An Explainable Statistical Method for Seizure Prediction Using Brain Functional Connectivity from EEG.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Informatics

Background:

• Epilepsy is a neurological disorder characterized by recurrent seizures.
• Accurate seizure prediction can significantly improve patient quality of life.
• Current deep learning methods for seizure prediction lack interpretability and are computationally intensive.

Purpose of the Study:

• To develop a novel, interpretable, and computationally efficient two-stage statistical method for seizure prediction.
• To overcome the limitations of existing deep learning-based seizure prediction techniques.

Main Methods:

• The proposed method involves two stages: estimating dynamic brain functional connectivity networks and using derived network predictors for seizure prediction.
• The method utilizes scalp electroencephalography (EEG) data to estimate brain networks.

Main Results:

• The method achieved high performance on two datasets: FH-PKU (AUC 0.963, sensitivity 93.1%, FDR 7.7%) and CHB-MIT (AUC 0.940, sensitivity 93.0%, FDR 11.1%).
• The proposed statistical approach outperformed existing state-of-the-art seizure prediction methods.

Conclusions:

• The study presents an explainable statistical method for seizure prediction based on brain network estimation from EEG.
• The developed method offers a viable, interpretable, and efficient alternative to deep learning models for epilepsy seizure prediction.