Self-Supervised Learning With Adaptive Graph Modeling for EEG-Based Epileptic Seizure Classification

This paper presents ASGPF, a self-supervised learning framework that uses graph neural networks and gated recurrent units to classify epileptic seizures from EEG signals with improved accuracy even with limited labeled data. The method achieved weighted F1-scores of 83.8% on four-class and 73.5% on eight-class seizure classification tasks on the TUSZ dataset, demonstrating robustness to data scarcity.