al. Appl. Neurophysiol., 1987, 50, 57. 19. C. Bakker, M. Moerland, R. Bhagwandien, et al. Magn. Reson. Imag., 1992, 10, 597. 20. J. Fitzpatrick, H. Chang, M. Willcott, et al. Proc. IXth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1990, p. 426. 21. T. Takizawa, Stereotact. Funct. Neurosurg., 1993, 60, 175. 22. R. Maciunas, R. J. Galloway, J. Fitzpatrick, et al. Stereotact. Funct. Neurosurg., 1992, 58, 108. 23. A. Kato, T. Yoshimine, T. Hayakawa, et al. J. Neurosurg., 1991, 74, 845. 24. P. Ascher, E. Justich, and O. Schrottner, Acta Neurochir., 1991, 52, 78. 25. T. H. Khan, et al. JMRI, 1998, 8, 160. 26. S. Blond, D. Caparros-Lefebvre, and F. Parker, et al. Neurosurg., 1992, 77, 62. 27. L. Laitinen, M. Harriz, and T. Bergenheim, J. Neurosurg., 1992, 76, 53. 28. D. LeBihan, J. Delannoy, and R. L. Levin, Radiology, 1989, 161, 401. 29. R. Dickinson, A. Hall, A. Hind, et al. J. Comput. Assist. Tomogr., 1986, 10, 468. 30. D. Parker, IEEE Trans. Biomed. Eng., 1984, 31, 161. 31. E. Vining, G. Duckwiler, R. Udkoff, et al. J. Neuroimag., 1991, 1, 146. 32. R. A. Gatenby, W. H. Hartz, P. F. Engstrom, et al. Radiology, 1987, 163, 172.

9

33. N. Higuchi, A. Bleier, F. Jolesz, et al. Invest. Radiol., 1992, 27, 814. 34. J. Robinson, R. Lufkin, D. Castro, et al. Eur. Radiology, 1992, 2, 24. 35. Y. Anzai, R. Lufkin, D. Castro, et al. JMRI, 1991, 1, 553. 36. D. J. Castro, R. E. Saxton, L. J. Lay®eld, et al. Laryngoscope, 1990, 100, 541. 37. F. A. Jolesz, A. R. Bleier, P. Jokab, et al. Radiology, 1988, 168, 249. 38. Y. Anzai, R. Lufkin, S. Hirschowitz, et al. JMRI, 1992, 1, 671. 39. F. Jolesz, G. Moore, R. Mulkern, et al. Invest. Radiol., 1989, 24, 1024. 40. K. Matthewson, P. Coleridge-Smith, J. P. O'Sullivan, et al. Gastroenterology, 1987, 93, 550. 41. R. Matsumoto, R. Mulkern, S. Hushek, et al. JMRI, 1994, 4, 65. 42. R. Stollberger, M. Fan, E. Ebner, et al., Proc. XIIth Annu. Mtg. Soc. Magn. Reson. Med., New York, 1993, p. 156. 43. H. Cline, J. Schenck, K. Hynynen, et al. J. Comput. Assist. Tomogr., 1992, 16, 956. 44. F. Tomlinson, C. Jack, and P. Kelly, J. Neurosurg., 1991, 74, 579. 45. S. Hassenbusch and P. Pillary, Neurosurgery, 1990, 27, 220. 46. K. Matsumoto, F. Shichijo, and T. Fukami, J. Neurosurg., 1984, 60, 1033. 47. P. Kelly and F. Gillingham, J. Neurosurg., 1980, 53, 322. 48. J. Siegfried, Surg. Neurol., 1977, 8, 126. 49. L. Organ, Appl. Neurophysiol., 1976, 39, 69. 50. H. Rosomoff, C. Brown, and P. Sheptak, J. Neurosurg., 1965, 23, 639. 51. S. Aronow, J. Neurosurg., 1960, 17, 431. 52. W. Sweet, V. Mark, and H. Hamlin, J. Neurosurg., 1960, 17, 213. 53. Y. Anzai, A. Desalles, K. Black, et al. Radiographics, 1993, 13, 897. 54. S. Matsumoto, F. Shima, K. Hasuo, et al. Nippon Igaku Hoshasen Gakkai Zasshi, 1992, 52, 1559. 55. K. Black, A. DeSalles, Y. Anzai, et al. J. Clin. Oncol., 1999, in press. 56. R. Foster, R. Bihrle, N. Sanghvi, et al. Eur. Urol., 1993, 23(Suppl. 1), 29. 57. G. Vallancien, M. Harouni, B. Veillon, et al. J. Endourol., 1992, 6, 173. 58. R. Silverman, B. Vogelsang, M. Rondeau, et al. Am. J. Ophthalmol., 1991, 111, 327. 59. F. Fry, and L. Johnson, Ultrasound Med. Biol., 1978, 4, 337. 60. A. Burov, Dokl. Akad. Nauk. SSSR, 1956, 106, 239. 61. H. Cline, K. Hynynen, C. Hardy, et al. Magn. Reson. Med., 1994, 31, 628.

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

Neurosurgical Procedures Monitored by Intraoperative MRI Terence Z. Wong Duke University, Durham, NC, USA

and Richard B. Schwartz, Arya Nabavi, Richard S. Pergolizzi Jr, Peter M. Black, Eben Alexander III, Claudia H. Martin, and Ferenc A. Jolesz Brigham and Women's Hospital, Boston, MA, USA

1 INTRODUCTION Image guidance has historically been an integral part of neurosurgery. The introduction of the operating microscope was a major development that enabled the surgeon to decrease the size of the craniotomy and provided superior surface visualization through magni®cation and perfect illumination. Nonetheless, its advantages for surgical navigation were con®ned to the visible surface, and the distance to underlying structures had to be estimated. Subsequently, interactive stereotactic image-guided systems were introduced to yield information on the subsurface structures. This allowed the position of the surgeon's tools or visual ®eld to be interactively correlated anatomically with preoperative computed tomographic (CT) or MR images.1±8 However, image-guided systems that are based on preoperative imaging cannot account for tissue shifts that occur during the course of surgery.9 This has ultimately led to the development of intraoperative imaging techniques. One modality that can be used is ultrasound. Although the image quality of intraoperative ultrasound has improved in recent years, it is still inadequate as the sole means for guiding neurosurgical procedures. Intraoperative MRI would appear to be an ideal imaging modality for guiding neurosurgical procedures. The inherent multiplanar capability of MRI facilitates interactive selection of imaging planes in near-realtime. Conventional T1- and T2weighted images provide exquisite sensitivity for identifying abnormalities within the brain. In addition, specialized imaging sequences are available. These include MR angiographic techniques for mapping blood vessels, dynamic gadoliniumenhanced imaging, functional imaging to map cortical activity, and optimized gradient-echo sequences for early identi®cation of blood products.10,11 The design of new open-con®guration MRI systems combined with the development of MR-compatible surgical instrumentation, anesthesia equipment, and monitoring devices has made intraoperative MRI-guided neurosurgery a reality. At Brigham and Women's Hospital, over 400 neurosurgical procedures have been performed using intraoperative MRI guidance. Design and operational features of the intraoperative

1

MRI system will be described, and our clinical experience with neurosurgical procedures summarized.

2

INTERVENTIONAL MRI DESIGN

Several magnet con®gurations are currently used for MRIguided procedures. These include modi®cation of the conventional long-bore con®guration, a short-bore design, a horizontal gap open con®guration, and a vertical gap open con®guration. While the conventional closed long-bore con®guration will provide the highest image quality, it offers the least patient access. However, noninvasive energy sources such as high-energy focused ultrasound can be integrated into long-bore MRI systems; the imaging capabilities of the 1.5 T magnet allow prelocalization with subtherapeutic temperatures prior to highenergy ablation.12,13 The closed short-bore con®guration with ¯anged bore openings provides improved patient access and maintains the high image quality of a high-®eld system; however it is not easily amenable for perioperative applications. The open-con®guration horizontal gap MRI design allows access to the patient from the sides, permitting most radiologic percutaneous procedures to be performed. The low ®eld minimizes the need for shielding of equipment but also potentially limits imaging capability. Currently, the overlying magnet structure precludes open surgical procedures while the patient is within the imaging ®eld. Both conventional closed-bore and horizontal open con®guration MRI designs can be used intraoperatively by performing surgery outside of the magnet and physically moving the patient into the imaging ®eld for scanning. This must be accomplished while maintaining sterility and is further complicated by the presence of equipment for anesthesia and monitoring the patient. In 1998, Hall et al. reported on perioperative MRI using a conventional 1.5 T system for guiding pediatric neurosurgery.14 Although the patient had to be repositioned in the magnet for each set of images, these authors point out that the use of the high-®eld imager allowed state-ofthe-art pulse sequences to be used and enabled images of the highest quality to be obtained. In 1997, Tronnier et al. reported on the use of a 0.2 T horizontal-gap open-con®guration magnet for intraoperative neurosurgical guidance on 27 patients.15 The advantages of direct MRI visualization during biopsies, cyst aspiration, and catheter placement procedures were realized. For open craniotomy procedures, however, the patient had to be transferred into the magnet for each imaging set. Newer horizontal-gap MRI units are being designed with improved access, which may permit intraoperative imaging without moving the patient. In 1993, a vertical gap mid-®eld strength MRI system was developed as a collaborative effort by General Electric (Signa SP, Milwaukee, WI, USA) and Brigham and Women's Hospital (Figure 1). By allowing vertical access to the patient, this is currently the only interventional MRI design that permits imaging during surgical procedures without moving the patient. The MRI system has been previously described in detail,11,16± 22 and consists of two vertically oriented superconducting coils (55 cm inner bore diameter) in a modi®ed Helmholtz con®guration separated by a 56 cm gap. This spacing permits access to the patient from both sides and allows the surgeon and assistant to work simultaneously in the operating ®eld. Other design

2 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

Figure 1 Interventional/intraoperative MRI system (GE Signa SP) features a vertical gap for surgical access, and is sited within an operating room. (a) Side view: liquid crystal display monitors mounted at eye level on either side of the magnet allow interactive visualization of the MR images. Connections for headlights, suction, navigation, and other instrumentation can be seen on the side of the magnet structure. (b) End-on view: MRI-compatible anesthesia and patient monitoring equipment are shown to the right

features included the use of niobium tin as the superconducting material, which has a higher transition temperature and allows the static ®eld to be maintained without a liquid helium bath, thus permitting a wider gap for patient access. The design results in an imaging volume having a 0.5 T static ®eld de®ned by a 30 cm sphere in the center of the open magnet bore. Flex-

ible transmit/receive coils have been designed to be placed on the patient in close proximity to the imaging volume, while simultaneously allowing surgical and interventional access. The MRI system is maintained in an operating room environment, with adjacent clean and scrub areas. Disposable sterile drapes have been designed to ®t within the surgical space of the mag-

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

net (Baxter, Deer®eld, IL, USA). The design considerations for the intraoperative MRI interventional suite have been previously reported in detail.20 Considerations for safety and imaging compatibility in the MRI environment have been reviewed,20,23 and appropriate materials are available that have low magnetic susceptibility and minimal effect on local MRI.24 Anesthesia equipment, surgical instruments, patient monitors, and electrocautery devices have already been modi®ed by the manufacturers to be MRI compatible. Specialized devices such as the neurosurgical head holder (May®eld; Ohio Medical Instrument Co., Cincinnatti, OH, USA), ¯exible Bookwalter arm (Codman Inc., Burlington, MA, USA), pneumatic neurosurgical drill (Midas Rex, Ft Worth, TX, USA), ultrasonic aspirator for neurosurgical resection (Selector; Elekta, Atlanta, GA, USA), and MR-compatible microscope (Studor Medical Engineering, Rhenfaal, Switzerland) have also been developed and are used routinely at our institution.16 The environment of the MRI system permits it to be used for both interventional and intraoperative applications.20 Although open surgery is by de®nition not a minimally invasive procedure, image guidance can, with accurate targeting and presurgical planning, minimize the damage to healthy tissue and maximize removal of the abnormality; this reduces the invasiveness of the surgical intervention. We summarize below our experience with MRI-guided interventional and intraoperative intracranial procedures using the vertical gap 0.5 T MRI system. To date, approximately 300 tumor resection procedures and 100 intracranial biopsies have been performed by our group with intraoperative MRI guidance. Other procedures have included evaluation of cystic lesions by injection of dilute gadopentetate dimeglumine,25 laser ablation of intracranial tumors,26 and endoscopic sinus surgery.27,28 2.1 Interactive Image Guidance During the MRI-guided neurosurgical procedures, the patient's head is immobilized with a head frame and maintained in the same position throughout the procedure. This provides several important advantages for image guidance. First, patient movement is minimized during imaging. Second, identical slice positions can be obtained and compared during the procedure. Finally, other three-dimensional imaging data sets, such as preoperative MRI, SPECT (single photon emission computed tomography), or PET (positron emission tomography) scans can be co-registered with the MR images and utilized for planning approach and therapy.29±35 It should be emphasized, however, that although the skull is ®xed in position, we have observed that spatial shifts and deformations occur within the brain during the course of surgery, and anatomic co-registration to preoperative imaging studies alone can be subject to signi®cant errors. This problem of tissue shifts may be minimized if access for biopsy or therapeutic probes is limited to small burr holes. In addition, computer models are under development to account for this problem.36,37 2.2 Interactive Imaging Tools Optical, acoustic, and rf techniques have been used to establish the location of biopsy needles or interventional probes in three-dimensional space, and this technology is commonly applied to frameless stereotactic surgery where target coordi-

3

nates are de®ned based on preoperative imaging.1±8. The intraoperative MRI unit features built-in optical (Flashpoint; Image Guided Technologies Inc., Boulder, CO, USA) and rf (CRD; GE Medical Systems, Inc., Schenectady, NY, USA) tracking capabilities. Tracking by rf can be implemented using miniature coils mounted on the surgical device.38,39 Unlike optical or acoustic tracking, intervening tissues are not a problem with rf localization, and this technique is ideal for endoscopic or intravascular procedures. Currently, the optical tracking system is used for most procedures at Brigham and Women's Hospital. Three chargecoupled device (CCD) cameras are mounted over the imaging ®eld in the MRI unit. These detect infrared light-emitting diodes (LEDs) that are, in turn, attached to the interventional probe. Since the spatial relationship between the LEDs and interventional probe is ®xed, the position of the LEDs can be used to de®ne an imaging plane either containing or perpendicular to the probe. Knowledge of the length of the probe allows localization of the probe tip or its projected location. Coordinates of the probe are determined through an interactive workstation (Sun Microsystems, Mountain View, CA, USA), which, in turn, is used to prescribe imaging planes to the SP scanner. Images are subsequently acquired and displayed to the surgeon or interventionalist on liquid crystal display (LCD) monitors which are mounted at eye level on either side of the work area for the MRI (Figure 1). Images are updated every few seconds, re¯ecting adjustments in probe position and providing visual feedback in near-realtime. For interactive imaging, the interventionalist or surgeon uses the interactive probe to navigate through the region of interest, in a manner similar to that utilizing ultrasonography. This technique is particularly useful for establishing the relationship between surface landmarks and underlying structures, and for planning the approach for biopsy or surgical exposure. For needle biopsies or placement of therapeutic probes, the targeting mode is particularly useful. Annotation is added to the acquired images in the form of a `virtual needle', which enables the operator to view the proposed needle path and tip. The trajectory and target can then be veri®ed by imaging in three orthogonal planes, and the operator is provided with near-realtime visual feedback. When the desired trajectory and target point has been established, the interventional probe can be advanced in a stepwise fashion and observed on the MR images as they are acquired in near-realtime; this is termed the tracking mode and con®rms satisfactory positioning of the biopsy needle or therapeutic probe. Two optical tracking handpieces are available for attachment to interventional probes (Figure 2): one has three localization LEDs in a triangular con®guration that form a plane perpendicular to the probe. This device can be ®xed to the surgical table with the Bookwalter clamp and is the handpiece of choice for intracranial biopsies. One of the three LEDs is offset relative to the other two, and this LED along with the biopsy needle de®nes a reference imaging plane that contains the needle trajectory. Other planes containing the needle can be selected by physically rotating the handpiece and/or by specifying a relative rotation through the interactive navigation software. This is done at the workstation under the direction of the radiologist and MR technologist. In addition, the software allows the plane perpendicular to the needle to be imaged at

4 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

Figure 2 Navigational tools feature either two or three LEDs (white arrows) for optical tracking by three cameras mounted overhead in the MRI unit. The two-LED device (left) is used to establish the trajectory of a biopsy needle attached to the end of the handle and is generally used for interventions involving the abdomen, pelvis, or extremities. In the three-LED handpiece (right), the probe is positioned perpendicular to the handle and is typically immobilized using a Bookwalter clamp during neurosurgical procedures. Near-realtime annotation is used to establish the needle trajectory (Figure 3), after which the biopsy needle is advanced under MRI observation

any desired depth relative to the needle tip. The needle length is speci®ed, and software annotation provides a projected tract, or `virtual needle' that can be interactively directed to the target lesion by positioning the handpiece. The second handpiece for optical navigation (Figure 2, left) features two LEDs mounted on a handle that de®nes the trajectory of the interventional probe. This device is used most frequently for non-neurosurgical applications, such as interventions in the abdomen, pelvis, or soft tissues.19 With the twoLED handpiece, the reference imaging plane is de®ned to be the plane containing the interventional probe and perpendicular to the ground. Other planes can be selected through interactive software prescription.19 Although used routinely for biopsy procedures, the optical navigation system can be similarly utilized to position other catheters or needles (e.g. for cyst evaluation or drainage25) or to place thermal ablation probes.26

3 MRI-GUIDED BIOPSY PROCEDURES Interactive MRI guidance offers several advantages over conventional stereotactic biopsy techniques. Conventional stereotactic procedures require the placement of a head frame on the patient followed by a set of preprocedure CT or MR images to de®ne target coordinates. A signi®cant advantage of interactive guidance is that it eliminates the need for this twostep process. The MRI sequence best de®ning the target tissue

can be selected for image guidance; gadolinium contrast can be injected prior to biopsy; this is usually done after the burr hole or craniotomy has been made and the dura exposed. Perhaps the most important advantage of interactive MRI guidance is that the actual path of the biopsy needle can be followed as it is advanced toward the lesion, and small compensatory changes can be made until the sampling port is positioned within the target. Imaging can then con®rm in all three dimensions that the biopsy site lies within the tissue having the abnormal MR signal. The ability to make realtime adjustments of the biopsy needle, along with con®rmation of biopsy location, may signi®cantly reduce sampling error, as well as decrease the number of needle passes. This can lead to higher diagnostic yield with reduced morbidity. Of the nearly 100 MRI-guided intracranial biopsies performed to date at Brigham and Women's Hospital, approximately two thirds were performed under general anesthesia with the remaining one third performed using monitored conscious sedation. Average procedure time was less than 2 h if general anesthesia was used, and approximately 1.5 h when conscious sedation was utilized. For navigational MRI, sequences were selected to de®ne the target best. For enhancing lesions, 10±20 ml intravenous gadopentetate dimeglumine was injected, and T1-weighted fast spin-echo (FSE) images used (updated every 14 s). For nonenhancing lesions, T2weighted FSE imaging was usually used (updated every 15±30 s). More recently, a SSFSE (single-shot fast spin-echo) protocol has been developed that allows T2-weighted images to be acquired every 4 s. An example of a biopsy performed using

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

5

Figure 3 Interactive MRI for biopsy of a right frontal lesion using the 3-LED handpiece and T1-weighted images following intravenous gadolinium enhancement. (a) Para-axial image perpendicular to needle in the plane of the biopsy port allows targeting of enhancing margin. (b) Para-sagittal image in the plane of the needle shows the `virtual needle' depicted as dashed lines directed at the target. Note the small signal void at the cortical surface from the biopsy needle tip prior to insertion. (c) As the titanium needle is advanced, near-realtime imaging allows observation of the needle tract along the virtual needle annotation. This allows ®ne adjustments to be made accordingly and enables the position of the actual biopsy needle to be con®rmed prior to sampling. (d) Actual needle tract is con®rmed in the perpendicular para-coronal plane. The tip of the virtual needle corresponds to the level of the sampling port of the biopsy needle; as a result the actual needle tip extends a few millimeters beyond the dashed annotation. Note focal signal void within the enhancing rim from biopsy sample

MRI-based navigation is illustrated in Figure 3. In this case, contrast-enhanced T1-weighted imaging was used for guidance. After the patient has been positioned in the headholder on the MRI operating table, the handpiece may be used in the interactive imaging mode to plan the incision location and biopsy trajectory. Prior to inserting the biopsy needle, the targeting mode is utilized, with annotation used to de®ne a virtual needle. The target position can be con®rmed in three orthogonal planes (Figure 3). With the handpiece immobilized with the Bookwalter arm, the titanium biopsy needle is advanced in a stepwise manner as MR images are acquired. The needle is observed as a signal void overlying the virtual needle annotation (Figure 3c,d). It should be noted that the tip of the virtual needle is set to correspond to the location of the sideport of the biopsy needle; therefore, the actual needle tip extends several millimeters beyond the virtual needle annotation. This con®rms that the biopsy needle is traveling along the expected trajectory

and to the desired target depth. If necessary, the annotation can be removed to allow the actual needle location to be seen more clearly and recon®rmed in all three planes. The ability to con®rm that the biopsy specimen was obtained within the MRde®ned abnormality minimizes sampling error and potentially reduces the number of needle passes and associated complications.

4

MRI GUIDANCE FOR OPEN CRANIOTOMIES

Since 1996, approximately 270 open craniotomies have been performed at Brigham and Women's Hospital for MRIguided tumor resection. Patient selection, demographics, and preliminary clinical results have been previously reported.11,16,18,22 In preparation for surgery, the patient's head is

6 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI positioned in the MRI with the head holder to allow the best surgical access. As a result, the patient's anatomic frame of reference no longer corresponds to conventional MRI planes. To overcome this problem, a new frame of reference is established to allow true axial, sagittal, and coronal images to be obtained. This can be easily accomplished using the navigational tools previously described; interactive imaging is used to de®ne an appropriate reference imaging plane from which conventional axial, sagittal, or coronal images can be prescribed. Since the patient is immobilized, the reference imaging procedure need only be performed once and is typically done immediately after the patient has been positioned within the intraoperative MRI. The use of conventional imaging planes facilitates accurate localization and identi®cation of critical anatomic structures as the surgery progresses. During the procedure, the radiologist prescribes the appropriate imaging planes and sequences that will best address speci®c questions of the surgeon as the surgery progresses. For example, for initial identi®cation of tumor boundaries, either conventional spin-echo gadolinium-enhanced T1-weighted images or T2weighted images may be most useful. Other specialized imaging sequences have been developed, including SSFSE for rapid T2-weighted imaging, dynamic contrast-enhanced imaging, susceptibility sensitive sequences to evaluate early hemorrhage,10,11 and rapid volumetric image acquisition techniques. In patients who have been previously treated with high-dose radiotherapy (stereotactic radiosurgery), it is not possible to distinguish active tumor recurrence from postradiation necrosis on conventional delayed gadolinium-enhanced images. Dual-isotope 201T1/99mTc-HMPAO (hexamethylpropyleneamine oxime) SPECT scanning has been found to be useful for making this important distinction,33,34 and these images can be co-registered with intraoperative MR images. Alternatively, a dynamic contrast-enhanced MR study can be performed. These techniques have been useful for estimating tumor grade as well as for distinguishing recurrent tumor from post-treatment change.40±42 Our intraoperative technique is to obtain twodimensional fast spoiled gradient-echo images through the tumor volume, with each slice sampled every 11±15 s for 2±3 min following rapid bolus contrast injection. Tumor recurrence, with associated neovascularily, will tend to enhance early, while post-treatment necrosis, related primarily to blood±brain barrier breakdown, will tend to show delayed enhancement. Distinguishing active recurrence from radionecrosis is valuable both for selecting appropriate biopsy sites and for planning the surgical resection. This has been very useful in the intraoperative setting. In an evaluation of a series of 24 patients with intraoperative dynamic MRI,4314 of 15 lesions demonstrating early enhancement had recurrent tumor by pathology; out of the nine lesions that did not show early enhancement, eight had only reactive post-treatment changes. In this study, early enhancement was determined visually when the appearance of contrast within the lesion occurred at the same time as in vascular structures or the choroid plexus. For all intraoperative MR, an initial baseline set of twodimensional slices is obtained through the volume of interest. Imaging is repeated at intervals during the surgery to verify the present location in relation to the remaining tumor to be excised. Additionally, the location of the surgical resection cavity relative to adjacent eloquent cortical pathways is always of

major importance. Once the most appropriate imaging sequence and imaging plane have been preloaded into the MR console, identical image slices can be acquired simply by rescanning. This allows direct serial comparison in the same imaging plane. Acquisition time for each imaging data set varies depending on the imaging sequence but takes approximately 1±2 min. Prior to imaging, electrical equipment and metallic instruments are removed from the surgical ®eld. Although these devices are MRI-compatible they can produce signi®cant imaging artifacts if they are left in place. Imaging is possible, however, with the operating microscope in position. In our experience, intraoperative MRI guidance enables the surgeon to verify his or her planned approach and surgical procedure and also allows him or her to change the strategy during the procedure according to the newly acquired information. Figure 4 illustrates an MRI-guided tumor resection. In this case, T2-weighted images were used for surgical guidance. Serial images from a single axial slice are shown in this example. Early in the procedure, the resection site was de®ned by a locator. The tissue anterior to this (white arrow, Figure 4(a)) was felt to represent non-neoplastic changes after previous surgery, and this was con®rmed by intraoperative histopathology. The remaining tumor was represented by a high T2 signal posterior to the pointer and just anterior to the motor strip. Stepwise resection of the tumor is illustrated in Figure 4b,c, with complete resection shown in Figure 4d. Prior to closing, the cavity was irrigated with saline (Figure 4e), and multi-echo gradient-echo images were obtained to evaluate for signi®cant hemorrhage. The MRI characteristics of hyperacute intracranial hemorrhage have been recently described.25 In our experience, T2-weighted images (FSE, TR/TE=4000/100 ms; Figure 4e) in combination with gradient-echo sequences (30 ¯ip angle, TR=600 ms) is sensitive for identifying hyperacute hemorrhage and for distinguishing this from postsurgical ¯uid. Figure 4f,g shows postsurgical gradient-echo images with TE of 9 and 60 ms, respectively. A typical postoperative appearance is illustrated here, with irrigation saline (bright T2) within the resection cavity, and a thin rim of low signal bordering the resection cavity on the short TE image (Figure 4f) that `blooms' on the second echo (TE=60 ms; Figure 4g). This rim is felt to represent deoxyhemoglobin content within acute blood products at the resection border. No hematoma, mass effect, or extra-axial collection was noted in this patient. Hematomas tend to have a non¯uid central area (lower T2 signal), with a rim of susceptibility that also blooms on gradient-recalled echo imaging. Intraoperative MRI-guided tumor resection provides several advantages over conventional techniques. Image guidance is interactive. Speci®c sequences and imaging planes can be applied during the procedure to optimize selection of involved tissue. For example, dynamic or steady-state contrast-enhanced images can be used to delineate active tumor, and T2-weighted images can be used to identify nonenhancing tumor. Neoplastic tissue having abnormal MR signal characteristics but normal appearance visually can be identi®ed and resected. Tissue shifts during surgery have been observed and are easily compensated for with intraoperative imaging. Critical anatomical and functional structures can be identi®ed during resection and, at crucial points in the procedure, images can be obtained more frequently in multiple planes if necessary to re®ne the resection margin. Finally, imaging capability allows early detection of

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

7

Figure 4 Intraoperative imaging during neurosurgical resection: serial T2-weighted axial images. (a) Early in the surgery, the neoplastic tissue is posterior to the surgeon's ®nger. Anterior tissue (arrow) represents postoperative change from previous surgery and was present on prior MR study (b,c) Serial resection, (d) complete resection of the T2-abnormal tissue. Some tissue shift is evident. (e) Following resection, the cavity is ®lled with saline. (f,g) Postsurgical gradient-echo images allow evaluation of hyperacute hemorrhage. These show typical postsurgical changes. (Reprinted with permission from Wong et al.22)

8 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

Figure 5 Intraoperative parasagittal image (note intraoperative pointer, white arrow) co-registered with the patient's preoperative three-dimensional brain model. Operative skin ¯ap in the lower portion of the image on the left. The preoperative model includes conventional MRI with segmentation of tumor (green) and ventricles (light blue). In addition, MR angiography was performed to display large vessels (red). Functional MRI was used to identify cortical activation during complex speech tasks (turquoise and yellow) and motor tasks (magenta)

any unexpected complication, such as hemorrhage. Out of some 400 neurosurgical procedures, postsurgical hematomas were observed in four patients. In all cases, imaging permitted immediate identi®cation, and facilitated removal in three patients. In the fourth patient, serial MR images were used to con®rm stability of a small hematoma, and removal was unnecessary. In one patient, a hematoma was noted in the hemisphere contralateral to the surgery and was felt to represent a spontaneous hematoma, possibly from blood pressure instability. This would not have been identi®ed without immediate postoperative imaging; a second craniotomy was performed using MRI guidance to evacuate it. 5 FUTURE PERSPECTIVES Specialized MRI sequences are available on higher-®eld MRI units that are not yet available on the intraoperative MRI

system, and sophisticated interactive methods have been developed for preoperative surgical planning.44 In addition, other three-dimensional imaging modalities can provide important information not available by MR techniques. Therefore, we have developed co-registration software to integrate preoperatively acquired multimodality information with the intraoperative imaging. Examples of specialized studies that can be performed prior to surgery include MR angiography, PET, SPECT, and functional MRI. MR angiography can de®ne the relationship of the tumor to major vascular structures during the surgical procedure. PET imaging with [18F]-¯uorodeoxyglucose re¯ects metabolic activity of gliomas, and areas of higher uptake correlate with higher tumor grade and poorer prognosis.29±31 Dualisotope 201T1/99mTc-HMPAO SPECT is useful for distinguishing tumor recurrence from radiation necrosis,33 and ®ndings also correlate with histopathological ®ndings and survival.34 Image registration of MRI with PET or SPECT images can, therefore, be useful for directing biopsy to regions most repre-

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI

9

Figure 6 Surgical planning computer display. Preoperative three-dimensional model (right, upper frame) demonstrating tumor (green) and anatomic relationship to superior sagittal sinus (red), corpus callosum (brown), and ventricle (blue). The bottom row displays three intraoperative sagittal images in the same imaging plane. On the left is the unenhanced T1-weighted image prior to craniotomy. The middle image demonstrates enhancing tumor prior to resection and is reproduced in the large upper left frame. The image on the bottom right demonstrates the surgical resection cavity following tumor removal

sentative of tumor behavior; this may reduce sampling error. For image-guided tumor resection, the addition of PET or SPECT data can help the neurosurgeon to select most aggressive portions of tumor for resection. Functional MRI can be

used to map areas of cortical activation during motor tasks, enabling critical sensorimotor areas to be identi®ed and allowing the neurosurgeon to avoid eloquent areas during tumor resection.45

10 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI The preoperatively acquired information is combined with the intraoperative imaging data set using registration algorithms, thus maximizing the information available during surgery. The grayscale images and three-dimensional models of speci®c structures are displayed to the surgeon and correspond spatially to the actual patient as positioned in the intraoperative MRI (Figure 5). The combination of the preoperative information and the intraoperative renewal of the grayscale information allows updated neuronavigation. With progress of the surgical intervention and increasing brain deformation, the validity of the preoperative information decreases. We are currently evaluating the de®nition of a deformation ®eld from the intraoperatively acquired grayscale images to align the threedimensional models and their preoperative information to the updated situation. Other models to account for brain deformation are being developed.36,37 The combination of pre- and intraoperative information, their alignment and the tracking of instruments in this de®ned space comprise the capabilities of this system for realtime interactive image-guided surgery (Figure 6).

6 CONCLUSIONS Intraoperative MRI-guided neurosurgical procedures are a collaborative effort, involving neurosurgeons, neuroradiologists, anesthesiologists, MR technologists, nurses, and engineers. MRI provides excellent contrast between abnormal and healthy brain tissue, and intraoperative imaging permits these areas to be de®ned relative to the surgeon's ®eld of view. This improves the accuracy of intracranial biopsies and surgical resection of tumors. Early identi®cation of unexpected complications is another bene®t. To date, MRI guidance has been utilized in selected complex cases or repeat surgery. Based on our current experience, there is little question that MRI guidance for neurosurgical procedures augments the technical capability of the neurosurgeon to remove neoplastic tissue selectively. However, evaluation of cost effectiveness and clinical outcomes relative to conventional neurosurgical procedures will require a prospective, randomized clinical trial. In the future, technological developments will continue to reduce the degree of invasiveness required for tumor resection. De®nition of the tissue targeted for resection and its relationship to critical functional structures will continue to be improved through multimodality presurgical planning. Improvements to MRI will include faster acquisition times and re®nement of imaging sequences for neurosurgical guidance. Currently available `near-realtime' imaging will become closer to realtime. Image processing will continue to become faster, and techniques to compensate for intraoperative tissue shifts and removal are under development. Surgical image guidance will ultimately be a synthesis of anatomical/functional preoperative images with updated co-registered intraoperative imaging. Image guidance will allow neurosurgeons to treat benign and malignant lesions using minimally invasive tools. Thermal ablation techniques using laser, rf, ultrasound, and microwave energy sources are currently being investigated, as is cryotherapy. Central to the success of these minimally invasive therapies is image guidance to position the ablation probes and monitor the thermal therapy. The navigational tools available

for MRI-guided biopsies described above can be used to position thermal ablation probes accurately. In addition, MRI is uniquely suited to reveal subtherapeutic as well as irreversible thermal effects, and development of improved, faster imaging sequences to monitor these tissue effects will parallel the re®nement of energy sources to deliver controlled, directed thermal ablation.

7

RELATED ARTICLES

ESR Probes as Field Detectors in MRI; MR-Guided Biopsy, Aspiration, and Cyst Drainage; MR-Guided Therapy in the Brain; Thermal Therapies in the Body Monitored by MRI.

8

REFERENCES

1. D. W. Roberts, J. W. Strohbehn, J. F. Hatch, W. Murray, and H. A. Kettenberger, J. Neurosurg., 1986, 65, 545. 2. R. J. Maciunas, M. S. Berger, B. Copeland, M. R. Mayberg, R. Selker, and G. S. Allen, Neurosurg. Clin. North Am., 1996, 7, 245. 3. D. Kondziolka and L. D. Lunsford, Intraoperative navigation during resection of brain metastases, Neurosurg. Clin. North Am., 1996, 7, 267. 4. P. K. Doshi, L. Lemmieux, D. R. Fish, S. D. Shorvon, W. H. Harkness, and D. G. Thomas, Acta Neurochir. Suppl. (Wien), 1995, 64, 49. 5. C. R. Maurer Jr, J. M. Fitzpatrick, M. Y. Wang, R. L. Galloway Jr, R. J. Maciunas, and G. S. Allen, IEEE Trans. Med. Imag., 1997, 16, 447. 6. D. G. Thomas, P. Doshi, A. Colchester, D. J. Hawkes, D. L. Hill, J. Zhao, N. Maitland, A. J. Strong, and R. I. Evans, Sterotact. Funct. Neurosurg., 1996, 66, 81. 7. E. Watanabe, Y. Mayanagi, Y. Kosugi, S. Manaka, and K. Takakura, Neurosurgery, 1991, 28, 792. 8. C. Schaller, B. Meyer, D. van Roost, and J. Schramm, Comput. Aided Surg., 1997, 2, 162. 9. C. R. Maurer Jr, D. L. Hill, A. J. Martin, H. Liu, M. McCue, D. Rueckert, D. Lloret, W. A. Hall, R. E. Maxwell, D. J. Hawkes, and C. L. Truwit, IEEE Trans. Med. Imag., 1998, 17, 817. 10. S. W. Atlas and K. R. Thulborn, Am. J. Neuroradiol., 1998, 19, 1471. 11. R. B. Schwartz, L. Hsu, T. Z. Wong, D. F. Kacher, A. A. Zamani, P. M. Black, E. Alexander, P. E. Stieg, T. M. Moriarty, C. A. Martin, R. Kikinis, and F. A. Jolesz, Radiology, 1999, 211, 477. 12. K. Hynynen, N. I. Vykhodtseva, A. H. Chung, V. Sorrentino, V. Colucci, and F. A. Jolesz, Radiology, 1997, 204, 247. 13. K. Hynynen, W. R. Freund, H. Cline, A. H. Chung, R. D. Watkins, J. P. Vetro, and F. A. Jolesz, Radiographics, 1996, 16, 185. 14. W. A. Hall, A. J. Martin, H. Liu, C. H. Pozza, S. O. Casey, E. Michel, E. S. Nussbaum, R. E. Maxwell, and C. L. Truwit, Pediatr. Neurosurg., 1998, 29, 253. 15. V. M. Tronnier, C. R. Wirtz, M. Knauth, G. Lenz, O. Pastyr, M. M. Bonsanto, F. K. Albert, R. Kuth, A. Staubert, W. Schlegel, K. Sartor, and S. Kunze, Neurosurgery, 1997, 40, 891. 16. P. M. Black, T. Moriarty, E. A. Alexander III, P. Stieg, E. J. Woodard, P. L. Gleason, C. H. Martin, R. Kikinis, R. B. Schwartz, and F. A. Jolesz, Neurosurgery, 1997, 41, 831. 17. J. F. Schenck, F. A. Jolesz, P. B. Roemer et al., Radiology, 1995, 195, 805. 18. T. M. Moriarty, R. Kikinis, F. A. Jolesz, P. M. Black, and E. Alexander III, Neurosurg. Clin. North Am., 1996, 7, 323.

NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI 19. S. G. Silverman, B. D. Collick, M. R. Figueira, R. Khorasani, D. F. Adams, R. W. Newman, G. P. Topulos, and F. A. Jolesz, Radiology, 1995, 197, 175. 20. S. G. Silverman, F. A. Jolesz, R. W. Newman, P. R. Morrison, A. R. Kanan, R. Kikinis, R. B. Schwartz, L. Hsu, S. J. Koran, and G. P. Topulos, Am. J. Roentgenol., 1997, 168, 1465. 21. F. A. Jolesz, Radiology, 1997, 204, 601. 22. T. Z. Wong, R. B. Schwartz, PMcL Black, E. Alexander III, and F. A. Jolesz, Semin. Intervent. Radiol., 1999, 16, 23. 23. J. F. Schenck, Med. Phys., 1996, 23, 815. 24. F. A. Jolesz, P. R. Morrison, S. J. Koran, R. J. Kelley, S. G. Hushek, R. W. Newman, M. P. Fried, A. Melzer, R. M. Seibel, and H. Jalahej, JMRI, 1998, 8, 8. 25. R. B. Schwartz, L. Hsu, P. M. Black, JMRI, 1998, 8, 807. 26. J. Kettenbach, S. G. Silverman, N. Hata, K. Kuroda, P. Saiviroonporn, G. P. Zientara, P. R. Morrison, S. G. Hushek, P. M. Black, R. Kikinis, and F. A. Jolesz, JMRI, 1998, 8, 933. 27. M. P. Fried, G. Topulos, L. Hsu, H. Jalahej, H. Gopal, A. Lauretano, P. R. Morrison, and F. A. Jolesz, Otolaryngol. Head Neck Surg., 1998, 119, 374. 28. L. Hsu, M. P. Fried, and F. A. Jolesz, Am. J. Neuroradiol., 1998, 19, 1235. 29. M. W. Hanson, M. J. Glantz, J. M. Hoffman, A. H. Friedman, P. C. Burger, S. C. Schold, and R. E. Coleman, J. Comput. Assist. Tomogr., 1991, 15, 796. 30. J. B. Alavi, A. Alavi, J. Chawluk, M. Kushner, J. Powe, W. Hickey, and M. Reivich, Cancer, 1988, 62, 1074. 31. M. Levivier, S. Goldman, B. Pirotte, J. Brucher, D. Baleriaux, A. Luxen, J. Hildebrand, and J. Brotchi, J. Neurosurg., 1995, 82, 445. 32. T. G. Turkington, R. J. Jaszczak, C. A. Pelizzari, C. C. Harris, J. R. MacFall, J. M. Hoffman, and R. E. Coleman, J. Nucl. Med., 1993, 34, 1587. 33. R. B. Schwartz, P. A. Carvalho, E. A. Alexander III, J. S. Loef¯er, R. Folkerth, and B. L. Holman, Am. J. Neuroradiol., 1991, 12, 1187. 34. R. B. Schwartz, B. L. Holman, J. F. Polak, B. M. Garada, M. S. Schwartz, R. Folkerth, P. A. Carvalho, J. S. Loef¯er, D. C. Shrieve, P. M. Black, and E. Alexander III, J. Neurosurg., 1998, 89, 60. 35. B. L. Holman, R. E. Zimmerman, K. A. Johnson, P. A. Carvalho, R. B. Schwartz, J. S. Loef¯er, E. Alexander, C. A. Pelizzari, and G. T. Chen, J. Nucl. Med., 1991, 32, 1478. 36. P. J. Edwards, D. L. Hill, J. A. Little, and D. J. Hawkes, Med. Image Anal., 1998, 2, 355. 37. K. D. Paulsen, M. I. Miga, F. E. Kennedy, P. J. Hoopes, A. Hartov, and D. W. Roberts, IEEE Trans. Biomed. Eng., 1999, 46, 213. 38. D. A. Leung, J. F. Debatin, S. Wildermuth, N. Heske, C. L. Dumoulin, R. D. Darrow, M. Hauser, C. P. Davis, and G. K. Schulthess, Radiology, 1995, 197, 485. 39. S. Wildermuth, J. F. Debatin, D. A. Leung, C. L. Dumoulin, R. D. Darrow, G. Uhlschmid, E. Hoffman, J. Thyregod, and G. K. von Schulthess, Radiology, 1997, 202, 578. 40. H. J. Aronen, I. E. Gazit, and D. N. Louis et al., Radiology, 1994, 191, 41. 41. H. F. Aronen, J. Glass, F. S. Pardo et al., Acta Radiol., 1995, 36, 520. 42. J. D. Hazle, E. F. Jackson, D. F. Schomer, and N. E. Leeds, JMRI, 1997, 7, 1084. 43. R. B. Schwartz, L. Hsu, D. F. Kacher et al., JMRI, 1998, 8, 1085. 44. R. Kikinis, P. L. Gleason, T. M. Moriarty, M. R. Moore, E. Alexander III, P. E. Stieg, M. Matsumae, W. E. Lorensen, H. E. Cline, P. M. Black, and F. A. Jolesz, Neurosurgery, 1996, 38, 640. 45. M. F. Nitschke, U. H. Melchert, C. Hahn, V. Otto, H. Arnold, H. D. Herrmann, G. Nowak, M. Westphal, and K. Wessel, Acta Neurochir. (Wien), 1998, 140, 1223.

11

Biographical Sketches Terence Z. Wong. b 1955. B.S.E. biomedical engineering, 1977, Duke University, M.D. and Ph.D., 1990, Dartmouth College and Dartmouth Medical School. Residency in Diagnostic Radiology, Deaconess Hospital, Harvard Medical School, 1992±96, Fellowship in Nuclear Medicine, Deaconess Hospital 1995±96, Fellowship in Cross-sectional Imaging, Beth Israel Deaconess Medical Center 1996±97, Fellowship in Interventional MRI, Brigham and Women's Hospital 1997±98. Assistant Professor of Radiology, Assistant Professor of Radiation Oncology, and Assistant Professor of Biomedical Engineering, Duke University Medical Center 1998±present. Current research interests: image-guided therapy, multimodality metabolic and functional imaging, selective oncologic imaging and therapy, thermal therapy. Richard B. Schwartz. b 1956. B.S., 1978, M.D., 1982, Ph.D., 1985, Neuroanatomy, University of Pennsylvania, USA. Neuropathology internship, Tufts-New England Medical Center, Boston MA, 1984±85. Radiology residency, Brigham and Women's Hospital, Boston MA, 1985±89. Neuroradiology fellowship, Brigham and Women's Hospital, Boston MA, 1989. Certi®cate of Additional Quali®cation, 1996. Instructor, assistant professor, and associate professor of radiology, Harvard University, 1989±present. Assistant Director of the Neuroradiology division, Brigham and Women's Hospital, Boston MA, 1992± present. Director of Functional and Navigational Neuroimaging, 1999. Approx. 80 publications in CT, MR and SPECT scanning of the brain. Current research interests: intracranial intervention under MR guidance, acute hypertensive disorders of the brain, neurovascular applications of spiral CT, and functional imaging of brain tumors using MR and SPECT. Arya Nabavi. b 1966, M.D. University of Kiel/Germany. Neurosurgeon, Research Fellow at the Department of Neurosurgery, Brigham and Women's Hospital, Harvard Medical School. Current research interests: application of MR in neurosurgery, image-guided surgery, intraoperative brain deformations. Richard S. Pergolizzi Jr. b 1964. B.Sc. Chemistry, 1987, The State University of New York at Stony Brook, M.D. 1993, Northeastern Ohio Universities College of Medicine. Residency in Diagnostic Radiology at University of Cincinnati, University Hospital 1993±97, Fellowship in Neuroradiology, University Hospital Health Science Center, Syracuse NY 1997±98, Fellow in Neurointerventional Radiology, Brigham and Women's Hospital and Massachusetts General Hospital, Harvard University, 1998±present. Current research interests: clinical applications for MR image-guided therapy. Peter M. Black. b 1944, A.B., 1966, Harvard College, M.D. 1970 McGill University, Montreal, Canada, Ph.D., 1978, Georgetown University, USA. Research fellow in Neuro-oncology, Massachusetts General Hospital, 1976, Asst. Professor of Surgery, Harvard Medical School, 1980±84, Associate Professor of Surgery, Harvard Medical School, 1984±86. Franc D. Ingraham Professor of Neurosurgery, Harvard Medical School, 1987±present. Neurosurgeon-in-Chief, Brigham and Women's and Children's Hospitals, 1987±present. Chief of Neurosurgical Oncology, Dana Farber Cancer Institute, 1987±present. Chairman of the Board, Brain Tumor Center, BWH, CH, DFCI, 1988± present. Chairman, Department of Surgery, CH, 1994±98. More than 40 visiting professorships and lectureships. Approx. 155 original articles, 7 books, 150 reviews and book chapters, and several clinical communications. Current major research interests: advanced imaging techniques for surgery including the intraoperative MRI and brain mapping, and in the laboratory, molecular biology of brain tumors, speci®cally the relation between glial development and glioma formation, and hormonal in¯uences on meningioma progression. Eben Alexander III. b 1953. A.B., 1975, Chemistry, University of North Carolina, Chapel Hill. M.D., 1980, Duke University School of Medicine, Durham, North Carolina, Intern in General Surgery, Duke University Medical Center, 1980±81. Resident, Neurological Surgery,

12 NEUROSURGICAL PROCEDURES MONITORED BY INTRAOPERATIVE MRI Duke University Medical Center, 1981±83. Research Fellow in Neurosurgery, Massachusetts General Hospital, Harvard Medical School, 1983±84. Acting Resident in Neurology, Massachusetts General Hospital, Boston, Massachusetts, 1985. Resident in Neurological Surgery, Duke University Medical Center, 1985±87. Senior Registrar and Cerebrovascular Fellow, Neurosurgical Service, Newcastle General Hospital, Newcastle-Upon-Tyne, England, U.K., 1987. Instructor in Surgery, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts 1988±90. Assistant Professor of Surgery (Neurosurgery), 1990±94, and Radiation Oncology, Harvard Medical School, 1990±99. Associate Professor of Surgery (Neurosurgery), Harvard Medical School, 1994±99. Director of the Stereostatic and Functional Neurosurgery Program, Brigham and Women's Hospital. Approx. 150 publications in neurosurgery, especially radiosurgery and image-guided surgery. Current research speciality: development of advanced synergistic radiation and surgical techniques through image-guidance. Claudia H. Martin. b 1959. B.Mus., 1984, M.D., 1988, McGill University, Canada: Surgical Internship. Yale University, New Haven, 1989± 90, NIH postdoctoral fellow in neurosurgery, Bethesda, Maryland, 1990±91, Neurosurgery resident, Yale University, New Haven, 1991± 95, Neurosurgery clinical fellow, Brigham and Women's Hospital, Harvard University, Boston, 1996±99. Chief resident Brigham and Women's Hospital, Boston, 1999±2000. Approximately 10 publications in image guided neurosurgery.

Ferenc A. Jolesz. b 1946. M.D., 1971, Semmelweis Medical School, Hungary. Resident in Neurosurgery, Medical School Pecs, Hungary, 1971±72. Research Fellow, Biomedical Engineering Computer Sciences, K. Kando College, Hungary, 1972±73. Resident in Neurosurgery, Institute of Neurosurgery, Hungary, 1975±79, Research Fellow, Neurology, Massachusetts General Hospital, Boston, MA, 1979± 80, Milton Research Fellow, Physiology, Harvard Medical School, Boston, MA, 1980, Research Fellow, Physiology, Harvard Medical School, Boston, MA, 1980±81, Research Associate, Physiology, Harvard Medical School, Boston, MA, 1981±82, Resident, Radiology, Brigham and Women's Hospital, Boston, MA, 1982±85, Assistant Professor, Radiology, Harvard Medical School, Boston, MA, Director, Neuro MR Imaging Section, Brigham and Women's Hospital, Boston MA, 1987±88, Director, Division of Magnetic Resonance Imaging, Brigham and Women's Hospital, Boston, MA, 1988, Associate Professor of Radiology, Harvard Medical School, Boston, MA, 1989±96, Director, Image-guided Therapy Program, Brigham and Women's Hospital, Boston, MA, 1993±present. Professor of Radiology, Harvard Medical School, Boston, MA, 1996. B. Leonard Holman Professor of Radiology, Harvard Medical School, Boston, MA, 1998. Member of the Institute of Medicine, National Academy of Sciences. Approx. 200 publications. Current research speciality: interventional and intraoperative MRI, quantitative neuroimaging, focused ultrasound surgery.

TEMPERATURE MEASUREMENT IN VIVO USING NMR

Temperature Measurement In Vivo Using NMR

3

TEMPERATURE DEPENDENCE OF T1

Theory predicts a relationship between temperature and T1 of the general form:

Ian R. Young

1   T1 T

The Robert Steiner MRI Unit, Hammersmith Hospital, London, UK

1 INTRODUCTION Philosophically, the desirability of using imaging to monitor therapy as it happens seems indisputable, with the potential this offers for reducing any uncertainty in treatment and for permitting adjustment of treatment as it proceeds to optimize outcome. Allied to the advantages of minimally invasive therapy ± reduced patient trauma, shorter duration in hospital ± the case for imaging in conjunction with the ablative therapies has seemed excellent. In this respect, because MR has so many potentially useful temperature dependencies, which might allow quantitative measurement of what is happening, it has seemed the modality of choice, and signi®cant research effort has been expended in efforts to convert that perception to reality. Most of this work has gone into attempts to exploit the thermal dependencies of T1 and the chemical shift of water; a much smaller resource has been applied to the temperature sensitivity of the diffusion coef®cient. Attempts to use T1 changes have included studies of the relaxation behavior of other nuclei than hydrogen; while chemical shift strategies have employed spectroscopy as well as imaging techniques. All of these approaches will be reviewed in the article. Each method will be covered in a brief section that will include a description of attempts made to use it.

A prime concern in thermal ablations is to control the extent of any unwanted damage to normal tissue around the lesion being treated. While the temperature in the target region for the ablation may well exceed 60 C, and there is little need to measure its actual level with great accuracy, it is important that temperatures of up to perhaps 45 C be observed with some precision. If precedents from low-temperature rf hyperthermia are applied (in which considerable care was exercised to avoid collateral damage) then the targets for measurement are:1 temperature accuracy of ‹0.5 C, volume of interest 5 mm3, and temporal resolution of 30 s. Temporal resolution is desirably substantially better in therapies such as focused ultrasound, where signi®cant quantities of energy can be deposited in a small volume very quickly, with consequently swift changes in temperature. Spatial resolution cannot be allowed to deteriorate signi®cantly as tissue can support quite considerable temperature gradients (certainly in the range of 5 C cmÿ1 or more).

…1†

where  is the macroscopic viscosity, and T the absolute temperature. Over the limited temperature excursions of tissue temperature for which real accuracy is important (see above), there is effectively a linear relationship between T1 and T. Energy level considerations predict a temperature coef®cient of around 1.3%  ÿ1 C for sensitivity of the measurement of T1 for water;2 because this approach to the measurement has been perceived as being the simplest of those that are available, it was the ®rst approach to be investigated.3,4 The usual approach that has been adopted4 is to acquire a short TE spin echo, or gradient recalled echo, image with long TR for use as a reference (data set A), then follow this with short TR data acquisitions using the same TE as before whenever needed (data set B). From this T1 is calculated using the relationship: SB 1 ÿ exp…ÿTRB =T1B † ˆ SA 1 ÿ exp…ÿTRA =T1A †

…2†

when SA and SB are the respective signals from the two data sets; T1A and T1B are the T1 values at the times of the A and B acquisitions; and TRA and TRB are the respective TR values. The methods has been successfully evaluated many times in phantoms and in vitro.5±7 However, it is clear that the relationship in Equation (2) is defective in at least one respect ± it assumes that the intrinsic signal is independent of temperature, which, via the Curie relationship, it is not:8 Mˆ

2 TARGETS OF PERFORMANCE FOR IN VIVO TEMPERATURE MEASUREMENT

1

N 2  h2 I…I ‡ 1† Ho ˆ o Ho 3kT

…3†

where M is the magnetization arising from an applied ®eld Ho, o is the nuclear susceptibility, N Avogadro's number; , h and k are, respectively, the gyromagnetic ratio, Planck and Boltzman constants; I is the spin quantum number. More exactly, Equation (2) should include an extra ratio on its right-hand side (MB/MA) where MA and MB are the respective tissue magnetizations; this effectively reduces to (TB/TA) using Equation (3), and allowing for the possibility that T2 is also temperature sensitive, though it seems to be only weakly so in near-normal tissue;9 consequently, the ratio in Equation (2) becomes SA TA …1 ÿ exp…ÿTR1B =T1B ††exp…ÿTE=T2B † : ˆ SB TB …1 ÿ exp…ÿTR1A =T1A ††exp…ÿTE=T2A †

…4†

The problems of obtaining an accurate measurement are compounded by other changes that are likely to take place during actual therapy. For example, when tissue is cooled10 or heated during thermal ablation,11 it can swell, and this can affect the measurements still further. As noted by Young et

2 TEMPERATURE MEASUREMENT IN VIVO USING NMR al.,12 the type of relationship that can be required to describe the changes with real precision (even if a two-component system (p, r subscripts) only is used) can become very complex, as illustrated by Equation (5).

possibility of there being other components) is tractable. The third curve (proton density) reveals what is happening. It is a plot obtained from the long TR data only, which shows that this is changing at around 1%  Cÿ1, indicating the possible for-

n

        o ÿTRB TRB TE TE x …1 ÿ exp † 1 ÿ exp ÿ exp ÿ ÿ …1 ÿ x exp ÿ B B T2Bp T2Br T1Bp T1Br SB         o ˆ TA =TB n  ÿTRA TRA TE SA xA 1 ÿ exp T1Ap exp ÿ T2Ap ÿ …1 ÿ xA † 1 ÿ exp ÿ TaAr exp ÿ TTE 2Ar

mation of edema and/or increased tissue perfusion. Equation (2) does not then re¯ect what is happening, and the very low coef®cient of T1 dependence on temperature obtained using it (and plotted as T1(N)) is not a true value. The value of T2 was assumed to be unaffected by temperature in this work. Nevertheless, the changes associated with variation in T1 are easy to demonstrate using TurboFLASH or similar sequences,5,6 which are available on most scanners. These are probably capable, for example with methods in which the local temperature can be changed very rapidly, such as focal ultrasound,13 of a measurement accuracy in the range 3±5 C. As such, they may well not be capable of providing useful data to help to avoid collateral damage, though it will be possible to use them to make sure the temperature of a target region has exceeded a lethal level for cells present there.

4

TEMPERATURE DEPENDENCE OF T2

The possibility of T2 being the basis of a temperature scale has been previously noted, through largely discounted. It does not seem to have a useful dependence in vivo in near normal tissue9 and has been little investigated as such. There is, however, one circumstance when it could be very signi®cant. This

350

300

330

T1 (ms)

320

280

Proton density (arbitrary scale)

where x is the effective NMR visible fraction of component p present ± which is, of course, temperature sensitive itself. Note that x is, using this de®nition, not, in general, the actual physical fraction of p that is present and so cannot be assessed by purely analytical methods. The potential impact of temperature changes on the fraction of the hydrogen nuclei in a component that actually contribute usefully to the NMR signal is another matter altogether. Clearly such relationships contain so many variables that can only be known with low, if any, precision at all that T1 cannot provide an accurate temperature scale within the requirements noted in Section 2. Figure 1 gives some measure of the impact of these complexities on the measurement temperature in human calf muscle, in an experiment in which the tissue was cooled using water circulated through bags on either side of the leg.12 The curve T1(R) was plotted from the data set assuming the relationship in Equation (2) was valid, and that the reference measurement (A) was made initially and not checked thereafter. The resultant coef®cient of T1 versus temperature is rather larger (at about 1.6%  Cÿ1) than would be expected. The second T1 set of data (T1(N)) was plotted using Equation (2) again, but, using long and short TR data taken at the same temperature; consequently, to an adequate extent TA=TB, and the relationship (ignoring the

…5†

310 30

31

32

33 34 35 Temperature (ßC)

36

37

38

Figure 1 Temperature and proton density from a region of calf muscle undergoing temperature stress in a volunteer. The curves ®tted to the points are T1(R) (~), T1(N) (*) (both left-hand scale), and proton density curve (&). (Reproduced with permission from Young et al., Magn. Reson. Med., 1994, 32, 366.)

TEMPERATURE MEASUREMENT IN VIVO USING NMR

is in cryoblation, where its application has been extensively studied.14 The problem with this approach is that it is not suf®cient just to freeze the tissue in order to damage it; the temperature must be reduced substantially below 0 C. Conventionally, it would be assumed that once the tissue had solidi®ed no further useful MRI signals would be obtained, and any attempts at tracking further temperature changes would need to be done by computer modeling.15 However Daniel et al. showed how it is possible to use variations in T2 of the unfrozen water fraction in the tissue as a means of monitoring temperature.14 5 TEMPERATURE DEPENDENCE OF THE DIFFUSION COEFFICIENT Le Bihan et al. were the ®rst to propose the temperature dependence of the diffusion coef®cient as a viable in vivo scale, pointing out that its theoretical sensitivity is usefully larger (at 2.2%  Cÿ1) than that of T1.2 The relationships they exploited are simple. Data from the sequences (A and B) with differing degrees of diffusion weighting (b) are used. If, as before, the signals for these are SA and SB respectively, and the diffusion weightings are, respectively, bA and bB then, in more complete form:      TR TE SB MB 1 ÿ exp ÿ T1B exp ÿ T2B ÿ bB D…TB †   …6†    ˆ SA MA 1 ÿ exp ÿ TR exp ÿ TE ÿ bA D…TB † T1A

T2A

where D(T) is the diffusion coef®cient at the temperature at which the measurements were acquired. From this, b is given by:  
ÿ …7† b ˆ 2 G2  2 3 where is the gyromagnetic ratio, G is the gradient strength in each of two identical pulses of effective duration  and having their leading edges
apart in time, using the classic pulsed gradient spin echo (PGSE) experiment of Stejskal and Tanner.16 If both data acquisitions are made at substantially the same temperature (as described in Cline et al.13) then Equation (6) can be simpli®ed to: SB ˆ exp…D…T †…bA ÿ bB †† SA

…8†

In vivo applications of the use of diffusion weighting to monitor temperature are few and far between,9,17 though the implementation now of echo planar imaging (EPI)18 in many machines may change that. Diffusion experiments are notoriously vulnerable to motion artifact, and rapid acquisition of data is highly desirable. As with T1, phantom and in vitro data yield very good calibrations of D versus T. Unfortunately, it is probable that some of the problems noted in connection with T1 measurement will also affect diffusion-weighted studies. Diffusion in tissue is anisotropic;19 as a result, the changes in edema and/or perfusion already noted will cause variations in the observed value of D that are not necessarily related to temperature at all. It is possible that calculating the full diffusion tensor20 may allow variations caused

3

by temperature change to be sorted out from those resulting from tissue structure alterations, but this has yet to be demonstrated, and, at best, will require more images (12, in two sets of 6 at least) to be obtained at each temperature.

6

TEMPERATURE DEPENDENCE OF THE CHEMICAL SHIFT OF WATER

Hindman ®rst demonstrated, and characterized, the chemical shift of water as a means of measuring its temperature.21 He suggested that the small effect he observed (of the order of ÿ0.0108 ppm  Cÿ1) was the result of temperature dependence of the shielding constant of the electrons in the water molecule. He based his analysis on the relationship: i ˆ o ‡ d ‡ a ‡ F

…9†

where i is the total shielding constant, o is the intramolecular shielding constant for an atom in an isolated molecule, d is the bulk diamagnetic susceptibility component, a is shielding caused by anisotropy in the susceptibility of neighboring molecules, and F is the contribution to the shielding by local electric ®eld effects. Some of these component contributions are temperature sensitive.21 This phenomenon was ignored in human MR terms until Ishihara et al. revived interest in it.22 Since then it has been exploited in two quite different ways: through imaging and through spectroscopy. These will be discussed separately below. 6.1

Approaches Using MRI to Exploit the Temperature Dependence of the Chemical Shift of Water

The method Ishirara et al. proposed involved observing the phase differences in the image data as the temperature varied.22 The experiment is a gradient recalled echo (GRE) system in which TE (the echo time) is extended to obtain the necessary sensitivity for the desired temperature scale. If !(T) is the temperature-dependent resonant frequency of the water, the phase difference developed between the signals at two different temperatures, (
'T), assuming nothing else is changed, is:
'T ˆ …!TB ÿ TA †TE ˆ k…TB ÿ TA †

…10†

where k is the phase coef®cient of temperature for the sequence and ®eld used. If the aim is a measurement accuracy of 0.5 C, then the desirable phase shift sensitivity should be no less than perhaps 30 C per 360 of phase if the phase noise is in the range 3±4 . Even in a 1.5 T magnet (remembering that != Bo) this means that TE has to be around 48 ms. The practicality of the method has been demonstrated at a variety of ®elds, including as low as 0.2 T,23 but it is, at the least, a technique that makes substantial demands on machine performance. A major issue is the possibility of susceptibility effects, which can arise from the tissue itself or as artifacts from neighboring objects.24 A very small susceptibility difference indeed is suf®cient to cause substantial errors. If
is an effective susceptibility change that develops, then the phase error caused by it is:

4 TEMPERATURE MEASUREMENT IN VIVO USING NMR
's ˆ
Bo TE

…11†

De Poorter et al. were concerned about the nature of possible chemical shift-like changes in lipid-rich tissue as well as in water-based ones,25 but, in practice, effects in fatty tissues seem to be minimal and, indeed, as will be described in the next section, can even be used as a reference. Since the effects sought are so small, correction for various machine drifts is normally essential.26 This can be done by a variety of methods but involves at least one means of monitoring any drift in the Bo ®eld. This can be done by phantoms, the temperature of which is controlled, or, in principle, by other methods. Figure 2 is a phase map typical of those obtained during temperature measurement experiments. The image shows a volunteer's calf muscle during cooling by water pads, which can be seen on either side of the leg. These appear black since, in the image, calculated from an initial data set obtained prior to cooling and one taken as the leg temperature was dropped, the

temperature of the water in the pads is much lower than it was at the start of the experiment. The phantoms used to enable corrections for ®eld drift to be made are apparent round the leg. They are insulated from the cooling pads, and the temperatures of two of them were monitored. The fat in the tissue retains the mid-gray color indicative of no phase change, while the muscle near the pads inside it is colder than it was at the start. The data in Figure 2 were generated using a protocol involving stabilization of the leg temperature at around 37 C for 30±40 min followed by a rapid reduction in the temperature demand for the water in the pads to 5 C. This command was followed by the water temperature with some delay; consequently when the temperature was reset to 37 C again, typically after 1 h, it had reached a minimum temperature of perhaps 10 C. After resetting, the leg was monitored for a further 40 min. Optical probes on the phantom, on the surface of the skin, and implanted in the muscle under the pads to a depth of 15 mm were used to monitor temperature (Luxtron 3000, Luxtron Inc., Santa Clara, CA, USA).

Figure 2 Image of a phase map of a leg of a volunteer during cooling. The phantoms and fat retain the mid-gray tone indicative of a very small change in phase, while the water pads (seen outside the leg) are very black, as the water is very cold. The muscle temperature has been reduced to some degree. (Reproduced with permission from Young et al., Magn. Reson. Med., 1994, 36, 369.)

TEMPERATURE MEASUREMENT IN VIVO USING NMR

5

(b)

(a)

130

120

0

Proton density (arbitary units)

Chemical shift phase (º)

20

–20

–40

110

100

–60 34

36

38

40

42

90 34

36

38

Temperature (ºC)

40

42

Temperature (ºC)

Figure 3 The effect of temperature cycling in a volunteer's calf muscle. Data were obtained from directly under a cooling pad, in the muscle (&) and at the top of the leg in muscle but not directly cooled (*). (a) Chemical shift of water versus temperature; note the hysteresis observable in these data. (b) Proton density values at changing temperature. (Reproduced with permission from Young et al., Magn. Reson. Med., 1994, 36, 372.)

This temperature cycling produced results of which those shown in Figure 3 are typical. Data from two points are shown: one under one of the pads, and the other at the top of the leg away from any direct cooling. Two pairs of curves are shown, those in Figure 3a being the chemical shift data, while Figure 3b shows proton density data analogous to that shown in Figure 1. It is apparent that the data from the point under the pads both show an unexpectedly large change in phase (as the likely expected change at 1 T with a TE of 60 ms was about 9.3 phase  Cÿ1) and also reveals marked hysteresis. The proton density data from the same region also shows substantial changes, and hysteresis. Both sets of data from the other point are much more convincing. Because of suspicions about possible changes in the blood content of the tissue as the body responded to the cooling, it was thought that such changes could have been caused by susceptibility differences,26 though later data (see below) suggests this may well not be correct. The hysteresis effects in the chemical shift phase data are not easy to explain if this factor is excluded; however it seems likely that it will involve changes in the lipid content of the voxels owing to physical movements and composition changes arising as the tissue swells. Since the chemical shift of lipids is around 3.5 ppm relative to water, ®rst sight suggests that small alterations only are needed to explain errors in a measurement where the scale is ÿ0.0108 ppm  Cÿ1. However, though the scale differences are dramatic, the worst case occurs when the lipid and water components are out

of phase by 90 or 270 C, and the many Hertz differences between the two resonant frequencies cannot cause larger errors than those angular differences allow. If the fraction of lipid components in a voxel is initially A and ultimately B at the same time as the temperature changed from TA to TB, at which values the observed phase differences (relative to a starting reference value) were 'A and 'B and TE was such that the lipid component was initially 90 out of phase with the water signal, then ' A ÿ 'B ˆ

…1 ÿ B †sin kTB ‡ B sin  …1 ÿ B †cos kTA ‡ B sin  ÿ

…1 ÿ A †sin kTA ‡ A sin  …1 ÿ A †cos kTA ‡ A sin 

…12†

which simpli®es to: 'A ÿ 'B ˆ …k…TB ÿ TA † ‡

 B ÿ A …1 ÿ B †…1 ÿ A †

…13†

if it is assumed that kTA and the  values are small;  is the phase angle difference between the two components determined by the sequence. Equation (13) suggests that the involvement of lipid changes cannot be the only, or even, it is likely, the dominant cause of the calibration errors observed. The fractional change in the

6 TEMPERATURE MEASUREMENT IN VIVO USING NMR value of  would need to be too large for this, for example a 1% change in  (from 4 to 5%) would result in a change of perhaps 0.6 of phase. At this time, the phenomena seen in Figure 3a have no convincing explanation. While the true accuracy of the chemical shift method in vivo may not be as good as the many phantom experiments using it have suggested, nevertheless it is a reliable indicator of larger temperature changes and is being studied by a number of groups to this end.27,28 6.2 Chemical Shift Temperature Measurements Using MRS Cady et al. were the ®rst to suggest using the chemical shift of water to measure temperature in vivo using spectroscopic rather than imaging methods.29 They exploited earlier in vitro data which showed that many metabolites other than water had no temperature coef®cient of chemical shift;30 as a result, their signals could be used as internal references in human studies. If there was to be susceptibility, or small main ®eld alterations, this would affect the reference moieties as well as the water, and any differences in the chemical shift of the latter could be readily assigned to temperature changes. While Cady et al. were interested in measuring the temperature of the neonatal brain during hypothermia to stabilize infants who had been seriously brain damaged at birth,29 the data to be described here were obtained from volunteer calf muscle using the same kind of protocol as described previously.31 Typical examples of the data obtained from the spectroscopic studies are shown in Figure 4. Figure 4a shows a typical spectrum and illustrates some of the problems with the

method. Since the water peak is essential for the measurement and since its phase and chemical shift are key issues, it cannot be suppressed, and TE has to be long enough to allow it to be very substantially attenuated relative to the signals of any long T2 moieties that might act as a reference (or source of comparison). This attenuation has to be suf®cient for the dynamic range of the data to be les than that of the receiver system in the spectrometer, with inevitable attenuation of other signals and a relatively poor signal-to-noise ratio. In spite of these problems, performance is good enough for the results shown in Figure 4b to be obtained. This shows data from several other moieties, creatine, choline and lipid (a number of partly resolved peaks treated as a single entity), as well as water. Analysis by Young et al. shows that these components are really very stable, suggesting that susceptibility changes were not a factor in this instance and that the impact of tissue swelling, which occurred as usual in the experiment from which these data were obtained, was minimal using this strategy.31 The data shown in Figure 4 were derived from a voxel 15 mm3 in size, with a temporal resolution of 7±8 min. In both these respects, it fails the criteria discussed in Section 2. Detailed analysis of imaging data overlapping the spectroscopic region of interest showed the same type of pattern that is illustrated in Figure 3. Spectroscopy appears, on the basis of relatively skimpy data,29,31,32 to have robustness for use as a thermal scale that is lacking in the imaging data. At least partly this is because of the large voxel (with the much greater spatial averaging inherent in its use). It is interesting to speculate on how, and if, the two approaches can be used to support each other to produce data that both are accurate and have adequate temporal and spatial resolutions.

(b) 5 Chemical shift (Hz)

4 3 2 1 0 —1 —2 —3 0

40

80

120

160

Time (min)

Figure 4 Magnetic resonance spectroscopy of temperature changes in a volunteer's calf muscle. (a) Spectrum obtained in the study. The small amplitude of the creatine and choline peaks relative to the water and lipid one is very obvious. (b) Plots of the chemical shift of a number of moieties against temperature. The continuous line is the chemical shift that might be expected from modeling the temperatures to be expected in the tissue using the thermoheat Equation (15), then assuming that the temperature coef®cient of chemical shift of muscle is the same as that of water. The chemical shift change for water (*) is derived from the data. The minimal chemical shift of the creatine (&), choline (&) and lipid (Y) peaks versus temperature is very apparent. (Reproduced with permission from Young et al., JMRI, 1998, 8, 1116, 1117.)

TEMPERATURE MEASUREMENT IN VIVO USING NMR

7

clear that though substantial image processing can improve matters, it does not transform the situation completely.

VARIATIONS IN T1 OF NUCLEI OTHER THAN HYDROGEN

Many components, such as the chemical shift of methyl groups in complex molecules including paramagnetic nuclei, vary with temperature, and suggestions have been made as to their utility for temperature measurement.33,34 The ¯uorocarbons also have properties that can be exploited,35 though, in this case, the T1s of the ¯uorocarbon lines are also pH sensitive. The relationship between the various quantities is of the form: 1=T1 ˆ A ‡ B…pH† ‡ CT ‡ D…pH†T

9 CONCLUSION Temperature measurement in vivo using MRI and MRS has become a matter of considerable interest to the scienti®c community, particularly since the usefulness of interventional MRI has come to be appreciated. It seems unlikely that it will reach the performance target suggested in Section 2 of this article. However, if the temperature target were to be reduced so that the accuracy were to be ‹3 C, then it is likely that MRI will become of real value as a temperature measurement. This will not allow as much appreciation of the risks, and extent, of collateral damage to normal tissue round a target lesion as is desirable. If this is acceptable clinically, then we can look forward to the increasingly widespread application of direct temperature measurement in interventional MRI.

…14†

where A, B, C, and D are constants for any one spectral line; T is the absolute temperature, and (pH) is the tissue pH. As described by Mason et al., from a spectroscopy experiment, the behavior of the multiple lines can allow the solution of a set of simultaneous equations and resolve the differences between changes caused by temperature and pH.35 As yet none of the methods suggested has been used in vivo.

8

7

10 RELATED ARTICLES Relaxation Measurements in Whole Body MRI: Clinical Utility; Thermal Therapies in the Body Monitored by MRI.

IMAGE PROCESSING

A problem with temperature measurement in vivo, which distinguishes it from any phantom experiment, is that tissue is mobile and responds to thermal stress by a range of potential changes. Without identifying its source, cooling the leg, for example, results in swelling of the tissue. Then, if no action is taken, the tissue content of a region of interest in a voxel determined by the machine gradients, and not adjusted in any way, will change continuously. It is not surprising, therefore, that content variations can result in substantial errors in small voxels, with relatively low signal-to-noise ratio, and be less apparent in the much larger voxels used in spectroscopy. The extent of the effects is demonstrated in Table 1 (line 1) and the results of the efforts to improve the situation using registration are described by Hajnal et al.,36 while data correction using other strategies is described by Saeed et al.37 The latter includes attempts to track the mean intensity of the voxel by selecting a closest local match to the standard deviation of the original gray-scale voxel. Another variant combined that approach with a best match to distances from a number of landmarks surrounding it. All the data in Table 1 are from a T1 experiment, and it is

11 REFERENCES 1. International Collaborative Hyperthermia Group, Int. J. Radiat. Oncol. Biol. Phys., 1996, 35, 731. 2. D. LeBihan, J. Delannoy, and R. L. Levin, Radiology, 1989, 171, 853. 3. D. L. Parker, V. Smith, P. Sheldon, L. E. Crooks, and L. Fusell, Med. Phys., 1983, 10, 321. 4. R. J. Dickinson, A. S. Hall, A. J. Hind, and I. R. Young, J. Comput. Assist. Tomogr., 1986, 10, 468. 5. D. L. Parker, IEEE Trans. Biomed. Eng., 1984, BME-31, 161. 6. C. J. Hardy, H. E. Cline, and R. D. Watkins, J. Comput. Assist. Tomgr., 1994, 18, 476. 7. K. Hynynen, A. Darzazanli, E. C. Unger, and J. F. Schenck, Med. Phys., 1993, 20, 107. 8. A. Abragam, `The Principles of Nuclear Magnetism', Clarendon Press, Oxford, 1961, pp. 2±53, 289±305, 566±569. 9. A. S. Hall, M. V. Prior, J. W. Hand, I. R. Young, and R. J. Dickinson, J. Comput. Assist. Tomogr., 1990, 14, 430. 10. I. R. Young, J. W. Hand, A. Oatridge, M. V. Prior, N. Saeed, and G. R. Forse, Magn. Reson. Med., 1994, 31, 342.

Table 1 Changes in tissues as a result of thermal stress: the level of areas in temperature measurement that is observed in a typical in vivo measurement of T1 using the temperature protocol described in the text and in Hajnal et al.36 Tracking/registration method

No tracking at all Using registration Using gray scale matching only Using gray and landmark tracking a

Region of interest numbera A

B

C

D

E

F

20.0 6.9 5.6 5.6

1.7 2.8 11.7 5.5

13.5 6.7 16.6 16.6

4.8 5.2 5.1 4.2

9.3 4.6 4.6 2.7

10.6 4.3 6.4 6.1

Values for typical regions of interest for the equivalent temperature spread using a temperature coef®cient of 1.3%  Cÿ1. The maximum true excursion of the temperature was less than 5 C.

8 TEMPERATURE MEASUREMENT IN VIVO USING NMR 11. N. M. deSouza, R. Flynn, G. A. Coutts, D. J. Gilderdale, A. S. Hall, R. Puni, M. Chui, D. N. F. Harris, and E. A. Keily, Am. J. Roentgenol., 1995, 164, 1429. 12. I. R. Young, J. W. Hand, A. Oatridge, and M. V. Prior, Magn. Reson. Med., 1994, 32, 358. 13. H. E. Cline, J. F. Schenck, R. D. Watkins, K. Hynynen, and F. A. Jolesz, Magn. Reson. Med., 1993, 30, 98. 14. B. L. Daniel, K. Butts, and W. F. Block, Proc. VIth Annu. Mtg. (Int.) Soc. Magn. Reson. Med., Sydney, 1998, p. 353. 15. H. H. Pennes, J. Appl. Physiol., 1948, 1, 93. 16. E. O. Stejskal and J. E. Tanner, J. Chem. Phys., 1965, 42, 288. 17. D. Morvan, A. Leroy-Willig, A. Malgouyres, C. A. Ceunod, and P. Jehenson, Magn. Reson. Med., 1993, 29, 371. 18. P. Mans®eld, J. Phys. C., Solid State Phys., 1977, 10, L55. 19. M. E. Moseley, J. Kucharczyk, H. S. Asgari, and D. Norman, Magn. Reson. Med., 1991, 19, 321. 20. S. Mori and P. C. M. van Zijl, Magn. Reson. Med., 1995, 33, 41. 21. J. D. Hindman, J. Chem. Phys., 1966, 44, 4582. 22. Y. Ishihara, A. Calderon, H. Watanabe, K. Okamoto, Y. Suzuki, K. Kuroda, and Y. Suzuke, Magn. Reson. Med., 1995, 34, 814. 23. S. Sinha, T. Oshiro, U. Sinha, and R. Lufkin, JMRI, 1997, 7, 918. 24. J. de Poorter Proc. IInd Annu. Mkg. (Int.) Soc. Magn. Reson. Med., San Francisco, 1994, p. 426. 25. J. de Poorter, C. de Wagter, Y. de Deene, C. Thomsen, F. StaÊhlberg, and E. Achten, Magn. Reson. Med., 1995, 33, 74. 26. I. R. Young, J. V. Hajnal, I. G. Roberts, J. X. Ling, R. J. HillCottingham, A. Oatridge, and J. A. Wilson, Magn. Reson. Med., 1996, 36, 366. 27. H. E. Cline, K. Hynynen, E. Schneider, C. J. Hardy, S. E. Maier, R. D. Watkins, and F. A. Jolesz, Magn. Reson. Med., 1996, 35, 309. 28. T. Harth, T. Kahn, M. Rassek, B. Schwabe, H.-J. Schwarzmaier, J. S. Lewin, and U. MoÈdder, Magn. Reson. Med., 1998, 38, 238. 29. E. B. Cady, P. C. D'Souza, J. Penrice, and A. Lorek, Magn. Reson. Med., 1995, 33, 862.

30. G. Aras, Y.-C. Chang, and M. Barany, J. Magn. Reson., 1985, 63, 369. 31. I. R. Young, J. D. Bell, J. V. Hajnal, G. Jenkinson, and J. Ling, JMRI, 1998, 8, 1114. 32. R. J. T. Corbett, A. R. Laptook, and P. Weatherall, Proc. Vth Annu. Mtg. (Int.) Soc. Magn. Reson. Med., Vancouver, 1997, p. 403. 33. T. Frenzel, K. Roth, S. Kossler, B. Raduchel, H. Bauer, J. Platzek, and H. J. Weinmann, Magn. Reson. Med., 1996, 35, 364. 34. S. Aime, M. Botta, M. Fasano, E. Terreno, P. Kinschesh, L. Calabi, and L. Paleari, Magn. Reson. Med., 1996, 35, 648. 35. R. P. Mason, F. M. Jeffrey, C. R. Malloy, E. E. Babcock, and P. P. Antich, Magn. Reson. Med., 1992, 27, 310. 36. J. V. Hajnal, N. Saeed, A. Oatridge, E. J. Soar, I. R. Young, and G. M. Bydder, J. Comput. Assist. Tomogr., 1995, 19, 289. 37. N. Saeed, J. V. Hajnal, A. Oatridge, and I. R. Young, JMRI, 1998, 8, 182.

Acknowledgements The author would like to acknowledge the help and support of his colleagues and the Robert Steiner MRI Unit, Imperial College School of Medicine, Hammersmith Hospital, London, in the work which comprises the basis of much of the material discussed in this article.

Biographical Sketch Ian R. Young. b 1932. B.Sc., 1954, Ph.D., 1988, Aberdeen. No involvement with NMR until joining EMI in 1976, where charged with leading the engineering group developing ®rst a resistive, then a cryogenic machine. This was installed at Hammersmith Hospital in 1980±1. Approx. 150 papers and 30 patents in whole body NMR and related aspects. Research interests are in obtaining better images and spectra in vivo.

Therapy Monitoring by MRI D. R. Hamilton, Y. Anzai, K. L. Black, K. Farahani and R. B. Lufkin UCL School of Medicine, Los Angeles, CA, USA

1 2 3 4 5 6 7

Introduction Basic Principles of MR Therapy Monitoring Therapeutic Techniques and Clinical Applications Specialized Equipment Future Directions Related Articles References

1

INTRODUCTION

1 1 3 4 5 6 6

In its short history, MRI has been championed over Xray techniques as a noninvasive, noninterventional, diagnostic tool that provides exquisite anatomical detail and soft tissue contrast. Ironically, many of the advantages contributing to the diagnostic success of MRI are now being used to expand the role of MRI into the interventional or therapeutic arena (Figure 1). 1.1

Visualization

Much of surgical technique has been developed to provide exposure or visualization of the surgical field. The actual treatment is often a small portion of the entire procedure compared with the surgical steps necessary for visualization. MRI has the ability to acquire high-contrast, submillimeter resolution detail in two or three dimensions in any orientation. This feature has been used since the mid 1980s to locate and guide biopsies of deep target tissues with limited accessibility. Stereotactic frames with MR visible markers have traditionally been used to register the target location on the MR images with an external reference. The MR image not only provides an accurate location of the target but also demonstrates the relationship of the target to surrounding vital structures including nerves and vessels using MRA techniques. This information is essential to planning the least invasive trajectory for a biopsy or therapeutic device. Although large-scale clinical acceptance of MR-guided biopsy has been slow due to the high cost and inaccessibility of conventional MR units, the reported clinical applications are growing and now include head and neck, brain, prostate, liver, and mediastinum.1 – 7 One new area that is expected to become a large application for MR-guided procedures is the evaluation of breast lesions detected by MR that are invisible on mammography. Ultrasound and computerized tomography (CT) have been used in the past to guide biopsy procedures. MR provides soft tissue detail superior to CT and higher spatial resolution than ultrasound. The impact of MR guidance on surgical procedures is to reduce both the time required for visualization and the potential for related complications, ultimately improving patient management.

Figure 1 T 2 -weighted fast spin echo image demonstrating MRcompatible rf therapy probe placed on target in patient with metastatic brain tumor

1.2 Therapy Monitoring

While facilitated visualization is a key aspect of improving an interventional procedure, MR is uniquely qualified to participate in the therapeutic process itself. Once the probe has been placed on target and the therapy administered, MR imaging can demonstrate thermal changes as well as tissue characteristics secondary to the treatment. Sequential MR acquisition, coupled with image subtraction and display in the examination room, provide an environment in which the therapy session becomes an iterative process. Projects are currently underway to develop accurate and reproducible methods of thermal dosimetry using MR, providing the physician with feedback on therapeutic effect during the procedure.8

2 BASIC PRINCIPLES OF MR THERAPY MONITORING

Any imaging technique, by definition, makes use of physical differences between neighboring tissues to create a visual map of tissue interfaces (e.g. density in CT, acoustic impedance in ultrasound, uptake of radioisotope in nuclear medicine, etc.). The MRI signal depends on the structure and dynamics of water–macromolecular interactions. Since these interactions are affected by energy absorption, the MRI technique provides a means of visualizing therapeutic administration. 2.1 T 1 Temperature Monitoring

As a tissue is heated, the distribution and mobility of water and lipid molecules change. These changes alter the tissue relaxation times, T 1 and T 2 . T 1 dependence on temperature can be exploited to detect temperature changes during and after therapy using MRI9 (Figures 2 and 3).

2 THERAPY MONITORING BY MRI

Figure 2 Series of time-lapsed T 1 -weighted MR images acquired before (a), every 25 s during (b)–(j), and after (k)–(p) laser irradiation of sheep brain in vitro. Open arrow designates tip of laser fiber optic and bold arrow shows temperature probe 1.5 cm from laser

2.2

T 2 Edema Monitoring

In a typical thermal ablation procedure, the tissue temperature is raised to 60–80 ◦ C in order to produce protein denaturation and coagulation necrosis. Tissue changes associated with these processes are detectable using T 2 -weighted MRI. Basic animal studies have characterized the acute tissue effects of laser heating, for example, and their MR appearances.10 In addition to a high sensitivity to edema, T 2 -weighted images reveal a clear-cut MR appearance of thermal-induced lesions: concentric rings which closely reflect the actual pathologic changes of central cavitation, coagulation necrosis, and tissue edema (Figures 4 and 5).

Figure 3 Relationship between temperature (T ) and T 1 signal intensity (S ) during and after laser irradiation

Figure 4 T 2 -weighted images of a laser-induced lesion in an animal model immediately after laser exposure (a) and up to 40 min later (f). Note the expansion of the lesion (arrow)

Although the regions of cavitation and coagulation necrosis clearly represent irreversible cell damage and death, the ultimate disposition of the edema layer is important in the quantification of the tissue damage. Preliminary data from chronic studies in animal models suggest that the area of cell death may expand up to 44% into the edema region depending on a number of factors as the healing process occurs.11

2.3 Position Monitoring

While most of the new developments in interventional MR involve therapy monitoring, there is also a need for

Figure 5 Gross pathology of the laser-induced lesion showing concentric regions of cavity, coagulation necrosis, and edema

THERAPY MONITORING BY MRI

3

3.2 Thermal Ablation

Manipulation of temperature can cause tissue destruction either by freezing and cell lysis at low temperature or protein denaturation at high temperature. Hyperthermia, tissue heating to temperatures in the range of 43–45 ◦ C, can sustain sublethal injury that becomes irreversible after extended exposure, depending on the type of tissue and experimental or therapeutic conditions. Thermal ablation at higher temperatures (60–90 ◦ C) is more useful for MR-guided therapeutic purposes because it is quicker to deliver. Unlike alcohol injection, which follows fascial planes, thermal ablation results in an ellipsoidal area of tissue destruction, depending on local thermal conductivity. Neighboring vessels, for example, can act as heat sinks, decreasing the heating effect on surrounding tissues. There are two main techniques currently being used clinically for administering thermal interstitial therapy: laser ablation and radiofrequency ablation. In addition, a new technique using focused ultrasound is currently under investigation.

Figure 6 Simulation of real-time position tracking of catheter using an MR angiography road-map, real-time MR scanning and an rf-based tracking system

real-time position tracking of therapeutic instruments. Techniques are under development that will allow tracking of invasive devices (e.g. catheter) in real time using magnetic resonance.12 By incorporating one or more small rf coils into a device, MR signals will be detected only from excited tissue in the immediate vicinity of the device. An MR pulse sequence using only read-out gradient pulses can localize the position of the receiver coil in the direction of the applied gradient. Repeating this technique in three orthogonal directions permits localization of the rf coil in three dimensions. Such a sequence can be repeated rapidly to provide position information in real time. This position information can be superimposed on a prospectively acquired anatomical ‘road-map’ (Figure 6). This technology may prove instrumental in making a field such as interventional MR angiography a reality.

3

THERAPEUTIC TECHNIQUES AND CLINICAL APPLICATIONS

Techniques of interstitial therapy can be separated into two mechanistic categories: chemical and thermal. 3.1

Chemoablation

Various strategies for chemoablation under imaging guidance have been used for years in angiography and other areas of classical (endovascular) interventional radiology. Many of these approaches will have applications interstitially as well. For example, injection of ethanol into tissues in sufficiently high concentration will result in local dehydration necrosis. Animal studies confirm the ability of MR to demonstrate lesion size and shape.13 Initial human studies have been reported in liver applications.14

3.2.1 Laser Ablation

This technique involves a fiber-optic-laser delivery system such as a neodymium–yttrium-aluminum-garnet (Nd:YAG) laser placed interstitially. The technique has been used in the past for treatment of hepatic tumors and head and neck lesions.15,16 Since the absorption coefficient of laser energy is variable, depending on the tissue characteristics, the prediction of tissue volume that will be coagulated with a given power output or exposure time setting is difficult in the clinical setting without a monitoring technique such as MRI. Laser therapy is well suited for MR guidance since fiber optic probes are MR-compatible. Laser-induced lesions show the characteristic zonal segmentation described earlier: cavitation, coagulative necrosis, and edema. In a study of laser lesions in muscle tissue of an animal model, the area of coagulative necrosis resulting from laser irradiation increased in size up to 7 days posttreatment. Progressive resolution resulted in absorption of injured tissue and subsequent replacement with fibrous scar tissue.11 MRI findings correlated well with pathologic findings in this chronic model. 3.2.2 Radiofrequency Ablation

A radiofrequency probe may also be used to produce tissue destruction by means of remote heating. Applications of this technique using MR monitoring were made possible with the development of MR-compatible rf-generating systems and probes.17 By defining rf dose based on temperature exposure in seconds (e.g. 80 ◦ C for 60 s), lesions created using radiofrequency ablation can be more accurately controlled. Using a thermocouple to monitor tissue temperature adjacent to the rf probe during ablation, MR images are acquired to assess acute tissue changes secondary to rf therapy. A clinical series involving treatment of primary and metastatic brain tumors using MR-guided rf ablation is currently underway with a 12 month follow-up completed18,19 (Figure 7). Patients are treated under MR guidance with local anesthetic, accessing the brain through a 2 mm-twist drill hole

4 THERAPY MONITORING BY MRI

Figure 7 Serial postcontrast MR images of a patient treated with MR-guided rf ablation of a metastatic adenocarcinoma to the brain. The images were acquired (a) immediately following rf ablation, (b) 1 week, (c) 1 month, and (d) 5 months posttreatment

instead of an open craniotomy (Figure 1). Since the patients are awake and communicating with the physician, neurological symptoms can be monitored during the entire procedure, a significant advantage over open craniotomy procedures requiring general anesthesia. The procedure may be done as part of a biopsy procedure and can be used in previously fully radiated patients and/or in instances of recurrence. These types of interventional MR procedures are ideal for patients who are elderly and/or have systemic diseases that preclude the option of general surgery. 3.2.3 Focused Ultrasound Ablation

New techniques in which focused high-power ultrasound beams are used to create tissue necrosis under MRI guidance are being investigated.20,21 This technique is capable of creating focal lesions in deep tissues which are clearly depicted by MRI, particularly using T 2 -weighted sequences (Figure 8). Although this approach has the distinct advantage of not requiring a needle, catheter, or any direct skin violation, it does require an ultrasonic path to the lesion (i.e. a path free of air and bone). Projects are underway to develop an MR-compatible, focused, ultrasound hyperthermia system to improve the clinical utility of this technique.

4 4.1

Figure 8 Focused ultrasound ablation. T 2 fast spin echo image showing lesions (arrows) induced in rabbit thigh muscle. These lesions were produced with the rabbit in the MR imager. (Courtesy of K. Hynynen et al., University of Arizona Health Sciences Center)

SPECIALIZED EQUIPMENT MR-Compatible Instruments

The first step toward practical interventional MRI was the development of satisfactory instruments that could be used in the high magnetic field necessary for clinical MRI. Even ‘low’-field MRI systems produce significant untoward effects on many standard interventional instruments.

Instruments made from ferromagnetic metals result in geometric distortions of the image causing large artifacts. A more dangerous effect of many of these metals is the tendency for them to experience torque and be accelerated in the magnetic field, thus risking damage to patients and imaging team personnel. By substituting high nickel content stainless steel or other nonferrous materials, researchers have developed new MRI-compatible instruments at reasonable costs.1,17 These devices are safer for the patient and imaging

THERAPY MONITORING BY MRI

5

Figure 9 (a) MR-compatible rf therapy probes (Radionics, Inc.), (b) susceptibility artifact caused by stainless steel biopsy needle, (c) reduced artifact from MR-compatible needle (E-Z-EM, Inc)

team personnel and result in reduced artifacts relative to standard instruments (Figure 9). 4.2

Open MR Systems

In 1993, the first of a new generation of MR scanners specifically designed for MR-guided interventional procedures was introduced (Figure 10). An interventional magnet must have open patient access, flexible patient positioning, and open rf coil designs that provide sufficient image quality for MR therapy monitoring without impeding therapeutic access. 4.3

Advanced Localization and Display Systems

Although very accurate and reliable, frame-base stereotactic techniques are uncomfortable and invasive. The development of powerful new frameless localization systems using articulated arms and optic sensors are enhancing the impact of MR guidance on minimally invasive surgical procedures, improving efficiency and patient acceptance (Figure 11). These devices obviate the need for stereotactic frames that fit inside rf coils, opening up more applications of interventional MR. The MRI scanning paradigm used over the last decade in which the technologist stands outside the scanner room and

Figure 11 Operating Arm SystemTM (Radionics/RSA, Inc.) allows registration of the surgical field to three-dimensional MR data. Ultimately, the physician will be able to define scan plane orientation and initiate acquisition interactively during the interventional procedure

obtains images on a console is currently being rethought with interventional MR. Rapid three-dimensional segmentation and rendering of patient image information is necessary for treatment planning and monitoring during the procedure. Image subtraction, dosimetry modeling, and calculation routines are necessary for proper treatment administration and should be displayed in the MR scan room. In addition, the need for interactive scan plan definition and sequence selection is required in the MR scan room to improve efficiency of the therapeutic procedure. New interventional MR workstations with powerful graphics capabilities are currently under development to address these issues.

5 FUTURE DIRECTIONS

Figure 10 An example of one of a new generation of MR instruments specifically designed for interventional procedures. This system is from Siemens and operates at 0.2 T

Interventional MR is clearly in its early stages of development. With dedicated interventional MR units now available and sophisticated workstations under development, the next step is to identify clear-cut clinical indications and begin large clinical trials to define the ultimate value of interventional MR. Even with current technology, it is possible to imagine dedicated MR therapy units for combined radiological and surgical approaches in what may be the archetype for operating rooms of the 21st century.

6 THERAPY MONITORING BY MRI 6

RELATED ARTICLES

Resistive and Permanent Magnets for Whole Body MRI; Temperature Measurement In Vivo Using NMR.

7

REFERENCES 1. R. B. Lufkin, L. Teresi, and W. Hanafee, Am. J. R., 1987, 149, 380. 2. G. Duckwiler, R. B. Lufkin, L. E. Teresi, E. Shukler, J. Dion, F. Vinuela, J. Benrson, and W. Hanafee, Radiology, 1989, 170, 519. 3. A. A. Saltiel, T. A. S. Matalon, and D. A. Turner, J. Interventional Radiol., 1990, 5 165. 4. S. D. Herman, A. C. Friedman, P. D. Radecki, and D. F. Caroline, Am. J. R., 1986, 146, 351. 5. D. G. Thomas, C. H. Davis, S. Ingram, J. S. Olney, G. M. Bydder, and I. R. Young, Am. J. N. R., 1986, 7, 161. 6. G. Duckwiler, R. B. Lufkin, and W. N. Hanafee, Radiol. Clin. North Am., 1989, 27, 255. 7. T. Trapp, R. B. Lufkin, E. Abemeyer, L. Layfield, W. Hanafee, and P. Ward, Laryngoscope., 1989, 99, 105. 8. Y. Anzai, R. B. Lufkin, D. Castro, R. E. Saxton, H. Fettermen, K. Farahani, L. J. Layfield, F. C. Jdesz, W. N. Hanafee, and D. J. Castro, Laryngoscope, 1991, 101, 755. 9. K. Farahani, Doctoral dissertation, UCLA Department of Radiological Sciences, 1992.

10. Y. Anzai, R. B. Lufkin, D. J. Castro, K. Farahani, B. A. Jabour, L. J. Layfield, R. Udkoff, and W. N. Hanafee, J. Magn. Reson. Imaging, 1991, 1, 553. 11. Y. Anzai, R. B. Lufkin, S. Hirschowitz, K. Farahani, and D. J. Castro, J. Magn. Reson. Imaging, 1992, 2, 671. 12. C. L. Dumoulin, S. P. Souza, and R. D. Darrow, J. Magn. Reson. Med., 1993, 29, 411. 13. Y. Anzai, R. B. Lufkin, D. J. Castro, et al. J. Comp. Radiol., (in press). 14. Y. Kubota, T. Nakano, T. Seki, S. Kitagawa, T. Rizuno, Y. Sameshima, T. Katoh, and Y. Tanara, Hepatogastroenterology, 1989, 36, 262. 15. D. Hashimoto, M. Takami, and Y. Idezuki, Gastroenterology, 1985, 88, 1663. 16. D. J. Castro, R. B. Lufkin, R. E. Saxton, A. Nyerges, J. Soudant, L. J. Layfield, B. A. Jabour, P. H. Ward, and H. Kangarloo, Laryngoscope, 1992, 102, 26. 17. S. Sinha, J. Parker, Y. Anzai, et al. MR compatible RF system: initial experience (abstr)., Soc. Magn. Reson. Imaging, 1993. 18. Y. Anzai, K. L. Black, A. A. F. DeSalles, D. R. Hamilton, K. Farahani, and R. B. Lufkin, Am. J. N. R., (in press). 19. K. L. Black, A. A. F. DeSalles, Y. Anzai, G. Nelson, E. Behnke, K. Farahani, S. Sinha, D. P. Becker, and R. B. Lufkin, J. Clin. Oncology, (in press). 20. K. Hynynen, A. Darkazanli, E. Unger, and J. F. Schenck, Med. Phys., 1993, 20(1), 107. 21. H. E. Cline, J. F. Schenck, R. D. Watkins, K. Hynynen, and F. A. Jolesz, Magn. Reson. Med., 1993, 30, 98.

THERMAL THERAPIES IN THE BODY MONITORED BY MRI

Thermal Therapies in the Body Monitored by MRI Margaret A. Hall-Craggs, S. Smart, and A. Gillams The Middlesex Hospital, UCLH, London, UK

1 INTRODUCTION The purpose of minimally invasive therapy (MIT) is to deliver effective treatment with the minimum disruption and damage to normal surrounding tissue. Consequently the morbidity of the procedure may be reduced compared with an open operation. Measures of this may be such factors as reduced scarring and deformity (as in the breast), more rapid postprocedural recovery, and reduced hospital stay (as in intraabdominal procedures). In some cases, therapy may be used where there is no other treatment option, as in tumor metastases to multiple lobes in the liver. Treatment using MIT is fundamentally different in benign and malignant disease. In benign disease, total eradication of the target tissue may not be necessary and size reduction may be adequate. This is the case with breast ®broadenomata and also in uterine ®broid disease where amelioration of menorrhagia or infertility is the treatment aim. With malignant disease, surgical principles preside; in general, tumor resection/destruction with an adequate margin is the treatment aim. Further MIT must not impose additional risks such as increasing the rate of local recurrence or distant disease spread, or reducing disease survival. A host of MIT techniques have been developed since the mid-1970s; these include the thermal ablative techniques to be discussed in this chapter. Other are percutaneous excision, photodynamic therapy, interstitial radiotherapy/brachytherapy, chemical ablations, laparoscopic surgery, lithotripsy, and intraluminal/intravascular procedures such as stenting, embolization, dilatation, etc.

2 METHODS OF THERMAL ABLATION There are four principal methods of thermal ablation. Three of these involve heating of the sampleÐinterstitial laser photocoagulation (ILP), radiofrequency (rf)-induced coagulation (RFC) and focused ultrasound (FUS)Ðwhile the fourth, cryoblation, uses cooling. 2.1 Interstitial Laser Photocoagulation For ILP, ®ber optic cables that emit light only from their tips are placed into the target region. These are coupled to a laser source and the light is turned on for the duration of the treatment. Theoretically the frequency of the light has to match an absorption band of some (unspeci®ed) chromophore within

1

the target region so the light energy can be converted into heat. There have been a number of approaches to improving the coupling of light to the tissue and these include the use of a focusing lens at the ®ber tip and diffuser tip ®bers. However it appears that laser energy deposition is primarily a thermal process. Pre-charring the optical ®ber causes the ®ber to behave as a point heat source (paradoxically reducing light transmission) and increases the size of the laser burn.1 The use of low laser powers (<5W) allows the slow and controlled development of lesions over extended periods (300±1000s). In our practice, we have used the semiconductor diode laser as it offers a number of advantages. It is small, compact, and easily transportable; it does not require a cooling system and operates from a conventional phase electrical source. 2.2

Radiofrequency

RFC uses rf alternating current source applied through needles tipped with rf electrodes. Frictional heating is induced via ionic agitation arising from the impedance of the tissue to alternating currents. The rf source typically operates at up to 500 kHz and will often interfere with the MR acquisition at all ®eld strengths. Such interference with the MR measurement can be reduced or eliminated with an appropriate choice of electronic ®lters, or by gating the rf source off during MR signal reception. This latter approach can easily be performed by simple modi®cation of pulse sequence code. To complete the electric circuit for the applied AC, grounding pads must be applied to the patient. Care must be taken with these to prevent the possibility of local skin burns at the site of the pads. RFC can form lesions faster than ILP; however, because the deposition of the heat depends upon the electrical characteristics of the tissue, the formation of the lesion may be inhomogeneous, especially in regions of tissue boundaries. 2.3

Focused Ultrasound

FUS uses an array of ultrasound sources, typically operating at 1.5 MHz, the energy of which converges at a focus within the body. In this region, a signi®cant amount of acoustic energy is converted into heat. The volume of the hyperthermia is very well de®ned for FUS; as a result, a sharp boundary may be observed between normothermic regions and those bound for necrosis. The duration of the treatment depends upon the ultrasound parameters but is typically in the region of tens of minutes. Lesions as small as 1 mm3 can be created; consequently, high-resolution imaging techniques, such as MRI, are necessary for their accurate detection and monitoring. 2.4

Cryotherapy

Cryosurgery will typically involve the use of a needle that is continuously cooled by liquid nitrogen and which is insulated along the length of its shaft except for at the tip. The tip of the needle acts as a heat sink for the tissue and so acts as a point source for cooling. Thermal conduction increases the volume of cooled tissue. There are a variety of mechanisms responsible for the subsequent necrosis, and these depend upon the organ undergoing the cryosurgery. As for FUS, there is a well-de®ned boundary between normal and treated tissues.

2 THERMAL THERAPIES IN THE BODY MONITORED BY MRI 3 ROLE OF IMAGING IN THERMAL ABLATION The principal roles of imaging in ablative treatments are to de®ne the extent of disease, to guide instruments, to monitor treatment during the therapeutic procedure, and to show the effect of therapy. In practice, a combination of imaging modalities are used to achieve these aims, including computed tomography (CT), ultrasound, and MR. However, there are a number of speci®c properties that have made MR the subject of much research interest. The exploitation of temperature-sensitive sequences can potentially be used to map areas of tissue damage during an ablative procedure, and this may facilitate treatment modi®cation at the time of delivery to improve ef®cacy. Other advantages include multiplanar imaging, which facilitates the guidance and accurate placement of tools. The lack of ionizing radiation is of importance to both the patient and the operator with lengthy therapeutic procedures.

4 MAGNET AND EQUIPMENT CONSIDERATIONS Interventional procedures have been described both in conventional tubular high-®eld systems and in lower-®eld magnets designed to have more open access. In open systems, the tissue of interest can be located within the `active volume' of the magnet for MRI, and the surgeon may have access to this volume from several sides. The use of the more conventional tubular high-®eld magnets is hindered by limited bore diameter, which limits the size of interventional devices, and by the restricted access owing to the magnet length, which prevents realtime MR-monitored adjustments of interventional tool position. On open access systems, rapidly and dynamically acquired MRI scans can assist with the positioning of interventional devices. Fast MRI may also allow near realtime visualization of physical changes within the tissue during interventional procedures. Typical procedures include MR-guided biopsies, tissue excisions, and thermal ablations. In this article we shall describe aspects of thermal ablation treatment in the human body as monitored by MRI. All interventional devices located within the region of treatment must be made of MR-compatible materials. Such tools are available and are usually made of alloys or ceramics that present a low magnetic susceptibility difference to the body. However residual susceptibility artifacts, dependent upon the applied magnetic ®eld strength and on the orientation of B0 relative to the interventional device, persist and may compromise image quality in their locale, especially when using gradient echo methods. The materials used in the manufacture of these MR-compatible devices are often soft, too ¯exible, and are dif®cult to tool to sharp and cutting tips. They are also expensive. For ILP therapy it is possible to use MR-compatible needles to position the optical ®bers and then to withdraw the needles leaving the ®ber in place; this eliminates the susceptibility problem. For RFC and cryotherapy, the needles must remain in place; consequently susceptibility artifacts are unavoidable. This is not necessarily a severe problem for FUS as the sonic transducers are positioned around the body not within the target region of tissue.

5

MR METHODOLOGY

Fast imaging methods are required for interventional therapies to facilitate tool guidance and to monitor therapy. Long scan times are also precluded by gross patient motion arising from respiration or peristalsis. The use of navigator echoes may allow the correction of some but not all of these motions. Prospective ordering of the phase-encode steps (e.g., respiratory ordered phase encoding, ROPE) can reduce the degree of motion artifact but requires good calibration and is sensitive to any respiratory irregularity. The use of breath-holding, especially when controlled under general anesthesia, allows imaging free of motion artifacts for scan times of up to about 30 s. Open-access scanners tend to operate at lower ®eld strengths than those used for routine clinical imaging and this reduces the possible signal-to-noise ratio (SNR). The rf and gradient hardware of the open-access scanner also tends to be of lower speci®cation than that of a high-®eld research-grade scanner. Some fast imaging protocols, e.g., echo planar imaging, may, therefore, not be available. Further the inherently low SNR realistically achievable with low-®eld scanners and the persistence of susceptibility artifacts, arising from interventional tools, may make echo planar imaging impractical for some of these applications. Rapid image reconstruction and display is necessary for realtime monitoring. This may be achieved more quickly by by-passing the scanner and using custom built hardware and software. Methods such as keyhole imaging and local-look scanning have also been used to enhance image update rates. 5.1

Sources of Contrast for Monitoring Thermal Ablations

Successful monitoring of hyperthermic ablations requires a temperature-dependent source of contrast for the heated region. It is important that any contrast seen is a direct consequence of the heating effect rather than resulting from a subsequent effect such as coagulation. Most successful studies demonstrating realtime monitoring of the effects of temperature in thermal ablations have, to date, been performed in model systems and generally using high-®eld systems. Currently favored mechanisms of temperature-dependent contrasts are T1 based, where a focal change in the signal amplitude is seen when running under conditions of partial saturation, and proton resonance frequency, where changes in the signal phase with temperature are seen. The proton resonance frequency method is essentially a chemical shift effect and so is most successful at high ®eld. Molecular diffusion has also been proposed as a method for introducing temperature-dependent contrast. However, this method has not been used widely as it is extremely sensitive to bulk patient motion, and because of the potential for anisotropic diffusion to introduce orientation-dependent contrast. For cryogenic ablations, rapid and accurate monitoring of the region of treatment is possible. As the tissue water solidi®es, its T2 time is greatly reduced; as a result, no echo signal can be measured using conventional MRI hardware. The border of frozen and unfrozen tissues is extremely well de®ned and for situations where multiple cooling probes are used, any pockets of normothermy can be readily identi®ed allowing repositioning of probes to encompass a given volume of interest. However, direct and unequivocal MR thermometry is

THERMAL THERAPIES IN THE BODY MONITORED BY MRI

precluded, at all ®eld strengths, by the absence of signal from those regions that are frozen. 6 MR-MONITORED THERMAL ABLATION PROCEDURES 6.1 Interstitial Laser Photocoagulation in Breast Disease Most work has been performed using the semiconductor diode laser (805 nm) and Nd:Yag (1064 nm) pulse lasers.2±5 Studies have shown that a power of 2±2.5 W applied for about 500 s will generate lesions of up to 10 mm. Lesion size and reproducibility can be improved by pre-charring the ®ber. Larger burns (of up to 40 mm) can be generated by using multiple ®bers or pull-back techniques. In a dose escalation study, Akimov has shown that at higher power (over 3 W), tumor vaporization can lead to tissue explosions.4 In most studies, patients are treated with local anesthesia and sedation alone; general anesthesia is unnecessary. 6.1.1

Benign Breast Tumors

In our institution we have treated just under 50 benign ®broadenomas of the breast with ILP. Diagnosis is made before treatment by triple assessment and either cytology (C2) or core biopsy. Lesions of up to 4 cm diameter have been treated with between one and four ®bers at 2±2.5 W (1000±4000 J). Follow-up for at least 1 year by clinical and ultrasound examination is available in 14 of these patients; in each case the mass has become impalpable by 1 year. In 13 patients, the tumor could not be seen on ultrasound by 12 months' followup and in the remainder a 9 mm remnant persisted.6 MR has a very limited role to play in the monitoring of ILP therapy to benign tumors. Fibroadenomas have variable enhancement characteristics; some enhance very intensely and others not at all. Consequently contrast-enhanced MR is not of consistent value. Those lesions that do enhance before treatment show areas of devascularization after treatment, but this is not of any clear clinical value. Ultrasound is a quick method of localizing needles within lesions and the treatment parameters are relatively standard. Unlike malignant disease, the treatment of margins is not critical; therefore, accurate monitoring of therapy is not essential. 6.1.2

Breast Cancer

All studies have reported that ILP can cause cell necrosis in breast cancers.3±5 Treatment of small relatively spherical tumors can be complete. Treatment of larger lesions is more challenging as modeling the treatment area to conform with irregular large masses and ensuring treated margins is much more dif®cult. Most studies have examined ILP where treatment has been followed by early surgical excision of the tumor and there are very few long-term follow-up data. There are small numbers reported where ILP has been the sole method of treatment. Akimov treated seven women with ILP alone. Two of these died of other causes and without tumor recurrence; a third had disease-free survival of over 5 years. In our own recent experience in a cohort of selected elderly women who are undergoing ILP and delayed excision at 3 months, we have

3

been able to show complete tumor ablation sustained over the 3-month period in small numbers. Complete tumor ablation and treatment of an adequate margin is essential in the management of breast cancer. The surgical literature has shown that incomplete tumor excision is a major risk factor for local tumor recurrence. Yet it is the margins that are most dif®cult to treat with ILP because, when the ®ber is placed in the center of a lesion, the periphery does not reach as high a temperature as the center of the tumor. In one patient treated with ILP alone, Akimov has shown peripheral tumor recurrence. We have found that when there is inadequate tumor ablation, viable cells are invariably found around the periphery of the mass. Unless margins can be reliably and consistently treated, ILP cannot be accepted as a reasonable form of primary therapy in breast cancer. Contrast-enhanced MR imaging has an essential role to play before therapy to map accurately the extent of the tumor and thus facilitate appropriate case selection. Treatment changes can be seen at high ®eld during ILP using T1-weighting sequences. Areas of low signal develop around the ®ber tip within a minute of treatment starting and these areas broadly correlate with the histologic extent of tumor necrosis.3,5 Consequently, MR does appear to have the potential to monitor treatment in close to realtime. Ideally, areas of signal change should be matched to the tumor. This is not easy in practice as the majority of tumors are only visible on contrast-enhanced scans. Repeated use of contrast causes enhancement of normal breast parenchyma and the loss of contrast between normal and abnormal breast tissue. Following treatment, delayed contrast-enhanced MR can be used to show the effects of treatment. Areas of devascularization (nonenhancement) occur within the treated area. Occasionally rim enhancement may occur around an adequately treated lesion as a result of peripheral in¯ammation. The only signi®cant complications of ILP reported are collateral burns to skin and muscle. The former can be avoided by irrigating the skin with chilled saline during treatment. Burns to the pectoralis muscle can be avoided by accurate placement of the ®ber tip. 6.2

Radiofrequency, Cryotherapy, and Focused Ultrasound in Breast Disease

Experience using these thermal treatments in breast disease is very limited. Preliminary reports of the use of rf in vivo7 and in vitro8 have shown that histologically con®rmed areas of cell necrosis occur. MR monitoring of rf therapy during treatment is limited by the need to have the metallic probe in position during the procedure. There are limited reports of the use of FUS in breast tumors.9,10 Prolonged treatment times reduce patient acceptance of the procedure, and patient movement compromises treatment. Cryotherapy has been shown to produce tissue necrosis in breast cancers11 but no signi®cant clinical studies have been reported. 6.3

Bone and Soft Tissue Tumors

There are limited reports of the application of MIT to bone and soft tissue tumors. ILP and rf are being used routinely in the ablation of benign osteoid osteomas. As the tumors are best

4 THERMAL THERAPIES IN THE BODY MONITORED BY MRI seen on CT and treatment parameters are standard, MRI has no role. There has been some preliminary experience with the treatment of bone metastases with MIT, but it is dif®cult to justify these experimental therapies when local radiotherapy is effective, widely available, and well tolerated. We have limited experience treating soft tissue metastases from soft tissue and bone sarcomas with PDT, rf, and ILP. In these patients (with chemo- and radio-insensitive tumors who have all undergone multiple previous operations), we have had variable results ranging from complete tumor ablation in patients with small volume disease to no effect in patients with massive recurrence. MR has been useful in these patients to stage local disease and for showing post-treatment changes. For complex tumors, needle placement is facilitated by MR as frequently the tissue/ tumor contrast is not suf®cient to enable accurate tumor localization with CT guidance. 6.4 Fibroid Disease ILP and rf have been used to treat uterine ®broids (leiomyomata) percutaneously, endoluminally, and at laparoscopy. The advantage of the last approach is that the peritoneum, which is very susceptible to thermal damage, can be directly visualized. The main role described for MR is in the monitoring of treatment response: ®broids frequently exhibit an iceberg effect and the extent of local damage cannot be accurately assessed by direct visualization. There is no published experience of perprocedural MR guidance to our knowledge. 6.5 MR-Guided Ablation of Liver Tumors The four principal techniques used to achieve percutaneous hepatic tumor ablation are ILP, RFC, cryotherapy, and percutaneous ethanol injection. A number of centers including our own are now routinely using rf and ILP to ablate hepatic metastases. In our own center most procedures are performed under local anesthesia and conscious sedation, with approximately 15±20% requiring general anesthesia. The 0.2 T Siemens Viva open-interventional MR system provides good access for either a subcostal or intercostal approach to the right lobe of the liver, or an anterior oblique approach to left lobe lesions with the patient in the supine position. A body ¯ex coil provides adequate coverage of the area of interest and good access for needle positioning. Initial breath-hold Ti-weighted gradient echo images (repetition time (TR); echo time (TE) 152 ms, 9 ms; ¯ip angle 70 ) scans are performed to locate the tumor. If tumor visualization is not adequate on baseline images, then liver-speci®c contrast agents are given (Figure 1). Either super-paramagnetic iron oxide particles (Ferumoxide, 15 mmol kgÿ1, Guerbet) or T1-shortening manganese-based agents (Mangafodipir trisodium Teslascan, 0.5 ml kgÿ1, Nycomed Amersham) will provide a long period of enhancement, facilitating tumor visualization. Needle placement can be performed entirely using MR guidance or by using a combination of ultrasound and MR. Generally we have favored the use of ultrasound for speed of needle placement, reserving MR for tumors not easily visualized by ultrasound and for con®rmation or ®nessing of the ®nal needle position. If ultrasound guidance is used, the patient is positioned within the

Figure 1 Coronal section through the liver following Endorem during interstitial laser photocoagulation therapy to a colorectal metastasis. The large tumor is seen as a relatively bright mass in the low signal liver. Two 19 gauge MR-compatible needles are shown positioned within the mass

coil on the MR table, and then the table is undocked and moved across the 5 G line, approximately 2 m from the magnet, to the ultrasound machine. 6.5.1

Needle Visualization

Needle positioning is con®rmed using multiplanar, often double-oblique T1 or T2 gradient echo images (Figure 1). As needle visualization is reduced when it is aligned with the B0 ®eld, a direct vertical puncture is best avoided. Other mechanisms for increasing needle visualization include swapping the phase- and frequency-encoding directions. Greater magnetic susceptibility and, therefore, better needle positioning can be achieved with the frequency encode perpendicular to the needle direction. Other techniques include increasing the number of averages of using breath-hold imaging. In practice breath-hold imaging is very dif®cult to achieve in sedated patients and if general anesthesia is being performed will require the addition of muscle paralysis and intubation. For laser therapy, up to eight bare-tip ®bers are used inserted through 19 G hollow, MR-compatible needles (Cook, Europe). For RFC, MR-compatible single rf electrodes are inserted directly. Where electrode clusters are required, guide needles are used as MR-compatible electrode arrays are still in development. 6.5.2

Monitoring the Treatment

We use a T1-weighted fast inversion±recovery sequence (TR, 1068 ms; TE 48 ms; time to inversion, 80 ms; number of signals averaged, 1 in 59 s; collimation, 10 mm; matrix, 182256) to monitor the changes produced by thermal ablation (Figure 2). On this sequence, untreated tumor is seen as high signal intensity against lower signal intensity liver. Over time, an area of reduced signal intensity develops at the treatment site and becomes progressively lower in signal intensity and larger in size. Around the area of low signal intensity, a high

THERMAL THERAPIES IN THE BODY MONITORED BY MRI

5

sequences are performed at breaks in the treatment cycle. Steiner et al. found good correlation between phase frequency shift imaging at 0.5 T immediately at treatment end and ®nal ablation volume.15 However, liver cooling is rapid (of the order of 50 s) and, therefore, there is only limited opportunity to acquire the necessary information. For cryotherapy, MR-compatible cryoprobes have been developed. The iceball is well visualized but tends to underestimate the ®nal lesion size.18,19 Percutaneous ethanol injection, unlike the other ablation techniques, is reserved for the treatment of hepatocellular carcinoma. Although ex vivo experiments have suggested the use of water-suppressed T2 fast-spin echo images for monitoring the actual injection and animal experiments have suggested that ethanol produces low signal on all sequences, in practice a range of signal intensity changes are produced and MR assessment relies on the absence enhancement during arterial phase imaging to indicate complete ablation.20±23 6.5.3

Long-term Follow-up

Most groups prefer CT for long-term follow-up because of cost and availability. 6.5.4

Figure 2 Transverse short tau inversion recovery images through the liver before (a) and after (b) interstitial laser photocoagulation (ILP) to a hepatic metastasis, showing treatment effect. The high-signal peripheral metastasis shows a low-signal focus within the center of the tumor following 8-®ber ILP to the mass. These changes con®rmed a treatment effect but underestimated the changes seen on follow-up contrast-enhanced computed tomography

signal intensity ring of edema is also visualized. Although the area of low signal intensity correlates with the ®nal area of thermal ablation on contrast-enhanced CT at 18 h, in general the MR image underestimates the ®nal extent of necrosis. We perform monitoring scans at 10-min intervals throughout the treatment. Treatment-sensitive sequences could theoretically provide more accurate monitoring information but are dif®cult to implement in the liver in vivo. Minimal signal intensity changes requiring image registration and subtraction are inaccurate. The shape of the liver is not constant but changes on respiration. A 13% error in liver registration in the craniocaudal direction has been reported.12 Further thermal injury itself produces a change in the liver shape and contour. Motion artifact becomes increasingly problematic during treatment when the patient is sedated and unable to breath-hold. Therefore, although ex vivo and animal experimentation suggest that both T1 contrast and phase frequency shift will provide accurate temperature information13±15 both Kettenbach and Vogl report dif®culties with susceptibility and motion in vivo.16,17 Vogl uses phase frequency shift at 1.5 T with the relatively simple indicator of a 50% reduction in signal to terminate treatment; and this is considered suf®cient for tumor ablation by this group.17 The use of rf suffers from the additional problem of interference with MR image acquisition; as a result, monitoring

Conclusions

MR guidance offers improved multiplanar imaging compared with other imaging techniques; although quantitative temperature measurement is not yet a reality, MR does provide some monitoring information. 6.6

MR-Guided Ablation in the Prostate

Failure of traditional treatments and the associated morbidity has driven a multifaceted search for alternative therapies for prostate treatment. There are now a number of image-guided treatment options under evaluation for localized prostate cancer, including cryotherapy, brachytherapy, and photodynamic therapy. Laser therapy has been used in benign prostatic hypertrophy. Fluoroscopy, CT and transrectal ultrasound have all been used to guide prostatic interventions. Of these, CT and ¯uoroscopy provide no information about the internal architecture of the prostate and are unidimensional; transrectal ultrasound offers signi®cant detail of the internal prostatic architecture but does not show the needle trajectory from perineum to prostate. In addition, the presence of the probe in the rectum holds the rectal wall against the prostate gland. MR offers multiplanar imaging, any plane, any obliquity, good prostate detail, and images of the full needle trajectory and of neighboring structures without distortion of the anatomy. 6.6.1

Photodynamic Therapy in the Prostate

We have used interventional MR to guide needle placement for photodynamic therapy in the prostate. Necrosis is produced by the interaction of light of a particular wavelength with a photosensitizing agent. The exact mechanism is unknown but a popular theory is that short-lived oxygen radicals are produced and these cause cell death. Necrosis does not occur immediately and the oxygen radicals are MR occult; therefore, MR does not have a monitoring role during the treatment. However, MR is excellent in guiding needle placement (Figures 3 and 4). As many as eight, 19 G hollow needles are introduced trans-

6 THERMAL THERAPIES IN THE BODY MONITORED BY MRI

Figure 4 Transverse two-dimensional gradient echo image through the prostate during multi®ber photodynamic therapy to the gland. Multiple needles are positioned within the tumor in posterior and leftsided lobes of the gland

Necrosis is appreciated as an area of nonenhancement at the treatment site (Figure 5). In comparison, CT demonstrates limited tissue contrast and both gray scale and Doppler ultrasound do not show any speci®c features. 6.6.2

Brachytherapy

MR guidance of seed insertion has been reported in nine patients at 0.5 T.24 MR allowed con®rmation of seed position in three planes. Both T1-weighted fast gradient echo and proton density fast spin echo sequences have been used satisfactorily.25,26 6.6.3

Figure 3 Coronal two-dimensional gradient echo image during photodynamic therapy to the prostate. Two 19 G MR-compatible needles are inserted within the gland and MR is used to con®rm the position of the needles

perineally into the peripheral zone of the prostate gland. Multiplanar T1-weighted gradient echo images are obtained to con®rm needle position and spacing. Laser ®bers are introduced through the needles and the tissue is illuminated with red light. Approximately a 16 mm sphere of necrosis is produced around each laser ®ber. Treatment is targeted to areas containing tumor on mapping biopsy and away from the urethra, neurovascular bundles, and external sphincter. Imaging assessment of the extent of necrosis is performed using gadolinium-enhanced T1-weighted sequences at high ®eld (1.5 T).

Cryotherapy

During cryotherapy, the iceball can be well visualized, although actual temperature measurement is not available. MR has also been used in the follow-up of cryotherapy-induced changes. Contrast-enhanced T1-weighted images show necrosis as areas of nonenhancement. Expected changes on follow-up include a reduction in prostatic volume of 30±52%, loss of zonal architecture in 81±100%, and soft tissue thickening of the rectal wall (47%) or around the capsule and neurovascular bundles in the majority (89%).27±29 6.6.4

Laser-induced Thermotherapy in Benign Prostatic Hypertrophy

MR has been used for monitoring of thermal ablation in benign prostatic hypertrophy and for follow-up assessment. On-line monitoring is feasible using phase-frequency shift30 providing an accurate prediction of ®nal lesion size in ten of 13 patients in one study.31 Follow-up MR showed lack of gadolinium enhancement in necrotic areas and volume reduction.32

THERMAL THERAPIES IN THE BODY MONITORED BY MRI

7

Figure 6 Navigational pointing device. A number of highly re¯ective balls are mounted on the handle of the pointing device, the light being detected by the stereoscopic camera. Tools may be mounted onto the device to facilitate accurate placement within the target

developed currently. We are investigating the use of a passive navigation device for interactive de®nition of the slice selection plane. The device can potentially increase the speed with which needles are positioned as it is possible to de®ne the image slice automatically, during needle insertion, according to the orientation of the needle itself. The biopsy needle within its holder is attached to the passive navigation probe (Figure 6). The reference point for image slice selection can be set relative to the tip of the needle. The stereoscopic camera detects the position and rotation of the probe so as to de®ne the slice selection axis. In situations where multiple needles require accurate positioning, the device reduces any ambiguity in identifying needles and so increases the safety of the procedure.

8 Figure 5 Transverse contrast-enhanced T1-weighted image through the prostate 5 days after photodynamic therapy (PDT). A lobulated region of nonenhancement is seen within the gland that corresponds to the region of devascularization after PDT. Similar changes are seen after cryotherapy

6.6.5

Conclusion

MR guidance and follow-up is preferable to other techniques in prostate ablation; for cryotherapy and thermal techniques there is the added advantage of treatment monitoring.

Temperature-sensitive MRI can be achieved for the monitoring of thermal treatments in the body within a realistic scan time. Low-®eld open-access scanners offer good surgical access to the patient. However, the image quality is poor compared with that expected from high-®eld closed-bore magnet systems. Several of the temperature-dependent mechanisms of contrast are ®eld strength dependent; consequently, for these, high-®eld magnets should be preferable for the monitoring of thermal interventions in the body. The development of shorter and wider bore high-®eld magnets should also assist in the promotion and acceptance of MRI monitoring of thermal interventions in the body, head, and extremities.

9 7 NAVIGATIONAL DEVICE As indicated above, accurate placement of tools is essential for thermal therapies to ensure that lesions are accurately localized and damage to collateral tissue is avoided. Although freehand placement is usually adequate where lesions are super®cial and large, the same is not true for deep and small lesions. In these cases, tool placement may be facilitated by navigational devices. There are a number of systems being

SUMMARY

RELATED ARTICLES

MR-Guided Biopsy, Aspiration, and Cyst Drainage; Temperature Measurement Using In Vivo NMR.

10

REFERENCES

1. S. A. Harries, Z. Amin, M. E. F. Smith, W. R. Lees, J. Cooke, M. G. Cook, J. H. Scurr, M. W. Kissin, and S. G. Bown, Br. J. Surg., 1994, 81, 1617.

8 THERMAL THERAPIES IN THE BODY MONITORED BY MRI 2. H. Mumtaz, M. A. Hall-Craggs, G. Buonnacorssi, M. Kissin, and S. Bown, Radiology, 1996, 201(P), 177. 3. H. Mumtaz, M. A. Hall-Craggs, A. Wotherspoon, M. Paley, G. Buonaccorsi, Z. Amin, I. Wilkinson, M. Kissin, T. Davidson, I. Taylor, and S. Bown, Radiology, 1996, 200, 651. 4. A. B. Akimov, V. E. Seregin, K. V. Rusanov, E. G. Tyurina, T. A. Glushko, V. P. Nevzorov, O. J. Nevzorova, and E. V. Akimova, Lasers Surg. Med., 1998, 22, 257. 5. S. E. Harms, H. Mumtaz, V. S. Klimberg, C. L. Cowan, S. Korourian, and W. B. Hyslop, Radiology, 1998, 209(P), 468. 6. L. M. Lai, M. A. Hall-Craggs, H. Mumtaz, P. M. Ripley, M. Kissin, T. I. Davidson, C. Saunders, I. Taylor, and S. G. Bown, Breast, 1999, 8, 89. 7. R. L. Birdwell, S. S. Jeffrey, E. L. Kermit, D. M. Ikeda, and K. W. Nowels, Radiology, 1998, 209(P), 198. 8. T. Boehm, I. Hilger, W. Mueller, J. R. Reichenbach, M. Fleck, and W. A. Kaiser, Radiology, 1998, 209(P), 198. 9. H. E. Cline, J. F. Schenck, R. D. Watkins, K. Hynynen, and F. A. Jolesz, Magn. Reson. Med., 1993, 30, 98. 10. J. W. Hand, C. C. Vernon, and M. V. Prior, Int. J. Hypertherm., 1992, 8, 587. 11. R. W. Rand, R. P. Rand, F. Eggerding, L. DenBesten, and W. King, Surg. Gynecol. Obstet., 1987, 165, 392. 12. D. L. Wilson, A. Carrillo, and L. Zheng, JMRI, 1998, 8, 77. 13. T. Kahn, T. Harth, B. Schwabe, H. J. Schwarzmaier, and U. Modder, Roto. Fortschr. Geb. Rontgenstr. Neuen. Bildgeb. Verfahr., 1997, 167, 187. 14. J. Olsrud, R. Wirestam, S. Brockstedt, A. M. Nilsson, K. G. Tranberg, F. Stahlberg, and B. R. Persson, Phys. Med. Biol., 1998, 43, 2597. 15. P. Steiner, R. Botnar, B. Dubno, G. G. Zimmermann, G. S. Gazelle, and J. F. Debatin, Radiology, 1998, 206, 803. 16. J. Kettenbach, S. G. Silverman, N. Hata, K. Kuroda, P. Saiviroonporn, G. P. Zientara, P. R. Morrison, S. G., Hushek, P. M. Black, R. Kikinis, and F. A. Jolesz, JMRI, 1998, 8, 933. 17. T. J. Vogl, M. G. Mack, A. Roggan, R. Straub, K. C. Eichler, P. K. Muller, V. Knappe, and R. Felix, Radiology, 1998, 209, 381. 18. J. Tacke, G. Adam, and R. Speetzen, Magn. Reson. Med., 1998, 39, 354.

19. H. P. Klotz, R. Flury, A. Schonenberger, J. F. Debatin, G. Uhlschmid, and F. Largiader, Comput. Aided Surg., 1997, 6, 340. 20. D. S. Lu, S. Sinha, J. Lucas, K. Farahani, R. Lufkin and K. Lewin, JMRI, 1997, 7, 303. 21. H. Shinmoto, R. V. Mulkern, K. Oshio, S. G. Silverman, V. M. Colucci, and F. A. Jolesz, J. Comput. Assist. Tomogr., 1997, 21, 82. 22. T. Fujita, L. Honjo, K. Ito, K. Takano, S. Koike, H. Okazaki, T. Matsumoto, and N. Matsunaga, J. Comput. Assist. Tomogr., 1998, 22, 379. 23. C. D. Becker, M. Grossholz, G. Mentha, A. Roth, E. Giostra, P. A. Schneider, and F. Terrier, Cardiovasc. Intervent. Radiol., 1997, 20, 204. 24. A. V. D'Amico, R. Cormack, C. M. Tempany, S. Kumar, G. Topulos, H. M. Kooy, and C. N. Coleman, Int. J. Radiat. Oncol. Biol. Phys., 1998, 42, 507. 25. D. F. Dubois, B. R. Prestidge, L. A. Hotchkiss, W. S. Bice, Jr, and J. J. Prete, Int. J. Radiat. Oncol. Biol. Phys., 1997, 39, 1037. 26. M. A. Moerlaand, H. K. Wijrdeman, R. Beersma, C. J. Bakker, and J. J. Battermann, Int. J. Radiat. Oncol. Biol. Phys., 1997, 37, 927. 27. C. L. Kalbhen, H. Hrciack, K. Shinohara, M. Chen, F. Parivar, J. Kurhanewicz, D. B. Vigneron, and P. R. Carroll, Radiology, 1996, 198, 807. 28. H. Howighorst, J. Dorsam, M. V. Knopp, W. Jugowski, S. O. Schoenberg, M. Wiesel, M. Essig, and G. Van-Kaick, Roto. Fortschr. Geb. Rontgenstr. Neuen. Bildgeb. Verfahr., 1998, 168, 44. 29. A. D. Vellet, J. Saliken, B. Donnelly, E. Raber, R. F. McLaughlin, D. Wiseman, and N. H. Ali-Ridha, Radiology, 1997, 203, 653. 30. U. G. Mueller Lisse, A. F. Heuck, M. Thoma, R. Muschter, P. Schneede, E. Weninger, S. Faber, A. Hofstetter, and M. F. Reiser, JMRI, 1998, 8, 31. 31. R. A. Boni, T. Sulser, W. Jochum, B. Romanowski, J. F. Debatin, and G. P. Krestin, Radiology, 1997, 202, 232. 32. U. G. Mueller Lisse, A. Heuck, P. Schneede, R. Muschter, J. Scheidler, A. G. Hofstetter, and M. F. Reiser, J. Comput. Assist. Tomogr., 1996, 20, 273.