Epileptic seizures diagnosis and prognosis from EEG signals using heterogeneous graph neural network

Epilepsy, often associated with neurodegenerative disorders following brain strokes, manifests as abnormal electrical activity bursts in the cerebral cortex, disrupting regular brain function. Electroencephalogram (EEG) recordings capture these distinctive brain signals, offering crucial insights into seizure detection and management. This study presents a novel approach leveraging a graph neural network (GNN) model with a heterogeneous graph representation to detect epileptic seizures from EEG data. Utilizing the well-established CHB-MIT EEG dataset for training and evaluation, the proposed method includes preprocessing steps such as signal segmentation, resampling, label encoding, normalization, and exploratory data analysis. We employed a standard train-test split with stratified sampling to ensure class distribution and reduce bias. Experimental comparisons with long short-term memory (LSTM) and recurrent neural network (RNN) models highlight the GNN's superior performance, achieving a classification accuracy of 98.0% and demonstrating incremental improvements in precision and F1-score. These findings emphasize the efficacy of GNN in capturing spatial and temporal dependencies within EEG data, surpassing conventional deep learning techniques. Furthermore, the study highlights the model's interpretability, which is essential for clinical decision-making. By advancing EEG-based seizure prediction methods, this research offers a robust framework for enhancing patient outcomes in epilepsy management while addressing the limitations of existing approaches.