Deep learning for epileptic seizure prediction from EEG signals: A review

Abstract

Epilepsy, a chronic noncommunicable brain disease affecting nearly 1% of the global population across all ages, manifests through seizures caused by abnormal electrical activity in the brain. Electroencephalogram (EEG) records the spontaneous electrical activity of the brain which is more suitable for analysing Epileptic Seizure (ES) than other modalities such as functional Near-Infrared Spectroscopy (fNIRS) and functional Magnetic Resonance Imaging (fMRI). ES prediction aims to provide advanced warning to patients, allowing timely intervention and preventing dangerous situations. Deep Learning (DL) has emerged as a promising approach for ES prediction due to its superior noise removal capabilities, nonlinear feature representation, and strong classification ability. This paper presents a comprehensive review of DL-based approaches for ES prediction in last 5 years, highlighting current research trends, identifying existing challenges, and suggesting potential future research directions.