Spatio-Temporal Spiking Transformer for Epileptic Seizure Detection on EEG Signals
The automated recognition of epileptic events from long-term electroencephalogram (EEG) monitoring remains a clinically significant challenge, demanding high sensitivity and computational efficiency for real-world deployment. Spiking Neural Networks (SNNs) offer a biologically plausible and energy-efficient alternative to conventional deep learning. However, existing SNN architectures for seizure detection either lack spatial information across electrode channels or fail to model global dependencies across temporal positions. In this paper, we propose a dual-branch Spatio-Temporal Spiking Transformer (ST-Spikformer) to detect epileptic seizure segments from scalp long-term electroencephalogram (EEG) recordings. We evaluated on the CHB-MIT Scalp EEG dataset, which contains 23 recording cases. Experimental results demonstrate high performance against published SNN baselines, with an accuracy of 96.2%, sensitivity of 90%, and specificity of 99%. Our implementation is available at https://github.com/ngochien108/ST-Spikformer.