Multi-Task Path-Based Heterogeneous Graph Model for Functional Brain Network Analysis and Gender-Related Diseases...

Area of Science:

• Neuroscience
• Graph Neural Networks
• Medical Imaging

Background:

• Functional brain network analysis is vital for understanding brain mechanisms, aging, sexual dimorphism, and disorders.
• Resting-state functional Magnetic Resonance Imaging (rs-fMRI) measures blood-oxygen-level-dependent (BOLD) signals to map brain interactions.
• Current methods often overlook brain activity heterogeneity and focus on single-task analyses, despite shared underlying features.

Purpose of the Study:

• To develop a novel Multi-Task Heterogeneous Path graph Network (MT-HPN) for advanced functional brain network analysis.
• To address the limitations of existing methods by incorporating brain activity heterogeneity and enabling multi-task learning.
• To improve the accuracy of tasks like age regression, gender classification, and disease diagnosis by leveraging shared latent features.

Main Methods:

• Proposed a novel Path-Based Heterogeneous Graph Convolution (PB-HGC) to fuse compact edge features from heterogeneous graph paths.
• Introduced a Path-Based Cross-Attention Block (PB-CAB) for inter-task information exchange and task-specific feature emphasis.
• Developed a cross-attention transformer specifically designed for graph algorithms to enhance edge feature fusion and identify crucial paths.

Main Results:

• The MT-HPN was evaluated on the ADHD200 and ADNI datasets for gender and disease classification.
• The proposed method demonstrated strong capabilities in multi-task functional brain network analysis.
• Achieved significant performance in gender-related disease diagnosis, highlighting the model's effectiveness.

Conclusions:

• The MT-HPN offers a powerful framework for multi-task functional brain network analysis by addressing heterogeneity and enabling cross-task learning.
• The novel PB-HGC and PB-CAB components effectively capture complex brain network information.
• The approach shows promise for advancing the diagnosis and understanding of brain disorders.