GL-LFGNN:A Global-Local Dual-branch Causal Graph Neural Network Based on Liang-Kleeman Information Flow for EEG Emotion Recognition

2026-05-24

AI Summary

This study addresses a key limitation in existing graph neural network (GNN)-based approaches for EEG-based emotion recognition, which typically rely on symmetric adjacency matrices and thus fail to capture the directed causal interactions inherent in neural dynamics. To overcome this, the work introduces, for the first time, the Liang-Kleeman information flow theory to construct a neurophysiologically interpretable directed causal graph. Furthermore, it proposes a novel global–local dual-branch GNN architecture that jointly models whole-brain dynamic functional connectivity and region-specific anatomical-functional characteristics. Evaluated on the MEEG dataset, the method achieves state-of-the-art accuracy of 86.17% for arousal and 86.71% for valence classification, with only 37K parameters—approximately 10% of those in current leading models—demonstrating substantial improvements in performance, computational efficiency, and model interpretability.

Abstract

EEG-based emotion recognition holds significant promise for objective diagnosis of mood disorders. Graph neural networks (GNNs) have emerged as the dominant paradigm for modeling inter-channel dependencies in EEG, yet existing approaches rely on symmetric adjacency matrices derived from spatial proximity or functional correlations that fundamentally capture statistical associations rather than directed causal influences, which conflicts with the inherently asymmetric, causally-driven nature of neural information flow. To bridge this gap, we propose GL-LFGNN, a Global-Local Dual-branch Causal Graph Neural Network grounded in Liang-Kleeman information flow theory. Unlike Granger causality that merely assesses temporal precedence, our approach rigorously quantifies causal strength from a dynamical systems perspective, yielding neurophysiologically interpretable directed graphs. A dual-branch architecture further integrates whole-brain connectivity with region-specific processing aligned to established functional neuroanatomy. On the MEEG dataset, GL-LFGNN achieves 86.17% (Arousal) and 86.71% (Valence) accuracy with only 37K parameters -- approximately 10% of the current state-of-the-art -- demonstrating that principled causal modeling can simultaneously enhance interpretability, generalization, and computational efficiency. Code will be released.

Problem

• EEG emotion recognition
• causal influence
• graph neural networks
• asymmetric connectivity
• neural information flow

Innovation

• Causal Graph Neural Network
• Liang-Kleeman Information Flow
• EEG Emotion Recognition
• Directed Functional Connectivity
• Global-Local Dual-branch Architecture