Delineating the epileptogenic zone from ictal SEEG: A system identification and soft voting ensemble approach

Purpose: Brain surgery remains the only curative treatment for focal refractory epilepsy. Despite ongoing efforts to identify a reliable ”fingerprint” of the epileptogenic zone (EZ), current clinical practice still relies on visual analysis of ictal intracranial EEG signals (stereo-electroencephalography, SEEG). Exploration of alternative approaches is therefore welcomed. Method: We propose a fully automated method based on frequency response function (FRF) estimations to characterize interactions between SEEG contact pairs during ictal periods. Contact pairs are selected using a distance-based strategy within a fixed radius. From FRF estimations and within a relatively low frequency band (1-100 Hz), we extracted features such as phase variance and magnitude characteristics. Then, a logistic regression classifier with soft voting inferred individual contact epileptogenicity from pairwise predictions. Validation was performed on 31 patients rendered seizure-free after surgery (11 from internal dataset, 20 from a public external dataset), using the intersection of visually identified seizure onset zones and surgically treated areas as a proxy for the EZ. Results: Leave-one-subject-out cross-validation yielded 80±22% sensitivity and 79±11% specificity, while cross-dataset validation maintained high sensitivity (82±18%). Clinically, our method offers high-sensitivity screening, capturing 95% of EZ contacts while requiring review of only 58% of all contacts. Conclusions: The proposed method demonstrates encouraging performance in identifying the EZ, with promising generalization capabilities. This new cross-disciplinary approach paves the way for valuable further exploration on larger, more homogeneous cohorts.