ANISOTROPICALLY RESTRICTED DIFFUSION IN MRI
Anisotropically Restricted Diffusion in MRI Michael E. Moseley and Alex de ...
52 downloads
2104 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,