Simultaneous electroencephalography/functional magnetic resonance imaging (EEG/fMRI) recording can noninvasively map in the whole brain the hemodynamic response following an interictal epileptic discharge. EEG/fMRI is gaining interest as a presurgical evaluation tool. This study aims to determine how hemodynamic responses related to epileptic activity can help predict surgical outcome in patients considered for epilepsy surgery.
Thirty-five consecutive patients with focal epilepsy who had significant hemodynamic responses and eventually surgical resection, were studied. The statistical map of hemodynamic responses were generated and coregistered to postoperative anatomic imaging. Patients were classified into four groups defined by the relative relationship between the location of the maximum hemodynamic response and the resection: group 1, fully concordant; group 2, partially concordant; group 3, partially discordant; and group 4, fully discordant. These findings were correlated with surgical outcome with at least 12-month follow-up.
Ten patients in group 1 had the maximum t value (t-max) inside the resection; nine in group 2 had the t-max outside but close to the resection and the cluster with t-max overlapped the resection; five in group 3 had the t-max remote from resection, but with another less significant cluster in the resection; and 11 in group 4 had no response in the resection. The degree of concordance correlated largely with surgical outcome: a good surgical outcome (Engel's class I) was found in 7 of 10 patients of group 1, 4 of 9 of group 2, 3 of 5 of group 3, and only 1 of 11 of group 4. These results indicate that the partially concordant and partially discordant groups are best considered as inconclusive. In contrast, in the fully concordant and fully discordant groups, the sensitivity, specificity, positive predictive value, and negative predictive value were high, 87.5%, 76.9%, 70%, and 90.9%, respectively.
This study demonstrates that hemodynamic responses related to epileptic activity can help delineate the epileptogenic region. Full concordance between maximum response and surgical resection is indicative of seizure freedom, whereas a resection leaving the maximum response intact is likely to lead to a poor outcome. EEG/fMRI is noninvasive but is limited to patients in whom interictal epileptic discharges can be recorded during the 60–90 min scan.
Twenty percent to 30% of patients with epilepsy have drug-resistant epilepsy. For some, surgical treatment could offer the best opportunity of seizure freedom if the epileptogenic zone (EZ) can be defined and removed while preserving eloquent cortex.[2, 3] Various noninvasive presurgical diagnostic studies are first used for localization of the EZ. In some patients, intracranial EEG (iEEG) is indicated, but this approach is invasive and spatial coverage is limited. As part of presurgical evaluation, it is important if a method could provide information about the likelihood of good surgical outcome, which could guide decision making for surgery.
Simultaneous electroencephalography/functional magnetic resonance imaging (EEG/fMRI) recording can investigate noninvasively the brain region involved at the time of epileptic discharges on scalp EEG, by measuring the hemodynamic response (blood oxygenation level–dependent [BOLD] changes).[4, 5] BOLD responses related to interictal epileptic discharges (IEDs) can be confined to the region generating the focal IEDs, but sometimes involve a widespread network.[6, 7] EEG/fMRI has shown good concordance with other noninvasive diagnostic tools,[8, 9] as well as iEEG findings.[8, 10-12] Bénar showed that near the region having BOLD responses, there was at least one “active” iEEG electrode contact. As a presurgical tool, EEG/fMRI is gaining interest, and results suggest that EEG/fMRI helps in surgical planning or to predict surgical outcome. EEG/fMRI provided additional information to scalp EEG to delineate the epileptic focus.[9, 12] In addition, EEG/fMRI helped define a surgical plan in patients who were denied surgery based on other investigations. Finally, a few initial studies showed that EEG/fMRI was useful in delineating the resection site in adults and children with mixed etiology, or specifically in focal cortical dysplasia (FCD).
The current study aims to correlate BOLD responses related to epileptic activity with the localization of surgical resection and determine how EEG/fMRI could help predict surgical outcome in future patients considered for epilepsy surgery.
Materials and Methods
In our institution, EEG/fMRI remains mostly an experimental procedure; it is performed in patients with frequent IEDs (>10 IEDs/h) during routine EEG or telemetry monitoring; sometimes in patients whose focus localization was not clear even without a very active EEG. However, not all the admitted patients who meet the preceding criteria were recruited for an EEG/fMRI study: some patients denied participation, or for others there was limited availability of the MRI scanner.
All consecutive patients with focal epilepsy from our database of EEG/fMRI who underwent surgery after EEG/fMRI from April 2006 to December 2010 and with at least 12-month follow-up were included in this study. Candidates underwent routine presurgical evaluation, and EEG/fMRI was performed independently of other modalities. Initially, EEG/fMRI findings were not used for placing intracranial electrodes or for surgical decision. In few recent cases, EEG/fMRI results were taken into account for presurgical management.
This study was approved by the Montreal Neurological Institute and Hospital Research Ethics Board. All patients gave written informed consent.
EEG was recorded inside a 3 T MRI scanner (Trio; Siemens, Erlangen, Germany) with 25 MR compatible scalp electrodes placed according to 10–20 (reference FCz) and 10–10 (F9, T9, P9, F10, T10, P10) systems, using a BrainAmp system (5 kHz sampling; Brain Products, Munich, Germany). A T1-weighted anatomic image was acquired first using the following sequences: Until July 2008: 1-mm slice thickness; 256 × 256 matrix; echo time (TE), 7.4 msec; repetition time (TR), 23 msec; flip angle 30 degrees and from July 2008: 1-mm slice thickness; 256 × 256 matrix; TE 4.18 msec; TR, 23 msec; flip angle 9 degrees. T1 image was used for superimposition with functional images. Functional data were collected in 6-min runs lasting 60–90 min, with a T2*-weighted echo planar imaging (EPI) sequence: Until July 2008: TR, 1.75 s; TE, 30 msec; 64 × 64 matrix; 25 slices; voxel, 5 × 5 × 5 mm; flip angle 90 degrees and from July 2008: TR, 1.9 s; TE, 25 msec; 64 × 64 matrix; 33 slices; voxel, 3.7 × 3.7 × 3.7 mm; flip angle 90 degrees.
BrainVision Analyzer was applied to remove MR gradient artifacts. Ballistocardiogram artifacts were removed using independent component analysis. IEDs similar to those obtained outside the scanner were marked. IEDs with the same distribution but different morphology were grouped. IEDs with different distributions (e.g., patients with independent bilateral temporal IEDs) were considered as different event types and analyzed independently, but correlation between BOLD response and surgical outcome was obtained only for IEDs on the side of resection.
The method is identical to that used in prior studies.[7-9, 12, 18-20] fMRI images were motion corrected and smoothed (6-mm full width at half maximum). Temporal autocorrelations were accounted for by fitting an autoregressive model of order 1, and low frequency drifts were modeled with a third-order polynomial fitting to each run. Timing and duration of each IED were built as a regressor and convolved with four hemodynamic response functions (HRFs) peaking at 3, 5, 7, and 9 s. Motion parameters were modeled as confounds. All regressors were included in the same general linear model. A statistic t map was created for each regressor using the other regressors as confounds for each event type. A combined t map was created by taking, at each voxel, the maximum t value from the four t maps based on four HRFs. The single combined t map was used for comparison.
To be significant, a response required five contiguous voxels having a t value >3.1 corresponding to p< 0.01 for the individual analysis using each HRF, or equivalently p<0.05 for the combined analysis.[18, 22] In the t maps, a yellow–red scale corresponds to positive BOLD responses (activation) and a blue–white scale corresponds to negative responses (deactivation). Responses outside the cerebral cortex were excluded. Deactivations in the “default mode network” as defined previously were not used for further comparison.
Comparison of BOLD response with resection
Anatomic MRI was performed at least 3 months after surgery, and each anatomic MRI was linearly registered to individual IED–related t map. For patients with more than one surgery, we used the MRI after the last surgery. In a few patients, postsurgical MRI was not available, and postsurgical computerized tomography (CT) or operative protocol was used for location of resection after review by two neurologists.
Patients were classified into four groups defined by the degree of concordance between BOLD responses (activation or deactivation) and location of resection:
Group 1, fully concordant: The maximum t value (t-max) was inside the resection.
Group 2, partially concordant: The t-max was outside but close to the resection (within 2 cm from the margin of resection), and the BOLD cluster with t-max overlapped the resection.
Group 3, partially discordant: The t-max was remote from resection (>2 cm from the margin of resection), but an additional less significant BOLD cluster was in the resection.
Group 4, fully discordant: There was no significant BOLD response in the resection.
Surgical outcome was defined by Engel's classification according to the documents of the last hospital visit. Classes I and II were considered a good outcome and Classes III and IV a poor outcome.
Sensitivity and specificity evaluation
We analyzed the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) using surgical outcome as reference standard. A good surgical outcome was considered as the “ground truth.” Two analyses were done. The first was done in all patients, combining “fully concordant” and “partially concordant” together as “concordant” (positive), and “fully discordant” and “partially discordant” as “discordant” (negative). A second analysis was done considering that “partially concordant” and “partially discordant” were “inconclusive” results. Therefore, only “fully concordant” and “fully discordant” were included in the analysis.
Sensitivity was defined as (patients with good outcome who were classified as concordant/ patients with good outcome) × 100%; specificity as (patients with poor outcome who were classified as discordant/patients with poor outcome × 100%; PPV was defined as (patients classified as concordant who had a good outcome/ patients classified as concordant) × 100%; NPV was defined as (patients classified as discordant who had a poor outcome/ patients classified as discordant) × 100%.
Forty-seven patients had surgery after EEG/fMRI study, and 12 were excluded: 11 had no IEDs inside the scanner and one showed only deactivation in the default mode network. Thus, 35 patients were included (17 male; mean age at evaluation, 29.5 ± 11.6 years, range, 15–65). Clinical and electrophysiologic characteristics are in Table S1. These patients were studied over 57 months; they represent approximately 20% of the patients operated at our institution.
Presurgical anatomic MRI was normal in 9 patients, showed mesial temporal sclerosis (MTS) in 11, a malformation of cortical development in 9 (FCD in 4, multilobar polymicrogyria in 2, nodular heterotopia in 2, and hemimegalencephaly in 1), brain tumor in 3, cerebral atrophy in one, an occipital cyst and cortical atrophy in one, and a temporal horn cyst and bilateral mesial temporal shape changes in one. Nineteen patients were diagnosed with temporal lobe epilepsy (TLE), nine with frontal lobe epilepsy, five with posterior quadrant epilepsy, and two with frontotemporal lobe epilepsy.
Twenty-eight patients had one type of IED. Three patients had independent bilateral temporal lobe IEDs; only responses related to the IEDs over the side of resection were considered for further analysis. One patient had three electrographic seizures and IEDs; both event types showed the same BOLD response and were therefore analyzed together. One patient had no IEDs during scanning but three electrographic seizures that were grouped together. One patient had a typical aura only; in this patient the IEDs showed no BOLD response and we therefore analyzed the response to the aura. Finally, one patient had two types of IEDs, but only one showed a BOLD response and this one was analyzed. Therefore, each patient had one type of event with BOLD response for further comparison. For simplification, all subsequent BOLD responses will be labeled as IED-related, including those related to seizures. The number of IEDs recorded during the fMRI ranged from 2 to 1,451. Two of the 35 patients showed activations only, three presented deactivations only, and 30 had activations and deactivations.
Relationship between IED-related BOLD responses, resection, and surgical outcome
Six patients (patients 10, 21, 27, 30, 32, and 33) underwent two surgeries after the EEG/fMRI study. Thirty-three patients had a postsurgical anatomic MRI, one (patient 20) only a postsurgical CT, and one (patient 23) did not have imaging study after the surgery and therefore location and extent of the resection were defined through the operative protocol. Among the 26 lesional cases, 7 had lesion completely removed; 7 had lesion mostly removed, and 12 had lesion partly removed (details in Table S2). Postsurgical MRI of patients in groups 1 and 4 are shown as examples in Fig. S1, demonstrating that the resections in group 1 are not generally larger than those in group 4. The size of resection is therefore not a factor in the distribution of patients in the different groups. Mean time of postsurgical follow-up was 25 ± 13 months (range, 12–60).
Ten (28.6%) of 35 patients were classified in group 1, nine (25.7%) in group 2, five (14.3%) in group 3, and 11 (31.4%) in group 4. Fifteen patients (43%) had a good outcome. Detailed information on IEDs, BOLD responses, resection, and outcome is found in Table S2.
Group 1 (fully concordant)
All 10 patients had focal IEDs with a single BOLD cluster, except patient 2 who showed two BOLD clusters (Figs. 1 and 2). By definition, the voxel corresponding to the t-max was inside the resection; the BOLD cluster with t-max was found to be mostly within the resection in the 10 patients: in temporal lobe in 6 (patients 10, 11, 12, 13, 24, and 25), in frontal lobe in 2 (patients 1 and 2), and in posterior quadrant in 2 (patients 28 and 29). The t-max corresponded to activation in eight patients and to deactivation in three.
Seven (70%) of 10 patients achieved seizure freedom (class I), whereas the remaining three had a poor outcome (patient 13, Engel's class III; patients 2 and 25, Engel's class IV).
Group 2 (partially concordant)
A focal BOLD response was observed in four patients (patients 3, 5, 14, and 15, all with focal IEDs) and a widespread BOLD response was observed in five (patients 20 and 33 with focal IEDs, patient 8 with bifrontal IEDs, patient 32 with bioccipital IEDs, and patient 4 with an aura) (Fig. 3). The resection was in the frontal lobe in four patients (patients 3, 4, 5, and 8), in temporal lobe in three (patients 14, 15, and 20), in temporal and parietal lobe in one (patient 33), and in posterior quadrant in one (patient 32). The t-max corresponded to activation in six patients and to deactivation in three.
Four (44%) of nine patients achieved seizure freedom (Engel's class I), whereas the other five had a poor outcome (patients 14 and 33, class III; and patients 8, 20, and 32, class IV).
Group 3 (partially discordant)
All five patients had more than one BOLD cluster (patients 16, 17, 18, and 19 with focal IEDs, and patient 9 with generalized IEDs) (Fig. 4). The BOLD cluster with the t-max was remote from the resection, but another cluster with a lower t value was located inside the resection: in temporal lobe in those four patients with focal IEDs and in frontal lobe in patient 9. The t-max corresponded to activation in all five patients.
Three (60%) of five patients had a good outcome (class I), and the remaining two had a poor outcome (class III).
Group 4 (fully discordant)
All patients except patient 31 had focal IEDs with focal BOLD responses (Fig. 5). All the BOLD responses were outside the resection: in frontal lobe in 5 (patients 6, 7, 30, 34, and 35), in temporal lobe in 5 (patients 21, 22, 23, 26, and 27), and in posterior quadrant in one (patient 31). The t-max corresponded to activation in 10 patients and to deactivation in one.
Only patient 22 (1/11, 9%) achieved seizure freedom (class I) at 12-month follow-up. The other 10 patients had a poor outcome (patients 6, 23, and 27, class III; the others, class IV). The sensitivity, specificity, PPV, and NPV with respect to surgical outcome are shown in Table 1.
Table 1. Sensitivity, specificity, PPV, and NPV with respect to surgical outcome
“Fully concordant” and “partially concordant” were combined as “concordant”; “fully discordant” and “partially discordant” were combined as “discordant”.
Only “fully concordant” and “fully discordant” were included; “partially concordant” and “partially discordant” were considered as inconclusive results.
Sensitivity (95% CI)
Specificity (95% CI)
PPV (95% CI)
NPV (95% CI)
This study reports the correlation between EEG/fMRI response, location of resection, and surgical outcome in a large consecutive series of patients with focal epilepsy. EEG/fMRI appears to have potential clinical utility to delineate the epileptic focus and to contribute to the prediction of surgical outcome: 10 patients showed full concordance between BOLD response and resection, 70% becoming seizure-free; in contrast, 91% of the fully discordant patients (10/11) had a poor outcome. Of interest, 50% of the patients that showed some concordance between BOLD response and resection (7/14 patients from groups 2 and 3) also had a good outcome. When considering all categories of concordance, the sensitivity, specificity, PPV, and NPV with respect to surgical outcome was 73.3%, 60%, 57.9%, and 75%, respectively. When only fully concordant and fully discordant results were considered, the sensitivity, specificity, PPV, and NPV increased to 87.5%, 76.9%, 70%, and 90.9%, respectively.
Previous studies suggested that EEG/fMRI could provide additional information to conventional diagnostic investigations for noninvasive presurgical evaluation.[9, 11-13, 15] Our results demonstrate this in a large group of patients with varied pathologies. The sensitivity and specificity were computed based on the relationship of BOLD response and resection using surgical outcome as reference, as was done in a study evaluating dense array EEG source analysis. In this context we demonstrate the value of EEG/fMRI for predicting the outcome of a specific surgical plan made with standard presurgical evaluation. Our results are best used by classifying the partially concordant and partially discordant categories as inconclusive.
The PPV of full concordance was 70%, whereas the NPV of full discordance was 90.9%. This implies that complete removal of the IED-related BOLD response with t-max is indicative of seizure freedom, whereas a resection leaving the maximum BOLD response intact usually leads to a poor outcome, which is consistent with two studies dealing with relatively small number of patients operated and showing a BOLD response.[11, 14] In the three patients from the fully concordant group with a poor outcome, one (patient 13) was class III at 5-year follow-up. However, between surgery and seizure relapse, this patient remained seizure-free for 2 years, indicating that EEG-fMRI was not totally erroneous. The poor long-term outcome might be explained by a very small resection. In the second patient (patient 2), the IED resulted in two clusters within the spike field having almost the same t value. Both clusters could be epileptogenic, and the removal of only one might explain the unsatisfactory outcome. Finally, the maximum BOLD activation of the third patient (patient 25) was in the small part of neocortex, which was removed as part of a transcortical selective amygdalo-hippocampectomy. A larger neocortical removal may have been more effective.
In the partially concordant and partially discordant groups, some patients were seizure-free (7/14), but the results were not sufficiently discriminating to provide a valuable predictive tool. In the group with partial concordance, the four patients with a favorable outcome showed a focal confined BOLD cluster, suggesting that removal of a critical or sufficiently large portion of a focal BOLD response could lead to good outcome. Another possibility is that in some patients the BOLD response indicates the region but not the precise focus. In the same group, four of five patients with a poor outcome showed extensive BOLD responses, some of which were consistent with the EEG or lesion. In these cases, EEG/fMRI pointed to a complex epileptogenic network, only part of which was removed. This implied that not only the location but also the extent of the BOLD response could impact the outcome, which was also shown in a previous study. It is surprising that three of the five patients of the partially discordant group were seizure-free, since only a small fraction of the BOLD response was in the resected area. Here again, as in group 2 it is possible that BOLD indicates the region but not the precise focus. A larger group will be required to understand this situation.
Deactivation concordant with the spike field or probable EZ is an uncommon but consistent phenomenon.[19, 26, 27] In this series, six patients had deactivation fully or partially concordant with the resection, three with a good outcome. Even in the group with fully discordant results, patient 6 had deactivation outside the resection but concordant with the spike field. The neurophysiologic mechanism of deactivation in the focus has not yet been well explained; it may result from an earlier positive BOLD response peaking before the spike, or from disrupted neurovascular coupling. In addition, BOLD deactivation could also be detected in areas remote from the epileptogenic region. The most common such regions are in the default mode network in generalized and focal epilepsy;[4, 5, 7] they are not usually considered as the IED generator. Hence, we excluded deactivation in the default mode network from analysis. However, it is possible that such deactivations may be a predictor of surgical outcome. The meaning of deactivation in regions other than the epileptic focus (default mode or other network) is unknown. Other resting state networks may be deactivated following epileptic discharges, but this has not been reported.
The current study focused on the IED-related BOLD response, which represents the irritative zone, but the surgical decision was based on clinical evaluation, neuroimaging findings, and definition of seizure-onset zone by scalp or intracranial EEG. Prior evidence,[29, 30] however, supports that brain regions generating IEDs match relatively well the EZ, and our results support this concept. For practical reasons, it is not common to study seizures using fMRI; however, two patients had short electrographic seizures and one had aura during scanning. We included them because such seizures have been shown to provide valuable localization information.[31-33] This makes our study representative of the fact that seizures are occasionally recorded during EEG-fMRI studies.
It is difficult to formally validate the EEG/fMRI results, given that EEG/fMRI, like most imaging methods, does not always give a clear-cut conclusive result for each subject, and sometimes shows BOLD responses in the EZ and in remote areas. BOLD clusters also have different statistical t values and sizes, which are difficult to combine systematically and quantitatively as one criterion. The criteria used for validation of EEG/fMRI results vary from study to study.[8, 11, 12, 14, 15, 27] In this study, the relationship between the BOLD response with the maximum t value and the resection was taken as the central point to define the degree of concordance, and this was well correlated with the surgical outcome.
This study is retrospective and not all the patients who underwent presurgical evaluation at our hospital had an EEG/fMRI study; therefore, we could not draw the conclusion that EEG/fMRI helps predict surgical outcome for all surgical candidates. Patients were recruited from a tertiary epilepsy center with usually complex cases (see for instance the relatively low proportion of temporal lobe epilepsy). Many demonstrated active spiking, and frequent IEDs may indicate a severe condition. Regarding the relationship between completeness of lesion resection and surgical outcome, 26 of our 35 patients had a MRI visible lesion, including MTS. Fourteen had a lesion completely or mostly removed and nine did well. Twelve had their lesion partly removed and three did well. We therefore confirm the known correlation between extent of lesion removal and surgical success, although this correlation is far from perfect. The impact of this correlation on our study is not obvious, as the peak BOLD response is not always in the lesion. The complexity of presurgical evaluation and the moderately successful results that have been published by most centers indicate that there is no simple and safe rule to define the epileptogenic zone. Several localization methods are most often required; none is perfect, and this is one reason we keep looking for new methods to define the epileptogenic zone.
Our selection of patients for EEG-fMRI based on the presence of active interictal discharges imply that our results apply only to patients in whom IEDs can be recorded during the 60–90 min scan. Our results cannot provide information for patients without an active scalp EEG. This applied to the 11 patients who were excluded because no IEDs were identified during the MRI session, and to those who were not recruited for the EEG/fMRI study because of an inactive EEG. Recent methodologic developments such as using information from routine EEG or finding BOLD changes without the EEG may allow increasing the yield of EEG/fMRI in such cases. Larger, prospective, controlled studies are needed to have a systematic view of the utility of EEG/fMRI in presurgical evaluation.
This work was supported by the Canadian Institutes of Health Research [MOP-38079]. The authors thank Natalja Zazubovits for helping to collect and analyze the data.
None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.
Dongmei An, MD is a research fellow at Montreal Neurological Institute.