A multi-scale information fusion approach for brain network construction in epileptic EEG analysis

Epilepsy is characterized by complex, multi-scale disruptions in brain connectivity, yet most EEG-based network analyses focus on specific frequency bands or time scales, overlooking crucial cross-scale interactions. In this study, we propose a novel multi-scale information fusion (MSIF) framework that integrates connectivity across multiple frequency bands, temporal windows, and construction methods into a single, fused brain network. By employing Particle Swarm Optimization (PSO), our approach adaptively weights each component to maximize seizure–non-seizure discriminability while preserving stability in non-seizure phases. We validated the MSIF framework using two publicly available EEG datasets (CHB-MIT and Siena) and compared its performance against conventional methods. Our results demonstrate that the MSIF framework outperforms single-scale methods, achieving higher Comprehensive Sensitivity Scores (CSS) and more pronounced separation of seizure vs. non-seizure states. Compared to single-scale methods, the multi-scale fusion significantly enhances sensitivity to seizure-induced network reconfigurations, as evidenced by marked alterations in network metrics (e.g., global efficiency, clustering coefficient) during the seizure phase and a clear return toward baseline in post-seizure segments. These findings underscore the potential of multi-scale fusion to provide richer insights into epileptic network behavior and support more accurate seizure detection and monitoring. The proposed framework paves the way for network-based biomarkers in clinical settings, offering a scalable approach adaptable to diverse electrode configurations and patient populations.