EEG microstate-derived dynamic network biomarkers for lateralization and structural etiology in temporal lobe epilepsy.

OBJECTIVE: Temporal lobe epilepsy (TLE) is the most common focal epilepsy but remains highly heterogeneous across hemispheric and structural etiology. This study aimed to characterize microstate-based network dynamics in TLE and evaluate their diagnostic value for seizure lateralization and structural etiology using machine learning. METHODS: Resting-state electroencephalography (EEG) recordings from 150 patients with unilateral TLE (71 right, 79 left) and 65 healthy controls (HCs) were analyzed. EEG signals were segmented into canonical microstates (A, B, C, D), and microstate-specific spatial, and temporal dynamic functional connectivity (dFC) variability metrics were extracted using phase lag index analysis. After two-step feature selection, the appropriate number of features were derived and input into Random Forest, XGBoost, and Support Vector Machine (SVM) classifiers to distinguish: TLE vs HCs, left vs right TLE, and magnetic resonance imaging (MRI)-negative (MRI-neg) vs hippocampal sclerosis (HS) TLE subtypes (TLE-HS). Model performance was evaluated on independent hold-out validation set using receiver operating characteristic analyses. RESULTS: Compared with HCs, patients with TLE exhibited increased duration and occurrence of microstate D and reduced expression microstate B, reflecting maladaptive attentional overactivation and visual suppression. Spatial variability was globally decreased, most prominently in left TLE. SVM achieved excellent performance for TLE detection (area under the curve [AUC] = .98) and lateralization (AUC = .97), whereas classification between MRI-neg TLE and TLE-HS was limited (AUC = .58). SIGNIFICANCE: EEG microstate-derived dFC metrics provide reliable, non-invasive biomarkers for identifying and lateralizing TLE using short duration resting-state EEG recordings. This framework advances understanding of TLE heterogeneity and supports the development of individualized electrophysiological tools for precision diagnosis.