EEG-based epilepsy detection with graph correlation analysis.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy detection relies on neurophysiological signals like electroencephalogram (EEG).
• Current EEG methods often focus on time-frequency features, neglecting inter-channel correlations.
• Abnormal inter-channel correlations may indicate epileptic activity.

Purpose of the Study:

• To develop a novel epilepsy detection method using graph correlation analysis (EEG-GCA) on EEG signals.
• To leverage inter-channel correlations for identifying abnormal channels and segments indicative of epilepsy.
• To introduce a method that does not require seizure data during training, outperforming existing supervised methods.

Main Methods:

• Utilizing a graph neural network (GNN) with weight sharing to capture and aggregate channel information.
• Employing Kullback-Leibler (KL) divergence regularization to align channel information distributions.
• Detecting anomalies in channels and segments by measuring correlations between different data views during testing.

Main Results:

• The proposed EEG-GCA method achieves state-of-the-art performance in epilepsy detection.
• EEG-GCA outperforms all relevant supervised methods in experimental evaluations.
• The method accurately estimates epilepsy detection by identifying abnormal inter-channel correlations.

Conclusions:

• EEG-GCA offers a reliable and novel approach for epilepsy detection using EEG signals.
• The method's ability to detect epilepsy without seizure training data marks a significant advancement.
• Analyzing inter-channel correlations provides a powerful new dimension for understanding and detecting epilepsy.