Enhancing electroencephalogram signal quality in epileptic patients using bidirectional stochastic long short-term memory network

Abstract

Artifacts frequently disrupt electroencephalogram (EEG) signal recordings, originating from diverse sources such as eye-blinks and muscle twitches. These artifacts present significant challenges when employing automated systems for diagnosing neurological disorders. In this research, we introduce an innovative architectural solution designed to effectively eliminate these artifacts from EEG signals acquired from individuals with epilepsy. Our proposed framework combines bidirectional long short-term memory networks with bidirectional stochastic configuration networks (BSCN). This integration empowers the model to discern intricate patterns within both past and future time steps of the EEG signal. Furthermore, the non-iterative training characteristic of the BSCN-based classifier enhances training efficiency. To assess the effectiveness of our approach, we conducted experiments on four epilepsy datasets and a sleep dataset. The performance of our novel technique was evaluated using a range of performance metrics, and the results unequivocally indicate its superiority over existing artifact removal methods. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.