2.1 Study population

We identified consecutive patients with focal epilepsy and sufficient clinical information for assessment (i.e., seizure semiology, EEG, MRI, PET, and neuropsychological assessment) from a database of patients who underwent EEG-fMRI from January 2014 to December 2016. At our institute, we generally consider performing EEG-fMRI in cases where the EZ is difficult to identify or ambiguous by standard localization methods. We reviewed the patient's EEG to check if there were frequent IEDs to determine the suitability for EEG-fMRI. Frequent IEDs were defined as more than 5–10/h. Therefore, the cases with an EEG study with few or no IEDs before scanning did not undergo an EEG-fMRI. Each participant provided written informed consent in accordance with the Research Ethics Committee of the Montreal Neurological Institute and Hospital.

2.2 EEG-fMRI acquisition

The imaging acquisition methods were as described in previous studies.^{15, 20, 22, 23} EEG was recorded inside a 3 Tesla Siemens Trio MRI scanner with 25 MR-compatible electrodes placed on the scalp, according to the 10–20 (reference FCz) and 10–10 (F9, T9, P9, F10, T10, P10) electrode systems, using a Brain Amp system (Brain Products, 5 kHz sampling). fMRI functional images were collected in 6-minute runs for a total scan time of 60–90 minutes with the patient at rest, using the following T2*-weighted echo planar imaging sequences: repetition time (TR), 1.9 seconds; echo time (TE), 25 ms; 64 × 64 matrix; 33 3.7 mm slices; 3.7 × 3.7 × 3.7 mm³ voxel; flip angle 90°. The IEDs were grouped into types, based on spatial distribution and morphology, and each type constituted a single study. IED types were spike, sharp wave, spike, and wave complex or paroxysmal slow wave, and those IEDs were identified as a burst event when it lasted longer than 1 s; otherwise, it was recognized as a single event. Thus, some patients had more than one study.

fMRI data were analyzed as an event-related design,²⁴ each type of IED was built as a regressor, and all regressors were included in the same general linear model and convolved with four hemodynamic response functions (HRFs) peaking at 3, 5, 7, and 9 seconds.²⁵ A statistic t-map was created for each regressor using the other regressors as confounders for each event type. A combined t-map was created by taking the highest t-value at each voxel from the four t-maps created with the four HRFs (t-value = contrast value divided by the standard error). A threshold at an uncorrected t > 3.1 was applied to the single combined t-map, and the responses with a t-value higher than the significance threshold corresponding to a corrected whole-brain topological false discovery rate (FDR) of 0.05 were identified.²⁶

2.3 Clinical evaluation protocol

The evaluation pipeline is shown in Figure 1. The assessment consisted of a pre-EEG-fMRI evaluation, EEG-fMRI image evaluation, post-EEG-fMRI evaluation, and an assessment of the impact of EEG-fMRI on the clinical evaluation. First, we conducted a pre-EEG-fMRI evaluation according to the clinical history, seizure semiology, EEG, MRI, and PET. Two epilepsy experts independently evaluated the epilepsy syndrome, presumed epileptic focus, and surgical candidacy. SEEGs were included in the evaluation if they were performed within 1 year of the EEG-fMRI scan. Regarding epilepsy syndromes, patients were grouped into temporal lobe epilepsy (TLE), frontal lobe epilepsy (FLE), posterior-quadrant epilepsy (PQE), focal epilepsy, multifocal epilepsy, and undetermined. The definition of “focal epilepsy” was used in cases of suspected focal epilepsy where the region of presumed focus cannot be established, while “undetermined” was used when it cannot be determined whether there is focal or generalized epilepsy. Regarding the epileptic focus, the evaluators selected a lobe that seemed most likely to include the presumed focus from clinical information as mentioned above; patients were grouped into right (R), left (L), or bilateral (B), and grouped into frontal lobe, temporal lobe, posterior-quadrant region, multifocal, or undetermined location. Regarding surgical candidacy, patients were grouped into “yes” (a surgical candidate) or “no” (not a surgical candidate). For each item (epilepsy syndrome, presumed focus, and surgical candidacy), the confidence level was evaluated at three levels (high, moderate, and low). Definitions of estimated confidence level included: high, almost all information (clinical history, all examinations) was concordant with the diagnosis of epileptic syndrome and presumed epileptic focus; moderate, general information was concordant, some information was discordant; low, some information indicates the diagnosis, and the diagnosis is uncertain but more likely than others. Epilepsy syndrome, presumed focus, and surgical candidacy were “evaluation items” in this study.

Not all findings are consistent, and it is often unclear what type of information such as imaging, semiology, EEG, and other findings is most important. The decision on how to weigh the different, possibly contradictory factors reflects clinical judgment and experience. Some clear semiological signs may provide stronger evidence than uncertain imaging results, but the reverse can also occur. The combination of evidence is the essence of the expert's work.

The BOLD signal of EEG-fMRI was evaluated by the degree of concordance between the BOLD response and presumed epileptic focus of the pre-EEG-fMRI evaluation, using the following criteria: concordant, presumed focus and highest positive BOLD response in the same lobe; partially concordant, any BOLD response, including negative BOLD responses, intersected the presumed focus but the highest response was not in the same lobe as the presumed focus; discordant, no BOLD response intersected the presumed focus.

After the BOLD maps were visually inspected, a post-EEG-fMRI evaluation was performed, including epilepsy syndrome, presumed epileptic focus, and surgical candidacy, as for the pre-EEG-fMRI evaluation. We reported “no effect” when the level of confidence was unchanged between pre- and post-EEG-fMRI evaluations, and an “increase” when the BOLD map helped to increase evaluation confidence. “Increase” included instances when the estimated confidence level changed from low to moderate, low to high, or moderate to high in any evaluation item in epilepsy syndrome, presumed focus, or surgical candidacy. On the other hand, we reported a “decrease” if the estimated confidence level was changed upon EEG-fMRI evaluation from high to moderate, high to low, or moderate to low. We reported “changed evaluation only” when the confidence level was unchanged from pre- to post-EEG-fMRI evaluation, but at least one evaluation item was changed (for instance, focus location changed from temporal to frontal, without a change in confidence level).

2.4 Confidence level of EEG-fMRI image

2.5 Statistical analysis

Data were analyzed using SPSS (SPSS Inc.). Continuous variables were compared using the Mann–Whitney U-test, while categorical variables were compared using Pearson's chi-square test or Fisher's exact test. For all analyses, P < 0.05 was considered statistically significant.

3.1 Clinical profiles

In total, 92 consecutive EEG-fMRI studies, performed on 89 patients from January 2014 to December 2016, were included in this study (Figure 2). Three patients underwent EEG-fMRI studies twice. Among the 92 studies, the following were excluded: 23 studies due to the absence of IEDs during the scan (25.0%), two due to the presence of severe artifacts, seven due to generalized epilepsy being diagnosed in the pre-evaluation, and two due to missing information. We excluded patients with presumed generalized epilepsy because this study emphasized the localization value of EEG-fMRI. Therefore, 58 studies from 55 patients were included in the EEG-fMRI evaluation. BOLD maps revealed that seven studies (12.1%) did not have a BOLD response above threshold. Finally, 51 studies from 48 patients were included in the evaluation (Table S1). The mean and standard deviation of the number of analyzed IEDs was 128.1 ± 216.6 (n = 51) in the group with a significant BOLD response, which was significantly higher than in the no significant BOLD response group with 23.7 ± 47.2 (n = 7) (P = 0.005).

The background of the patients was as follows: 15 males, 33 females; median age at the time of scanning, 28 years; median age at onset, 15 years. We performed an EEG and MRI on all patients. Thirteen patients had a pre-existing surgical resection with another intervention under consideration. Regarding the etiology of epilepsy, nine patients suffered from focal cortical dysplasia, five from periventricular nodular heterotopia, four from tumor, one from hippocampal sclerosis and 32 were unknown. SEEG was performed in 13 cases. The high proportion of unknown etiologies reflects that EEG-fMRI studies are most often requested in “difficult” patients.

3.2 Pre- and post-EEG-fMRI evaluation

The pre-EEG fMRI case evaluation was performed while considering standard elements of a phase 1 pre-surgical evaluation including EEG, PET, semiology, etc. It is noteworthy that some patients had undergone invasive studies and/or surgery prior to this assessment. The results of pre- and post-EEG-fMRI evaluation are given in Table 1A. The impact of EEG-fMRI results on the clinician's confidence level in identifying the epilepsy syndrome, the epileptic focus, and the patient's surgical candidacy are presented in Table 1B. Three cases are detailed next to demonstrate these findings.

Case 1 (Figure 3A) is an example in which EEG-fMRI increased the confidence level of the evaluators when assessing the various elements of the pre-surgical investigation. The patient was a 24-year-old woman with nocturnal seizures manifesting with hyperventilation and fear. Interictal EEG showed occasional spikes in the right frontal region (F4, Fz). Her MRI was non-lesional and ictal EEG was not recorded. Therefore, the evaluators assumed the presumed focus was right frontal with low-to-moderate confidence because her seizure semiology did not indicate any laterality, and there was little information indicating the localization of the EZ other than the location of IEDs. The BOLD map showed the maximum positive BOLD response in the right frontal region (peak t-value, 7.9), which was concordant with the clinical information. In the post-EEG-fMRI evaluation, both evaluators increased their confidence in the localization of the presumed focus and in surgical candidacy categories.

Case 2 (Figure 3B) is an example in which EEG-fMRI decreased the evaluators' confidence. The patient was a 35-year-old man with seizures manifesting with right oculo-cephalic deviation and vocalization. Interictal EEG showed spikes and fast activity mostly in the right frontal regions and rarely right temporal regions independently. His MRI was non-lesional and ictal EEG was not recorded. Therefore, the evaluators considered the presumed focus in the frontal region as the most likely possibility with a low confidence in laterality, but the maximum positive BOLD response was in the right posterior-quadrant and BOLD responses were widely scattered in other regions in the hemisphere. Therefore, the BOLD map raised doubts about the presumed focus and suggested the possibility of an EZ including a wider right hemispheric area, and thus lowered the confidence level of the presumed focus and surgical candidacy.

Case 3 (Figure 3C) is an example where the diagnosis of epilepsy syndrome was changed to generalized epilepsy due to the EEG-fMRI findings. The patient was a 27-year-old woman with seizures manifesting with right oculocephalic deviation as well as elevation and abduction of the right arm, occasionally evolving to focal to bilateral tonic–clonic seizures. Interictal EEG showed diffuse spikes or poly-spike and wave complexes predominantly seen in the bifrontal head regions. Her MRI was non-lesional. Ictal EEG was controversial because the only change prior to the seizure becoming bilateral tonic–clonic was a low-amplitude fast activity over bilateral frontal head regions. Therefore, the evaluators considered that the seizure semiology indicated focal epilepsy, particularly FLE; however, EEG findings indicated bi-hemispheric involvement. The pre-evaluation was focal epilepsy, but the negative BOLD response was in the default mode network (DMN) and the positive BOLD response was in the thalamus bilaterally. There were positive BOLD responses in the right mesial frontal cortex and bilateral dorsolateral frontal activations. Such features of subcortical regions are observed in generalized epilepsy.²⁷ Considering the fact that the seizure semiology of generalized epilepsy can show focal manifestations, the BOLD map raised the possibility of changing the diagnosis to generalized epilepsy and decreased the confidence level of surgical candidacy.

3.3 EEG-fMRI effect on the evaluation

The main results are shown in Figure 4. Of the 51 studies completed, EEG-fMRI results influenced the clinician's confidence level in 66.7% of cases. In 47.1% of studies, the confidence level was increased, and in 11.8% of studies, the confidence level was decreased. There was a change in evaluation items without a change in confidence level in 7.8% of cases. The confidence level was increased with regards to the classification of epilepsy syndrome in 28.5%, identification of presumed focus in 33.9%, and determination of surgical candidacy in 29.4%. (These figures include duplicates, i.e., cases with the increase of confidence level on two or more evaluation items). The results of surgical candidacy in pre- and post-EEG-fMRI are shown in Table 1A,B. The BOLD map increased the confidence of surgical candidacy in 29.4% of cases (percentage of increased confidence cases in the “yes” group), and 17.7% were confirmed as poor surgical candidates (percentage of increased confidence cases in the “no” group), due to the uncertain epileptic focus localization or the BOLD map demonstrating the presence of multiple foci.

3.4 BOLD signal concordance

The distribution of BOLD signal concordance with the presumed focus was as follows: 27.5% were concordant; 43.2% were partially concordant; 28.5% were discordant. Among the concordant group (concordant and partially concordant), 59.5% showed increasing confidence in any of the evaluation items; among the discordant group, 18.2% showed decreasing confidence. The mean and standard deviation of the number of IEDs in all studies was 128.1 ± 216.6. The mean and standard deviation of the number of IEDs in the concordant group were 126.1 ± 229.5 and 134.4 ± 178.6 in the discordant group. There was no significant group difference in the number of IEDs (P = 0.90).

3.5 Confidence level of BOLD response

Confidence levels of the BOLD maps were analyzed (reflecting the strength of the BOLD response); 32 BOLD maps had high BOLD signal confidence, and 19 BOLD maps had low signal confidence. Among the 32 maps in the high confidence group, 20 were positive and 12 were negative BOLD response. Among the 19 maps in the low confidence group, 12 were positive and seven were negative. There was no association between these positive and negative BOLD responses and etiology. The increases in the confidence level in the clinical assessment were found in 54.7% of the high BOLD signal confidence group and in 34.2% of the low BOLD signal confidence group. There was no significant difference between these two proportions. We evaluated the relationship of the number of IEDs to the BOLD signal confidence: high confidence group, mean 150.6 ± 268.6; low confidence group; mean 91.3 ± 110.3. There was no significant group difference in the number of IEDs (P = 0.28).

4.1 EEG-fMRI effect on epileptic syndrome and localization

Approximately 50% of EEG-fMRIs added confidence regarding epileptic syndrome, presumed focus, and/or surgical candidacy in the present study. A previous report shows that EEG-fMRI contributed to the localization of the epileptic focus in half of the patients in comparison to EEG alone.¹¹ Another study reported that EEG-fMRI was concordant with the spike location of 8/9 patients and it may provide different types of information supporting topography, concordance with PET and SPECT, and structural peculiarities for patients with non-lesional FLE.¹⁸ We used several modalities in addition to EEG in the pre-evaluation, and we confirmed that EEG-fMRI could add information to a relatively standard clinical evaluation.

4.2 EEG-fMRI effect on surgical candidacy

EEG-fMRI increased the estimated confidence of surgical candidacy in 29.4% of patients. A study by Kowalczy et al²⁹ reported that EEG-fMRI results critically influenced the decision to offer surgery in 17% (10/59) of the patients in their cohort. This led to resection in one patient and opened new prospects of surgery in three others. Their results align with our study. On the other hand, there is some evidence^{10, 35} that EEG-fMRI supports the classification of patients as non-surgical candidates in cases when the BOLD response has multiple activation regions. For example, a case in which EEG-fMRI indicated the possibility of multiple EZs was not considered a surgical candidate and would not be reflected in previous studies of surgical outcome. Our study shows that 17.7% of cases were no longer considered surgical candidates, due to uncertainty of the focus or the presence of multiple foci after this was clarified by EEG-fMRI. Therefore, EEG-fMRI may be useful for the localization of the epileptic focus prior to surgical treatment but also in the generation of supportive information for exclusion of a surgical candidate.

4.3 Confidence level of BOLD response

We assessed the confidence of the BOLD response according to our previous report.²³ A formula was used to classify the images into high and low BOLD signal confidence. In our previous study, if the region of the peak response of the high confidence level was not resected, no patient had a good outcome. For medium and low confidence levels, only 18% and 28% had a good outcome, respectively. The authors concluded that the negative predictive value remained high for all confidence levels, and that resection of the primary cluster is necessary but not sufficient to obtain a good outcome because the positive predictive value was low.²³ In the present study, the concordance and confidence level of the BOLD response did not affect the results. The reason may be that some cases of concordant BOLD responses only confirmed the evaluation but did not change it, whereas some discordant cases were considered to have provided important information resulting in a changed evaluation. This suggests that EEG-fMRI may be useful whether or not the BOLD response is concordant with the presumed focus. Another reason may be the limitation of the evaluation of the level of confidence, which is calculated by the difference of t-values between primary and secondary clusters, since even secondary clusters were consistent with EZ in previous reports.^{22, 36}

4.4 Limitations

The key limitation of our study is that it was a retrospective single-center study. A prospective randomized study is desirable, and postoperative outcomes may be a good indicator of whether the EEG-fMRI results lead to better EZ localization and better surgical decisions. A single-center study may contain bias with respect to the patients' background. There may be a selection bias affecting our results since EEG-fMRI is typically requested in complex patients. However, this bias, which exists in all surgical candidacy studies, is seemingly unavoidable. Our successful EEG-fMRI rate (55.4%; 51/92) was similar to that of another study (50%).²⁹ In our investigation, 25% of studies reported inactive IEDs during recording and 7.6% were excluded due to the absence of a significant BOLD cluster. A previous report excluded 46% of scans due to a lack of epileptic activity in the scalp EEG during the recording.²⁹ A lower rate of exclusions in our study may be due to patient selection. Performing a scalp EEG is important to determine whether a patient has active or inactive IEDs, and long-term EEG recording assists in this conclusion.²⁹ At our institute, inactive IED patients were not EEG-fMRI candidates, and this different selection of patients may have affected our results. Conventional IED-related EEG-fMRI is only analyzed when IEDs are recorded during scanning; thus, one of the limitations of EEG-fMRI is that it is only adapted to patients with active IEDs.³⁷ The number of BOLD responses appears to correlate to the number of IEDs analyzed with fewer BOLD responses recorded in patients with few IEDS and vice versa, a phenomenon previously reported in the literature.³⁷ Our results were similar, with significantly fewer IEDs in the no BOLD group than in the group with BOLD responses.

Although not critical for positive identification of the EZ, a test with high negative predictive value could improve patient selection of surgical candidacy, decreasing cost and exposure to unnecessary risk of surgery by this non-invasive technique. Thus, the strength of this study is that it included cases that were not considered for surgery, and it assessed both the increase and decrease of the confidence level by adding EEG-fMRI to a standard clinical evaluation for focal epilepsy.