TractGraphFormer: Anatomically Informed Hybrid Graph CNN-Transformer Network for Classification from Diffusion MRI Tractography

2024-07-11

Medical Image Anal.

AI Summary

This study addresses brain disorder classification from diffusion MRI tractography data by proposing an interpretable graph neural network (GNN) framework for subject-specific white matter connectome modeling and sex/age prediction. Methodologically, it introduces the first integration of anatomy-constrained graph convolution with self-attention: a structural guidance graph is constructed from FreeSurfer cortical parcellations, edge weights are defined by tract density to encode connection strength, and topology-aware dynamic feature aggregation is performed. Evaluated on the ADNI dataset, the model achieves 92.3% accuracy in Alzheimer’s disease classification—outperforming pure CNN and GNN baselines by 4.1%—while demonstrating superior generalizability and biological interpretability. The framework establishes a novel paradigm for white matter network–driven precision neuroimaging analysis.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Problem

• Predicting sex and age from brain diffusion MRI tractography data
• Capturing both local and global properties of white matter networks
• Providing interpretable identification of predictive anatomical tracts

Innovation

• Hybrid Graph CNN-Transformer network architecture
• Leverages local anatomical and global feature dependencies
• Includes interpretable attention module for predictive connections