Physics-informed model of epileptic seizure dynamics

Area of Science:

• Computational Neuroscience
• Machine Learning
• Epilepsy Research

Background:

• Epileptic seizure prediction is challenging due to complex, nonlinear brain dynamics.
• Traditional machine learning models lack generalization; physics-based models lack patient-specific adaptability.
• Current methods struggle to reliably forecast seizure onset.

Purpose of the Study:

• To develop a hybrid physics-informed machine learning framework for improved seizure prediction.
• To integrate the Kuramoto model with neural ordinary differential equations (ODEs) for enhanced EEG signal analysis.
• To improve the robustness, interpretability, and generalizability of seizure forecasting models.

Main Methods:

• Developed a hybrid framework combining the Kuramoto coupled-oscillator model with neural ODEs.
• Utilized multi-channel electroencephalogram (EEG) data from the Temple University Seizure Corpus (TUSZ).
• Compared the hybrid model against purely data-driven and physics-based baseline models.

Main Results:

• The hybrid model demonstrated superior performance in predicting future EEG signals compared to baseline methods.
• Achieved enhanced robustness and generalizability across diverse epileptic seizure patterns.
• Outperformed traditional approaches in predicting seizure dynamics from EEG data.

Conclusions:

• The proposed hybrid physics-informed machine learning framework offers a promising approach for accurate and adaptable seizure prediction.
• This method enhances computational modeling of neural activity and contributes to neuroscience applications.
• The findings suggest potential for improved patient-specific seizure forecasting and early warning systems.