Localization of brain function using magnetic resonance imaging

Localization of brain function using magnetic resonance imaging

T h e future 'When lunatics attempt to write, there is a perpetual recurrence of one or two favourite ideas, intermixed with phrases which convey scar...

1MB Sizes 8 Downloads 174 Views

T h e future 'When lunatics attempt to write, there is a perpetual recurrence of one or two favourite ideas, intermixed with phrases which convey scarcely any meaning either separately, or in connection with the other parts... patients will run their ideas in the very same track for many weeks together' (John Ferrier, 1795, see Ref. 21). This quotation, which shows for how long perseveration has been recognized as a symptom of mental illness, is intended not as a comment on any particular neuroscientist but as an observation on the DA hypothesis and thinking about schizophrenia in recent years. Dopamine is involved in schizophrenia but if the neurobiology of this grim disease were bounded simply by this, the game would have been up long ago. Changes are taking place: the ideas being generated about the development of the frontal cortex and its relations with other structures have provided a framework for research that has been missing for a long time. The work with animals

(1986) Arch. Gen. Psychiatry 43, 31-35 8 Benes, F., McSparren, J., Bird, E., Sangiovanni, J. and Vincent, S. (1991) Arch. Gen. Psychiatry 48, 996-1001 9 Akbarian, S. eta/. (1993) Arch. Gen. Psychiatry 50, 169-177 10 Akbarian, S. eta/. (1993) Arch. Gen. Psychiatry 50, 178-187 11 Scheibel, A. B. and Kovelman, J. A. (1981) BioL Psychiatry 16, 101-102 12 Arnold, S. E., Hyman, B. T., Van Hoesen, G. W. and Damasio, A. R. (1991) Arch. Gen. Psychiatry 48, 625-632 13 Falkai, P., Bogerts, B. and Rozumek, M. (1988) Biol. Psychiatry 24, 515-521 14 Roberts, G. W. (1990) Trends Neurosci. 13, 207-211 Selected references 15 Bloom, F. E. (1993) Arch. Gen. 1 Bleuler, E. (1913) Dementia praecox Psychiatry 50, 224-227 or the group of schizophrenias in 16 Fuster, J. (1993) Curt. Opin. NeuroThe Clinical Routes of the Schizobiol. 3, 160-165 phrenic Concept (Translated by 17 Burns, L. H., Robbins, T. W. and Cutting, J. and Shepherd, M., eds), Everitt, B. J. (1993) Behav. Brain Res. Cambridge University Press 55, 167-183 2 Crow, T. J. (1980) Br. Med. J. 280, 18 Geyer, M. A., Swerdlow, N. R., 1-9 Mansbach, R. S. and Braff, D. L. 3 Liddle, P. (1987) Br. J. Psychiatry 151, (1990) Brain Res. Bull. 25, 485-498 145-151 19 Seeman, P., Guan, H-C. and Van Tol, 4 Liddle, P. (1987) PsychoL Med. 17, H. H. M. (1993) Nature 365, 49- 57 441-445 5 Liddle, P. and Barnes, T. (1990) Br. J. 20 Frith, C. D. and Done, D. J. (1983) Psychiatry 157, 558-561 Psychol. Med. 13, 779-786 6 Weinberger, D., Berman, K., Suddath, 21 Frith, C. D. and Done, D. J. (1990) in R. and Torrey, E. (1992) Am. J. Neurobiology of Stereotyped BePsychiatry 149, 890-897 haviour (Cooper S. J. and Dourish, 7 Benes, F., Davidson, J. and Bird, E. C. T., eds), Oxford University Press

- anatomical, electrophysiological, developmental and behavioural - is generating hypotheses about the normal structure and function of the frontal cortex, and new techniques for post-mortem analysis of human tissue allied to an improving ability to visualize living human brain, offers the opportunity to integrate work with experimental animals and patients. More than anything else, this meeting emphasized that there is hope of progress.

l ,echniques o

Z

...........111

.....

. . . . . .

Localizationof brain fund/on usingmagneticresonance imaging M a r k S. Cohen and Susan Y. Bookheimer

MarkS. Cohenis at the Depts of Neurology and Radiology, Brain Mapping Division, UCLA Schoolof Medicine, 710 Westwood Plaza, Los Angeles, CA 90024, USA,andSusan Y. Bookheimeris at the Dept of Psychiatry, BrainMapping Division, UCLA Schoolof Medicine, 710 Westwood Plaza, LosAngeles, CA 90024, USA.

268

When nuclear magnetlc resonance images (MR/s) of the brain are acquired in rapid succession they exhibit small differences in signal intensity in positions corresponding to focal areasof activation. These signal changes result from small differences in the magnetic resonance signal caused by variations in the oxygenation state of the venous vascu/ature. Using this non-invasive functiona/ MRI (fMRI) method, it is possible to localize functional brain activation, in norma/ individuals, with an accuracy of millimeters and a temporal resolution of seconds. Though numerous technical challenges remain, fMRI is increasingly becoming a key method for understanding the topographical organization of the human brain. A remarkable feature of the vertebrate brain is the anatomical specialization of cortical regions for the processing of different types of information. Since the late 19th century, it has been recognized that restricted lesions of the human brain result in location-specific sensory, motor or cognitive deficits 1. Few tools are available to understand how activities in these distinct neural processing regions are orchestrated to perform complex tasks such as reading, memory and spatial visualization. High-resolution structural data collected using magnetic resonance imaging (MRI) have an estab© 1994.ElsevierScienceLtd

lished place in the neurosciences. For example, the presence and localization of lesions correlated with behavioral or cognitive deficits suggest structurefunction relationships in cognitive skills. Until recently, human functional data have been constrained by severely limited spatial resolution, as provided by electrical recording methods, or by the need for radionuclide imaging (for example, positron emission tomography or PET) involving complex apparatus and radio-pharmaceuticals, even then achieving only moderate ( - 5 - 1 0 m m ) spatial resolution. A confluence of MRI developments, particularly those involving ultra-fast imaging, has resulted recently in techniques by which activity in the human brain can be observed non-invasively with spatial resolution of a few millimeters and temporal resolution of less than a second. The MRI approach is technically challenging, expensive and less than two years old, yet the publications on both methods and results are already too extensive to summarize fully in a short review. These new techniques, generically termed functional MRI (fMRI) have led already to an improved understanding of the neural TINS, Vol. 17, NO. 7, 1994

processing of higher-level information, and they will contribute substantially to the ability of the neuroscientist to explore the higher-level workings of the human mind.

Principles of magnetic resonance To understand the fMRI method, investigators should be familiar with the physical principles of magnetic resonance that determine its signal characteristics, and through which it is possible to form images. In overview the process is as follows: (1) The subject is placed into a strong and homogeneous magnetic field. Various atomic nuclei, particularly the proton nucleus of the hydrogen atom (from here, we will consider the proton only), align themselves with this field and reach a thermal equilibrium. The subject is thereby 'magnetized'. (2) The proton nuclei precess about the applied field at a characteristic frequency but at a random phase (or orientation) with respect to one another. (3) Application of a brief radio frequency (RF) electromagnetic pulse disturbs the equilibrium, and introduces a transient phase coherence to the nuclear magnetization that can, in turn, be detected as a radio signal, and formed into an image. Signal changes with blood oxygenation The rate at which the magnetic resonance (MR) signal decays, T2*, depends upon a variety of physiological and physical factors. Variations in precession frequency among the excited nuclei result in signal loss (from spin dephasing - see Box 1). One of the principal mechanisms for this is the presence of local variations in magnetic-field strength caused by the presence of particles or tissues with differing magnetizability or 'susceptibility'. As early as 1936, Pauling noted that the magnetic susceptibility of oxyhemoglobin and deoxyhemoglobin differed slightly 2. Thulborn predicted, and in 1982 demonstrated 3, that the signal decay rate of deoxyhemoglobin is mere rapid than that of its oxygenated counterpart. Effects of blood oxygen on T2* were first reported in MR images by Ogawa and Lee who noted that cortical blood vessels became more visible as blood oxygen was lowered 4. They interpreted the cause to be the creation of local magnetic-field inhomogeneities, and thus signal losses, from deoxyhemoglobin and termed it the BOLD (blood oxygenationlevel dependent) method. Turner s demonstrated that with ultra-fast echo-planar imaging he was able to observe the timecourse of these oxygenation changes while an animal breathed an oxygendeprived nitrogen atmosphere. Shortly thereafter, Kwong and colleagues6 reported observing similar changes in humans during breath-holding. Functional activation Alterations in blood 02. Activated areas of the

human brain show localized increases in blood flow. These increases are exploited in functional imaging with PET (Ref. 7). Blood volume increases during sensory stimulation also8 and, using another MRI method (that measures blood volume based on TINS, Vol. 17, No. 7, 1994

signal changes following contrast-agent injections9), these volume changes were used by Belliveau and colleagues to make the first functional MR images of brain activation 1°. The increases in blood flow seem to exceed increases in oxygen consumption 11. Thus, the oxygen content of venous blood increases during brain activation, resulting in an increase in the intensity of the MR signal (Fig. 1). Using blood as an endogenous contrast agent, it was demonstrated that the transient changes in MR signal that accompany these hemodynamic events could be observed with rapid MRI methods 12'~3. This has revolutionized the field of fMRI, forming the basis for practical, non-invasive observation of the hemodynamic changes accompanying neuronal activity. Signal changes with blood flow. The prolonged rate (T1) at which the MR signal approaches equilibrium (see Box 1) can also be used to detect vascular signal changes. In Kwong and colleagues' original publication ~2, it was noted that the increased inflow of blood into the imaging volume can also result in a signal change due to T1 effects. By using a relatively strong RF excitation pulse (a 180 ° pulse), the signal available from a slice of tissue can be much reduced, so that the flow of fresh blood into that tissue slice, or volume, results in a signal increase. The flow-dependent signal difference between baseline and stimulated behavioral conditions might also be used for fMRI. Through manipulation of the MR-signal acquisition, the newly proposed EPISTAR technique 14 further amplifies the usable MR-signal change. These flow-based techniques are of special interest because they might, in principle, be used to quantify change in blood flow. Furthermore, while the T2* effects probably reflect signal changes in the venous system

,

,,

Normal flow

High flow

Oxyhemoglobin e

Deoxyhemoglobin

Fig. 1. During periods of neuronal activity, local blood flow and volume increase with little or no change in oxygen consumption. Consequently, the oxygen content of the venous blood is increased, resulting in an increase in the intensity of the magnetic resonance signal. 269

Box 1. Magnetic-resonance The proton magnetization can be broken down into the sum of a stationary (longitudinal) and a rotating (transverse) component (Fig. 2). Each proton within a magnetic field thus yields a tiny field that rotates about that applied field. The rotating field from individual nuclei is, in general, aligned at random with respect to other protons in the subject or sample. In macroscopic systems, the average rotating field will be effectively zero, since that arising from any individual nucleus is canceled by another neighbor that is oppositely oriented. In nuclear magnetic resonance (NMR), a second magnetic field is applied that is orthogonal to the static field, and which rotates about the static field at the precession frequency of the atomic nuclei. When the rotating field is present, the nuclei will precess about it, forcing the magnetization away from equilibrium, and causing the ensemble of protons to precess together, or to 'in-phase'. The combined rotating magnetic moment thus produced by the ensemble of protons can be observed as a time-varying electromagnetic (radio) signal. The second, rotating magnetic field is applied at radio frequencies and is therefore known as an 'RF' pulse. These fundamental principles were elucidated more than 40 years ago; among the seminal contributions were those of Bloch and colleaguesa,b and Hahn c.

The proton nuclei of hydrogen atoms possess a small magnetic moment. When placed within a magnetic field, a torque will be exerted upon them, resulting in a slight energetic advantage of one orientation (parallel to the field) over another (the anti-parallel orientation). Over time, random atomic collisions and other perturbations enable the system to reach a magnetic and thermal equilibrium with an excess of protons aligned with the magnetic field. The combined alignment of all of these protons results in a net magnetic moment; a subject placed within a magnetic field thus becomes 'magnetized'. In biological tissues, this magnetization is exceedingly small, and generally not observable. In addition to their magnetic moment, atomic nuclei possess angular momentum - a quantum property known as 'spin'. Because of this angular momentum, rather than simply aligning with magnetic fields, the individual nuclei precess about it, much as a spinning top or gyroscope might when placed in the earth's gravitational field (Fig. 1A). The precession rate, or frequency, is characteristic of the atomic nucleus (for example, protons) and is proportional to the strength of the magnetic field (Fig. 1B), a property crucial to the process of image formation. With the magnetic-field strengths used by typical magnetic resonance (MR) imagers, the precession frequency is between 10 MHz and I O O M H z - just below the FM radio range.

A Precession

84

v

f Spin

Applied magnetic field

"~

•.,

42

0 0

i 1

i 2

Field strength (Tesla)

Fig. 1. (A) Magnetic properties of the proton nucleus of the hydrogen atom. The hydrogen proton possesses the quantum property of 'spin' or angular momentum, and has a small magnetic dipole moment. When placed in a magnetic field, a torque is exerted on the particle, causing it to precess about the appfied field. (B) The precession frequency of the protons is directly proportional to the magnetic-field strength. Protons precess at approximately 43 MHz per Tesla.

mostly, the T1 changes might be biased toward the arterial supply.

Characteristics of the fMRI signal Magnitudes. Blood accounts for a very small percentage (approximately 6%) of gray matter (and even less of white matter): the hemodynamic signal changes that occur in MR during brain activation are extremely small, from 2 - 5 % at moderate magneticfield strengths (1.5 Tesla) to approximately 15% at very high fields (4Tesla). Nevertheless, with ad270

equate signal-to-noise ratio (SNR) in the MR images, they are clearly visible. Figure 2A shows signal changes in the human primary visual cortex (V1) as signal-difference images during visual (photic) stimulation. The accompanying graph (Fig. 2B) plots the signal intensity as a function of time in a region within V1, showing the excellent SNR that can be obtained with this method. In darkness, the intensity of the MR signal preceding the first period of visual stimulation fluctuates slightly. It is not yet known how much of TINS, VoL 17, No. 7, 1994

signal formation Fig. 2. Vector description of proton magnetization. The rotating magnetic moment of the proton can be broken down into a longitudinal component, along the applied magnetic field, and a transverse component orthogonal to it and precessing around it.

"c°ng i:nat /

Magnetic dipole moment Rotating component (MT)

Signal characteristics Two fundamental temporal parameters are used to describe the MR signal. The longitudinal relaxation rate, T1, is the rate at which nuclei, once placed in a magnetic field, exponentially approach thermal equilibrium, so that the magnetization (M) is described by the formula: M(t) = Mo (1 - e -t/r1) where Mo is the equilibrium magnetization. In biological tissues, the proton TI is quite long: from tens of milliseconds to seconds. Differences in the T1s of tissues are one of the primary basis of contrast in clinical MRI. A second time-constant parameter describes the rate at which the MR signal decays. Once an MR signal is formed (that is, after an RF pulse) it fades quickly; small variations in the local magnetic field, for example, those caused by neighboring magnetic nuclei, cause the protons to precess at slightly different rates and therefore to become out of phase with one another. Interactions among the magnetized protons, and motion in inhomogeneous fields, for example, caused by diffusion, results in signal dephasing also. The observed signal decay rate (T2*) generally ranges from a few milliseconds to tens of milliseconds and, to a reasonable approximation, follows first-order kinetics. The MR signal S(t) decays according to the formula: S(t) = So e- t/t2. where So is the signal strength immediately following

the RF excitation pulse. The observed T2* decay is the net effect of all the dephasing terms: I/T2" = I/T2 + I/T2m + I/T2D + other terms... where T2m represents the dephasing due to magneticfield inhomogeneities and T2D is the diffusion-related signal loss. Like T1, the T2 signal decay rates differ among body tissues. For most current fMRI, T2 is the dominant contrast mechanism. In addition, blood oxygen content strongly effects the observed signal decay rate (see Box 2). By waiting for a short period, 'TE', following the RF excitation pulse, differences in the signal decay rate become evident as differences in the MR-signal intensity: tissues with longer T2s will have stronger signals than those with short T2s, whose signals decay more rapidly (Fig. 3). Modifications to the pattern of RF excitation (the 'pulse sequence') can modulate the contributions of the various relaxation processes to the resulting MR signal. In particular a 'spin echo' pulse sequence can be used to almost eliminate the T2m contribution, increasing the relative contributions of other factors, such as proton diffusion, to the image contrast.

References a Bloch, F. (1946) Phys. Rev. 70, 460-474 b Bloch, F., Hansen, W. W. and Packard, M. (1946) Phys. Rev. 70, 474-485 c Hahn, E. (1950) Phys. Rev. 80, 580-594

Fig. 3. Spontaneous decay of transverse magnetization (signal). Immediately following a radio frequency excitation pulse, the coherent rotation of the ensemble of protons forms a detectable signal. This signal decays spontaneously, with first-order kinetics, at the characteristic rate, T2. At the time the MR signal is sampled (TEl, the signal intensity from tissues with a long T2 (for example, cerebrospinal fluid) will be greater than that from tissues with a short T2, for example, fat. Differences in effective T2 form the basis of the contrast for most fMRI methods.

this fluctuation is the result of variations in physiological signal, and how much is simply the consequence of instabilities in the MR instrument 15. Furthermore, the relative contributions of these components can be expected to vary among MR scanners, requiring that signal intensities be expressed in non-physiological units such as contrastto-noise ratio. Response latencies. The fMRI signal takes several seconds to reach its peak following the onset of the stimulus presentation. Kwong fitted the response TINS, Vol. 17, No. 7, 1994

0.5' CJ)

0 0

25

50

75

100

Echo time (ET)

increase to a monoexponential function and determined the time constant to be approximately 4.4 s for images of this kind, while the more flowsensitive techniques seemed to have response latencies that are slightly longer. The characteristic response delay differs across brain regions and stimulus regimens (for example, see Refs 16 and 17 or Fig. 5 of Ref. 12). Nevertheless, the delay is substantially slower than the neural or psychophysical response. Thus, the temporal resolution of fMRI seems to fall between that of electrical 271

be able to resolve signal changes that occur within a few tens of milliseconds. Thus, the temporal resolution of fMRI would seem to be limited by the phenomenon (detection of vascular signal changes) rather than by the 'camera'. Temporal fluctuations. The intensity of the fMRI signal during activated periods might be quite variable, even with constant stimulus intensities. The response not only takes some time to appear, but begins to decrease before cessation of the stimuli and, furthermore, seems to fluctuate during the stimulus presentation (Fig. 2B). Compared with the signal variations in the absence of stimulation, those during activated periods are often larger.

A

f M R I results

B on

off

6050

on

~m*o,,'%%~

° n

=

off

5900

I,

°~

C

01 5750

5600 0

. 60

.

. . 120

.

. 180

7- - 240

Time (s)

Fig. 2. (A) MR signal-difference map during photic (visual) stimulation. In an image aligned along the calcarine fissure, the signal intensity increases visibly during presentation of a photic stimulus, consisting of an 8 Hz patterned flash. The image at the upper left was acquired in darkness, and the four images which follow were subtracted from this. The local signal increases can be seen along the calcarine fissure. The pixel-intensity scale is normalized by the standard deviation of the signal intensity during the initial baseline period in order to account for local differences in baseline function. The pixel intensities are thus in units of 'contrast-to-noise ratio '. (B) Signal-intensity changes within the visual cortex. The signal in a small ( - 6 0 m m 2) area near the calcarine fissure during exposure to an 8 Hz patterned flash. Images were acquired once every 3 s. Note the signal decrease following cessation of the stimulation. Signal intensity is in arbitrary units, and data are taken from a different subject than that in A. Reproduced, with permission, from Ref. 11. recording methods such as EEG or direct cellular recordings and that of PET. Using a non-imaging nuclear magnetic resonance (NMR) method, Hennig and Ernst ~8 recently reported small changes in the strength of the MR signal with latencies of approximately 500ms to the presentation of a visual stimulus that might prove useful for advancing the temporal resolution of fMRI. These response latencies probably represent repeatable physiological delays. The MRI methods should, in principle, 272

Primary sensory and motor activation. In individuals, fMRI responses to visual stimuli 12'16'19-23, somatosensory or motor activity 24-27 and acoustic stimuli 28 have been reported. It has been shown that the magnitude of the fMRI response seems to be scaled to the stimulus intensity ~2 but the linearity of this scaling, and the ultimate sensitivity to lowintensity stimuli, are still unknown. It has been suggested that averaging the responses to repeated low-intensity stimuli might improve this sensitivity 29,3°. Higher level function. (A) Language tasks. An important challenge for fMRI (or any other functional-imaging technique) is to detect signal changes during subtle cognitive tasks. In PET activation studies, blood-flow changes in association areas (for example, during language performance) are more difficult to detect than those seen in primary visual or motor cortices 31. In several studies, fMRI has demonstrated successfully activation during covert word generation 32-35 in the inferior frontal lobe, a probable language-association region. The task-related changes reported during word generation replicate and extend previous PET studies 36. Recent work on single-word reading 28 has also demonstrated activation in Broca's area, as well as visual pre-striate cortex. (B) Motor and visual imagery. Among the more intriguing results in fMRI is the localized signal change observed during covert mental activity (mental imagery or ideation). In the hands of several investigators, and in visual, somatosensory and motor systems, the MR signal increases when the subject imagines a visual stimulus 37, or imagines performing a motor task 27. Such blood-flow changes were observed using non-tomographic techniques more than 15 years ago 38, and, more recently, using PET (Ref. 39). However, with fMRI it becomes possible to examine the locus of such activity on single subjects with a high degree of reliability. Using this technique, such questions as whether the mental image of a visual stimulus is represented in pseudo-retinotopic form on the primary visual cortex during a recall task - a question of abiding concern in the study of mental imagery 4 ° - can be examined. While the temporal resolution of fMRI is slow compared with neuronal firing, it might be appropriate for the study of a variety of physiological TIN& Vol. 17, No. 7, 1994

Box 2. Image formation

A method first proposed in 1977 by Mansfield, The suggestion that the nuclear magnetic resonance known as echo-planar imagingd (EPI), performs all signal could be used to form images was made first by required spatial encoding during the several tens of Lauterbur a in 1973. Reasoning that the precession milliseconds that the MR signal is present, without frequency of the atomic nuclei depended upon the resorting to repeated excitation-sampling cycles. This local magnetic field, he proposed that by forming a technically challenging method that was brought into magnetic field that varies spatially, it would be possible practice in 1984 (Ref. e), and to high-magnetic-field to separate the signal from different locations accord- whole-body imaging in 1987 (Refs f and g), makes it ing to frequency. For example, with a sample placed possible to form complete MR images in as little as within a linear magnetic-field gradient, the Fourier 20ms. Various modifications to EPI were developed, transform of the signal would show its strength at each over the years that followed, to bring high resolution frequency, and thus at each position. Presently, and controllable contrast to the technique, enabling a magnetic resonance imaging (MRI) instruments use wide variety of novel medical and scientific applithree mutually orthogonal sets of electromagnetic cationsh'i. Consequently, commercial devices having 'gradient coils' to encode the three spatial co-ordinates EPI capability are now available from several manuof the MR signal. facturers. Both FLASH scanning (in approximately 6s Imaging speed. Detecting small differences in fre- per image) and EPI (in approximately 0.1 s per image) quency (which in MRI are equivalent to small differ- achieve excellent functional MRI results. While the ences in position) requires sampling the signal for a FLASH method enables good control of both TI and relatively long time - the smaller the frequency T2* contrast, the EPI method is more flexible in difference, the longer the time needed. In commercial controlling the relative contributions of T2m and T2D to MRI, one of the challenging tasks is to switch the large the images. magnetic-field gradients needed for adequate spatial encoding on and off in the limited time that the MR signal is available. In conventional imaging instru- References ments, this problem is handled by performing, in a Lauterbur,P. C, (1973) Nature 242, 190-191 effect, only part of the spatial encoding at any one b Frahm,J. etaL (1990) Magn. Reson. Med. 13,150-157 time, and later re-exciting the MR signal to perform c Haase,A. (1990) Magn. Reson. Med. 13, 77-89 further encoding, repeating this process as many as d Mansfield,P. (1977)J. Phys. C10, L55-L58 several hundred times to form a complete image. For e Mansfield, P. (1984) Br. Med. Bull. 40, 187-190 f Pykett, I. and Rzedzian, R. (1987) Soc. Magn. Reson. this reason, MRI times have, traditionally, been exMed. Abstr. 10 tremely long-from 3-15 min for an imaging series. By g Rzedzian, R. and Pykett, I. (1987) Am. J. Roentgenol. minimizing the perturbation of the magnetization from 149, 245-250 its equilibrium, a method known as FLASH (Refs b and h Cohen,M. S. and Weisskoff, R. M. (1991) Magn. Reson. c) enables the time between successive excitation Imaging9, 1-37 pulses to be reduced, making imaging times of less i Stehling, M. K., Turner, R. and Mansfield, P. (1991) than 1 s possible, with some penalty in total contrast. Science 254, 43-50

Spatial encoding with gradients

In a study of pediatric epilepsy, the spread of seizure activity through its blood-flow effects was demonstrated using fMRI (Jackson, G., Connelly, A., Cross, H., Gordon, I. and Gadian, D., pets. commun.). Such studies might become valuable both in increasing the understanding of the physiological basis of the disease, and in its therapeutic management. processes.

final MR image. Unfortunately, in MRI, reductions in voxel volume reduce the available signal per voxel, while the noise (per voxel) remains essentially constant. Thus, the SNR scales with the third power of the linear voxel dimensions (or feature resolution). Since the method typically exploits signal changes of only a few percent, the SNR must be quite high for such changes to be observed. The spatial resolution in fMRI must therefore be someTechnical issues what coarse compared with the theoretical resolving Spatial resolution. Hemodynamic-response data power of conventional MRI techniques. obtained by optical methods 41'42 suggest that the The fMRI technique is presumably sensitive cortical vascular responses might be localized to the mostly to changes in the signal from venous blood. columnar level. Because the fMRI response appar- As the voxel volume content of blood increases, less ently corresponds to local changes in blood flow, it change in blood oxygen is needed to produce the might, in principle, be possible to obtain fMRI maps same fMRI-signal change. Some investigators have of cortical columns. Furthermore, the feature res- suggested that much of the fMRI-signal change olution of standard MRI can be brought readily to might be seen within brain areas having little or no 100 ~m or so - the appropriate size range to assess neural tissue, being instead images of the venous columnar anatomy 43,44. To date, no such fMRI vasculature 45. By implication, such signal changes results have been reported. In practice, a variety of would be displaced spatially from the activated factors limit the useful spatial resolution of the fMRI neural tissue. At this time, the relationship between method. vessel size and vascular territory are poorly underThe MR signal is intrinsic, arising from the tissues stood. A variant of MRI, known as magnetic of the brain, and quite small. Increases in the spatial resonance angiography 46 (MRA) images blood resolution (decreases in the image feature size) vessels of only a few hundred micrometres. Used result in smaller MR-signal energy per pixel in the properly, it is possible to exclude pixels imaging TINS, VoL 17, No. 7, 1994

273

B

A

C

4000...... Signal

~'2000

i''"':'""Fi~'i!ii:i~'r~

Error

o

-2000 Position

Fig. 3. (A) Raw 'functional' image of the visual cortex of a human subject. (B) Difference image created from the subtraction of the image at left from the identical image offset by one pixeL The calculated image appears as a rim of dark and light pixels, similar to a pattern of cortical activation. (C) Graph of the signal intensity (arbitrary units) of the baseline, and difference images along the line indicated in A and B. Note that a single pixel shift can appear as a large increase in signal Typical activation signals would be in the order of 3-15%. these vessels from the functional-image analysis, thereby mitigating this problem somewhat. Theoretical and experimental work by Fisel and coworkers 47 and later by Weisskoff and colleagues48 suggest that it might be possible to 'tune' the sensitivity of the MR method to vessels of a certain size range, such that signals from microvasculature less than a few tens of microns are more effective at modulating the MR signal than are large vessels. This feature arises when so-called 'spin echo' as opposed to 'gradient echo' MR methods are used. While the latter methods are sensitive to magnetic-field inhomogeneities within voxels (as described above), the former show large signal changes only when the protons are able to diffuse a relatively large distance compared with the blood-vessel size. In the time scales appropriate for MR imaging, the protons can only diffuse a distance comparable to the size of the capillaries. Therefore, these microscopic vessels will have a disproportionately large effect on the MR signal. Even with these limitations, fMRI appears to have excellent spatial sensitivity compared with other functional neuroimaging methods. As shown in Fig. 2A, the activation maps conform to the basic shape of the cortical surface, at least in primary visual 274

cortex. It would be reasonable to anticipate a resolving power of 1-2 mm However, there is still considerable work to do, both theoretical and practical, to understand the limiting spatial resolution of fMRI. Sensitivity. Among the most important advantages of fMRI is its ability to detect relatively large signal changes in single individuals, elicited by a wide variety of stimuli and cognitive tasks. Because of this, the activation maps of multiple individuals need not be combined to achieve sufficient sensitivity, and it is therefore not necessary to transform the coordinate system of one brain to conform to that of another. It would be difficult to overstate the advantage of this; neither the morphological or functional topography of the brain will be identical across individuals and therefore the combination of spatial data across subjects results necessarily in a reduction of signal and obfuscation of individual differences. The value of single subject analysis is well demonstrated in Watson and colleagues' recent PET study of area V5 in the human 49. There are many unanswered questions regarding the sensitivity of fMRI. Only a few reports exist, for example, assessing the magnitude of the fMRI response as a function of stimulus intensity, be it the brightness of a flashing light or the complexity of a cognitive task. As in most functional PET, EEG, or magnetic encephalography (MEG) studies, fMRI responses are presented typically as the normalized difference in signal intensity between control and activated conditions. Interpretation of such results assumes a graded response to the stimulus conditions. The contribution of a brain region that activated 'all or none' in a task might be difficult to assess, or even detect. A more subtle question (common between fMRI and PET, and perhaps MEG and EEG) concerns the relationship between the magnitude and extent of cortical activation. Being generally SNR-limited, these techniques are much better at detection of large activation of a small region than of small activation of a large region. This introduces a bias into the results and their interpretation when finding localized processing centers. Many questions of this sort will challenge the field, for example, how do we handle the effects of training; and will the activation of a region be systematically reduced (to the point of invisibility) with repeated or continued exposure? Experimental

design

and

statistical

issues.

Traditional statistical analysis of PET activation images involves averaging across a group of subjectsS°'51; this has been used to improve the SNR of the blood-flow response and to localize the results, usually to some common co-ordinate system52,53. One of the great advantages of fMRI is its built-in localization power, and the high SNR, enabling detection of change within a subject. The disadvantage of a single-subject approach is that there is no obvious way to combine data across subjects to establish reliability of the results. Statistical approaches used in fMRI generally include normalization of response-to-baseline variance, and performing Student's t tests of activation TINS, VoL 17, No. 7, 1994

blocks versus rest. This approach has the same statistical effect as PET in which responses within an Brain activation condition are averaged, -1 - effectively losing temporal resolution. Another approach models the fMRI response to repeated ~.~ Map - 2 -rest-activation cycles as a sinusoid l:: (effectively fitting the first ¢) -3-moment - a first-order exponenN tial)3°; this approach appreciates .~ Column that the protocol alternates bec:~ Layer- 4 - 0 tween periods of rest and actiu vation but otherwise closely Neuron - 5 resembles the Student's t test design. Both approaches assume Dendrite ° 6 -an a priori model of cortical activation: that blood flow increases during activation and decreases Synapse - 7 "" during resting in the relevant I I I I I I I I I I I brain regions. There is some -3 -2 -1 0 1 2 3 4 5 6 7 evidence that a simple increase in ms s min h d blood flow is one possible response type only; a change in the Non-invasive Invasive Log time (s) variance of the signal intensity without a change in the mean intensity would not be detected with the mean-comparison ap- Fig. 4. The temporal and spatia/ resolution of methods for the study of brain function. The temporal proach, yet this response pattern and spatial resolution of methods for the study of brain function, in this diagram, are related to the has been observed 26. An alter- size scale of neural features and to the 'invasiveness' of the methods. Abbreviations.. MEG, native statistical method, devel- magneto-encephalography. ERP, event related potentials; fMRI, functional magnetic resonance oped as an fMRI method by imaging,, and PET, positron emission tomography. Adapted from Ref. 59. Weisskoff and co-workers (pers. commun.) at the Massachusetts General Hospital the quiescent MR signal appear to be larger at NMR center, compares activation and rest periods higher fields 3,12,13,s4. Thus, the contrast sensitivity using the Kolmogorov-Smirnov statistic, which is that can be achieved does not appear to depend on nearly as sensitive to changes in the mean as the field strength as strongly as might otherwise have Student's t test but which also detects changes in been hoped. Because the cost of the MR instrument skew and variance. Such a method (or some variant) increases rapidly with increases in field strength, this might be better suited to exploratory brain-imaging is a crucial issue in the design of practical, dedicated methods where the local response change is not fMRI units. known beforehand. Head motion. The high spatial resolution of MRI, Where data across subjects have been combined, coupled with its high intrinsic contrast, results in the very few studies have tested the reliability of localized task-related activation. One approach applies a non- disadvantage that, when activation-related signal parametric measure of the number of subjects show- changes are very small, even slight misregistration creates significant artifacts following baseline subing activation of a cortical region, and the number of traction. Head motion not only reduces SNR in regions activated in each hemisphere as a measure activated regions but also produces spurious pseudoof lateral asymmetry 32. Reference to a common activations, especially at borders at the edge of the coordinate system, although possible, has been little brain, and between large fissures. Since there is a used in fMRI studies. need for stringent head motion control, the choice of subject responses available for measurement is Problems limited. Experiments focused on primary sensory Field strength. From the basic principles of magsystems can rely on the stimulus, such as auditory netic resonance, we expect the magnitude of the input or visual flashes, to produce activation without MR signal to increase with greater magnetic-field strengths, offering a net SNR advantage at higher the need for a behavioral response. In motor fields. Furthermore, because the magnitude of the experiments, small hand or foot movements need magnetization difference between oxyhemoglobin not elicit excessive head motion. Nevertheless, and deoxyhemoglobin should increase with field studies of higher cognitive functions such as strength also, researchers have suggested that the language and memory might be severely curtailed, SNR of the fMRI method might scale with the since speaking aloud is difficult to accomplish square of the magnetic field. Both magnitude without significant head movement. One alternative differences and the spontaneous signal variation of is to have subjects perform a speech task covertly, as in silent word generation; some have found this m

TINS, Vol. 17, No. 7, 1994

275

sufficient 32'3~ while others have failed to show Selected references 1 Broca, P. and Brown-Sequard, C. E. (1855) Proprietes et expected activation when performing the task fonctions de la moelle epiniere: rapport sur quelques exsilently55. Covert performance requires a high level periences de M. Brown-Sequard: lu a la Sod#t~ de biologle le of co-operation from subjects, and while this might 21 juillet 1855, Bonaventure et Ducessois be feasible for highly motivated (for example, paid) 2 Pauling, L. and Coryerl, C. D. (1936) Proc. Natl Acad. 5ci. USA 22, 210-216 normal volunteers, it is less feasible when studying 3 Thulborn, K. R., Waterton, J. C., Matthews, P. M. and Radda, patients or children. Furthermore, to determine K. (1982) Biochem. Biophys. Acta 714, 265-270 whether the study was successful (that the subjects 4 0G. g a w a , S. and Lee, T. M. (1990) Magn. Reson. Med. 16, performed the task), it is necessary to know the 9-18 correct answer (for example, this paradigm pro5 Turner, R., Le Bihan, D., Moonen, C. P., Despres, D. and Frank, J. (1991) Magn. Reson. Med. 1, 159-166 duces activity in inferior frontal gyrus), not an ideal 6 Kwong, K. eta/. (1992) Soc. Magn. Reson. Med. 11thAnnu. requirement for performing original research. Meeting Abstr. 301 Ideally, a concurrent, observable and measurable 7 Fox, P. T., Mintun, M. A., Raichle, M. E. and Herscovitch, P. behavioral response, such as a yes or no bar-press (1984) J. Cereb. Blood Flow Metab. 4, 329-333 8 Grubb, R. L., Raichle, M. E., Eichling, J. O. and Ter-Pogossian, response, measuring accuracy or reaction time, M. M. (1974) Stroke 5, 630-639 should verify task performance. 9 Rosen, 8., Belliveau, J. and Chien, D. (1989) Magn. Res. It might be hoped that a time series of images Q. 5, 263-281 based retrospectively on surface features 56, on 10 Belliveau, J. W. etaL (1991) Science 254, 716-719 cortical landmarks 57 or on overall image intensity 58 11 Fox, P. T. and Raichle, M. E. (1986) Proc, NaflAcad. Sci. USA 83, 1140-1144 could be registered; such methods have met with 12 Kwong, K. K. eta/. (1992) Proc. Natl Acad. Sci. USA 89, some success. However, these approaches are 5675- 5679 limited in that the inherent contrast in MR images, 13 Ogawa, S. et a/. (1992) Proc. Nail Acad. Sci. USA 89, for example, between gray and white matter, can be 5951-5955 quite large so that adjacent pixels might differ in 14 Edelman, R., 5ievert, B., Wielopolski, P., Pearlman, J. and Warach, S. (1994) Soc. Magn. Reson. 1st Annu. Meeting signal intensity by more than 20%. Consequently, 301 misregistrations of a fraction of a pixel can swamp 15 Abstr. Jezzard, P. eta/. (1993) Soc. Magn. Reson. Med. 12th Annu. the functional contrast (Fig. 3). Meeting Abstr. 1392 Perspective The temporal and spatial resolving power of a variety of methods for the study of brain function is depicted in Fig. 4. When fMRI is added to this framework, it seems to provide a satisfying level of spatial resolution, near to that of cortical columns, but a disappointing (by neural-processing standards) temporal resolving power of seconds. In addition to the resolution axes, this Fig. superimposes 'invasiveness', that is, the risk of harm to the subject, for each method. Here, fMRI holds a special position of apparently complete safety (with the exception of pacemakers and certain metal implants). Using fMRI, it will be possible to perform longitudinal studies on individual subjects advancing substantially the practical spatial resolution of functional imaging and enabling vastly more complex experimental designs to be implemented. The mapping of cortical and subcortical function in the human brain will ultimately require methods having the appropriate balance of temporal and spatial resolution, coupled with risk to the subject that is low enough to justify repeated experimentation on normal volunteers. Furthermore, the absolute locus of activation, and its relation to anatomical structure must be known and, ideally, the temporal relationship of its activation to that of other areas involved in processing of the same cognitive or sensory information should also be known. Functional MRI has moved us closer to achieving these goals. Though few neuroscientific research groups will be able to afford MR devices of their own, with thousands of installed units, and intensive creative effort, fMRI will have an active and expanding role in the understanding of brain function. 276

16 Bandettini, P. A., Wong, E. C., Hinks, R. S., Tikofsky, R. S. and Hyde, J. S. (1992) Magn. Reson. Med. 25, 390-397 17 DeYoe, E., Neitz, J., Bandettini, P., Wong, E. and Hyde, J. (1992) Soc. Magn. Reson. Med. 1 lth Annu. Meeting Abstr. 1824 18 Hennig, J., Ernst, T., Speck, O. and Laudenberger, J. (1993) Soc. Magn. Reson. Med. 12th Annu. Meeting Abstr. 12 19 Blamire, A. S. O. eta/. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 1821 20 Frahm, J., Bruhn, H., Merboldt, K. and H~nicke, W. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 306 21 Frahm, J., Bruhn, H., Merboldt, K. and H~nicke, W. (1992) 5oc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 1820 22 Menon, R. eta/. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 309 23 Turner, R. et aL (1992) Soc. Magn. Reson. Med. 11th Annu, Meeting Abstr. 304 24 Kwong, K. eta/. (1992) 5oc. Neurosci. Abstr. 18, 1265 25 Stern, C. eta/. (1992) Soc. Neurosci. Abstr. 18, 1265 26 Stern, C., Kwong, K., Belliveau, J., Baker, J. and Rosen, B. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 1821 27 Rao, S. eta/. (1993) Neurology43, 2311-2318 28 Benson, R. eta/. (1993)Soc. Magn. Reson. Med. 12th Annu. Meeting Abstr. 1398 29 Blamire, A. M. et al. (1992) Proc. Nail Acad. Sci. USA 89, 11069-11073 30 Bandettini, P. A,, Jesmanowicz, A., Wong, E. C. and Hyde, J. S. (1993) Magn. Reson. Med. 30, 161-173 31 Petersen, S. E. and Fiez, 3. A. (1993) Annu. Rev. Neurosci. 16, 509- 530

32 Cuenod, C. eta/. (1993)Soc. Magn. Reson. Med. 12thAnnu. Meeting Abstr. 1414 33 Hinke, R. M. et a/. (1993) NeuroReport 4, 675-678 34 McCarthy, G., Blamire, A. M., Rothman, D. L., Gruetter, R. and Shulman, R. G. (1993) Proc. Nat/Acad. Sci. USA 90, 4952 -4956 35 Rueckert, L. et aL J. Neuroimagmg (in press) 36 Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M. and Raichle, M. E. (1988) Nature 331, 585-589 37 Le Bihan, D., Turner, R., Jezzard, P., Cuenod, C. and Zeffiro, T. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 311 38 Ingvar, D. H. and Philipson, L. (1977) Ann. Neurol. 2, 230-237 TINS, Vo/. 17, No. 7, 1994

39 Fox, P. T., Pardo, J. V., Petersen, S. E. and Raichle, M. E. (1987) Soc. Ncurosci. Abstr. 13,398.10 40 Kosslyn,S. M. (1981) Cognition 10, 173-179 41 Frostig, R. D,, Lieke, E. E., Tso, D. Y. and Grinvald, A. (1990) Proc. Natl Acad. Sci. USA 87, 6082-6086 42 Tso, D. Y., Frostig, R. D., Lieke, E. E. and Grinvald, A. (1990) Science 249, 417-420 43 Powell, T. and Mountcastle, V. (1959) Bull. Johns Hopkins Hosp. 105, 133-162 44 Hubel, D. and Wiesel, T. (1969) Nature 221, 747-750 45 Lai, S. etaL (1993) Magn. Reson. Med. 30, 387-392 46 Laub, G. A. and Kaiser, W. A. (1988) J. Comput. Assist. Tomogr. 12, 377-382 47 Fise[,C. R. eta/. (1991) Magn. Reson. Med. 17, 336-347 48 Zuo, C., Boxerman, J. and Weisskoff, R. (1992) Soc. Magn. Reson. Med. 11th Annu. Meeting Abstr. 866 49 Watson, J. etaL (1993) Cerebral Cortex 3, 79-94 50 Fox, P. T., Mintun, M. A., Reiman, E. M. and Raichle, M. E. (1988) J. Cereb. Blood Flow Metab. 8, 642-653 ....

....

51 Fox, P. T. and Pardo, J. V. (1991) Ciba Found. Symp. 163, 125-140 52 Talairach,J. et aL (1967) Atlas d'Anatomie Stereotaxique du Telencephale, Masson 53 Fox, P. T., Perlmutter, J. S. and Raichle, M. E. (1985) J. Comput. Assist. Tomogr. 9, 141-153 54 Frahm, J., Merboldt, K. and H~nicke, W. (1993) Magn. Reson. Med. 29, 139-144 55 Blamire,A. et aL (1993) Soc. Magn. Reson. Med. 12th Annu. Meeting Abstr. 1413 56 Pelizzari,C. A., Chen, G. T., Spelbring, D. R., Weichselbaum, R. R. and Chen, C. T. (1989) J. Comput. Assist. Tomogr. 13, 20-26 57 Evans,A. C., Beil, C., Marrett, S., Thompson, C. J. and Hakim, A. (1988) J. Cereb. Blood Flow Metab. 8, 513-530 58 Woods, R. P., Mazziotta, J. C. and Cherry, S. R. (1993) J. Comput. Assist. Tomogr. 17, 536-546 59 Belliveau, J., Cohen, M., Weisskoff, R., Buchbinder, B. and Rosen, B. (1991) J. Neuroimaging 1, 36-41 ........

wpolnt

Theenigma of myelin-associatedgrowthinhibitorsin spontaneouslyregeneratingnervoussystems T o m e r Sivron and M i c h a l Schwartz Recent results shed new light on how some nervoussystems can regenerate after injury while others cannot. Until recent/y, it was widely believed that the mare difference between systems that regenerate and those that do not hes in the normal state of their permlssivenessto the regeneratingaxons. Thus, while nonregenerafire systems, such as the rat optic nerve, were shown to contain myelin-associatedgrowth mhlbitors, regenerativesystems, such as the hsh optic nerve, were thought to have no such inhibitors. However, it hasnow beendemonstratedthat spontaneouslyregenerating systemsdo containgrowth inhlbitors, though their levelsseem to be lower than in nonregenerative systems. The mare difference, however, appears to reside m the system'sresponseto in/ury. Th~s artlcle discussesthe involvement of myelin-associatedgrowth mhlbitors in the spontaneously regeneratingnervoussystem of fish, tracesthe apparentdiscrepancy,and showshow it has been resolved

recently. Axons in the CNS of mammals do not spontaneously regenerate after injury, unlike those of phylogenetically lower vertebrates, such as fish and amphibians, or the peripheral nerves of mammals ~-7. Since injured mammalian CNS axons can, under certain conditions, grow for considerable distances 8-1°, it is generally believed that the environment surrounding the axons plays a crucial role in determining regenerative capacity. In the past, the ability of nerves to regenerate has been correlated with the ability of sections of these nerves to support neuronal attachment and axonal growth co-cultured in vitro. Thus, sections of rat sciatic nerve and fish optic nerve, both of which are capable of regeneration, support neuronal attachment and axonal growth of neuroblastoma cells or embryonic neurons, while sections of the nonregenerative rat optic nerve do not 1~-~3. Concurrently, myelin-associated growth inhibitors have been shown to be present in the mammalian CNS (Refs 14 and 15). It was therefore proposed that in spontaneously regenerating systems such inhibitors are absent 14,~6,17 and, accordingly, that the differences in the ability of some nervous systems to TINS, Vol. 1Z No. 7, 1994[

regenerate is largely due to the presence or absence of myelin-associated growth inhibitors. This proposal, however, failed to explain several experimental findings, among them the observation that when fish retinal axons encounter rat CNS myelin or mature GalC (galactocerebroside)-positive rat oligodendrocytes in vitro, they either circumvent them or collapse ~7. This suggests that growth cones of fish axons possess not only receptors that recognize the rat myelin-associated growth inhibitors, but also the intracellular machinery to respond to them, It seems unlikely that fish growth cones would possess such receptors or machinery unless the fish axons encounter similar inhibitors in their own surroundings. Moreover, fish retinal axons collapse in response to molecular cues in the optic tectum 18. Such growth inhibition or repulsion is thought to be part of the developmental guidance mechanism ~9.

TomerSivronand MichalSchwartzare at the Dept of Neuroblo/ogy, The WeizmannInstitute of Science,PC)Box 26, 76100Rehovot, Israel

Myelin-associated growth inhibitors similar to those of rat CNS are present in fish CNS Further evidence that seems to contradict the above proposal comes from studies showing that myelin of fish CNS is barely permissive to growth of adult fish axons 17'2° (Fig. 1). These investigations demonstrated that fish myelin, although somewhat more growth permissive than rat CNS myelin is, nevertheless, far less growth permissive than either rat or fish CNS membranes. Bastmeyer and colleagues 17 observed that fish myelin permitted some axonal growth, and although much less permissive than some other substrates, it was considered as devoid of growth inhibitors. However, it was recently shown unequivocally that IN-1 monoclonal antibodies, which were directed originally against rat-myelin inhibitors 21, neutralize the growth inhibition of fish myelin 2° (Fig. 2). This demonstrates not only that fish myelin does contain growth inhibitors but also that its inhibitors are similar to or identical with those of rat CNS myelin. Accordingly,

© 1994.EfsevierScierlceLtd

277