The Application of Functional MRI of Memory in Temporal Lobe Epilepsy: A Clinical Review


  • H. W. Robert Powell,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author
  • Matthias J. Koepp,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author
  • Mark P. Richardson,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author
  • Mark R. Symms,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author
  • Pam J. Thompson,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author
  • John S. Duncan

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College of London and MRI Unit, National Society for Epilepsy, Buckinghamshire, U.K.
    Search for more papers by this author

Address correspondence and reprint requests to Dr. R. Powell at


Summary:  Functional MRI (fMRI) is a useful tool for noninvasively localizing areas in the brain involved in specific cognitive functions. Since its introduction, there has been considerable speculation regarding the role it may play in the presurgical assessment of temporal lobe epilepsy (TLE). This review considers the progress made to date in using fMRI to investigate memory processing in the medial temporal lobe in normal subjects and in those with TLE. Results so far suggest that fMRI will be incorporated into the presurgical assessment of TLE in the coming years to improve definition of eloquent cerebral areas, with the objective of minimizing the adverse cognitive sequelae of anterior temporal lobe resection.

Investigating the effects of brain lesions can inform us which brain areas are necessary for normal memory function, but it is less specific for defining the precise functional roles of these areas. Functional neuroimaging has the potential to offer more precise localization of cognitive processes than do lesion studies and can address the fundamental question of why some experiences are remembered rather than forgotten (1). Conversely, functional imaging tends to reveal a network of regions, of which only a subset are necessary for the task in question, and therefore does not indicate whether the integrity of a certain brain region is essential for normal function.

Functional neuroimaging methods, notably positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), are based on the measurement of hemodynamic responses to specific stimuli, and the signals obtained are believed to reflect indirectly the underlying neuronal activity (2). The differing magnetic properties of oxyhemoglobin and deoxyhemoglobin provide the basis for the increased signal seen in activated brain regions using fMRI. Since the discovery of this blood-oxygen-level–dependent (BOLD) contrast (3), fMRI has become widely used for the noninvasive mapping of human brain function (4–6).

This review considers the progress made using fMRI to investigate normal memory processes and the effects of temporal lobe epilepsy (TLE) on these. We focus on some of the neuropsychological and technical factors that have made fMRI of memory, in particular with regard to medial temporal lobe (MTL) function, more difficult to study than other cognitive functions. We then consider how fMRI may be incorporated into the presurgical assessment of TLE patients in the future.


Neuropsychological issues

The term episodic memory refers to the cognitive processes that enable the explicit recollection of unique events and the context in which they occurred (7). These include the transformation of an experience into an enduring memory trace (memory encoding) and the subsequent recollection of the event at a later time (memory retrieval), and neuropsychological evidence suggests that these processes depend on a number of circumscribed, interconnected brain regions including the MTL and prefrontal cortex (PFC) (8).

The MTL consists of the hippocampus, amygdala, and parahippocampal regions. Bilateral injury to these areas leads to a characteristic amnesic syndrome, as shown by Scoville and Milner's famous patient, HM. After surgical resection of both temporal lobes, HM was unable to acquire any new information for later recall despite memory from his childhood remaining relatively intact (9). Similar findings have been seen in lesion studies in animals (10).

Standard fMRI experiments initially used block-design paradigms looking for regions of the brain showing greater activation during task blocks (e.g., hand tapping or word generation) compared with rest blocks. Although these types of paradigms are very efficient for simple motor and language tasks, localizing memory function has proved more challenging. This is partly because of the different components involved in memory processing, such as encoding and retrieval, and also because the nature of the material being encoded or retrieved influences which brain areas are used.

A further problem faced when designing memory fMRI experiments was how to separate brain activity due specifically to memory from that due to other cognitive processes being used in the task. Early fMRI studies of memory encoding used blocked experimental designs to contrast tasks promoting differing memory performance, using the “depth of encoding” principle (11). This states that if you manipulate material in a “deep” way (e.g., make a semantic decision about a word), then it is more likely to be recalled successfully than material manipulated in a “shallow” way (e.g., make a decision of whether the first letter of a word is alphabetically before the last letter). The results of these studies [e.g., intentional learning vs. reading (12) and semantic versus nonsemantic classification (13,14)] tended to show consistent activation in left prefrontal cortical regions (15) along with less reliable MTL activation.

Similar assumptions underlie the use of novelty paradigms in probing memory encoding. During these experiments, alternating blocks of novel and repeated stimuli are presented, with the hypothesis being that more memory encoding takes place while viewing a block of novel stimuli than when viewing the same repeated stimulus (16).

The advantage of blocked designs is that they are efficient in detecting differences between two conditions. The main problem in their interpretation, however, lies in the inference that the effects shown by these contrasts reflect differences in memory encoding, rather than any other differences between the two conditions (e.g., response to novelty, semantic processing) that are independent of differences in memory encoding.

Attempts were made to overcome this problem by using parametric block designs. In this type of experimental design, the magnitude of the fMRI signal during a block is correlated with the number of items remembered from that block on a subsequent memory test. This type of study, first applied to fMRI by Fernandez et al. (17), could claim to examine successful memory encoding but was soon superseded by the advent of event-related studies.

Event-related fMRI is defined as the detection of transient hemodynamic responses to brief stimuli or tasks (18–21). This technique, derived from those used by electrophysiologists to study event-related potentials, enables trial-based rather than block-based experiments to be carried out. Trial-based designs have a number of methodologic advantages (22), one of which is that trials can be categorized according to a subject's performance on a subsequent test to obtain fMRI data at the individual item level. Therefore when we study memory encoding, activations for individual items presented can be contrasted according to whether they are remembered or forgotten in a subsequent memory test. This type of analysis allows the identification of brain regions showing greater activation during the encoding of items that are subsequently remembered compared with items subsequently forgotten (subsequent memory effects), which are then taken as candidate neural correlates of memory encoding (1).

In summary, significant developments have been made in the design of the experiments used to investigate the neural correlates of memory encoding. Blocked designs using novelty and depth of processing paradigms have been superseded, initially by parametric block designs, and finally by event-related designs. Whereas the latter may be more vulnerable to alterations in the hemodynamic response function (e.g., due to pathology), they have the crucial advantage of being able to examine specifically subsequent memory effects.

Technical issues

Functional MRI requires images to be acquired at high speed, and the sequence most commonly used for this is echo planar imaging (EPI). The main technical problem when using EPI sequences, especially within the temporal lobes, is that of susceptibility artefact. Ideally in the absence of an applied gradient, the magnetic field would be homogeneous throughout the bore of an MRI scanner. Unfortunately, the different magnetic properties of bone, tissue, and air introduce inhomogeneities in the field when a head is introduced into the bore. Brain regions closest to borders between sinuses and brain or bone and brain are most affected and therefore especially likely to suffer geometric distortions or loss of BOLD signal (23).

Geometric distortions of the EPI data make it difficult to overlay fMRI activations directly on co-registered high-resolution scans. They can be unwarped by using techniques that map the local field in the head (24), though it has been shown that approaches of this kind can introduce extra noise into the corrected EPI data (25). Alternative acquisition sequences that do not experience geometric distortions are available (26), although these rarely have the temporal resolution or high signal-to-noise ratio (SNR) per unit time of EPI.

The second artefact in EPI data is more serious, as signal loss leads to sensitivity loss, which is unrecoverable by image-processing techniques (see Fig. 1). It has been demonstrated that this artefact is most prominent in the inferior frontal and inferolateral temporal regions (27), and as the hippocampus rises from anterior to posterior, one would expect greater susceptibility-induced signal loss in the anterior (inferior) relative to posterior (superior) hippocampus. This may have been one reason for the relative lack of anterior hippocampal activation in early fMRI studies of memory (28).

Figure 1.

High-resolution echo-planar imaging axial slice demonstrating distortion and signal loss in the inferior frontal lobes.

Studies have shown that signal loss due to susceptibility artefact can have significant effects on MTL activation. In a direct comparison between PET and fMRI using an identical semantic task, Devlin et al. (29) demonstrated temporal lobe activation with PET, but not with fMRI. Cordes et al. (30) showed susceptibility-induced signal loss in the parahippocampal gyrus and amygdala while a subject mentally rehearsed a gymnastics routine.

One recent study directly examined the effects of susceptibility artefact on hippocampal activation by demonstrating its differential effect on the anterior versus the posterior hippocampus. The averaged resting voxel intensity in an anterior hippocampal region of interest (ROI) was significantly less than in a posterior hippocampal ROI, and intensity decreases were substantial enough to leave many voxels below the threshold where BOLD effects could no longer be detected (28). On top of this, it has been shown that the sensitivity to BOLD changes is proportional to signal intensity at rest, so that voxels with a lower baseline signal (such as those in anterior hippocampal regions) would be more difficult to activate than would those with higher baseline signals (31). Some of these artefacts can be corrected by shimming, a process whereby the static magnetic field is made more homogeneous over the ROI (23). Some will remain, however, leading to distortion and dropout in EPIs. Other approaches to removing dropout often involve acquiring extra images, leading to a loss of temporal resolution (32), but more recent work has shown that dropouts and distortions can be reduced without incurring time penalties if regions of reduced spatial extent are imaged (33).

In spite of these problems, the past 5 years have seen a number of fMRI studies showing hippocampal activation in the context of episodic memory. A wide range of stimuli have been shown to cause hippocampal activation, including words, faces, line drawings, patterns, objects, scenes, and routes, and experimenters have looked at both memory encoding and retrieval. The data so far have given us important insights into the way memory processing occurs in the human brain. The distribution of MTL activation seen varies both in lateralization and localization, depending on the type of material being used (i.e., verbal or nonverbal), and the stage of memory processing (i.e., encoding or retrieval).


Material-specific lateralization of memory processing

Patients with unilateral MTL lesions have provided evidence of dissociation in function between the dominant (usually the left) hippocampus, mediating verbal memory (34), and nondominant (usually the right) hippocampus, mediating nonverbal or visual memory (35). A study of patients after unilateral anterior temporal lobe resections showed deficits in topographic memory after right temporal lobe resection and episodic memory deficits after left temporal lobe resection, suggesting a material-specific lateralization of function in MTL structures (36). These observations have led many researchers to use functional imaging studies to look for lateralization of cerebral activation patterns during episodic memory processes. Several early PET and fMRI studies revealed greater activation in left PFC during verbal encoding (13,37–39). Other more recent studies have shown material-specific lateralization in prefrontal regions, with right inferior prefrontal activation during encoding for nonverbal information, such as abstract visual patterns (40), and checkerboards (41).

By using a blocked experimental design comparing intentional and incidental encoding, Kelley et al. (12) investigated memory encoding for three stimulus types: written words, nameable line-drawn objects, and unfamiliar faces. Word encoding versus fixation on a cross-hair produced left-lateralized activation, face encoding produced right–lateralized activation, and object encoding produced bilateral activation. These results were observed in both dorsal frontal cortex and MTL. They indicated regions in both hemispheres that could be differentially engaged, depending on the nature of the material being encoded, although in this study, only five normal subjects were tested, and subsequent memory effects were not specifically investigated.

Golby et al. (16) compared activations for the encoding of four kinds of materials—words, faces, scenes, and abstract patterns—by using a novelty paradigm. Verbal encoding resulted in left-lateralized activation of the inferior PFC and MTL. Pattern encoding activated right inferior PFC and right MTL. Scenes and faces resulted in approximately symmetrical activation in both regions. These results suggested that the lateralization of encoding processes is determined by the verbalizability of the stimuli; however, the disadvantages of this type of paradigm have already been discussed.

Not all the studies linking MTL activation with episodic encoding have shown such clear lateralization of function. Fernandez et al. (17) found significant positive correlation between the number of recalled words and activation in posterior hippocampal regions, although no difference between left and right was revealed. Bilateral parahippocampal activation also has been demonstrated for the encoding of pictures, spatial environments, and color photographs of everyday scenes, all of which involve both verbal and nonverbal memory processes (42,43).

Two early landmark studies used event-related fMRI to show subsequent memory effects in medial temporal regions. Brewer et al. (44) showed in six normal subjects that the magnitude of activation in bilateral parahippocampal and right frontal areas during the study of complex, color photographs predicted which photographs were later remembered or forgotten. In addition, greater parahippocampal activity occurred during encoding for pictures that were remembered specifically than for pictures that just seemed familiar, and also greater parahippocampal activity for familiar pictures compared with forgotten pictures, although no clear lateralization of activation was seen.

In an article published simultaneously, Wagner et al. (14) reported two experiments, one using a blocked design and one using an event-related design to examine verbal encoding processes. In the blocked-design experiment, 12 subjects performed alternating task blocks consisting of semantic processing, nonsemantic processing, and visual fixation. As expected, subsequent memory performance was significantly higher for semantic than for nonsemantic blocks, and fMRI data analysis showed greater activation during semantic processing in left PFC, left parahippocampal, and left fusiform gyri. In the second experiment, 13 subjects performed a word-encoding trial that was categorized according to whether a word was subsequently remembered or forgotten. Analysis of the fMRI data showed event-related activity in left prefrontal, left parahippocampal, and left fusiform gyri that correlated with subsequent memory.

Process-specific lateralization of memory function

A further hemispheric asymmetry emerged from early PET studies showing lateralization due to the separate components involved in memory processing, a finding seemingly at odds with lesion-deficit models. Data showed that during encoding, left PFC tended to show greater activation than right PFC, whereas during retrieval, right PFC tended to show greater activation than left. This pattern of results was dubbed the Hemispheric Encoding/Retrieval Asymmetry (HERA) model (45,46). These findings have been demonstrated for both verbal (47) and nonverbal materials (48). Similar findings have since been demonstrated in some fMRI studies (49).

Others have argued that this asymmetry of PFC activations may reflect the material used rather than the memory processes. In addition, the HERA effect has not been demonstrated in the MTL.

Process-specific localization of memory function within the MTL

In addition to the functional division between left and right, evidence also exists for further dissociation of function within each MTL. The amygdala, lying anterior to the hippocampus, is a structure that has been linked to emotional memory (50), and patients with isolated amygdala damage show a specific impairment of recall of emotional stimuli (51). Functional imaging has also demonstrated amygdala activity correlating with subsequent memory for emotional material (52).

Further evidence points to a functional localization within each hippocampus that is dependent on the stage of memory processing. A meta-analysis of 54 published PET studies of memory by Lepage et al. (53) demonstrated this functional dissociation, with activations associated with memory encoding located in anterior hippocampal regions, and activations associated with retrieval located more posteriorly. They referred to this pattern of encoding and retrieval activations as the HIPER (Hippocampal Encoding/Retrieval) model. Evidence from fMRI experiments has proved contradictory, with many early studies refuting the HIPER model (54,55). More recent event-related studies, however, have supported the PET model of anterior hippocampal activation during memory encoding (56).

Further functional segregation within the hippocampus has been suggested, with differential anteroposterior activation in relation to stimulus familiarity (57) and dissociable perirhinal, hippocampal, and parahippocampal activations during free recall of words presented during scanning (58). With high-resolution structural and functional MRI, Zeineh et al. (59) “unfolded” the hippocampal cortex, revealing the entirety of each hippocampal region and adjacent neocortical regions in a single “flat map” and were able to demonstrate differential activation of hippocampal subregions during memory encoding. Although these results would suggest that different subdivisions within the hippocampus make distinct contributions to new memory formation, the combination of fMRI's limited spatial resolution and the anatomic distortions seen in the MTL would suggest that such conclusions should be treated cautiously.

In summary, it is clear that many factors may affect the anatomic correlates of memory as seen with functional neuroimaging in normal subjects. These include material type, the nature of the encoding strategy adopted by the subject, the type of subsequent memory test used (recognition, free recall, cued recall), and other item-specific qualities such as distinctiveness (both semantically and physically). All these sources of variability may affect both the magnitude and also, possibly, the distribution of the subsequent memory effect seen. Nevertheless it would appear that an important functional relation between prefrontal and medial temporal structures lies at the heart of both memory encoding and retrieval.


Much speculation has occurred regarding the possible role of fMRI of memory in epilepsy, in particular with regard to the presurgical evaluation of patients with TLE. It has potential both in replacing the intracarotid amytal test (IAT) and in providing additional data regarding memory lateralization, currently assessed primarily by baseline neuropsychological assessment.

The primary aim of the IAT is to screen for postoperative amnesia. By temporarily anesthetizing the affected temporal lobe, it mimics the proposed surgery and allows the mnestic capacity of the contralateral temporal lobe to be assessed. It also provides information regarding the dysfunctional temporal lobe by comparing the memory capacity of each hemisphere. Some centers will use the degree of memory impairment after injection of the dysfunctional side to predict postoperative memory outcome. The IAT is less useful for the lateralization of memory, as deactivation of the language dominant hemisphere will cause increased errors on verbal memory testing.

IAT testing has a number of disadvantages, notably the fact that it is an expensive, invasive procedure with significant variations between centers in many aspects of its methodology. Whereas fMRI can also vary between centers in both methods and equipment, it is nevertheless cheaper, noninvasive, and repeatable. However, important caveats exist when considering the role of fMRI. First, areas activated by a particular fMRI paradigm are not necessarily crucial for the performance of that task. Second, it does not necessarily follow that all areas involved in a task will be activated by a particular fMRI paradigm. Third, the extent of activation seen in a task may bear no relation to the competence with which that task is performed.

Caution also will be needed in the interpretation of results, bearing in mind that fMRI techniques, although useful for the localization of cognitive function, cannot be used to assess the capacity of unilateral temporal lobe structures (60). This is in contrast to the IAT, which is useful in safeguarding against a severe postoperative amnesic syndrome, but less relevant in predicting the types of material-specific memory impairments seen following surgery (i.e., impaired verbal memory after left anterior temporal lobe resection, and impaired nonverbal, or visuospatial memory after right anterior temporal lobe resection). Indeed, with the improvements in structural MRI, the risk of a postoperative amnesic syndrome has been dramatically reduced, and of all the postoperative unilateral temporal lobe amnesic patients written up in the literature, none would have had an entirely normal contralateral hippocampus in terms of size and structure on modern MRI scans (61).

A number of studies have used fMRI to look at the lateralization of memory in patients with TLE compared with that seen in normal subjects and also compared the findings with the results of the IAT. To date, all but one of these have used block-design paradigms (see Table 1). As mentioned previously, there are advantages in using block designs in that they are less vulnerable to any alterations in the hemodynamic response function due to pathology, but it is only by using event-related studies that subsequent memory effects can be specifically examined.

Table 1. Studies performed in patients with TLE showing patient groups studied, experimental design used, and principal findings
AuthorsPatient groupMethodsFindings
  1. LHS, left hippocampal sclerosis; TLE, temporal lobe epilepsy; IAT, intracarotid amytal test; MTL, medial temporal lobe.

Detre et al. 1998Eight normal controls, 10 consecutive patients undergoing presurgical assessmentBlock design.
Complex visual scenes vs. abstract pictures
Symmetric MTL activation in normal subjects. Lateralization of memory concordant with IAT in nine of 10 subjects
Bellgowan et al. 199828 patients with TLE, 14 left TLE, 14 right TLEBlock design.
Semantic decision vs. auditory perception task
Greater activation in the left MTL in right TLE compared with left TLE group
Dupont et al. 200010 normal controls, seven patients with LHSBlock design.
Verbal encoding and retrieval vs. fixation on the letter A
Left occipitotemporoparietal network activated in controls. Reduced MTL activation and increased activation in left dorsolateral frontal cortex in patients
Jokeit et al. 200117 normal controls, 30 consecutive patients undergoing presurgical assessment.Block design.
Roland's Hometown Walking vs. baseline
No asymmetry of MTL activation in controls, greater activation in the MTL contralateral to seizure focus in 90% of patients
Golby et al. 2002Nine consecutive TLE patients undergoing presurgical assessmentBlock design comparing novel vs. repeated stimuli.
Four encoding stimuli used: patterns, faces, scenes, and words
Group level: greater activation in the MTL contralateral to seizure focus for all encoding stimuli
Single subjects: lateralization of memory concordant with IAT in eight of nine subjects
Richardson et al. 200312 normal controls, 24 patients with LHSEvent-related verbal encoding taskLeft hippocampal activation in normal subjects. Reorganization of function to right hippocampus and parahippocampal gyrus in patients

Detre et al. (62) used an encoding task during which subjects were shown alternating blocks of complex visual scenes and abstract patterns. Although subjects were told to memorize the scenes for subsequent testing, subsequent memory effects were not looked at, and an assumption of incidental memory encoding was made. The symmetry of the fMRI activations in nine patients with TLE were compared with the results of IAT performed as part of their preoperative evaluations. Task activation, which was located posteriorly in the hippocampal formation, was near symmetrical in normal subjects, whereas in patients with TLE, significant asymmetries were observed. Furthermore, in all nine patients, the asymmetry of the activation concurred with the assessment of hemispheric memory skills from the IAT, including two patients with paradoxical IAT memory lateralization ipsilateral to seizure focus. Task-correlated activity was seen to be higher in controls and patients with good performance, although successful lateralization was seen even in patients who complained that the task was too difficult. Correlating the results of a single fMRI memory paradigm with those from an IAT in which multiple memory tests were performed may, however, be problematic.

Dupont et al. (63) found that patients with left hippocampal sclerosis produced different functional activations on a verbal episodic memory task compared with normal subjects. The experimental design for memory encoding and retrieval consisted of blocks of word learning and silent free recall, respectively, compared with visual fixation on the letter A. Again an assumption was made that the contrast seen was due to memory encoding, but subsequent memory effects were not looked at. Whereas controls demonstrated bilateral activation of the parahippocampal gyrus, among other areas, during memory retrieval, the patients demonstrated slightly less parahippocampal activation but significantly increased left prefrontal activations during the encoding and retrieval phases of the task. No hippocampal activation was detected. Memory test performance was significantly worse in patients, with a mean recall of 3.1 of 17 words, than in normal subjects, whose mean recall was 10.7. This result was interpreted as a dysfunctional response due to the epilepsy and left hippocampal sclerosis (63). They subsequently examined activation patterns during a 24 h–delayed retrieval of the word list in the same group of controls and left TLE patients. Controls showed a similar left occipitotemporofrontal network activated during both immediate and delayed retrieval conditions but additional right hippocampal activation during the delayed retrieval. Memory test performances for this group were broadly similar over a 24-h period. In the patients with left TLE, this distributed neocortical and MTL network was very poorly activated during delayed retrieval. The authors suggested that this reflected an inability to reactivate areas important for retrieving stored information, although it is worth noting that memory test performance was not significantly worse for delayed versus immediate retrieval and that both were significantly worse for patients compared with the control group (64).

A complementary study was performed by using the same paradigm in patients with right TLE. Surprisingly, verbal memory performances also were significantly impaired for these patients compared with controls, and their activation patterns showed a global reduction of left hemisphere activations compared with controls. This suggested bilateral functional consequences of unilateral hippocampal sclerosis on memory processing (65).

Other experiments have used memory tasks that produce unilateral temporal lobe activations in normal subjects. Bellgowan et al. (66) used a block-design paradigm comparing deep and shallow encoding, known to cause extensive left prefrontal and temporal activation in normal right-handed subjects, to investigate activation during verbal encoding in 28 patients with TLE. They found that patients with right TLE showed much stronger activation in the left MTL, including hippocampus, parahippocampal gyrus, and collateral sulcus, than did patients with left TLE, who showed little significant activation in these regions. Activation of language areas in the frontal and parietal lobes was similar in both groups. Neither group showed activation of any of these regions in the right hemisphere. Its success in discriminating left TLE from right TLE would suggest a possible role for fMRI techniques in contributing to the prediction of the side of seizure focus in patients with MRI-negative TLE (66).

Jokeit et al. (67) also investigated the hemispheric asymmetries of MTL activation in 30 patients with TLE using a task employing mental navigation and recall of landmarks based on retrieval of individually familiar visuospatial knowledge [Roland's Hometown Walking Test (68)]. This task was shown to activate MTL structures reliably, even in a child of 7 years and a patient with an IQ of 51. Asymmetry ratios of activation were calculated from significantly activated voxels. This activation was bilateral and symmetric in both parahippocampal gyri in normal subjects, but in 90% of patients with unilateral TLE, the activation was reduced on the side of seizure onset, as judged by absolute voxel numbers and asymmetry ratios. No inferences were made about whether the MTL activation seen was related to memory or memory-independent visuospatial abilities; however, it can be seen how such lateralizing and localizing data may provide complementary information for presurgical evaluation, in particular when results of other investigations have been contradictory.

Golby et al. (69) studied memory lateralization in nine patients with TLE undergoing presurgical evaluation. Lateralization was calculated by using an index of asymmetry for significantly activated voxels within the MTL ROI. In eight of these, lateralization was concordant with that obtained from the IAT. At a group level, greater activation also was demonstrated in the MTL contralateral to the seizure focus, such that in the left TLE group, verbal encoding engaged the right MTL and in the right TLE group, nonverbal encoding engaged the left MTL.

A recent event-related fMRI study looked at verbal memory encoding in nonamnesic right-handed patients with left hippocampal sclerosis. Verbal memory encoding involved activation of the left hippocampus in normal subjects, but was associated with reorganization to the right hippocampus and parahippocampal gyrus in the patient group when compared with the normal group. In addition, the presence of left amygdala sclerosis resulted in the reorganization of encoding of emotional verbal material to the right amygdala. This remains the only fMRI study of memory in TLE patients to use an event-related experimental design (56).


The studies discussed here represent important new clinical applications in the presurgical assessment of the memory functions in TLE patients. The patterns of activation seen in normal subjects are often not seen in these patients when group comparisons are made, although as yet no evidence supports this observation at a single subject level. Nevertheless, it seems likely that fMRI will be incorporated into the presurgical assessment of TLE patients in the coming years, both in assessing the risk of postsurgical language and memory deficits and, to a lesser extent, with regard to determining the side of seizure onset.

The advantage fMRI holds is that paradigms can be designed specifically to look at material-specific memory activation and in particular its location and laterality within MTL structures. Although it is clear that event-related experimental designs and item-by-item analysis are required to demonstrate subsequent memory effects, other paradigms not explicitly testing memory function may prove useful as clinical tools (62,67).

Much of the data discussed in this review, in particular with regard to healthy volunteers, refers to group data, as these provide more robust results. For an fMRI memory paradigm to be a useful investigation to incorporate into the presurgical assessment, however, meaningful data are required at the single-subject level. From this point of view, the concordance seen between fMRI and IAT results in TLE patients by some of the studies is encouraging (62,69). A further consideration is whether a reduction in BOLD activation in an atrophic hippocampus may just reflect the loss of tissue (and hence increase in CSF) within that structure rather than any reorganization of function.

To date, one PET study in patients having selective amygdalohippocampectomy has shown a correlation between preoperative activation in the ipsilateral MTL and postoperative memory impairment (70).

In the same way that language laterality determined by fMRI has been used to predict postoperative language deficits after temporal lobe resection (71), it is possible that the distribution of activation between left and right hippocampus on tests of memory encoding would predict memory outcome postoperatively. Future studies looking at correlations between this activation asymmetry and detailed postoperative neuropsychological assessment will provide us with important information, which may be used to predict the memory deficits seen in patients after unilateral temporal lobe resection. For example, we would hypothesize that patients showing greater verbal memory–encoding activation in the left hippocampus compared with the right hippocampus would have greater verbal memory deficits after left anterior temporal lobe resection.

This type of information, in combination with structural MRI looking at preoperative hippocampal volume and baseline neuropsychology, will enable preoperative prediction of the material-specific memory impairment seen after unilateral anterior temporal lobe resection to be made with greater accuracy. As a result, it will be possible to modify surgical approaches in those patients most at risk and to improve preoperative patient counseling. It seems likely that as experience grows in the interpretation of patterns of preoperative investigation results associated with good and bad postoperative outcomes, a gradual reduction will be seen in the need for the IAT.


Acknowledgment:  We are grateful for the support of the Wellcome Trust (R.P., M.S. supported by Programme Grant No. 067176), the National Society for Epilepsy (M.J.K., P.T., J.D.), and the Medical Research Council (M.R.).