1 Montreal Neurological Institute and Hospital, McGill University, Montréal, Québec, Canada.
2 Department of Neurology, Duke University Medical Center, Durham, North Carolina, USA.
3 Department of Biomedical Engineering, Duke Pratt School of Engineering, Durham, North Carolina, USA.
4 Department of Neurology, University Hospital of Nancy, Lorraine University, Nancy, France.
5 Research Center for Automatic Control of Nancy (CRAN), Lorraine University, CNRS, UMR, Nancy, France.
6 Brno Epilepsy Center, First Department of Neurology, St. Anne's University Hospital, Faculty of Medicine, Masaryk University, Brno, Czech Republic.
7 Neurophysiology Unit, Institute of Neurosurgery Dr. Asenjo, Santiago, Chile.
8 Epilepsia Helsinki, Full member of ERN EpiCare, Department of Clinical Neurophysiology, HUS Diagnostic Center, University of Helsinki and Helsinki University Hospital, Helsinki, Finland.
9 Department of Neurology, The Ohio State University Wexner Medical Center, Columbus, Ohio, USA.
10 Department of Clinical Neurophysiology, Aarhus University Hospital, Aarhus, Denmark.
11 Department of Clinical Neurophysiology, Danish Epilepsy Center, Dianalund, Denmark.
12 CHU Grenoble Alpes, Univ. Grenoble Alpes, Inserm, U1216, Grenoble Institut Neurosciences, Grenoble, France.
13 Multimodal Functional Imaging Lab, Department of Physics, Concordia School of Health, Concordia University, Montréal, Québec, Canada.
14 Multimodal Functional Imaging Lab, Biomedical Engineering Department, McGill University, Montréal, Québec, Canada.

Objective: Epilepsy surgery needs predictive features that are easily implemented in clinical practice. Previous studies are limited by small sample sizes, lack of external validation, and complex computational approaches. We aimed to identify and validate visually stereo-electroencephalography (SEEG) features with the highest predictive value for surgical outcome, and assess the reliability of their visual extraction.

Methods: We included 177 patients with drug-resistant epilepsy who underwent SEEG-guided surgery at 4 epilepsy centers. We assessed the predictive performance of 10 SEEG features from various SEEG periods for surgical outcome, using the area under the receiver operating characteristic curve, and considering resected channels and surgical outcome as the gold standard. Findings were validated externally using balanced accuracy. Six experts, blinded to outcome, evaluated the visual reliability of the optimal feature using interrater reliability, percentage agreement (standard deviation ± SD) and Gwet's kappa (? ± SD).

Results: The derivation cohort comprised 100 consecutive patients, each with at least 1-year of postoperative follow up (40% temporal lobe epilepsy; 42% Engel Ia). Spatial co-occurrence of gamma spikes and preictal spikes emerged as the optimal predictive feature of surgical outcome (area under the receiver operating characteristic curve 0.82). Applying the optimized threshold from the derivation cohort, external validation in 2 datasets showed similar performances (balanced accuracy 69.2% and 73.2%). Expert interrater reliability for gamma spikes (percentage agreement, 96% ± 2%; ?, 0.63 ± 0.16) and preictal spikes (percentage agreement, 92% ± 2%; ?, 0.65 ± 0.18) were substantial.

Interpretation: Spatial co-occurrence of gamma spikes and preictal spikes predicts surgical outcome. These visually identifiable features may reduce the burden of SEEG analysis by reducing analysis time, and improve outcome by guiding surgical resection margins. ANN NEUROL 2025.