Automated detection of the epileptogenic zone in stereoelectroencephalography for drug-resistant epilepsy using...

Neurosurgery is a viable treatment option for patients with drug-resistant epilepsy (DRE), where accurate localization of the epileptogenic zone (EZ) is crucial for surgical success. This study aims to develop an interpretable machine learning (ML) framework that integrates electrophysiological features to enhance EZ localization.

We retrospectively reviewed patients with DRE who underwent stereoelectroencephalography (SEEG) exploration between January 2020 and December 2023. Multiple epileptogenic biomarkers, including the Epileptogenicity Index (EI), spike rate, ripple rate, and fast ripple rate, were computed. For improved EZ localization, model development was conducted using various machine learning algorithms that integrated these interictal and ictal electrophysiological features. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). SHapley Additive exPlanations (SHAP) were employed to interpret the ML models.

A total of 38 patients with 1671 SEEG channels were included in the final analysis. Four ML models were tested, achieving AUCs ranging from 0.767 to 0.798 for predicting the EZ in patients with DRE. Among these, the deep learning model demonstrated the highest performance, with an AUC of 0.798, and was selected as the optimal predictive model. SHAP analysis identified spike rate and the EI as the most influential features, underscoring their dominant role in the model's decision-making.

Machine learning is a reliable tool for predicting the epileptogenic zone in patients with drug-resistant epilepsy. The use of SHAP methods to interpret the deep learning model offers clinically relevant insights and may assist clinicians in optimizing patient-specific management strategies.