Predicting medical refractoriness of patients with temporal lobe epilepsy: EEG-based parameter optimization and network analysis

Abstract

The early identification of refractory epilepsy is important to provide surgical treatment. However, limited studies have used electroencephalography (EEG)-based features to predict medical refractoriness. In this study, we employed feature-based machine learning algorithms to analyze resting-state EEG data to predict drug refractoriness in patients with temporal lobe epilepsy (TLE). This retrospective observational multicenter study included consecutive unilateral TLE patients treated with monotherapy at the time of the first EEG acquisition. Multiple EEG features were extracted from the EEG. The optimal features and frequencies were identified to predict drug refractoriness. Classification was conducted using random forest, extreme gradient boosting, and light gradient boosting models. The features were selected using filter methods and the wrapper method. Graph measurements were compared between the groups. Among the 48 participants, 34 (70.8%) were responsive, while 14 (29.2%) were refractory over a mean follow- up duration of 38.5 months. Coherence feature within the gamma frequency band exhibited the most favorable performance. The light gradient boosting model, employing the mutual information filter-based feature selection method, demonstrated the highest performance (AUROC = 0.821). Interchannel coherence displayed larger values in the refractory epilepsy. Graph theory measurements were higher in the refractory group than in the responsive group. Our study has demonstrated a promising method of identifying the early identification of refractory TLE, a population that may benefit from surgical intervention.