NeuroTree: Hierarchical Functional Brain Pathway Decoding for Mental Health Disorders.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Mental disorders are prevalent global health issues.
• Functional magnetic resonance imaging (fMRI) is vital for studying brain networks in mental disorders.
• Existing fMRI-based graph neural networks (GNNs) struggle with complex relationships between brain regions and demographics.

Purpose of the Study:

• To introduce the NeuroTree framework for improved analysis of mental disorders using fMRI data.
• To overcome limitations of current GNNs in capturing intricate brain network dynamics.
• To enhance the understanding of disease-related brain subnetworks and neural mechanisms.

Main Methods:

• Developed NeuroTree, integrating k-hop AGE-GCN, neural ordinary differential equations (ODEs), and contrastive masked functional connectivity (CMFC).
• NeuroTree decodes fMRI network features into tree structures to capture high-order pathway features.
• Utilized contrastive learning to refine brain region distance similarities and dissimilarities.

Main Results:

• NeuroTree achieved state-of-the-art performance on two mental disorder datasets.
• The framework effectively identified hierarchical neural behavioral patterns.
• Demonstrated improved capture of high-order brain regional pathway features.

Conclusions:

• NeuroTree offers a powerful new approach for analyzing fMRI data in mental disorders.
• The framework provides insights into age-related deterioration patterns and their neural underpinnings.
• NeuroTree enhances the identification of disease-specific brain subnetworks.