EEG VMamba: Vision Mamba for seizure prediction based on EEG

Epileptic seizure prediction algorithms based on EEG signals can help epilepsy patients take timely measures to avoid risks. However, EEG signals possess high dimensionality, nonlinearity, and strong temporal dependencies, making it difficult for models to integrate global and local features and capture long-term dependencies. To address these issues, we propose EEG VMamba, which use the Visual State Space (VSS) block as the backbone and introduce Convolutional Neural Network (CNN) at the later stage. This approach fully combining the local perception capability of CNN and the global modeling capability of VSS. It was evaluated on the publicly available CHB-MIT dataset to demonstrate its effectiveness in seizure prediction, achieving a sensitivity of 91.1% and an AUC of 0.914. Compared to the seizure prediction method based on Vision Transformer, the EEG VMamba demonstrates superior performance, evidenced by higher sensitivity and AUC score.