ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI
Anisotropically Restricted Diffusion in MRI Michael E. Moseley and Alex de ...
52 downloads
2107 Views
95MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
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 2T
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 S
0eÿbD
2
where the b value is de®ned12,13 by equation (3): b 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 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 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 expÿbD
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
' 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 Gz2
2
where z2 is the spin position during the second pulse. The net transverse magnetization Mxy is: Mxy=M0 expi
'2 ÿ '1 expi G
z2 ÿ z1
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 d z1 P
z1 expi G
z2 ÿ z1 P
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 expÿ 2 G2 D2
ÿ =3
@P
z2 ; z1 ; t=@t 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 expÿ
G2 D expÿ
G2 hz2 i=2
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 ;
4Dÿ1=2 expÿ
z2 ÿ z1 2 =4D
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 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 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 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 1=n1 exp2D
b1 ÿ b2 =n2 1=2 =
SNRD
b1 ÿ b2
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 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 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 (i6j ), of the diffusion tensor (now expressed in the gradient frame), which re¯ect correlation between molecular displacements in perpendicular directions:85 ! X X A exp ÿ bij Dij
16 ix;y;z jx;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 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 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; expiqzP
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 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 sin
G a=2=
G a=22
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 expÿ
G2 R20 =5
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 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 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 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 hi2
2 ÿ hi2
3 ÿ hi2 = 3hi
RA
25
where hi
1 2 3 =3 FA
q 3
1 ÿ hi2
2 ÿ hi2
3 ÿ hi2 = q 2
21 22 23
VR 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
s X p
Dii ÿ hDi2
D2xy D2xz D2yz = 6hDi
28 ix;y;z
with hDi=
X
Dii/3.
ix;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
A
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 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 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 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 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
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 20
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 20
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 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 f0 !0
1 ÿ Y gbvl R2 f0 !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 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 25426 ms and 18123 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 636.2 ms at 4 T39 as opposed to 872 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 636.2 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 ÿlnS
t=S
t0 =
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 S
t0
1 ÿ expTRR1
t expTER2
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 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 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 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
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 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
tg FfFt R
t Ca
tg FfCt
tg ) Ft R
t Fÿ1 FfCa
tg
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 Ft
0
tj
Ca
R
tj ÿ d Ft t
j X
Ca
ti R
tj ÿ ti
i0
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 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 FtR(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 1
t!0
whereas lim R
t 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 E 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
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 ÿ
dR dt
14
and the distribution of ¯ows is by the central volume theorem14 given by w
f 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 Cev
t Civ
t K1 Z Ft
0
t
0
t
K1 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 A
t 0
Cc
d BCc
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
(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
tvt 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: 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 K abssin
VTAg
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 Ex 1 Ex 2 Ex 3 Ex 4
4
X flow component ÿEx 1 Ex 2 Ex 3 ÿ Ex 4
5
Y flow component ÿEx 1 Ex 2 ÿ Ex 3 Ex 4
6
Z flow component 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 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 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 0
1
tGr
t dt 0
2
and Z
TE 0
for the readout gradient, Gr and Z T
Gpe
t dt T Gpe
3
and Z T
tGpe
t dt 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: 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 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 contaminated ATP ÿ 0:5
ATP=DPG 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:
P
2sP WCPH 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
S
DPGWB CPH FDPG EDPG DPG 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
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 CR
2SCR W 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 6:77 log10
Pi ÿ 3:29=
5:68 ÿ Pi
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. P