Anxiety disorder identification with biomarker detection through subspace-enhanced hypergraph neural network

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Psychiatry

Background:

• Anxiety disorders (AD) are common mental illnesses with significant global impact.
• Accurate classification and biomarker identification are crucial for effective treatment.
• Existing deep learning methods, including graph neural networks (GNNs), show promise but can be improved.

Purpose of the Study:

• To propose a novel subspace-enhanced hypergraph neural network (seHGNN) for improved anxiety disorder classification.
• To enhance feature extraction in hypergraph neural networks (HGNNs) using a learnable incidence matrix.
• To integrate multimodal brain limbic system data for robust AD diagnosis.

Main Methods:

• Development of the subspace-enhanced hypergraph neural network (seHGNN) model.
• Utilization of a learnable incidence matrix to enhance hyperedge influence and feature extraction.
• Integration of multimodal brain limbic system data into a hypergraph framework.
• Application of an ensemble learning strategy to further boost classification performance.

Main Results:

• The seHGNN model achieved an accuracy of 84.46% for anxiety disorder classification.
• An ensemble learning strategy improved the classification accuracy to 94.1%.
• The proposed method outperformed existing deep-learning and GNN-based approaches.
• Identified discriminative AD biomarkers consistent with existing neuroscientific findings.

Conclusions:

• The seHGNN offers a powerful and interpretable tool for anxiety disorder classification.
• The integration of multimodal brain data and advanced HGNN techniques shows significant potential in psychiatric research.
• The identified biomarkers provide valuable insights into the neurobiological underpinnings of anxiety disorders.