Spatio-Temporal Spiking Attention Recurrent Neural Network for EEG-Based Motor Imagery Classification

Motor Imagery (MI), a primary research domain of brain-computer interface (BCI) studies, focuses on the recording and analysis of brain activity related to imagined movements, primarily using electroencephalography (EEG). Recent advance-ments in deep learning have significantly enhanced EEG-based MI classification. New challenges, raised with increasing interest in developing low-power neuromorphic devices, brought the demand for spiking neural networks (SNNs), which promise efficient MI classification with reduced energy requirements. However, the application of SNNs to EEG-based MI classification remains relatively unexplored. In this study, we propose a novel approach that integrates spatio-temporal spike encoding with an attention recurrent neural network (ARNN), referred to as spiking ARNN (SARNN). This architecture is designed to capture both the spatial dependencies across EEG channels and the temporal dynamics within the EEG signals, enabling more robust feature extraction and classification. Our method achieved an accuracy of 78.46% on the benchmark BCI Competition IV 2a dataset, demonstrating substantial improvements in MI classification performance and emphasizing the potential of SNNs in BCI applications.