RS-STGCN: Regional-Synergy Spatio-Temporal Graph Convolutional Network for emotion recognition.

Area of Science:

• Affective neuroscience
• Computational neuroscience
• Machine learning for neuroimaging

Background:

• Decoding emotional states from electroencephalography (EEG) is crucial for affective neuroscience.
• Existing methods struggle to integrate neurophysiological priors with dynamic brain activity.
• Accurate modeling of spatio-temporal brain dynamics is essential for emotion recognition.

Purpose of the Study:

• To present a novel framework, the Regional-Synergy Spatio-Temporal Graph Convolutional Network (RS-STGCN), for decoding emotional states from EEG signals.
• To bridge the gap between predefined neurophysiological priors and task-specific functional brain dynamics.
• To develop a method that achieves high accuracy and provides neuroscientific interpretability.

Main Methods:

• Introduced the Regional Synergy Graph Learner (RSGL) to integrate brain-region priors with task-driven optimization.
• Constructed a sparse, adaptive graph modeling connectivity at intra-regional (local) and inter-regional (long-range) levels.
• Applied a spatio-temporal network to capture evolving emotional features using the learned graph.

Main Results:

• Achieved state-of-the-art subject-independent recognition accuracies of 88.00% on SEED and 85.43% on SEED-IV datasets.
• Generated a neuroscientifically interpretable map of functional brain connectivity.
• Identified key frontal-parietal pathways, consistent with known attentional networks.

Conclusions:

• The RS-STGCN framework offers a powerful computational approach for investigating dynamic brain network mechanisms underlying human emotion.
• The model effectively captures evolving emotional features and provides data-driven insights into functional brain organization.
• This work advances the field of emotion recognition through improved EEG signal analysis and interpretable AI.