Article Text
PDF
Abstract
Background The main challenge in assessing patients with epilepsy for resective surgery is localising seizure onset. Frequently, identification of the irritative and seizure onset zones requires invasive EEG. EEG correlated functional MRI (EEG-fMRI) is a novel imaging technique which may provide localising information with regard to these regions. In patients with focal epilepsy, interictal epileptiform discharge (IED) correlated blood oxygen dependent level (BOLD) signal changes were observed in approximately 50% of patients in whom IEDs are recorded. In 70%, these are concordant with expected seizure onset defined by non-invasive electroclinical information. Assessment of clinical validity requires post-surgical outcome studies which have, to date, been limited to case reports of correlation with intracranial EEG. The value of EEG-fMRI was assessed in patients with focal epilepsy who subsequently underwent epilepsy surgery, and IED correlated fMRI signal changes were related to the resection area and clinical outcome.
Methods Simultaneous EEG-fMRI was recorded in 76 patients undergoing presurgical evaluation and the locations of IED correlated preoperative BOLD signal change were compared with the resected area and postoperative outcome.
Results 21 patients had activations with epileptic activity on EEG-fMRI and 10 underwent surgical resection. Seven of 10 patients were seizure free following surgery and the area of maximal BOLD signal change was concordant with resection in six of seven patients. In the remaining three patients, with reduced seizure frequency post-surgically, areas of significant IED correlated BOLD signal change lay outside the resection. 42 of 55 patients who had no IED related activation underwent resection.
Conclusion These results show the potential value of EEG-fMRI in presurgical evaluation.
- Eeg
- Epilepsy
- Epilepsy, Surgery
- Functional Imaging
Statistics from Altmetric.com
Introduction
In refractory focal epilepsy, surgical resection has the best chance of a good outcome if seizure onset is identified and remote from eloquent cortex.1 The challenge of presurgical evaluation rests in accurate delineation of these regions.
High quality structural MRI has increased the identification of underlying pathology in epilepsy but successful resective surgery is increasingly possible in the absence of MRI abnormalities,2 3 and the epileptogenic zone may extend beyond the margin of abnormal tissue where pathology is seen.1 Standard non-invasive investigations can fail to localise seizure onset and invasive EEG recording is often necessary, which is expensive and has associated morbidity.4 Intracranial recording requires careful patient selection and 70–90% of such patients will subsequently be offered surgical resection.5 There is a pressing need therefore for non-invasive techniques to identify these regions and assist in the planning of invasive recordings.
Fluorodeoxyglucose–positron emission tomography and ictal-interictal single photon emission CT are helpful but lack the spatial resolution of MRI. Neurophysiological approaches (high density EEG and magnetoencephalography) have shown concordance with intracranial recordings and postoperative outcome (ie, have some positive predictive value6 7) but despite excellent temporal resolution are limited by the accuracy of source localisation.
EEG correlated functional MRI (EEG-fMRI), whereby EEG and fMRI are acquired simultaneously, reveals regions of blood oxygen level dependent (BOLD) signal changes associated with interictal (IED) and ictal epileptiform discharges which may provide information about the epileptic network. The methodology combines the spatial resolution of MRI with the temporal resolution of EEG and applying EEG-fMRI to presurgical evaluation has been an important motivation for the technique's development.8–12
To date, clinical validation of EEG-fMRI has consisted of studies comparing IED correlated BOLD signal change with invasive and non-invasive methods of localising the seizure onset zone,9 11–15 reporting up to 70% concordance (colocalisation of areas of maximal positive BOLD signal activation and presumed seizure onset at a lobar level) between IED-related BOLD activations and electroclinical seizure onset in patients with focal epilepsy.
Comparison of novel localisation techniques in focal epilepsy with intracranial EEG, the current gold standard, is considered the best method for validation16 but the approach has drawbacks. Intracranial EEG records directly from regions of interest but has reduced spatial coverage owing to the limited number of electrodes which can be implanted. The problem of source reconstruction found in scalp EEG is not abolished as many regions of interest cannot be accurately sampled using current methods. Nevertheless, it remains one of the best methods of identifying the likely irritative and epileptogenic zones before resection. In EEG-fMRI research, concordance of activations with the irritative zone recorded during invasive monitoring have been reported in small groups, an important step in establishing the technique's clinical utility.12 14 17 One group specifically addressed the use of EEG-fMRI in surgical planning,18 carrying out studies in a group of 29 patients previously rejected for surgery with frequent IEDs. They reported useful EEG-fMRI results in eight patients, four of whom were considered for surgical resection, and suggested that EEG-fMRI may contribute to the surgical decision making process when standard methods did not identify a surgical target.18
Here we compared EEG-fMRI results in a group of patients undergoing surgery with postoperative outcome, assessing whether resection of a region exhibiting IED correlated BOLD activation was associated with postoperative seizure freedom.
Methods
Patients
Seventy-six consecutive patients with refractory focal epilepsy undergoing presurgical evaluation with IEDs recorded during video telemetry underwent EEG-fMRI between December 2005 and May 2008. The EEG-fMRI results did not form any part of the surgical decision making process and were undertaken independently from other investigations without any reduction in medication.
Clinical course
Patients underwent electroclinical assessment, including video EEG, clinical examination and structural MRI (National Hospital for Neurology and Neurosurgery Epilepsy protocol). The decision regarding electroclinical localisation and subsequent resection was made by the clinical team and undertaken with curative intent. Six patients underwent anterior temporal lobe resection and four underwent neocortical resection (two frontal, one parietal and one occipital).
The extent of resection, histopathological diagnosis and International League Against Epilepsy (ILAE) outcome19 were recorded 1 year postoperatively. ILAE outcome is measured by a graded scale summarised as follows: 1=seizure free, 2=auras only, 3=seizures on a maximum of 3 days per year, 4=>50% reduction in seizure frequency, 5=50% reduction to 100% increase in seizure frequency, 6=>100% increase in seizure frequency.
EEG-fMRI acquisition
All patients underwent EEG-fMRI for between 35 and 60 min at 1.5 or 3 T. Patients lay still in the scanner with their eyes closed and with no instruction regarding vigilance. EEG was recorded continuously during fMRI using MR compatible systems (Brain Products, Munich, Germany) along with a scanner synchronisation signal and ECG. Sets of 404 T2* weighted single shot gradient echo, echo planar images (EPI; TE/TR 30/3000 ms at 3 T; TE/TR 0.5/3000 ms at 1.5 T), flip angle 90° 43 (at 3T) and 21 (at 1.5T), interleaved slices (thickness: 3 mm at 3 T; 5 mm at 1.5 T), FOV 24×24 cm2, 642) were acquired continuously on GE MR scanners (GE Medical Systems, Milwaukee, Wisconsin, USA). Offline MRI and pulse related artefacts were removed from the EEG trace20 21 and events marked.
fMRI processing and analysis
The fMRI time series were realigned, spatially smoothed with a cubic Gaussian kernel of 8 mm full width at half maximum and analysed using a general linear model in SPM5 (http://www.fil.ion.ucl.ac.uk/SPM) to identify IED related BOLD changes. Separate sets of regressors were formed for each type of IED allowing identification of specific BOLD effects. Discharges were represented as zero duration events (unit impulse, or ‘∇’, functions) convolved with the canonical haemodynamic response function its temporal and dispersion derivatives, resulting in three regressors for each event type.22 Ictal events were modelled as three ‘blocks’ representing earliest electrographic change, clinical seizure onset and postictal change on the EEG.
Motion related effects were included in the general linear model as 24 regressors representing six scan realignment parameters and a Volterra expansion of these,23 and Heaviside step functions for large motion effects.24 Additional regressors were included for pulse related signal changes.25
F contrasts were used across three regressors corresponding to each event type with a threshold of p<0.05 corrected for multiple comparisons (family-wise error) considered significant. A T contrast (p<0.001 uncorrected for multiple comparison) assessed whether the haemodynamic response function was positive or negative. BOLD responses were considered positive when a positive haemodynamic response function (HRF) was plotted for a given cluster. A less stringent significance threshold was used to explore the data (p<0.001, uncorrected). EPI data were coregistered to the preoperative T1 weighted images to create activation map overlays.26 Clusters of significant BOLD change were labelled anatomically on high resolution EPI images and coregistered with preoperative structural T1 images.
Postoperative imaging
Postoperative T1 weighted MRI was acquired and coregistered with the preoperative images allowing visualisation of fMRI activation maps in relation to the area of resection. Concordance was defined for the cluster of BOLD activation containing the global maximum.
Results
Seventy-six patients underwent EEG-fMRI recordings. Of these 76 patients, 43 had temporal lobe epilepsy, 26 had frontal lobe epilepsy and seven had posterior epilepsies. Fifty-two (68%) of these subsequently underwent surgical resection. Thirty-four of 52 (65%) reached 1 year follow-up of whom 10 (33%) had significant activation on EEG-fMRI. A further 11 had significant activation on EEG-fMRI but did not undergo surgery owing to an extensive epileptogenic zone or overlap with motor function (n=5), intraoperative complications (n=1), patient choice (n=1) or awaiting further evaluation (n=4). In 36/52 (69%) patients who were operated and 13/24 (54%) who were not operated, no IEDs were recorded. The mean number of IEDs during the EEG-fMRI studies was 29.3 across all 76 patients (89.3 in those who had any IED). The median number of IEDs in the group who had BOLD activations and underwent resection was 329 (range 22–635).
Clinical data and EEG-fMRI results are summarised in table 1. Two representative cases are presented in figures 1 and 2. The remainder are available in the supplementary web material (available online).
Clinical data and EEG-fMRI results
Patient No 1. Patient with left mesial temporal lobe epilepsy and left hippocampal sclerosis on structural MRI at 1.5 T. Interictal EEG revealed left temporal spikes. EEG-functional MRI (fMRI) showed widespread activation in the left anterior temporal lobe. Comparison with the postoperative T1 weighted MRI showed the interictal epileptiform discharge (IED) related blood oxygen level dependent (BOLD) cluster within the resection margins. The patient is seizure free (International League Against Epilepsy class 1) 38 months post-surgery. (A) EEG-fMRI. Left temporal spike correlated activation overlaid on EPI in individual space (p<0.05 FWE corrected for multiple comparisons, SPM-T test z=6.89). Crosshair at global maximum. (B) Event related response for the same IED correlated BOLD at global maximum. Event (spike) related haemodynamic response +90% confidence interval (broken lines). Time=time in seconds after event. (C) Postoperative T1 weighted image showing resected region in same location as BOLD response.
Patient No 10. Right frontal lobe epilepsy. Patient with right frontal atrophy. Electroclinical localisation suggested right frontal seizure onset. During EEG-functional MRI (fMRI), bifrontal spike wave discharges were recorded most marked in F4. Electrographic seizures were recorded with similar EEG appearance. Maximal blood oxygen dependent level (BOLD) activation in the right mesial premotor cortex with further significant clusters in the right supplementary motor area and right orbitofrontal cortex was associated with both ictal and interictal activity. Limited right frontal resection was carried out guided by intracranial EEG, and the region of activation extended beyond resection margins. Seizure frequency was unchanged at 12 months (International League Against Epilepsy class 5). (A) Interictal correlated BOLD activation (SPM T-test, corrected for multiple comparisons FWE p<0.05, z=7.6). (B) Plotted response for the same events at global maximum. Event (spike) related haemodynamic response +90% confidence interval (broken lines). Time=time in seconds after event. (C) Postoperative resection seen on T1 weighted MRI. Crosshair in right hemisphere at global maximum between z=7.6) and (B).
Discussion
This series of patients with refractory focal epilepsy demonstrated good correspondence between the localisation of IED related BOLD changes, the area of resection and seizure outcome, with useful information obtained in 10/34 patients in whom resections were carried out and the requisite follow-up period reached. In six of the seven cases that were seizure free postoperatively, the area of resection included the locus of maximum BOLD signal increase. No IED was captured during EEG-fMRI in case Nos 6 or 9 but the area of maximum ictal correlated BOLD signal increase lay within the resection margin. In the remaining three patients who continued to have seizures after surgery, the areas of maximal BOLD activation did not overlap with the resected area. Although the clinical outcomes in these patients are expected given their diagnoses, these results support the contention that IED and ictal EEG-fMRI BOLD signal change are linked to the seizure onset zone in focal epilepsy. As EEG-fMRI data did not contribute to the surgical decision, this gives an unbiased evaluation of a potential role for the technique in presurgical evaluation.
Methodological considerations
EEG-fMRI yield and confounding factors
EEG-fMRI relies on the recording of events during the scanning period, a problem shared with both magnetoencephalography and standard EEG. Events were captured in 40% of patients in this group, which appears low, but previous studies of EEG-fMRI often exhibit selection bias, considering only patients with a very active resting EEG,11 12 in contrast with this study. We observed EEG-fMRI activations in one-third of patients who reached the 1 year follow-up, suggesting the technique is less sensitive than existing advanced imaging techniques such as fluorodeoxyglucose–positron emission tomography (shown in one series to have a diagnostic sensitivity of 44% in patients undergoing surgery for neocortical epilepsy27). However, EEG-fMRI is likely to be of most interest in specific patients who have complex epilepsy syndromes not localised by conventional means, and larger studies in selected patient groups are required to further assess its contribution to presurgical evaluation.
Our approach to fMRI data modelling is designed to ensure that regional BOLD changes explained by confounding factors are not considered as effects of interest, by incorporating these features into the model. We generally use a stringent threshold of p<0.05 FWE corrected for multiple comparisons but in one case have reported the uncorrected, but statistically significant, result as this is confirmed on intracranial recording. We employ rigid techniques to correct for physiological noise. While the statistical tools used to produce maps of activity are designed to control for rates of false positive findings, lack of significant BOLD activation essentially represents low signal to noise ratio (particularly owing to high noise reflecting variance in the baseline) which may be scanner related or physiological. While conservative measures to correct for physiological noise improve the specificity of the model, smoothing restricts the spatial resolution, as do distortions in the EPI data discussed below. We report the global maximum on each SPM, an objective measure of the most statistically significant region of BOLD signal change correlated with the events of interest, irrespective of cluster size or proximity to the region of seizure onset.
It seems clear that BOLD increases generally reflect increases in neuronal activity but in the case of epileptic activity recorded on EEG the relationship is not as clearcut. Previous reports suggested seizures and IEDs are associated with a positive BOLD response.28 Importantly, a lack of activation does not allow any firm inferences to be made on the level of brain activity and in particular lack of regional epileptic activity in this context.
Limits of an interictal study
Caution is required in extrapolating the results of any interictal investigation to make inferences about the epileptogenic zone although the development of an interictal method that may contribute to identification of the epileptogenic zone has advantages. It is clear that EEG-fMRI does not image the epileptogenic zone even in the context of ictal recordings but rather that these results are concordant with seizure onset in straightforward cases and may offer hypotheses for further evaluation in complex cases.
Comparison of EPI and T1 weighted imaging
Coregistration of T1 images and EPI for the localisation of BOLD signal change is problematic, particularly when complicated by changes in brain structure. We addressed this by comparing individualised SPMs of BOLD signal change with each subject's postoperative T1 volume scan, ensuring accurate anatomical localisation of the area of BOLD signal change. Spatial smoothing limits the resolution of EPI but we found visual comparison adequate to compare activations with the resected region.
Clinical significance
Previous studies in focal epilepsy demonstrated regions of IED related BOLD signal change, often concordant with the seizure onset zone determined by electroclinical localisation both in temporal and extratemporal lobe epilepsy, and our data support these findings.9 11 12 29
Validation of the technique's clinical use must now depend on demonstrating added value and can be achieved by comparing the EEG-fMRI findings with those of intracranial EEG. However, successful demonstration of a new localisation technique's capability to predict postsurgical outcome remains the gold standard against which all localisation methods are judged. In this series, we demonstrated that resections which completely removed the region in which IED correlated BOLD signal change were generally associated with seizure freedom, while the finding that areas of significant BOLD activation lying outside of the resection was predictive of poorer outcome in this unbiased sample. It can be argued that the outcomes are unsurprising given the diagnoses, especially in patients with hippocampal sclerosis, but this is a test that any new technique must pass. The results presented here suggest that the clinical utility of EEG-fMRI in presurgical evaluation is likely to be restricted to those patients with focal epilepsy in whom the ictal onset zone is not identified by conventional non-invasive means and who have frequent IEDs (>10/h) on interictal EEG.
In case No 5, bilateral activations were observed in relation to runs of left temporal IED while electroclinical evaluation suggested bilateral seizure foci. Despite the lack of seizures postsurgically, routine scalp EEG remained very active following resection, with frequent runs of IEDs, an independent predictor of poor outcome7 30 This lends support to the theory that multiple EEG-fMRI activations may be predictive of poor outcome in surgery.18 In case Nos 7, 8 and 10, the most significant activations lay outside the region of resection and the epileptogenic and irritative zones were confirmed to be extensive on intracranial recording, lending further support to this hypothesis. In case No 10, there was also rapid propagation of both interictal spikes and seizures from the mesial to the lateral right frontal lobe on intracranial EEG with a similar distribution to the BOLD signal clusters observed. The temporal resolution of fMRI prevents further evaluation of these clusters using EEG-fMRI alone but simultaneous electric source imaging in interictal EEG-fMRI studies demonstrates that widespread or multiple BOLD signal clusters may correspond to the regions of onset and propagation of interictal discharges.31 32
In case Nos 6 and 9, interpretation is more difficult as ictal activity was recorded during EEG-fMRI. BOLD activations were more widespread, but spatially concordant with seizure onset, similar to previous reports of ictal EEG-fMRI in focal epilepsy33 34 which suggested that such widespread activations may represent regions of ictal propagation. Although the areas of activation were more widespread than the resected area, it is notable that the area of most significant early ictal correlated positive BOLD signal change was within the resected tissue in both cases.
Deactivations
This study focused on positive BOLD signal changes, similar to previous investigations which assessed the concordance of BOLD signal change with the ictal onset zone. We observed deactivations remote from the seizure onset zone in seven of 10 cases. This work did not focus on BOLD deactivations, which evidence suggests may represent regions functionally connected with the irritative zone reflecting neuronal inhibition.35–37 Deactivations in these cases were predominantly limited to the contralateral hemisphere and the ‘default mode network’, similar to previous reports of IED associated EEG fMRI, which suggested that such deactivations may represent a subclinical suspension of the resting state related to interictal events.36
Non-resected group
In 24 patients resection was not carried out, including 11 who had activations on EEG-fMRI. The results are not discussed in detail here but it is notable that in those patients in whom the seizure onset zone was found to be extensive on intracranial recording, EEG-fMRI activations were also generally widespread. It is notable that fewer events were recorded in the group that underwent resection compared with those that did not, and in particular a greater proportion of patients who underwent resection had no IEDs compared with those who did not (69% vs 54% of patients with no IEDs). This reflects the case mix, in particular the higher proportion of patients in the resected group with mesial temporal lobe epilepsy in whom fewer IEDs are observed on scalp recordings in general.
Further work
These results suggest that the EEG-fMRI has potential use as a clinical tool, particularly in the subgroup of patients in whom the EEG is active, but localisation using conventional means is difficult. Further work is required to establish its validity in larger patient groups with specific syndromes, in particular those undergoing intracranial recording,18 and also to identify those subgroups in whom the technique adds most value to existing methods of presurgical assessment. Larger studies with longer postoperative follow-up periods will add to the evidence base. Improvement in EPI acquisition and coregistration between MRI modalities may be able to extend the usefulness of EEG-fMRI.
Conclusion
Our results suggest that localisation based on EEG-fMRI of interictal and ictal activity may be a useful adjunct to the preoperative workup of patients in whom surgery for focal epilepsy is being considered, particularly when standard data do not indicate a clearcut focus and patients have frequent IED on routine EEG. We demonstrated good concordance of IED correlated BOLD with the seizure onset zone, and the observation that the presence of IED correlated BOLD activations remote from the seizure onset zone is associated with poorer outcome is particularly interesting. These findings support the argument that EEG-fMRI may have a valuable role in presurgical evaluation in focal epilepsy.
Acknowledgments
HL was supported by the Deutsche Forschungsgemeinschaft (LA 1452/3-1) and the Bundesministerium für Bildung und Forschung (BMBF 01 EV 0703).
References
Supplementary materials
Web Only Data
Files in this Data Supplement:
Footnotes
Funding This work was funded by the Medical Research Council (G0301067). This work was undertaken at UCLH/UCL who received a proportion of funding from the Department of Health's NIHR Biomedical Research Centres funding scheme and supported by a grant from the UK Medical Research Council No. G0301067. We are grateful to the Big Lottery Fund, Wolfson Trust and National Society for Epilepsy for supporting the NSE MRI scanner. HL was supported by the Deutsche Forschungsgemeinschaft (LA 1452/3-1) and the Bundesministerium für Bildung und Forshung (BMBF 01 EV 0703). SV was supported by the “Fonds de Perfectionnement” of the University Hospital of Geneva, Switzerland. JD receives funding from the Wellcome Trust (067176) and the Medical Research Council (G9805089) and the National Society for Epilepsy.
Competing interests None.
Ethics approval This study was conducted with the approval of the National Hospital for Neurology and Neurosurgery and the Institute of Neurology Joint Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.