Objective It remains controversial whether interictal spikes are a surrogate of the seizure onset zone (SOZ). Electric source imaging (ESI) is an increasingly validated non-invasive approach for localising the epileptogenic focus in patients with drug-resistant epilepsy undergoing evaluation for surgery, using high-density scalp EEG and advanced source localisation algorithms that include the patient's own MRI. Here we investigate whether localisation of interictal spikes by ESI provides valuable information on the SOZ.
Methods In 38 patients with focal epilepsy who later underwent intracranial EEG monitoring, we performed ESI of interictal spikes recorded with 128–256-channel EEG. We measured the distance between the ESI maximum and the nearest intracranial electrodes in the SOZ and irritative zone (IZ, the source of interictal spikes). The resection of the region harbouring the ESI maximum was correlated to surgical outcome.
Results The median distance from the ESI maximum to the nearest electrode involved in the SOZ was 17 mm (IQR 8–27). The IZ and SOZ colocalised in most patients (median distance 0 mm, IQR 0–14), supporting the notion that localising interictal spikes is a valid surrogate for the SOZ. There was no difference in accuracy among patients with temporal or extratemporal epilepsy. In the 32 patients who underwent resective surgery, including the ESI maximum in the resection correlated with favourable outcome (p=0.03).
Conclusions Localisation of interictal spikes provides an excellent estimate of the SOZ in the majority of patients. ESI should be taken into account for the management of patients undergoing intracranial recordings.
- Functional Imaging
- Stereotaxic Surgery
Statistics from Altmetric.com
Surgical resection of the epileptogenic zone is a therapeutic option for patients suffering from drug-resistant focal epilepsy, leading to seizure freedom in a substantial proportion of cases.1 In order to maximise the chances of seizure freedom and minimise the risk of neurological deficit, the cortical area responsible for seizure generation must be localised as accurately as possible. Intracranial EEG of the seizure onset zone (SOZ) is the gold standard for this purpose.2 However, because intracranial EEG is invasive and cannot sample the activity from the whole brain, non-invasive approaches play a capital role in selecting patients and designing the implantation strategy.3
EEG-based electric source imaging (ESI) directly estimates the cerebral generators of surface-recorded electric fields.4–8 Using high-density EEG systems, head models that take into account the cerebral anatomy of the individual patient, and distributed inverse solutions, ESI of interictal spikes is among the most accurate non-invasive approaches for localising the epileptogenic zone, as validated by the concordance of ESI with the resected brain volume and surgical outcome.9
Despite these recent progresses, uncertainty persists on two issues. First, it still remains controversial whether the irritative zone (IZ), which produces interictal spikes, colocalises with the SOZ.5 If high-definition ESI of interictal spikes localises the underlying source reliably, and if the IZ is a valuable surrogate for the SOZ, then ESI should identify the SOZ. So far, no study has compared ESI of interictal spikes to the localisation of the SOZ by intracranial EEG. Second, in epilepsy surgery, the resection generally involves a larger volume than just the SOZ and may include non-epileptogenic tissue so that the exact concordance between ESI and the epileptic generators cannot be formally determined from postoperative comparisons alone. In the present study, we set out to compare ESI of interictal spikes to the localisation of the SOZ by intracranial EEG in a cohort of 38 patients with focal epilepsy, assessing the spatial accuracy of ESI on a subcentimetric scale.
This study was approved by the institutional ethical review committee of Geneva University Hospitals. Patients gave their written consent to participate. Between 2000 and 2011, we prospectively recruited 44 consecutive patients with drug-resistant epilepsy who had undergone high-density EEG recordings with at least 128 channels prior to intracranial video-EEG monitoring at the epilepsy unit of Geneva University Hospitals, Switzerland. High-density EEG failed to record epileptic activity in 6 patients (14%). Thus, 38 patients (17 females) were studied further (see online supplementary table S1). Nineteen (50%) had temporal lobe epilepsy (TLE; 14 medial (MTLE), 5 lateral (LTLE)) and 19 (50%) extra-TLE (ETLE). Patients with temporal polar epilepsy were classified in the MTLE group. Median age at evaluation was 24 years (range 3–51); median age at epilepsy onset was 10 years (range 0–33).
EEG recordings, spike selection and averaging
All patients underwent a comprehensive non-invasive evaluation, including neurological, neuropsychological and psychiatric examinations, 32-channel long-term video-EEG monitoring, 3D millimetric MRI, positrion emission tomography (PET). Subtraction ictal–interictal single photon emission computed tomography (SPECT) coregistered with MRI (SISCOM) was performed in 35 patients.
High-density EEG was recorded for 30–60 min using 128-electrode or 256-electrode Geodesic Sensor Nets (Electrical Geodesics Inc., Eugene, OR), as previously described.9 Impedances were kept below 20 kOhm. EEG was filtered (0.1–100 Hz) and digitised at 256–1000 Hz sampling rate with a vertex electrode as reference. A neurologist experienced in clinical EEG (VB, SV, MS) marked all spike peaks that were then averaged within a window of 500 ms prespike and postspike. In cases where more than one spike with different surface topographies were recorded (n=9), only the most frequent topography was retained for further analysis. Artefact-ridden channels were removed and interpolated.
Head model and inverse solution
To compute the forward model, we used a head model based on the individual patient's T1-weighted 3D millimetric MRI. The brain surface was extracted from this MRI, and the best-fitting sphere was calculated. The MRI was then warped according to the ratio of the sphere radius and the real surface radius. Depending on brain size, between 3000 and 5000 solution points were defined at regular distances within the grey matter including deep structures such as the amygdala and the hippocampus.10 The lead field matrix was then computed using the known analytical solutions for a 3-shell spherical head model with a conductivity ratio of 1 : 20 between skull and brain.11 ,12 A linear distributed inverse solution with biophysical constraints was used to calculate the 3D current density distribution.13 Finally, the result was back-transformed to the original head shape. This simplified anatomically constrained head model was shown to reveal similar localisations as a boundary element model.14 The localisation of the solution point with the maximal source amplitude at 50% of the rising phase of the spike was considered for subsequent analysis (ESI maximum). This time point most reliably localises the underlying electrical source, whereas the localisation at the peak of the spike is contaminated by spike propagation.15–17 When we observed rhythmic spike discharges or very close consecutive spikes, only the first spike was analysed. EEG and ESI analyses were performed using the free Cartool software (http://brainmapping.unige.ch/cartool).18
Grids and strips of subdural electrodes (in 13 patients), depth electrodes (in 12 patients) or combinations of both (in 13) were implanted (range 38–128 electrodes) according to the individual non-invasive presurgical work-up (see online supplementary table S1). Intracranial electrodes were localised on a postimplantation CT scan and coregistered with the preimplantation MRI using Analyze 9 (Biomedical Imaging Resource, Mayo Clinic, Rochester, Minnesota, USA). To compensate for the brain shift caused by subdural electrode implantation, these electrodes were projected back orthogonally to the brain surface of the preimplantation MRI.
Interictal spikes and seizures recorded with intracranial EEG were reviewed by board-certified neurologists and neurophysiologists with additional experience in intracranial EEG (SV, MS). Electrodes displaying interictal spikes formed the IZ. Likewise, electrodes displaying the earliest ictal activity formed the SOZ.19 Contacts involved only in the propagation of interictal spikes or seizures were not included in the IZ or SOZ.
Following intracranial video-EEG monitoring, surgery was carried out in 32/38 patients (84%). Follow-up was at least 1 year. Outcome was more favourable for patients with TLE (10/15, 67%, were Engel class I, ie, seizure-free) than for those with ETLE (5/17, 29%; χ2=4.4414, p=0.0351).
Euclidian distances were measured between the ESI maximum and the nearest intracranial electrodes involved in the IZ and the SOZ. The frequency of intracranial spikes was not a criterion in deciding which intracranial spike corresponded to the scalp spike. The fact that subdural electrodes lie at the surface of the brain over the gyral crowns spuriously increases the ESI-electrode distance if the ESI maximum is in fact buried in the depth of a sulcus. Therefore, when subdural electrodes were used, we corrected for this depth error by orthogonally projecting the ESI maximum onto the brain surface.
In the patients who underwent surgery, we measured the distance between the ESI maximum and the resection cavity on a postoperative MRI (or CT in two cases) coregistered to the preoperative MRI. ESI was considered concordant with the resection when the solution point with the ESI maximum was located within or on the margin of the resection revealed by postoperative imaging. A solution point was on the margin of the resection when one of its direct neighbours was inside the resection.
We used the χ2 test to evaluate differences between proportions, the Wilcoxon rank sum test to compare differences between distributions and the Spearman rank correlation coefficient to assess the significance of correlations. Graphical data representation and statistical analysis were performed using MATLAB 9 with the Statistics Toolbox (MathWorks, Natick, Massachusetts, USA). Values of p≤0.05 were considered significant.
ESI of interictal spikes and the SOZ
The ESI–SOZ distance in a given patient is determined by (1) the accuracy of ESI in localising the source of interictal spikes (the ESI–IZ distance) and (2) the degree to which the IZ overlaps with the SOZ (the IZ–SOZ distance). The median distance between the ESI maximum and the nearest IZ electrode was 15 mm (IQR 8–21; figure 1A). The distribution of ESI–IZ distances appeared unimodal, with the distances being smaller than 30 mm in all but one outlying patient and smaller than 20 mm in 71% of patients.
The median distance between the IZ and SOZ electrodes nearest the ESI maximum was 0 mm (IQR 0–14). The IZ and SOZ electrodes nearest the ESI maximum were the same electrode in 21/38 patients (55%).
The median distance between the ESI maximum and the nearest electrode involved in the SOZ was 17 mm (IQR 8–27; figure 1C). The distribution of ESI–SOZ distances was bimodal, with most distances (87%) being smaller than 35 mm. Figure 2A–C illustrates successful localisation of the SOZ by ESI.
Patients with discordant localisation of IZ and SOZ
Examining the five patients with the largest ESI–SOZ distances, we found that they all had ESI–IZ distances comparable to those of the other patients (see figure 1B). In these patients, all with multifocal IZs, the ESI in fact accurately localised a part of the IZ, but that IZ was not part of the SOZ. Only one spike topography was recorded during high-density scalp EEG in these patients. In Patient 1 (patient numbers refer to see online supplementary table S1), the ESI localised a right inferior frontal IZ, whereas the SOZ lay in the right medial temporal lobe. In Patient 7 (figure 2D), the ESI localised a left temporo-parietal IZ concordant with clinical video-EEG and PET results, whereas the SOZ was in the left basal temporal lobe. Patient 4 had bilateral temporal IZs in clinical scalp and intracranial EEG but only left-sided spikes were captured during the high-density EEG recording. The ESI identified a left temporal IZ, whereas the SOZ lay in the right temporal lobe. Similarly, Patient 38 had bilateral frontal IZs, but ESI only characterised the left frontal IZ, whereas the SOZ was located in the right frontal lobe. Finally, Patient 32 had a left frontal periventricular heterotopia from where spikes and seizures originated. However, the head model did not allow for isolated solution points to be placed in the small heterotopia and the ESI could therefore technically not be located there. It lay instead in the left middle frontal gyrus which showed propagation from the spikes generated in the periventricular nodular heterotopia. In four of these five patients, surgery either was not offered or did not bring persistent seizure freedom, illustrating the complex nature of their epileptic networks.
ESI and epilepsy subtype
Comparing patients with TLE and ETLE, we found no significant differences in the ESI–SOZ distance (Wilcoxon rank sum statistic W=413, p=0.2201, two-tailed). There were also no significant differences when comparing patients with medial TLE versus neocortical epilepsies (LTLE and ETLE) (W=312, p=0.244, two-tailed). Finally, we found no significant difference between patients with medial and lateral TLE (W=45, p=0.6868). Hence, ESI did not appear to perform better in any epilepsy subtype, although the small number of patients with LTLE (n=5) calls for caution in interpreting this comparison.
ESI and postoperative outcome
In the 32 patients who underwent resective surgery, the ESI maximum lay within the resection significantly more often in patients who became seizure-free (Engel class I outcome: 12/15 patients, 80%) than in those with less favourable outcomes (7/17 patients, 41%; χ2=4.9795, p=0.0256). This suggests that careful analysis of interictal spikes is useful in delineating the epileptogenic zone.
Outcomes in the patients in whom more than one spike topography was recorded during high-density EEG (n=9) were not different from those of patients with a single spike topography. Among the patients in whom the ESI–SOZ distance was less than 10 mm (n=11), the outcome was Engel class I in six and class III or IV in three (two patients did not undergo resective surgery).
In this study, we evaluated the accuracy of ESI of interictal spikes in delineating the SOZ defined by intracranial EEG. Our main finding is that the localisation of interictal spikes by ESI in the individual patient's own MRI provides an accurate estimate of the SOZ. Furthermore, including the source maximum in the resected brain volume is associated with a favourable postoperative outcome, indicating that ESI of interictal spikes helps delineating the epileptogenic zone. Importantly, ESI performs similarly well in temporal and extratemporal epilepsy. These results bolster the role of ESI of interictal spikes as a reliable tool to delineate the SOZ and add to the evidence that ESI has an important role to play in defining the strategy for the implantation of intracranial electrodes and for resective surgery.
As mentioned earlier, no study that we know of has compared ESI of interictal spikes to intracranial EEG localisation of the SOZ. A magnetoencephalographic source imaging study of interictal spikes found colocalisation with the SOZ at a sublobar level in 78% of cases of LTLE and 45% of cases of MTLE and ETLE.20 ,21 Comparing these encouraging results with ours is not straightforward, because sublobar regions are defined according to anatomical landmarks and can differ widely in size and shape. We estimated the average size of a sublobar region: the total surface of one cerebral hemisphere is 820 cm2, two-thirds of which are buried in sulci.22 Its outer surface is therefore about 270 cm2. Parcellating this surface into 13 circles of equal area (the number of regions per hemisphere used by Knowlton et al20 ,21) yields regions with a radius of about 25 mm. In our study, 71% of ESI–SOZ distances were below 25 mm, a performance that compares favourably to the aforementioned study.
Successful epilepsy surgery by definition entails resecting the entire epileptogenic zone.2 However, because this zone cannot be defined unambiguously before surgery, the IZ and SOZ must be used as surrogates in clinical practice. Of these, the SOZ is often considered to be the better one,23 but the vast majority of non-invasive localising techniques are based on interictal activity, as seizures can generally only be recorded serendipitously or with long-term recordings and are frequently accompanied by motion artefacts. Our finding that the IZ colocalised with the SOZ in most cases is in line with previous research and affects ESI and magnetic source imaging (MSI) or EEG-fMRI.16 ,24 ,25 In addition, including the ESI maximum in the resection volume was associated with a favourable postoperative outcome, confirming our previous findings.9 Similarly, good prognostic performance was also found with MSI.26 The clinical importance of delineating the IZ in addition to the SOZ is also underlined by findings that patients whose SOZ is completely contained in the resected brain volume, but whose IZ extends beyond the margins of the surgical resection, had poorer surgical outcome.27 The perilesional primary IZ (but not remote interictal generators) might be a better surrogate of the epileptogenic zone than a very focal SOZ corresponding to only a subset of the epileptic network.
In a minority of our patients, ESI localised an IZ that was not part of the SOZ as revealed by intracranial EEG.28 These patients all had complex epileptic networks with multifocal, sometimes bihemispheric IZs and poorer outcome if operated. As it is true of any investigation, ESI should not be interpreted in isolation, but must be integrated within the complete clinical, electrophysiological and neuroimaging picture for optimal management. Recording high-density EEG for longer periods of time, including during sleep, will likely increase the yield of ESI and the detection of multifocal activity. Further, imaging the source of each spike topography may help planning the implantation of intracranial electrodes in order to better sample the complex, multifocal networks generating spikes and seizures.
In our study, the accuracy of ESI was limited by the following methodological considerations. Its spatial resolution is limited by the distance between neighbouring solution points (about 5 mm in each anatomical plane). Additionally, we restricted our analysis of ESI results to the point of maximum intensity, not taking into account the spatial spread of the source analysis for which there is currently no established algorithm. Furthermore, intracranial EEG is an imperfect gold standard: its spatial sampling (the areas of the brain recorded by the intracranial electrodes) is always partial and its spatial resolution is rather in the range of 5–10 mm, so that the epileptogenic zone may be missed (eg, if it is buried in a deep sulcus).21 This is illustrated in our patients by the fact that a short ESI–SOZ distance was not unequivocally related to an excellent postoperative outcome. Keeping these limitations in mind, ESI–IZ and ESI–SOZ distances of 20 mm or below, as observed in the majority of our patients, can be considered very accurate.
An open question in ESI and MSI pertains to identifying the surface counterparts of intracranial EEG spikes, and particularly whether spikes generated by and confined to the medial temporal lobe can be recorded at the scalp surface. Probably the best way of tackling this issue is to compare simultaneously recorded scalp EEG or magnetoencephalography (MEG) with intracranial EEG.5 Studies which applied this methodology to small patient numbers yielded conflicting results: some found that a fraction of medial temporal spikes are visible at the scalp surface29–31 while others did not.32 Further research using simultaneous scalp and intracranial recordings in larger patient cohorts will be necessary to better address this important point.
A related concern is the possibility that the rapid propagation of spikes away from their initial generators towards distant brain regions16 ,17 might lead to erroneous electromagnetic source imaging results. ESI performed on the initial rising phase of spikes strongly reduces such contamination.15 Here as well, simultaneous scalp and intracranial EEG recordings will be helpful in understanding how spike propagation can affect the performance of ESI.
Our study illustrates that ESI of interictal epileptic spikes provides an accurate estimate of the seizure onset and epileptogenic zones. We suggest that the accuracy of ESI is sufficient to influence surgical decision-making in patients with drug-resistant epilepsy undergoing presurgical evaluation, similarly to findings obtained with MSI.3 ,33 ,34 ESI performs similarly well in patients with temporal or extra-TLE. The ease of use of current high-density EEG systems (bedside recording, no sedation required, feasible in young children and cognitively impaired patients) should increase the availability of ESI. This validation of ESI is also informative for other applications of the technique, in clinical neurology for localising and testing the function of sensory, motor and cognitive cortical areas and in cognitive neuroscience.35 ,36 Finally, our findings give support and clinical relevance to other techniques which use interictal epileptogenic discharges to identify the SOZ, such as MEG and simultaneous EEG-fMRI.4 ,37
The Cartool software (http://brainmapping.unige.ch/cartool) has been programmed by Denis Brunet from the Functional Brain Mapping Laboratory, Geneva, and is supported by the Center for Biomedical Imaging (CIBM) of Geneva and Lausanne, Switzerland.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online table
SV and MS contributed equally
Contributors PM: study design, data collection and analysis, drafting the manuscript, preparing the figures. LS: study design, data collection and analysis. MG, VB and SM: data collection and analysis. KS: study design, data analysis. CMM: study design, data analysis, drafting the manuscript. SV and MS: study design, data collection and analysis, drafting the manuscript. All authors reviewed the manuscript and approved the final version.
Funding Swiss National Science Foundation grants # 139829 to PM, # 122073 to KS, # 132952 to CMM, # 141165 to SV and # 140332 to MS.
Competing interests CMM receives honoraria from Springer as editor-in-chief of Brain Topography and receives royalties from Cambridge University Press as one of the editors of the book Electrical Neuroimaging. SV received speaker's fees from Electrical Geodesics, Inc., for an invited lecture in an industry symposium, and serves as a consultant in advisory boards for Eisai Pharmaceuticals and Desitin Pharma. MS has received speaker's fees from Electrical Geodesics, Inc., for an invited lecture in an industry symposium, and serves as a consultant in advisory boards for Eisai Pharmaceuticals, UCB Pharma and GlaxoSmithKline. Electrical Geodesics, Inc., the company that manufactures some of the equipment used to acquire high-density EEG in this study, played no role in study design, data collection and analysis, or in writing and submission of the paper.
Patient consent Obtained.
Ethics approval This study was approved by the institutional ethical review committee of Geneva University Hospitals.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.