Automatically Extracting and Utilizing EEG Channel Importance Based on Graph Convolutional Network for Emotion...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Graph Convolutional Networks (GCNs) are prevalent for Electroencephalography (EEG) emotion recognition using brain networks.
• Existing GCN models often neglect dimensionality reduction, potentially including irrelevant or interfering network information.
• Effective extraction and utilization of core brain network components are crucial for enhancing model performance.

Purpose of the Study:

• To propose and evaluate novel GCN models for EEG emotion recognition that incorporate core network extraction.
• To investigate the impact of dimensionality reduction on GCN performance in brain network analysis.
• To introduce CWGCN for data-driven core network and channel importance extraction, and CCSR-GCN for emotion recognition using this extracted information.

Main Methods:

• Development of a Core Network Extraction model (CWGCN) utilizing channel weighting and GCN principles.
• Introduction of a Channel Convolution and Style-based Recalibration GCN (CCSR-GCN) model that leverages CWGCN outputs.
• Experimental validation using the SEED dataset to assess model performance.

Main Results:

• Core network extraction demonstrably improves the performance of GCN models for EEG emotion recognition.
• The proposed CWGCN and CCSR-GCN models achieve superior results compared to current popular methods.
• The study highlights the efficacy of data-driven core network extraction in brain network analysis.

Conclusions:

• Dimensionality reduction through core network extraction is a beneficial strategy for GCN-based EEG emotion recognition.
• CWGCN and CCSR-GCN offer a novel and effective approach to analyzing brain networks for emotion recognition tasks.
• The proposed methodology provides a promising perspective for GCN applications in broader brain network analysis.