MSFR-GCN: A Multi-Scale Feature Reconstruction Graph Convolutional Network for EEG Emotion and Cognition Recognition

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Graph Convolutional Networks (GCNs) are effective for electroencephalogram (EEG) recognition by analyzing brain connectivity.
• Previous GCN approaches often overlook the intrinsic features of EEG data.
• Developing advanced methods for EEG analysis is crucial for understanding brain states.

Purpose of the Study:

• To propose a novel Multi-Scale Feature Reconstruction Graph Convolutional Network (MSFR-GCN) for enhanced EEG-based emotion and cognition recognition.
• To improve the utilization of EEG features within GCN frameworks.
• To explore the relationship between emotion and cognition for potential rehabilitation applications.

Main Methods:

• The proposed MSFR-GCN incorporates a Multi-Scale Feature Reconstruction (MSFR) module, comprising Multi-Scale Squeeze-and-Excitation (MSSE) and Multi-Scale Sample Reweighting (MSSR) sub-modules.
• MSSE and MSSR dynamically assign weights to EEG channels, frequency bands, and samples based on statistical information.
• A feature-pooling layer is integrated post-GCN to preserve crucial EEG channel information.

Main Results:

• MSFR-GCN demonstrated excellent performance in emotion recognition tasks on the SEED and SEED-IV datasets.
• The model achieved strong results in both emotion and cognition classification on a self-collected dataset (ECED).
• Analysis revealed an implicit relationship between emotion and cognition classification using the proposed method.

Conclusions:

• MSFR-GCN effectively enhances EEG recognition by integrating multi-scale feature reconstruction, outperforming previous methods.
• The model's success in classifying both emotion and cognition suggests its potential for brain-computer interfaces.
• Findings may inform emotional perspective-based rehabilitation strategies for cognitive impairments.