Abstracts

Rationale: EEG Source Imaging (ESI) of visually identified interictal  spikes has proven to be useful to localize the epileptogenic zone (EZ). Despite these encouraging results, ESI of spikes is only used in a limited number of epilepsy centers worldwide because of the technical skill needed to perform ESI. Ictal ESI is even more difficult than ESI of spikes because of the dynamics of EEG patterns and frequent contamination by movement-, muscle- and eye-blink-artefacts. Therefore, ictal ESI is almost never used in clinical practice to localize the epileptogenic focus during the presurgical evaluation. In this study we investigate the performance of automated ictal ESI to localize the EZ. This automated analysis would allow epilepsy centers that do not have ESI expertise to incorporate this in the presurgical evaluation. Methods: Sixteen patients who had single resective epilepsy surgery with Engel class I outcome were included. The EEG recorded during the presurgical long-term video-EEG monitoring with standard electrode setup (27 to 32 electrodes) was used for further analysis. In total, 75 seizures were analyzed. The electrophysiological beginning of the seizures was marked by epileptologists with EEG expertise. The automated analysis started by segmenting a fragment of 3s with high signal-to-noise ratio within the first 25s after the marked EEG seizure onset. Afterwards a continuous wavelet transform was done followed by a tensor decomposition to identify seizure components in the EEG. From each ictal fragment, two components were extracted, each consisting of an individual time series, frequency spectrum and EEG topography. The EEG topographies of the components were localized in the brain using ESI with a subject-specific head model generated from the patient’s MRI. For each component, the distance from the ESI localization to the border of the resection was calculated. Correct EZ localization was assumed if 1 of the components was localized in the vicinity of the resection (< 2cm). If 75% of the seizures were correctly localized in a single patient, we considered the method to be able to localize the EZ in that patient. Results: In 61% of the seizures (46/75), the automated method was able to correctly localize the SOZ in the vicinity of the EZ. Overall, the automated method was able to localize the SOZ in the vicinity of the EZ in 69% (11/16) patients.  The median distance to the resection over all patients was 13.2 mm. We present an illustrative case of the automated ictal ESI analysis in fig.1. First, a fragment of 3s is automatically segmented from of the ictal EEG, as depicted by the red lines. Second, EEG tensor decomposition is applied to the segment resulting in two components each with its’ own time series, frequency spectrum and topography. Third, the topographies of the components are localized in the brain using ESI with a patient specific model and the distance from this location to the resection (d) is calculated. The localization of component 1 was 12 mm from the resection, while that of component 2 was in the resection. Conclusions: We showed the potential of automated EEG analysis to localize the SOZ from ictal EEG recordings. This method does not require user-input except for marking the beginning of the seizures in the EEG and can therefore be used in all centers worldwide without any necessary technical skills. Funding: This project received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 660230. The study was supported by the Swiss National Science Foundation (MS: 163398, 140332, 146633, SV 169198).