Abstracts

This abstract has been invited to present during the Broadening Representation Inclusion and Diversity by Growing Equity (BRIDGE) poster session

Rationale: Only 50% of epilepsy patients become seizure free after surgery. Therefore, the search of intracranial EEG (iEEG) markers of the epileptogenic zone (EZ) is a hot topic. Ictal fast activity and the seizure-onset zone (SOZ) on one hand, and interictal spikes, spike/gamma and high frequency oscillations on the other hand, have all been identified as potential promising biomarkers of the EZ (Frauscher et al., Epilepsia2007). However, conflicting results are reported on their efficiency, probably because of to the lack of direct comparisons of both ictal and interictal activities. To capture a complete picture of the epileptic network, we aimed to identify the best features from both interictal and ictal activities that can predict the EZ and discriminate good (Engel Ia) and poor surgical (II-IV) outcomes.

Methods: Twenty-six patients (15 Engel Ia and 11 Engel II-IV), who underwent epilepsy surgery after iEEG investigation for focal drug-resistant epilepsy with ≥ 1-year post-surgical follow-up, were included. Six features were determined by visually analyzing the interictal and ictal iEEG (Figure 1). The interictal features included the primary irritative zone (PIZ)(Abdallah et al, Neurology 2022) and the spike/gamma zone (SGZ) (John et al, in rev). The ictal EEG features were the preictal zone, SOZ, postictal zone and the zone with the first postictal spikes. For comparison purposes with a more quantitative measure, we also included the list of channels exhibiting ictal fast activity as measured by the epileptogenicity index (EI) (Bartolomei et al, Brain 2008). To assess the predictive performance of each individual feature on the EZ localization, defined as the list of resected channels, we computed the overlap ratio, defined as the number of common channels between those features and the EZ normalized by the total number of channels. We then identified the significant features that discriminate good from poor outcome patients. Next, we divided the data into a training and testing set to evaluate the performance of combination of significant features of features using an ensemble decision tree.

Results: For paired comparisons, the overlap ratios of the following features were significantly higher in Engel Ia patients than in Engel II-IV patients: preictal (Cliff’s d=0.66, p=0.004), SOZ (d=0.61, p=0.009), SGZ (d=0.55, p=0.01) and postictal zones (d=0.50, p=0.003). We found no significant differences between good and poor outcome patients for the PIZ and the zone with the first postictal spikes and for the EI. After applying ensemble decision tree model, the preictal zone (d=0.69, p=0.02) was found as the best feature to predict postsurgical outcome. When combining individual significant features, only the preictal zone and SGZ slightly improved the predictive performance (d=0.74, p=0.01, Figure 2).

Conclusions: By using a novel approach based on both ictal and interictal EEG, our results suggest that visual analysis performs better than the EI in the localization of the EZ. Moreover, both the ictal and interictal EEG contain markers that can be used in the localization of the EZ. Specific attention should be paid to the preictal zone.

Funding: Savoy Foundation, CIHR PJT-175056, FRQS Salary Award