CoCL: EEG connectivity-guided contrastive learning for seizure detection

Abstract

Epilepsy is a neurological disorder characterized by repetitive seizures, making early prediction crucial for patient safety and quality of life. Traditional detection methods primarily rely on time–frequency information from EEG signals. However, since EEG signals are interconnected and abnormal activity spreads across brain regions, understanding their connectivity is essential. This study proposes CoCL, a novel representation learning approach that employs contrastive learning with EEG connectivity-guided supervision to capture these interconnections. When applied during pretraining and transferred to seizure detection, CoCL outperforms state-of-the-art methods and maintains high accuracy with only 6 EEG channels, reducing the need for numerous electrodes.