Epileptic Seizure Detection from EEG Signals Using Long Short-Term Memory-Transformer with Self-Supervised Learning

This study proposes SALT, a deep learning model combining LSTM-Transformer architecture with self-supervised pretraining to detect epileptic seizures from EEG signals without requiring large labeled datasets. The method achieved >98% sensitivity and accuracy on two public benchmark datasets using both segment-based and event-based evaluation protocols.