BiSCoT: Behavior-Informed Subgroup-Consistent Connectome Template for Interpretable Brain Network Analysis.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biostatistics

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals brain functional connectivity.
• Understanding brain network heterogeneity in patient cohorts is crucial for personalized medicine.
• Existing methods may not fully capture subgroup-specific connectivity patterns.

Purpose of the Study:

• To introduce BISCoT, a graph-based framework for learning interpretable functional subnetworks from rs-fMRI.
• To leverage multidimensional behavioral data for guiding the discovery of subgroup-consistent connectome templates.
• To enhance the performance of downstream prediction tasks using learned subnetworks.

Main Methods:

• Developed BISCoT (Behavior-Informed Subgroup-consistent Connectome Template), a graph information compression framework.
• Utilized graph convolution networks for local connectivity feature extraction.
• Implemented a subgroup-informed pooling process for extracting subnetworks guided by behavioral profiles.

Main Results:

• BISCoT successfully learned interpretable functional subnetworks from rs-fMRI data.
• The learned subnetworks significantly improved performance in multiple downstream prediction tasks.
• Identified consistent connectivity 'templates' at the subgroup level, enabling biomarker discovery.

Conclusions:

• BISCoT offers a powerful approach for analyzing brain connectivity and capturing cohort heterogeneity.
• The framework facilitates the identification of interpretable, subgroup-specific brain network biomarkers.
• BISCoT has potential applications in understanding neurological and psychiatric disorders, such as autism spectrum disorder.