BC-GTM: A Bidirectional Causal Graph Transformer Mapping for Modeling Brain Structural and Functional Connectivity.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Understanding the relationship between structural connectivity (SC) and functional connectivity (FC) is crucial for brain research.
• Deep learning, particularly graph neural networks (GNNs), shows potential for mapping SC-FC interactions.
• Existing GNN methods lack interpretability and struggle to reveal directed dependencies in SC-FC mapping.

Purpose of the Study:

• To develop a novel model for learning the bidirectional mapping between SC and FC.
• To address the limitations of existing GNNs in terms of interpretability and directed dependency identification.
• To enhance the reliability and understanding of SC-FC relationships using causal-inspired learning.

Main Methods:

• Proposing the Bidirectional Causal Graph Transformer Mapping (BC-GTM) model.
• Utilizing a Bidirectional Causal Graph Transformer module with Directed Acyclic Graph (DAG) learning to identify directed dependency subgraphs.
• Employing a Causal Invariance Graph Bidirectional Mapping module to improve the robustness of SC-FC mapping.

Main Results:

• The BC-GTM model demonstrated superior performance compared to state-of-the-art methods in bidirectional SC-FC mapping.
• The model successfully identified key brain regions involved in the SC-FC bidirectional mapping process.
• Experimental results confirmed the model's ability to provide an interpretable, causal-inspired framework.

Conclusions:

• BC-GTM overcomes the 'black-box' nature of traditional GNNs by integrating causal-inspired directed dependencies.
• The model achieves both high mapping accuracy and enhanced interpretability in SC-FC analysis.
• BC-GTM offers an effective and interpretable GNN framework for advancing research on SC and FC bidirectional mapping.