Feature Fusion Based on Temporal-Spatial Attention Model for Automatic Epileptic Seizure Detection

XiangLi, Zhiheng Zhang, Zhaonian Guo, Xin Wang, Ke Zhang, Shixiong Chen

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Epileptic seizure detection, EEG, Temporal- Spatial Attention, Deep Learning, Feature Fusion.

Abstract: Epilepsy, a widespread neurological disorder, poses challenges for timely seizure detection due to complex EEG signals, noise, and data imbalance. We propose the Temporal-Spatial Fusion Attention (TSFA) model, which integrates Bi-LSTM for temporal feature extraction and Pearson correlation for spatial dependencies, fused via an attention mechanism to enhance seizure detection. Evaluated on the CHB-MIT and Siena datasets, TSFA achieves a 10-fold cross-validation accuracy of 97.15% and F1-score of 96.45% on CHB-MIT, and a 10-fold cross-validation accuracy of 98.22% with a leave-one-out cross-validation (LOPOCV) accuracy of 91.65% on Siena, surpassing existing methods. Ablation studies highlight the synergy of temporal and spatial attention, ensuring robustness and generalization. TSFA shows promise for clinical applications, with future work targeting multimodal learning.

Track: 7. General Track

Registration Id: Y6NTJB48SZH

Submission Number: 73