Leveraging System Identification Techniques for SEEG-Based Epileptogenic Zone Detection

Accurate localization of epileptogenic tissue is paramount in epilepsy surgery. This paper proposes a novel approach using system identification techniques to analyze stereo-electroencephalographic (SEEG) recordings for improved epileptogenic zone (EZ) delineation. By treating adjacent SEEG contacts as coupled systems, we extract frequency response characteristics that capture local network dynamics. Transfer function features fed an XGBoost-based classifier to differentiate between electrode contacts within and outside the EZ. Using ictal data from 11 drug-resistant epilepsy patients who reached seizure freedom post-surgery, our system identification approach achieved 58% sensitivity and 70% specificity, outperforming reference biomarkers in accuracy, F1-score and Index of Balanced Accuracy. Moreover, our method delivers higher computational efficiency and operates independently of both signal amplitude and sampling frequency. These results suggest that system identification features could provide valuable complementary information for EZ delineation in clinical practice.