CLEP: Contrastive Learning for Epileptic Seizure Prediction Using a Spatio-Temporal-Spectral Network

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy diagnosis relies on accurate seizure prediction from electroencephalogram (EEG) signals.
• Current deep learning methods face challenges with data requirements and neglect multi-domain epileptic brain characteristics.
• Existing approaches often lead to suboptimal seizure prediction performance.

Purpose of the Study:

• To develop an effective seizure prediction method addressing limitations of current deep learning strategies.
• To incorporate spatio-temporal-spectral dependencies for enhanced epileptic brain analysis.
• To improve the practicality and efficiency of seizure prediction systems.

Main Methods:

• Proposed Contrastive Learning for Epileptic seizure Prediction (CLEP) using a Spatio-Temporal-Spectral Network (STS-Net).
• CLEP learns cross-subject epileptic EEG patterns via contrastive learning.
• STS-Net extracts multi-scale temporal and spectral features, utilizing a triple attention layer (TAL) and spatio dynamic graph convolution network (sdGCN).

Main Results:

• Achieved 96.7% sensitivity and 0.072 false predictions/hour on the CHB-MIT scalp EEG database.
• Validated on clinical intracranial EEG (iEEG) data, yielding 95% sensitivity and 0.087 false predictions/hour.
• Outperformed state-of-the-art methods, demonstrating the efficacy of the proposed CLEP-STS-Net.

Conclusions:

• The CLEP-STS-Net effectively predicts epileptic seizures by integrating multi-domain EEG features.
• The proposed method offers a more robust and efficient approach compared to existing techniques.
• This work advances the field of automated seizure prediction for clinical applications.