Joint Graph Convolution for Analyzing Brain Structural and Functional Connectome.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Brain's white matter architecture supports neuronal synchrony and functional connectivity.
• Relating structural (diffusion tensor imaging) and functional (resting-state fMRI) brain networks is a key challenge in systems neuroscience.
• Graph convolutional networks (GCNs) have been used for analyzing individual network types but not their inter-relationships.

Purpose of the Study:

• To develop a novel method, Joint-GCN, for analyzing coupled structural and functional brain networks.
• To investigate inter-network relationships by integrating structural and functional data within a single GCN framework.
• To assess the performance of Joint-GCN in predicting demographic variables like age and sex.

Main Methods:

• Coupling structural and functional brain networks by adding learnable inter-network edges between corresponding brain regions.
• Applying a single Graph Convolutional Network (GCN) to analyze the joint structure-function graph.
• Utilizing the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) dataset with 662 participants.

Main Results:

• The proposed Joint-GCN method effectively integrates white-matter micro-structural and functional brain networks.
• Learnable inter-network edge weights capture non-uniform structure-function coupling strengths across the brain.
• Joint-GCN demonstrated superior performance in predicting age and sex compared to existing multi-modal graph learning approaches.

Conclusions:

• Joint-GCN provides a powerful new approach for analyzing the intricate relationships between brain structure and function.
• This method advances multi-modal neuroimaging analysis by enabling direct integration of different network types.
• The findings highlight the potential of Joint-GCN for understanding brain development and individual differences.