Exploring physical and functional EEG connectivity with multilayer graph transformer convolutional networks for...

Area of Science:

• Affective Computing
• Neuroscience
• Machine Learning

Background:

• Electroencephalogram (EEG)-based emotion recognition offers objective, neural-level insights into emotional states.
• High-dimensional EEG data presents challenges for accurate emotion modeling due to complex spatial and functional characteristics.

Purpose of the Study:

• To propose a novel Multilayer Graph Transformer Convolutional Network (Multilayer-GTCN) for enhanced EEG-based emotion recognition.
• To effectively capture both local and global dependencies within EEG signals.

Main Methods:

• The Multilayer-GTCN framework utilizes a dual-graph approach: a physical proximity graph and a functional connectivity graph.
• Graph Convolutional Networks (GCNs) consolidate stable patterns, while Graph Transformer layers capture long-range dependencies.
• The architecture merges localized structure and global context for robust affective decoding.

Main Results:

• Achieved high accuracy rates on benchmark datasets: 98.24% on SEED, 95.82% on SEED-IV, and 93.35% (valence)/94.11% (arousal) on DEAP.
• Demonstrated the efficiency and flexibility of the Multilayer-GTCN across varied datasets.
• The proposed method provides a robust basis for affective decoding.

Conclusions:

• The Multilayer-GTCN framework effectively decodes emotions from EEG signals by integrating physical and functional connectivity graphs.
• This study lays a foundation for scalable affective computing systems and advances neural signal analysis.
• The approach offers a robust framework for future research in emotion recognition.