Exploring EEG Connectivity in Higher-Order Cognitive Tasks with Explainable Graph Neural Networks.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for studying brain activity.
• Existing models like EEGNet lack interpretability for identifying EEG channel connectivity.
• Higher-order cognitive tasks, such as spatial perspective-taking, require sophisticated analysis methods.

Purpose of the Study:

• To develop an interpretable Graph Neural Network (GNN) for EEG connectivity analysis.
• To compare the performance and interpretability of the GNN with EEGNet.
• To identify brain networks critical for spatial perspective-taking.

Main Methods:

• Utilized a GNN architecture based on Self-supervised Graph Attention Networks.
• Incorporated a fully connected graph structure with channel embeddings and a convolutional encoder.
• Employed visualization techniques to identify critical subgraphs and functional connections.

Main Results:

• The GNN achieved performance comparable to EEGNet while offering enhanced interpretability.
• Identified critical subgraphs linked to decision-making, highlighting the fronto-parietal network's role in spatial perspective-taking.
• Demonstrated that GNNs effectively capture long-range functional connections missed by correlation-based methods.

Conclusions:

• GNN-based methods show significant potential for analyzing EEG data in complex cognitive tasks.
• Enhanced model interpretability is crucial for advancing neuroscientific research.
• The fronto-parietal network plays a key role in demanding spatial perspective-taking tasks.