A Compact Graph Convolutional Network With Adaptive Functional Connectivity for Seizure Prediction.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy management requires accurate seizure prediction for improved patient monitoring and treatment.
• Existing seizure prediction models face challenges due to complex spatiotemporal EEG patterns and patient variability.
• Traditional methods often analyze spatial and temporal EEG features independently or rely on predefined connectivity.

Purpose of the Study:

• To develop a compact and effective model for seizure prediction using electroencephalogram (EEG) data.
• To address the limitations of traditional methods in capturing dynamic brain connectivity during seizures.
• To introduce the adaptive functional connectivity graph convolutional network (AFC-GCN) for enhanced seizure prediction.

Main Methods:

• Proposed a novel compact model: the graph convolutional network based on adaptive functional connectivity (AFC-GCN).
• AFC-GCN adaptively infers the evolution of functional connectivity in epilepsy patients during seizures using data-driven approaches.
• The model synchronously analyzes spatiotemporal responses of functional connectivity across multiple brain network topologies.

Main Results:

• AFC-GCN demonstrated accurate and robust seizure prediction performance on the CHB-MIT dataset.
• Achieved high performance metrics: AUC of 0.9820, accuracy of 0.9815, sensitivity of 0.9802, and a low false positive rate (FPR) of 0.0172.
• The model exhibits low computational complexity, making it suitable for practical applications.

Conclusions:

• The AFC-GCN model offers a promising approach for precise and reliable seizure prediction.
• The method effectively captures dynamic functional connectivity, overcoming limitations of previous techniques.
• This approach has significant potential for real-time seizure prediction during daily patient monitoring.