High‐Resolution Imaging in Epilepsy

The abnormalities causing intractable epilepsy are frequently small with little contrast from adjacent normal brain tissue. Detection of such abnormalities is enhanced by using specialized imaging protocols (1,2). Several groups have found conventional magnetic resonance imaging (MRI) studies to be ∼90% sensitive and 85% specific in the diagnosis of hippocampal sclerosis in a series of epilepsy patients undergoing temporal lobectomy (3,4). However, this still leaves ∼10% of cases with hippocampal sclerosis that are undetected with standard MRI studies. Malformations of cortical dysgenesis are even more difficult to detect, particularly without using high-spatial-resolution imaging (1,5,6). Based on the experience of our groups and others, many patients with cortical dysgenesis may be going undetected by conventional MRI with standard head coils (6–9). This is the rationale for using high-spatial-resolution techniques to image seizure patients. However, simply decreasing voxel size to increase spatial resolution is often not an attractive option, because decreasing voxel size with constant imaging time results in a decrease in the image signal-to-noise ratio (SNR), which in turn may result in a loss of tissue contrast. In this section, three techniques for improving the spatial resolution, image contrast, or both, of MRIs are discussed: use of phased-array surface coils, image averaging, and high-field-strength MRI. The first technique discussed for increasing spatial resolution without loss of SNR is the use of receive-only phased-array surface coils (Fig. 1). The phased-array dual-coil design allows a wider area of coverage than routine surface coils. Wald et al. (10) found that compared with the conventional quadrature head coil, the SNR at the brain surface was increased sixfold with four-coil phased-array surface detectors. The SNR decreases with distance from the surface coil, and thus the SNR was increased 1.67 times at the hippocampus compared with volume head coils (11), which is approximately equal to increasing the number of signal acquisitions by a factor of 3. This allows one to increase spatial resolution while maintaining a reasonable (or increased) SNR without requiring a concomitant increase in time and signal acquisitions (number of excitations). It also allows one to decrease acquisition time of long acquisition sequences by decreasing the number of signal averages needed for an adequate image based on SNR. Comparison of a standard quadrature head coil (A) and two 10-cm phase-array surface coils (B). Because the head coil is immobile after being locked into the stretcher, the head can be fixed (with tape and wedges) within the head-coil mechanism. The subject's head is more mobile with the surface-coil imaging. However, the head can be immobilized if placed in the head hold (not shown, without the head coil itself) and then secured with wedges and tape. Although the SNR is improved compared with that in head-coil imaging, surface-coil imaging can be associated with a number of shortcomings. Characteristic of all surface coils, there is a rapid decrease in signal intensity from the brain surface adjacent to the coil to the center of the head, making viewing and interpretation of images more difficult; when the brain surface is windowed optimally, deeper brain sections appear dark. However, this problem can be partially surmounted by saturating out the calvarium (and peripheral brain), by using an algorithm designed to compensate for the change in signal intensity with distance from the coil, or by viewing the images on an interactive monitor such as a picture-archiving computerized system (PACS) viewing station. Signal loss also may occur at the bregma. Because of the configuration of the coil, imaging cannot be performed with the coils perpendicular to the z-axis of the magnet (i.e., surface coils on the head vertex), so that the surface of the brain at the bregma cannot be placed as close to the coil as can the lateral portions of the cortex. However, the coils are flexible, and they also can be oriented ∼15–20 degrees from the z-axis of the magnet without significant loss in signal reception. This helps to optimize imaging of the bregma cortex. Other potential drawbacks associated with surface-coil imaging include limited brain coverage, potential motion problems, and an overall increase in scanning time. Because of the size of the surface coils (usually 10 cm with phased-array coils), only a limited amount of brain can be evaluated in the adult. In infants, conversely, imaging of the entire brain may be possible. Head movement may be a greater problem with surface coils than with head coils, because the head is more difficult to secure and immobilize. However, one method for overcoming this problem is to wedge the coils between the head and the sides of the head-hold used with the standard head coil supplied with our MR system (but with the head coil removed). Combining this with adhesive tape placed to secure the coils and the patient's forehead works very well to stabilize both the patient and the coils. Another limitation of surface-coil imaging relates to the length of the MR examination. The time needed to perform an imaging study may actually be increased, if both head coil and surface coil imaging are used. However, the added detail and information obtained with such a combined imaging protocol may be worth the extra time. Surface coil imaging may be useful for imaging (a) the infant brain; (b) the hippocampus to evaluate hippocampal sclerosis 2, 3; (c) secondary findings of hippocampal sclerosis, such as atrophy of the fornix; (d) the cortex to investigate cortical dysgenesis 4, 5; (e) surface lesions to distinguish intraaxial from extraaxial abnormalities (Fig. 6), and (f) peripherally based head-coil-imaging findings to better characterize these findings (Fig. 7)(6,12–15). The Yale group has found surface-coil imaging most useful for characterizing head-coil findings and assessing for cortical dysgenesis, whereas the University of Washington group has found it very useful for characterizing hippocampal pathology as well. Hippocampal sclerosis. A: Coronal high-resolution medullary inversion recovery suppression (MIST) image of patient with intractable temporal lobe seizures. There is minimal decrease in size of left hippocampus. There are, however, several subtle findings involving the left hippocampus (as annotated on the insert of the left hippocampus): thinning of CA1 region (white arrow) along with loss of normal architecture of the stratum radiatum and mild signal increase in CA1 and CA4 regions (black arrow). Compare this appearance with the normal right hippocampus in A and B. B: Referential anatomy is provided by this digitally enhanced and altered close-up of the right hippocampus. The stratum radiatum (*) in the white matter provides the hippocampus with an internal architecture, separating the CA1 (1) region from the CA4 (4) region. Neuronal loss and gliosis that occurs with hippocampal sclerosis is found most frequently in the CA1 and CA4 regions (f, fornix; t, temporal horn). Resection of left hippocampus revealed histologically proven hippocampal sclerosis. Hippocampal sclerosis. Coronal high-resolution phased-array coil scans of a patient with intractable temporal lobe seizures and histologically proven left hippocampal sclerosis. There is no significant decrease in hippocampal size, and the fast spin-echo T2-weighted image (A) does not show definite abnormality. However, the medullary inversion recovery suppression (MIST) image (B) shows subtle findings consisting of loss of normal architecture of the stratum radiatum on left compared with right side and an increase in signal intensity along CA1 region (arrowheads in insert). Note that loss of hippocampal architecture sometimes occurs in a normal hippocampus on a single slice, and it is assessed as being abnormal only if it occurs on two or more consecutive images, as occurred in this case (but other slices not shown). Cortical dysgenesis. This young girl had electroclinical findings of medial frontoparietal epilepsy. A: Axial spoiled gradient-recalled echo (SPGR) images in the head coil were interpreted as normal. Even in retrospect, it is difficult to detect the region of cortical dysgenesis (arrow) medial to the central sulcus (C). B: Inversion recovery image (video inverted) using the phased-array surface coil demonstrates blurring of the gray–white junction and thickening of the cortex (arrow) involving the posterior paracentral lobule, posteromedial to the central (C) sulcus. Cortical dysgenesis. This young male patient had left frontal seizures. A: Axial spoiled gradient-recalled echo (SPGR) image in the head coil demonstrates abnormal topography of the posterior left frontal lobe gyri and sulci. However, it is difficult to assess the actual cortical thickness. B: This axial SPGR image was produced with the same parameters as in A, except that phased-array surface coils were used. The improved signal-to-noise ratio (SNR) allows better definition of the cortex; one can see the irregularity of the dysplastic cortex on the left side. C: Contrast between gray and white matter occurs with this axial inversion recovery (video inverted) with a T1 of 350 (to null the white matter) produced with surface-coil imaging. The irregularity of the dysplastic (D) cortex is shown to better detail. A magnified view of this is depicted in D. Because of the relation of primary motor regions to this abnormal area, the patient underwent multiple subpial transections; no histologic correlation was possible. D, dysplastic sulci and cortices; SF, superior frontal sulcus; PrC, precentral sulcus; C, central sulcus; PoC, postcentral sulcus. Extraaxial lesion. This male patient without epilepsy had a peripherally based abnormality. A: This axial T2-weighed image demonstrates the peripherally located abnormality. Although it appears to be extraaxial, there was still doubt based on the conventional head-coil imaging. B: This surface-coil inversion recovery image definitely places this abnormality within the extraaxial space. Note the cerebrospinal fluid (large arrow) between the tumor (T) and cortical gray matter (small arrows). Surgery confirmed the extraaxial location of this unusual chondroma. Characterization of head-coil findings. A computed tomography (CT) scan was performed for a poorly described ictal event in this 2-year-old girl. The CT showed a left frontal abnormality. A: Coronal T2-weighted fast-spin-echo image confirms a left frontal lobe hyperintense signal abnormality (arrow), but does not fully characterize this signal change. B: Coronal inversion recovery sequence using surface coils again demonstrates the region of signal change, but higher resolution allows one to distinguish the multiple hyperintensities that combine to form the signal changes in A. These signal changes are isointense to cerebrospinal fluid on all pulse sequences and radiate perpendicular to the cortex, consistent with perivascular Virchow–Robin spaces. In a prospective study of 25 patients, Grant et al. (6) found that surface-coil imaging resulted in detection of new findings in three (12%) cases not found using head coils. These cases were a neoplasm, a cyst, and heterotopia (none of which was confirmed by histology). In four (16%) cases that were indeterminate with head-coil imaging, the surface coil changed the diagnosis to cortical dysplasia. Overall, surface-coil imaging reclassified a normal head-coil image or findings on a head-coil image in 15 (60%) of 25 cases and depicted findings of polymicrogyria better in an additional single case (additional 4%, or 64% in total). The authors used a coronal three-dimensional (3D) volumetric gradient-echo sequence with parameters of 33/7/0.75/35–45 (TR/TE/signal averages/flip angle), 256 × 224 matrix, 16- to 18-cm field of view, 0.7- to 1.0-mm-thick partition size, and 124 partitions. The resulting voxel size (0.31–0.57 mm2) was one third of that used with their head-coil images. Additional fast spin-echo sequence is used to provide T2 information with the following parameters: 4,000/112/4 (TR/TE/signal averages), 512 × 512 matrix, 8–16 echo-train length, 12- to 18-cm field of view, and 3-mm-thick sections. A review of the experience at Yale University with 22 patients found similar results. Ten subjects had negative MR studies using the standard head coil, and 12 had MR findings that needed further clarification. Of the 10 with normal head-coil images, seven had well-localized (lobar) epilepsy by electroclinical criteria, whereas three had poorly localized seizures. Surface-coil imaging detected cortical dysgenesis not seen with the head coil in one case: one of the seven lobar-localized cases (Fig. 4). Of those 12 subjects with abnormal head-coil images, the surface-coil sequences better characterized the MR abnormality and answered the clinical question in 10 (83%) of the 12 patients 5, 7. Overall, surface-coil imaging allowed the new detection of cortical dysgenesis in one (5%) of 22 cases or one (14%) of seven cases with electroclinical lobar seizures and normal conventional MRI. Surface-coil imaging added value to the examination (by either detection of the abnormality or better characterization) in 11 (50%) of the 22 cases. Imaging protocols consisted of 3- to 4-mm inversion recovery high spatial resolution sequences and/or 1- to 1.6-mm inversion recovery prepped spoiled-gradient-echo sequences. Some groups, such as the group at the University of Washington (UW), use phased-array coil imaging as part of their routine intractable epilepsy MRI protocol. Because of the increased potential for head movement during the latter portion of an MRI study, the UW group begins with high-resolution temporal lobe imaging with the phased-array coils. Coronal imaging of the hippocampus and parahippocampal structures is done using a fast spin-echo (FSE) T2, an FSE inversion-recovery sequence with a T1 of 350 ms to null signal from normal white matter (MIST or medullary inversion recovery suppression) (15). After these acquisitions, the patient is placed in the head coil, and the study is completed with whole-brain imaging using axial fluid-attenuated inversion recovery (FLAIR), FSE T2 and T1 sequences. Postgadolinium images are acquired only if a structural abnormality is identified on the noncontrast images. Both the Yale and Washington groups use inversion recovery prepped fast spoiled gradient recalled echo (IR-FSPGR) and/or a MIST inversion recovery sequence to evaluate the cortex for cortical dysgenesis (14). With an inversion time of 350 ms, the white-matter signal is suppressed, yielding an exquisite view of the gray matter–white matter junction. In summary, surface coil imaging is most beneficial when used in selective circumstances. One situation consists of a case of focal or lobar electroclinical localization in face of a normal head-coil MR study, provided that the MR was a high-quality study. Surface-coil imaging is frequently helpful for characterizing questionable or subtle MRI abnormalities detected by routine head-coil imaging, such as cortical thickening. The second technique discussed here is image averaging. In MRI, SNR is proportional to the square root of the number of signal averages. The SNR gained by signal averaging can be used to improve image contrast, or it can be “traded-off” to increase spatial resolution by decreasing voxel size. Signal averaging is routinely done in clinical MRI by performing back-to-back repetitions of the imaging pulse sequence. However, this approach is limited for studies with longer scan times by the inability of most patients to remain motionless for more than several minutes. Researchers developing digital atlases for brain mapping have combined scans of different subjects into a single image using interimage registration and averaging (16). This same approach can be used to improve the SNR of the MRI of a single patient; several MRI studies acquired at different times may be spatially registered and then combined into a single imaging study that has greater SNR than any one individual study. For practical purposes, initial work involved mainly the repetition of four T1-weighted GE scans. This requires only an additional 30 min of scan time. Postprocessing is performed with automated image registration (AIR) (17). This use of registration for averaging has been confirmed to provide a true improvement of SNR by twofold. Of the four scans, the subsequent three scans are co-registered to the geometric space of the first scan. Averaging is performed with a program provided in the AIR package called “soft mean.” In an initial test of methods using a normal control subject, eight scans were performed, a number that allowed a verified increase in SNR of 2.83. All example images from registration and averaging illustrate an obvious improvement in overall visual quality. Mostly this appears to be due to an increase in contrast, but also to some degree from a decrease in artifact-related background noise, particularly that from motion disturbance. A better delineation of fine detail is clearly appreciated in gray–white matter interfaces. In Fig. 8, native and averaged scans for a normal subject illustrate the obvious visual improvement. In particular is the questionable disturbance of internal architecture of the left hippocampus in the native scan versus the clearly normal hippocampal internal architecture visualized on the average scan. Figure 9 shows a neocortical, gray matter–based gliotic lesion. With the marked improvement in contrast on the average scan, the equivocal focal abnormality from the native scan becomes more conspicuous and confidently abnormal. Figure 10A illustrates the marked improvement in noise from a patient with a small focal area of cortical dysplasia. Comparisons between native and average scans in Fig. 10B show the result of improved SNR on the delineation of this subtle gray-matter abnormality. Finally, Fig. 11 is from a case of truly cryptogenic focal cortical dysplasia. Although this average scan provides higher image quality, it still does not reveal any abnormality. A co-registered fluorodeoxyglucose PET scan reveals unequivocal focal relative hypometabolism in the epileptogenic region, which was confirmed to show disorganized cortical pyramidal cell architecture and abnormal neurons on histopathology. This final case emphasizes the limitation that remains in the ability of this technique to detect microscopic-only pathology. Average of eight spoiled gradient-recalled echo (SPGR) scans: note the questionable appearance of disturbed internal architecture of the left hippocampus in the native scan (left image). Visualization of the averaged scan (right image) correctly reveals a well-defined cornu ammonis and absence of hippocampal abnormality. Right occipital neocortical lesion of uncertain significance in a 54-year-old woman with partial seizures of right hemisphere origin. The focal gliotic grey–white matter junction lesion (white arrow) is difficult to discern confidently on the native (left) image. Improved contrast in the averaged scan (right) delineates an unequivocal focal lesion. Image set in conventional grey scale reveals the result of improved signal-to-noise ratio on the delineation of a subtle grey-matter abnormality in the right insular and uncinate fasciculus region (white arrows). Right neocortical temporal lobe cryptogenic focal cortical dysplasia. Lesion is not detected with either native (not shown) or averaged magnetic resonance imaging scan (upper image slices). Focal defect in glucose uptake (white arrows) associated with the epileptogenic zone is visualized on co-registered fluorodeoxyglucose–positron emission tomography scan (lower image slices). Surgical histopathology revealed disorganized cortical pyramidal cell architecture and large abnormal neurons In summary, image registration and signal averaging can be easily implemented to improve delineation of subtle structural abnormalities on MRI. It is clear that enhanced conspicuity of abnormalities is achieved. This is especially the case with ambiguous or questionable gray matter–based lesions. Further work must address whether averaging-related enhancement of visual quality can be objectively demonstrated to improve diagnostic accuracy. It is emphasized that further research should systematically examine what other sequences and protocols might optimally benefit from averaging. The final technique discussed is use of high-field-strength MR magnets. High-field magnets are becoming more popular. The pursuit of increased image resolution and SNR has helped fuel a drive for higher and higher magnetic fields (3–8 Tesla) for human use. This effort has led to a revised terminology for MRI field strength. Many now refer to field strengths ≥3 T as “high field,” 1.0–2 T as “standard,” and <0.5 T as low field. Just slightly more than a decade ago, predictions about the practicality of very high field MRI were mainly pessimistic. Research during the past decade has shown, however, that very high magnetic fields can offer substantial benefits in many different areas of brain research. In the range from 1.5 to 7 Tesla, experimental results show an approximately linear increase in SNR with increasing field strength. By decreasing voxel size, this improvement in SNR can be traded for increased spatial resolution without any loss of image SNR relative to standard-field-strength images (Fig. 12). Furthermore, for anatomic imaging, dynamic range in T1 contrast can be increased using new pulse sequences that exploit the SNR advantage of higher fields. Many technical problems remain to be solved, although these are often less severe than originally predicted, and new methods can sometimes be developed to circumvent them. Radiofrequency (RF) power deposition is typically the most significant limitation. As a result, some pulse sequences commonly used in standard MRI examinations must be modified for use at high field. This is particularly true for sequences that require high RF pulse duty cycles like FLAIR images or others that use a multi-180-degree echo read-out train. A final limitation is nonuniform RF deposition concentrated in the center of the volume at field strengths >3 T. T-1 weighted axial image at 7 T. Due to the increased SNR provided by imaging at 7 T, voxel size could be decreased to 0.07 mm cubed while retaining good image contrast. This is greater than an order of magnitude improvement in spatial resolution compared to high resolution imaging parameters at 1.5 T (typically no better than 1mm cubed for a 3D volumetric acquisition).