TractGraphFormer: Anatomically informed hybrid graph CNN-transformer network for interpretable sex and age prediction from diffusion MRI tractography

Yuqian Chen, Fan Zhang, Meng Wang, Leo R. Zekelman, Suheyla Cetin Karayumak, Tengfei Xue, Chaoyi Zhang, Yang Song, Jarrett Rushmore, Nikos Makris, Yogesh Rathi, Weidong Cai, Lauren J. O'Donnell

Abstract: Highlights•We propose a novel hybrid two-stream network for prediction based on fiber cluster features.•A graph CNN stream learns local features by leveraging local anatomical characteristics.•Transformer extracts global features and builds long-range dependencies of clusters.•TractGraphFormer combines local anatomical relationships and global dependencies to enhance prediction.•A gated attention module enables interpretation of highly predictive fiber tracts.

External IDs:dblp:journals/mia/ChenZWZKXZSRMRCO25