Epileptic Seizure Prediction Using Spatiotemporal Feature Fusion on EEG

The proposed model achieved a segment-based accuracy of 98.71% for predicting epileptic seizures.

• The model integrates Graph Attention Network (GAT) for spatial feature extraction and Temporal Convolutional Network (TCN) for capturing temporal features.
• This approach allows for effective identification of the preictal state by leveraging the spatiotemporal relationships in multi-channel EEGs.
• Evaluation on the CHB-MIT database showed a specificity of 98.35% and sensitivity of 99.07%.
• An event-based sensitivity of 97.03% and a low False Positive Rate (FPR) of 0.03/h were also recorded.
• The system demonstrates improved seizure prediction performance without requiring extensive feature engineering.

AI simplified