Epileptic seizure prediction from EEG using continual learning with CNNs

Abstract

Epilepsy is a persistent neurological disorder that affects over 50 million people worldwide, with nearly one-third of patients remaining unresponsive to conventional therapeutic treatments. This study introduces a progressively adaptive seizure prediction framework designed to enhance early detection and clinical decision-making. The proposed model employs a deep learning strategy grounded in continual learning (CL) principles, using Convolutional Neural Networks (CNNs) in combination with knowledge distillation techniques. This enables the model to assimilate new data while retaining previously learned information. The approach was evaluated on the publicly available Bonn University EEG dataset, following a sequential learning process in which each successive model iteration improved prediction performance. The final model version (Model C) demonstrated significantly improved predictive performance, showing strong stability and generalization across sequential learning stages. Its results clearly outperformed conventional machine learning classifiers, including support vector machines, logistic regression, and random forests.The results demonstrate that continual learning architectures can effectively manage evolving EEG data streams, offering stable and accurate seizure prediction without retraining from scratch. Overall, this research emphasizes the benefits of continual deep learning in clinical applications, establishing a foundation for intelligent, scalable, and real-time seizure prediction systems for AI-assisted healthcare monitoring.