Abstract
INTRODUCTION: Temporal lobe epilepsy (TLE), the most common type of drug-resistant epilepsy (DRE), has a postoperative seizure-free rate of ~70%. Furthermore, precisely localizing the epileptogenic zone and determining the surgical resection area have been established as the key factors influencing surgical outcomes. Herein, we innovatively coupled the surgical resection area with characteristics of effective connectivity via intracranial electroencephalography (iEEG) to predict patients' surgical prognosis. METHODS: This study involved 56 patients who underwent TLE surgery and were followed up for over 1 year. All patients underwent stereo-electroencephalography (SEEG) electrode implantation and single-pulse electrical stimulation (SPES) tests. After comparing patients' RMS value of N1/N2 (Z-score standardized) from cortico-cortical evoked potentials (CCEP) with different surgical outcomes, an interpretable machine learning (ML) model based on support vector machine (SVM) for predicting patients' surgical prognosis was constructed. RESULTS: Patients with various surgical outcomes exhibited differences in effective connectivity. Furthermore, compared to the seizure-free group (Engel I), patients in the nonseizure-free group (Engel II-IV) exhibited stronger connectivity between the seizure onset zone (SOZ) and regions outside the surgical resection area. The nonseizure-free group also exhibited stronger connectivity between the surgical resection area and regions outside the resection area. Our prediction model demonstrated high-accuracy performance, with accuracy and area under the curve (AUC) values of 0.800 and 0.893, respectively. CONCLUSIONS: This study confirmed the potential value of integrating the surgical resection area and effective connectivity characteristics in predicting patients' surgical outcomes; offering a novel approach that could be leveraged to precisely determine the surgical resection area and improve TLE patients' surgical prognosis.