A multi-frame network model for predicting seizure based on sEEG and iEEG data.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Epilepsy diagnosis relies on electroencephalogram (EEG) signal analysis.
• Current deep learning seizure prediction models require complex preprocessing and struggle with raw data and real-time application.
• Existing single-frame models exhibit poor generalization due to EEG signal non-stationarity.

Purpose of the Study:

• To develop an end-to-end seizure prediction model for direct application to raw EEG data.
• To improve the accuracy and real-time applicability of seizure prediction models.
• To enhance the generalization ability of epilepsy prediction frameworks.

Main Methods:

• Proposed a multi-frame deep learning network for automatic feature extraction and classification.
• Introduced instance-based and sequence-based frames for simultaneous multi-modal feature extraction.
• Developed a model that bypasses complicated pre-processing steps, enabling direct application to raw EEG data.

Main Results:

• The multi-frame network demonstrated superior performance compared to existing models.
• Achieved higher accuracy, sensitivity, specificity, F1-score, and AUC in EEG signal classification.
• Validated the model's effectiveness and generalizability as a robust seizure prediction framework.

Conclusions:

• The multi-frame network offers a novel approach for improved seizure prediction.
• The model's ability to process raw data directly addresses limitations of current methods.
• Future research can integrate this multi-frame concept into state-of-the-art models for enhanced performance.