Adaptive Multi-Expert Graph Transformer for Interpretable EEG-Based Diagnostics

2026-07-20

AI Summary

This work addresses the limitation of conventional EEG analysis methods, which reduce dynamic neural synchrony to static features and thus struggle to capture spatiotemporal multiscale abnormal patterns. To overcome this, the authors propose an adaptive multi-expert Graph Transformer architecture that models EEG as a sequence of dynamic functional connectivity graphs. Time-varying connectivity is estimated using the weighted Phase Lag Index (wPLI), and hierarchical graph encoding aggregates information from electrodes to regional and global levels. A multi-expert Transformer combined with an adaptive gating mechanism enables subtype-aware anomaly prediction. Experiments on the TUAB dataset demonstrate superior performance in abnormal EEG detection, confirming the effectiveness and interpretability of dynamic graph modeling and adaptive expert fusion.

Abstract

Electroencephalographic (EEG) abnormalities arise from dynamic changes in neural synchrony across spatial and temporal scales, yet many computational approaches reduce these dynamics to static features. We present a Spatial Multi-Expert Graph Transformer that models each EEG recording as a sequence of dynamic functional connectivity graphs. Time-resolved connectivity is estimated using the weighted Phase Lag Index (wPLI), and hierarchical graph encoding aggregates information from electrode to regional and global levels. A multi-expert transformer architecture enables subtype-aware reasoning, with a gating mechanism adaptively fusing expert outputs for global abnormality prediction. Experiments on the TUAB dataset show competitive abnormal EEG detection performance and demonstrate the potential of dynamic graph modeling with adaptive expert fusion for interpretable, subtype-aware spatial--temporal analysis.

Problem

• EEG
• dynamic functional connectivity
• neural synchrony
• abnormality detection
• interpretable diagnostics

Innovation

• Dynamic Graph Modeling
• Multi-Expert Transformer
• Functional Connectivity
• Adaptive Gating
• Interpretable EEG Diagnostics