|Home | About | Journals | Submit | Contact Us | Français|
Partial removal of the anterior temporal lobe (ATL) is a highly effective surgical treatment for intractable temporal lobe epilepsy, yet roughly half of patients who undergo left ATL resection show decline in language or verbal memory function postoperatively. Two recent studies demonstrate that preoperative fMRI can predict postoperative naming and verbal memory changes in such patients. Most importantly, fMRI significantly improves the accuracy of prediction relative to other noninvasive measures used alone. Addition of language and memory lateralization data from the intracarotid amobarbital (Wada) test did not improve prediction accuracy in these studies. Thus, fMRI provides patients and practitioners with a safe, non-invasive, and well-validated tool for making better-informed decisions regarding elective surgery based on a quantitative assessment of cognitive risk.
The first reports of human functional brain mapping using MRI scanners appeared 20 years ago [1, 2]. Functional magnetic resonance imaging (fMRI) in its most common form, using endogenous blood oxygen-level-dependent (BOLD) contrast, is now practiced routinely at most medical centers. Despite extensive research and clinical experience, uncertainty persists over the use of fMRI in the presurgical evaluation for epilepsy. This article reviews evidence supporting the use of fMRI for predicting postoperative language and verbal memory deficits in patients undergoing elective anterior temporal lobe (ATL) surgery. This clinical setting continues to be the most common indication for the intracarotid amobarbital (Wada) test. Recent studies suggest that fMRI provides a valid noninvasive alternative to the Wada test for most patients.
Although the focus of this review is on lateralization of language and verbal memory functions, it should be noted that fMRI also provides detailed activation maps that can in some cases be used to guide surgical resections. For example, fMRI can localize primary and secondary motor areas, even in some patients whose brain anatomy has been profoundly distorted by developmental anomalies or mass lesions [3-8]. Together with diffusion tensor imaging (DTI) localization of corticospinal white matter pathways, these maps can be valuable in helping surgeons maximize a resection zone while avoiding critical motor areas . Similarly, fMRI can localize primary auditory, somatosensory, and visual cortex. In the case of visual cortex, the fMRI experiment can be designed to generate a retinotopic map showing the precise cortical representation of each region in the visual field [10, 11], allowing the surgeon to know with reasonable certainty what pattern of visual field loss will result from resection of a particular cortical zone. The utility of these sensory and motor activation maps, however, rests on well-established lesiondeficit relationships. The discovery of retinotopic maps, for example, was based on decades of careful observation in patients with focal occipital lesions , thus there is no doubt about the effects that can be expected from focal damage to these regions.
Compared to these relatively straightforward relationships, the relationships between focal lesions and specific language deficits are complex and incompletely understood. The traditional emphasis on Broca and Wernicke areas has given way in recent decades to a much more complex picture of the language system, with recognition that both production and comprehension of language involve widely distributed brain networks, including many regions outside the traditional Broca and Wernicke zones [13-21]. In addition, evidence suggests that the exact location of language areas varies from person to person [22, 23], perhaps accounting for some of the wide variation in aphasia outcome after focal lesions . Given this uncertainty and the highly distributed nature of language processes, important concerns have been raised about the meaning of activation foci identified by fMRI language experiments. Unlike with primary motor and sensory areas, the effect of removing an fMRI-defined “language area” is simply not known. Some areas identified by fMRI could participate in language functions but play a nonessential role. Because language tasks engage a variety of general cognitive processes, such as attention and working memory, some of the areas “activated” in an fMRI language study may represent these general cognitive processes rather than the language components of interest. Adding to the uncertainty surrounding fMRI language maps is the fact that there are many different task paradigms available, which produce markedly different patterns of activation (see [25-28] for examples of comparisons between paradigms). Thus, a brain area declared “not active” using one paradigm might very well turn out to be “active” using another. These legitimate concerns about the specificity and sensitivity of fMRI-defined language maps currently limit the usefulness of such maps for detailed surgical planning. Those who would use fMRI language maps in this way run two risks: sparing of “active” regions that are actually not critical for language, resulting in sub-optimal seizure control; and resection of critical language zones that are “not active” merely due to insensitivity of the particular fMRI protocol employed, resulting in post-operative language deficits. Only through carefully designed and systematic studies – in which resections are performed blind to the fMRI data, standardized procedures are used for assessing outcome, and quantitative measures are made of the anatomical and functional lesion – will the usefulness of fMRI language maps for planning surgical resections be determined.
Although in the author's opinion fMRI language maps should not yet be routinely used for planning resection boundaries, fMRI already has a clearly established role to play as an alternative to the Wada test. Temporal lobectomy is highly effective for seizure control [29-31], yet roughly half of patients undergoing dominant ATL resection experience postoperative language [32-37] or verbal memory decline [38-46]. The traditional role of the Wada test is to estimate the risk of decline by determining the patient's hemispheric dominance for language and memory. Risk assessment provides the patient and physician with additional information that can be useful in deciding whether to proceed with treatment in elective situations. This information can also be used to select high-risk patients for more invasive procedures such as electrical stimulation mapping.
The Wada test was originally developed to assess the risk of language decline in patients undergoing brain surgery , based on the assumption that operating on the language-dominant hemisphere entailed increased risk. Though the Wada test has been in use for over 50 years, until recently the relationship between Wada language asymmetry and postoperative language outcome had never been quantified. The historical reasons for this curious knowledge gap relate to the traditional view of language lateralization as dichotomous (left or right) or trichotomous (left, right, or bilateral). Under this schema, it was obvious that operating on a non-dominant hemisphere would be safer than operating on a language-dominant hemisphere. Several aspects of this formulation have changed in recent decades. First, language lateralization has come to be seen as a continuously graded rather than an all-or-none phenomenon, with relative degrees of dominance rather than distinct categories [48-54]. Thus, while the majority (~75%) of patients who undergo left hemisphere surgery for epilepsy are left-hemisphere dominant for language, there is variation within this group in terms of the degree of left dominance. This variability raises the question of whether graded degrees of language dominance are reflected in graded levels of risk. Second, quantitative neuropsychological evaluation for postoperative language deficits has become a more standard practice, resulting in a shift of the clinical focus away from prediction of severe aphasia (which is very rare after standard left ATL resection) and toward prediction of more moderate degrees of language decline.
Use of fMRI for predicting language outcome in epilepsy surgery is therefore motivated by two critical assumptions. First, it is assumed that patients show varying degrees of language (mainly naming) deficit after surgery, and that it is desirable to know before surgery what degree of decline can be expected. Second, it is assumed that the degree of decline will be related to the degree of language lateralization toward the surgical hemisphere. The goal of fMRI in this context is thus to provide a reliable and valid measure of language lateralization. Many fMRI language activation paradigms have been described, differing in the type of language stimuli, stimulus modality, language task, control stimuli, and control task used, raising the question of which of these paradigms, if any, is optimal. Though different paradigms have seldom been compared quantitatively, it is clear that they can produce very different, in some cases entirely non-overlapping, activation patterns. This variation is related primarily to the cognitive, sensory, and motor processes engaged by the tasks, and the degree to which the language and control conditions differ in engaging these specific processes .
Several simple criteria can be applied in assessing the usefulness of fMRI language paradigms. First, the pattern of activation obtained in healthy, right-handed adults should be lateralized to the left hemisphere, as almost all such individuals are left-hemisphere dominant for language [50, 56]. Second, the activation should be robust, i.e., it should be reliably obtained across individuals and in the same general brain regions. Third, there should be concordance between language lateralization measured with the fMRI paradigm and lateralization measured with other techniques, such as the Wada test, in the same individuals. Finally, it may be desirable that the paradigm produce activation in particular target brain regions. In the case of ATL surgery, for example, activation asymmetry in the temporal lobe might be more predictive of outcome than activation in the frontal lobe, thus a paradigm that activates the temporal lobe would have advantages over one that does not.
Lateralization of fMRI activation is typically expressed numerically in the form of a laterality index (LI). The first such LI was based on a simple count of the voxels that survived thresholding in each hemisphere . The formula (L - R)/(L + R), where L and R refer to the voxel counts in each hemisphere, yields a number that varies from +1 when all activated voxels are on the left side to -1 when all activated voxels are the right. LI values obtained with this method vary as a function of the threshold used for defining activated voxels, thus several authors have explored alternative asymmetry measures that do not require thresholding [27, 53, 54, 57-59]. No consensus regarding the optimal method for calculating activation asymmetry has yet emerged from these studies.
Figure 1 illustrates the critical importance of task selection in language fMRI studies. The figure shows average activation maps obtained while 26 right-handed subjects listened to spoken words and performed a semantic decision task [14, 28]. In the top panel, BOLD signal during this task is compared to a “resting” baseline. The activated regions are largely bilateral, including bilateral auditory, working memory, general executive, and attention networks. In the middle panel of the figure, the semantic decision task is compared to a nonlinguistic auditory control task. In this case the activated regions are strongly left-lateralized and include several left temporal, parietal, and prefrontal regions (indicated by blue arrows) that were not observed when the resting baseline was used. These data illustrate in dramatic fashion how activation patterns depend on the choice of control condition. In the second paradigm, the use of an active nonlinguistic control task “subtracts out” bilateral activation in early auditory, general executive, and attention networks, leaving activation in left-lateralized language networks. These results also demonstrate that many high-level language processing regions are active during the “resting” state and can only be observed when an active nonlinguistic control condition is employed [28, 60-64]. Graphs in the lower portion of Figure 1 show average activation volumes and laterality indexes for each task contrast, again illustrating the dramatic differences that can occur simply by changing the choice of control condition.
Many fMRI language paradigms have been compared to Wada language testing [25, 26, 49, 53, 58, 65-80]. These studies generally report high concordance rates, typically in the 80-90% range (for reviews, see [81, 82]). In assessing concordance, patients are usually assigned to categories such as “left dominant”, “right dominant”, or “mixed” on each test. The proportion of concordant cases depends strongly on how these arbitrary categories are defined.
With so many studies focusing on fMRI-Wada correlations, it is easy to forget that the actual aim of measuring language lateralization prior to brain surgery is prediction of language outcome. An fMRI procedure that reliably identifies patients at risk for postoperative naming deficits would be a valuable clinical tool, especially if the fMRI results added information over and above other available tests. Previous studies have identified demographic and behavioral variables that may predict outcome. For example, left ATL patients who develop seizures at an earlier age generally have a lower risk for postoperative language decline [34, 83, 84], presumably because earlier age at onset is associated with a higher probability of language shift to the right hemisphere . Better preoperative naming performance is associated with a higher risk for decline . It has long been assumed that Wada language testing is predictive of language outcome, though the actual evidence on this issue is limited to a few case reports [33, 47].
Sabsevitz et al.  studied 24 consecutively encountered patients undergoing left ATL resection. The fMRI paradigm used a contrast between an auditory semantic decision task and a nonlinguistic tone decision task (see Figure 1). A previous study had shown that asymmetry of activation with this task paradigm is correlated with language lateralization on the Wada test . For the Sabsevitz et al. study, separate LIs were computed for the whole hemisphere, frontal lobe, temporal lobe, and angular gyrus. All patients also underwent Wada testing and preoperative assessment of confrontation naming using the 60-item Boston Naming Test (BNT). The BNT was administered again at 6 months after surgery, and a change score was calculated as the difference between postop and preop scores. Surgeries were performed blind to the fMRI data but were tailored using intraoperative electrical stimulation mapping.
Compared to a control group of 32 right ATL patients, the left ATL group declined postoperatively on the BNT (p < .001), with an average change score of -9. Within the left ATL group there was considerable variability, with 13 patients (54%) showing variable degrees of decline relative to the control group. The temporal lobe fMRI LI was the strongest predictor of outcome (r = -.64, p < .001), indicating that language lateralization toward the left (surgical) temporal lobe was related to poorer naming outcome, whereas lateralization toward the right temporal lobe was associated with little or no decline. This fMRI measure showed 100% sensitivity, 73% specificity, and a positive predictive value of 81% in predicting significant decline. By comparison, the Wada language LI showed a somewhat weaker correlation with outcome (r = -.50, p < .05), 92% sensitivity, 43% specificity, and a positive predictive value of 67%. Notably, the frontal lobe fMRI LI was also less predictive (r = -.47, p < .05), suggesting that an optimal LI is one that indexes lateralization in the surgical resection area.
Sabsevitz et al. also created multivariate models to determine the contribution of fMRI relative to other noninvasive predictors. Both age at epilepsy onset (r = -.35, p = .09) and preoperative performance (r = -.39, p = .06) showed strong trends toward a correlation with outcome, and together these variables accounted for 27% of the variance in outcome. Adding the temporal lobe fMRI LI to this model accounted for an additional 23% of the variance, indicating a significant increase in predictive power (p < .01). Addition of the Wada language asymmetry score did not improve the model (1% increase in explained variance, p > .1).
These results show how preoperative fMRI can be used to stratify patients in terms of risk for language decline in the setting of left ATL resection, allowing patients and physicians to more accurately weigh the risks and benefits of the surgery. It is crucial to note, however, that these results hold only for the particular methods used in the study and may not generalize to other fMRI protocols, analysis methods, patient populations, or surgical procedures. Future studies should not only confirm these results using larger patient samples, but also test whether other fMRI protocols in current widespread use have similar predictive capability.
Verbal memory decline after left ATL resection is a consistent finding in group studies and is observed in 30-60% of such patients [38-46, 85-90]. A main focus of the preoperative evaluation in ATL surgery candidates is, therefore, to estimate the risk of verbal memory decline in patients undergoing left ATL resection. The Wada memory test was originally developed for the purpose of predicting global amnesia after ATL resection . Studies of its ability to predict material-specific verbal memory decline have been inconsistent, with several suggesting good predictive value [41, 87-89, 92] and others showing little or none, particularly when used in combination with non-invasive tests [42, 45, 46, 93-95]. Some authors have questioned the general validity and reliability of Wada memory results [96-102]. Others have emphasized the sensitivity of the test to certain details of the stimulus presentation, procedures used for recall, and other methodological factors [103-106].
As with any invasive test, a major concern with the Wada test is whether it adds information beyond that available from noninvasive measures. Structural MRI of the hippocampus is modestly predictive of memory outcome [42, 45, 107-109] as is inter-ictal positron emission tomography . Preoperative neuropsychological testing is one of the strongest predictors of outcome, in that patients with good memory abilities prior to surgery are more likely to decline than patients with poor preoperative memory [39, 42-46, 85, 86, 111-113]. Age at onset of epilepsy is also predictive, with decline more likely in those with later age at onset [43, 86, 111, 114]. Recent research has investigated whether fMRI might provide additional predictive information.
Medial temporal lobe (MTL) activation during memory encoding and retrieval tasks has been a subject of intense research with fMRI (for reviews, see [115-121]). FMRI of this region is not without technical challenges. The hippocampal formation is small relative to typical voxel sizes used in fMRI. Within-voxel averaging of signals from active and inactive structures may thus impair detection of hippocampal activity. Loss of MRI signal in the medial ATL due to macroscopic field inhomogeneity can affect the amygdala and occasionally the anterior hippocampus [122-124]. Finally, the baseline state employed in subtraction analyses is probably of critical importance. Human imaging evidence suggests that the hippocampus is relatively activated in the “resting” state [61, 125, 126]. Stark & Squire , for example, showed that the hippocampus and parahippocampus both show higher BOLD signals during “rest” than during active perceptual discrimination tasks. Activation of these MTL regions during encoding of pictures was detected using the perceptual discrimination tasks as a baseline, but not when “rest” was used as a baseline.
Several fMRI studies have examined the relationship between preoperative medial temporal lobe (MTL) activation and memory outcome after ATL surgery (Table 1). Rabin et al.  studied 23 patients undergoing ATL resection (10 left, 13 right) using a scene-encoding task that activates the posterior MTL bilaterally . Patients were tested for delayed recognition of the same pictures immediately after scanning. Delayed picture recognition was then tested again after surgery, and the change on this recognition task was used as the primary memory outcome variable. About half of the patients in both surgery groups declined on this measure. Preoperative fMRI activation lateralization toward the side of surgery was correlated with decline, as was the extent of activation on the side of surgery. These results were the first to demonstrate a relationship between preoperative fMRI activation asymmetry and outcome, yet they are of limited relevance to the problem of predicting verbal memory outcome. In the left ATL patients studied by Rabin et al., neither Wada memory nor fMRI activation asymmetry predicted verbal memory decline as measured by standard verbal memory tests.
Richardson and colleagues studied correlations between hippocampal activation and verbal memory outcome in three small studies [129-131]. Patients performed a semantic decision task with words during the fMRI scan and then took a recognition test after scanning. Words that were subsequently recognized were contrasted with words that were judged to be familiar but not recognized. In the first of these studies , the authors observed a focus in the anterior hippocampus where asymmetry of activation (i.e., left -right) predicted verbal memory outcome on a standardized word list learning test after left ATL resection. Greater activation in this region on the left side relative to the right side predicted greater decline. The second study by the same authors, however, showed correlations between outcome and hippocampal activation on either side . That is, greater activation unilaterally on the left or the right was associated with poorer outcome. The correlation between verbal memory decline after left ATL resection and activation in the right hippocampus is difficult to explain, as patients with greater activation in the right hippocampus preoperatively would be expected to have a better outcome, not a worse outcome . This finding was not replicated in the third study , which reported a correlation between left hippocampus activation and poor outcome but no correlation between right hippocampus activation and outcome. A methodological limitation in all of these studies is that they are based on fMRI activation values extracted from a small region of interest defined by searching the volume for voxels that show a correlation with outcome across a group of patients. As the coordinates of these correlated voxels have varied across the studies, it is not clear how this method of extracting activation values would be applied to a newly encountered patient.
Frings et al. studied the relationship between preoperative hippocampal activation asymmetry and verbal memory outcome in a small sample of patients undergoing left or right ATL resection . The fMRI protocol used a task in which patients viewed a virtual-reality environment containing colored geometric shapes and either memorized the location of these objects or performed a recognition decision following memorization. These “memory tasks” were contrasted with a control task in which patients saw two versions of a geometric object and indicated which one was larger. This fMRI contrast had been shown previously to activate posterior MTL regions (mainly posterior parahippocampus) bilaterally. A lateralization index was computed using the entire hippocampus as the region of interest. Verbal memory change was marginally correlated (1-tailed p = .077) with preoperative LI in the left ATL surgery group, but not in the right surgery group. A significant correlation (1-tailed p < .05) was obtained when the groups were combined, indicating greater verbal memory decline when preoperative hippocampal activation was lateralized more toward the side of surgery.
Köylü et al. examined correlations between preoperative MTL activation and verbal memory performance before and after ATL surgery . Average fMRI activation produced by a semantic decision - tone decision contrast was measured in left and right MTL regions of interest including the hippocampus and parahippocampus. The authors observed correlations between MTL activation and both preoperative and postoperative verbal memory. In the left ATL surgery group, postoperative memory was positively correlated with preoperative activation in the right MTL. Unfortunately, the analyses examined only pre- and postoperative scores in isolation and not pre- to postoperative change, which is the primary issue of clinical interest.
Finally, Binder et al.  measured hippocampal activation asymmetry in 30 left and 37 right ATL surgery patients using a scene-encoding task. When contrasted with a perceptual matching task, this paradigm activates the anterior hippocampus bilaterally . Activation asymmetry was correlated with side of seizure focus (p = .004) and with Wada memory testing performed in the same patients (p = .009). This activation asymmetry, however, did not predict verbal memory outcome.
Although preliminary, these studies are informative in several ways. Three studies [127, 133, 135] used complex scene encoding tasks that activate the MTL bilaterally on fMRI, a pattern that suggests activation of both verbal and nonverbal memory encoding systems. Prediction of verbal memory outcome using these paradigms seems to be weak at best. In contrast, the verbal memory fMRI paradigms used by Richardson et al. provide better predictive information regarding verbal memory outcome, at least when the analysis is confined to a specific region of the hippocampus. The persistent difficulty in applying the latter approach, however, is identifying a priori the small set of voxels that will be predictive in a given individual patient.
Binder et al. studied the relationship between preoperative language lateralization and verbal memory outcome . The premise underlying this approach is that the verbal episodic memory encoding system is likely to be co-lateralized with language. More generally, the authors proposed that the type of material preferentially encoded by the left or right MTL depends on the type of information it receives from the ipsilateral neocortex. According to this model, the MTL in the language-dominant hemisphere is more critical for supporting verbal episodic memory, and language lateralization should be a reliable indicator of verbal memory lateralization.
The study included 60 patients who underwent left ATL resection and a control group of 63 patients who underwent right ATL resection. The fMRI paradigm used a contrast between an auditory semantic decision task and a nonlinguistic tone decision task (Figure 1). Verbal memory was measured preoperatively and 6 months after surgery using the Selective Reminding Test, a word-list learning and retention test . Other neuropsychological testing included the story recall and visual reproduction subtests from Wechsler Memory Scale . Language LIs were computed from the fMRI data using a large region of interest covering the lateral two-thirds of each hemisphere . All patients also underwent preoperative Wada language and object memory testing.
The left ATL surgery group showed substantial changes in verbal memory, with an average raw score decline of 43% on word list learning and 45% on delayed recall of the word list. Of the individual patients in this group, 33% declined significantly on the learning measure and 55% on the delayed recall measure. In contrast, the right ATL surgery group improved slightly on both measures. Neither group showed significant changes on any nonverbal memory tests. Preoperative measures that predicted verbal memory decline in the left surgery group included the preoperative score, the fMRI language LI, the Wada language asymmetry score, the age at onset of epilepsy, and the Wada memory asymmetry score (Table 2, Figure 2).
In applying these results to real clinical situations, the main questions to resolve are: which tests make a significant independent contribution to predicting outcome, and how should results from these tests be optimally combined? Binder et al. addressed these questions in a series of stepwise multiple regression analyses. The first variables entered in all analyses were preoperative test performance and age at onset of epilepsy. The rationale for including these variables first is that they can be obtained with relatively little expense and at no risk to the patient. Next, the fMRI language LI was added, followed by simultaneous addition of both the Wada memory and Wada language asymmetry scores. The rationale for adding fMRI in the second step is that fMRI is non-invasive and carries less risk than the Wada test. The two Wada scores were added together in the final step because these measures are typically obtained together.
Preoperative score and age at onset of epilepsy together accounted for 49% of the variance in List Learning outcome and 54% of the variance in Delayed Recall outcome. The fMRI LI accounted for an additional 10% of the variance in List Learning outcome (p = .001) and 7% of the variance in Delayed Recall outcome (p = .003). Addition of the Wada language and memory data did not significantly improve the predictive power of either model (R2 change for List Learning = .025, R2 change for Delayed Recall = .017, both p > .1). When patients were categorized as showing decline or no decline based on a negative change score 1.5 standard deviations or more from the mean change score in the right ATL surgery group, the List Learning outcome model showed sensitivity of 90% and specificity of 80% for predicting decline on List Learning. The Delayed Recall outcome model showed sensitivity of 81% and specificity of 100% for predicting decline on Delayed Recall.
These results are interesting for several reasons. Most intriguing is the finding that language lateralization, whether measured by fMRI or the Wada test, is a better predictor of verbal memory outcome than Wada memory testing. The explanation for this apparent paradox rests on two hypotheses. One, mentioned above, is that verbal memory encoding processes tend to co-lateralize with language processes. The second hypothesis is that many tests of memory lateralization do not specifically assess verbal memory encoding. That is, visual stimuli such as objects and pictures can be dually encoded using both verbal and visual codes. Wada memory procedures that use such stimuli (including the Wada test used by Binder et al.) therefore do not provide a measure of verbal memory lateralization, but rather a measure of overall memory lateralization that includes both verbal and nonverbal encoding processes. Together, these two hypotheses suggest that language asymmetry may be a better indicator of verbal memory lateralization than Wada memory asymmetry (Figure 3). In particular, some patients with left temporal seizures show right-lateralized memory on the Wada test due to a strong nonverbal memory component in the right hemisphere, but are nevertheless at high risk for verbal memory decline because their verbal memory remains strongly lateralized to the left (Figure 3B).
These data also have direct implications for clinical practice. First, they confirm the utility of fMRI for predicting verbal memory outcome in patients undergoing left ATL resection. The fMRI language LI is a safe, noninvasive measure that improves prediction accuracy relative to other noninvasive measures. The finding that Wada memory lateralization is not a strong predictor of verbal memory outcome and adds no predictive value beyond these noninvasive measures confirms several previous studies that also examined multivariate prediction models [42, 45, 93-95]. Although Binder et al. found that Wada language asymmetry is a stronger predictor of verbal memory outcome than Wada memory lateralization, even the addition of both Wada tests together did not contribute additional predictive power after inclusion of available noninvasive data (including fMRI). These results call into question the routine use of the Wada test for predicting material-specific verbal memory outcome, particularly if a validated fMRI measure of language lateralization is available. Some practitioners value the Wada test as an indicator of risk for severe “global” amnesia, such as is known to occur after bilateral MTL damage [91, 139-141]. According to this theory, anesthetization of the to-be-resected MTL is necessary to discover whether the contralateral hemisphere is healthy enough to support memory on its own. Empirical observations, however, provide little support for such an approach. Cases of global amnesia following unilateral temporal lobe resection --especially modern, well-documented cases -- appear to be rare in the extreme [96, 97, 99, 100, 142, 143]. Furthermore, there is ample evidence that contralateral hemisphere “memory failure” on the Wada test suffers from poor test-retest reliability and does not reliably predict amnesia [96-102]. Given the availability of fMRI for predicting material-specific verbal memory outcome, perhaps use of the Wada test should be reserved only for those patients at greatest risk for global amnesia, i.e. patients undergoing unilateral ATL resection who have structural or functional evidence of damage to the contralateral MTL. Because it is noninvasive and requires fewer personnel, fMRI is also likely to be substantially less costly than the Wada test .
Recent studies demonstrate that preoperative fMRI can be used to predict postoperative naming and verbal memory changes in patients undergoing left ATL resection. Most importantly, two studies showed that fMRI significantly improves prediction accuracy when combined with other noninvasive measures, and that Wada testing does not add significant additional predictive power [37, 46]. Thus, fMRI provides patients and practitioners with a tool for making better-informed decisions based on a quantitative assessment of cognitive risk. The quantitative nature of these predictions represents something of a paradigm shift, in that traditional predictive models using the Wada test tended to be implemented as a dichotomous “pass or fail” judgment. The alternative approach followed in many recent studies involves the development of multivariate models that compute predicted change scores (Figure 4). These quantitative predictions provide a much more realistic picture of the actual outcomes, which are not dichotomous, but vary smoothly along a continuum. Ultimately, of course, the decision whether to undergo surgery is a categorical one, but the categorical nature of the decision does not obviate the need for precision regarding the factors that enter into the decision. A patient disabled by frequent seizures may be willing to tolerate a substantial decline in naming or verbal memory, whereas a patient who depends on such cognitive abilities for her livelihood may be willing to risk a small decline but not a large one.
In practice, implementation of fMRI methods for predicting outcome in epilepsy surgery will depend on the availability of a validated fMRI protocol and involvement of clinicians with the necessary clinical expertise. Fast T2*-weighted imaging capabilities necessary for fMRI are a standard feature on currently marketed clinical MRI systems, and fMRI is now available in some form at most medical centers. Implementation of fMRI protocols requires only installation of relatively low-cost audiovisual stimulation and response monitoring systems. Of course, fMRI is not suitable for all epilepsy patients. The largest outcome studies cited above [37, 46] included all patients with full-scale IQ >70, but those with more severe cognitive impairments may not be able to comply with task requirements. Wada testing will continue to play a role in determining language dominance in such patients, as well as in younger children who are unable to comply with tasks or to refrain from large movements during scanning. In the author's experience, even many cognitively disabled patients can be scanned successfully when given clear instruction and feedback by a professional with expertise in cognitive assessment. Similarly, the experience at many centers suggests that successful fMRI studies can be conducted in most children over age 10 given adequate instruction, encouragement, and feedback.
Thanks to Linda Allen, Thomas Hammeke, Wade Mueller, Conrad Nievera, Ed Possing, Manoj Raghavan, David Sabsevitz, Sara Swanson, and other personnel at the Froedtert-MCW Comprehensive Epilepsy Center for assistance with this research, which was also supported by National Institute of Neurological Diseases and Stroke grant R01 NS35929, National Institutes of Health General Clinical Research Center grant M01 RR00058, and the Charles A. Dana Foundation.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.