Multi-Head Graph Convolutional Network for Structural Connectome Classification.

Area of Science:

• Neuroimaging
• Machine Learning
• Graph Theory

Background:

• Brain connectivity analysis from diffusion magnetic resonance images (dMRI) is crucial for understanding neurological health and disease.
• Sex-based differences in brain connectomes are significant for personalized medicine and disease research.

Approach:

• A novel machine learning model inspired by graph convolutional networks (GCNs) was developed.
• The model utilizes a parallel GCN mechanism with multiple heads, focusing on both graph edges and nodes for comprehensive feature extraction.
• The model's efficacy was evaluated on sex classification tasks using two publicly available datasets (PREVENT-AD and OASIS3).

Key Points:

• The proposed GCN model effectively captures complementary and representative features from brain connectivity graphs.
• Experiments on PREVENT-AD and OASIS3 datasets demonstrated superior performance compared to existing machine learning algorithms.
• The model's architecture allows for thorough analysis of both nodal and edge-level information in brain networks.

Conclusions:

• The developed GCN model represents a significant advancement in analyzing brain connectivity for classification tasks.
• This approach enhances our understanding of sex-related variations in the human connectome.
• The model's high performance suggests potential applications in clinical diagnostics and neurological research.