Preliminary Epilepsy Screening Using Multi- dimensional Brain Network Features

Abstract Epilepsy is a neurological disorder caused by abnormal neuronal activity, resulting in brain dysfunction. Preliminary screening and diagnosis during non-seizure periods are crucial, as they enable patients to better understand their condition, prepare for potential seizures, and identify optimal intervention opportunities. However, compared to obvious behavioural abnormalities during seizures, patients with epilepsy exhibit behaviour comparable to healthy individuals during non-seizure periods, making early detection challenging. Electroencephalogram (EEG) signals from patients with epilepsy exhibit complex spatiotemporal dynamics across multiple interconnected brain regions, offering valuable insights for preliminary screening.This study constructed a brain network during the interictal period of epilepsy using EEG data and integrated time-domain, frequency-domain, and nonlinear EEG features to comprehensively characterize the brain activity for patients with epilepsy. We developed a multimodal-gated graph convolutional network (MG-GCN) based on graph convolutional neural networks (GCNs). By incorporating a structure-aware regularization term, we improved the model's sensitivity to graph-based structural information in EEG data. Additionally, a cross-attention mechanism was employed to effectively fuse EEG signals, and brain networks, enabling preliminary screening and diagnosis.We found that there were differences in brain network structure between the interictal period of epilepsy and the resting state of healthy subjects in terms of functional connectivity and topological structure. Compared to existing methods, our model demonstrated superior performance, achieving an accuracy, sensitivity, specificity, F1-score, and precision of 95.08%, 94.61%, 96.03%, 94.65%, and 95.08%, respectively.The method proposed in this study achieves strong classification performance in the preliminary screening of interictal periods in epilepsy and offers a practical approach for auxiliary diagnosis during non-ictal phases.