Explainable Multimodal Graph Isomorphism Network for Interpreting Sex Differences in Adolescent Neurodevelopment.

Area of Science:

• Neuroscience
• Computer Science
• Medical Imaging

Background:

• Understanding sex-related variability in healthy individuals is crucial for neuropsychiatric research.
• Functional magnetic resonance imaging (fMRI) is a key tool for identifying sex differences.
• Graph neural networks (GNNs) are effective for analyzing fMRI-derived brain networks.

Purpose of the Study:

• To introduce a multi-modal graph isomorphism network (MGIN) for detecting sex-based disparities in fMRI data.
• To enhance predictive capabilities by amalgamating brain networks from multiple scans.
• To improve the interpretability of sex difference analysis in adolescent brain networks.

Main Methods:

• Developed a multi-modal graph isomorphism network (MGIN) using task-related fMRI data.
• Integrated brain networks from multiple scans per individual to improve feature identification.
• Utilized GNNExplainer for interpretability, identifying pivotal sub-network structures for sex classification.

Main Results:

• The MGIN model demonstrated superior classification accuracy compared to other models.
• Combined fMRI paradigms enhanced predictive performance.
• Identified significant sex-related functional networks including DMN, VIS, CNG, FRNT, SAL, SUB, and SM.
• Achieved an 81.67% improvement in sex classification accuracy.

Conclusions:

• The MGIN model's strength lies in consolidating multi-scan data within an interpretable framework.
• The model enhances understanding of adolescent neurodevelopment by pinpointing critical functional connectivity subnetworks.
• This approach offers a powerful tool for exploring sex differences in brain function.