Residual graph transformer for autism spectrum disorder prediction

Yibin Wang, Haixia Long, Tao Bo, Jianwei Zheng

Abstract: Highlights•Our graph encoder surpasses linear FC matrix generation, capturing intricate brain node activity patterns.•We employ residual learning to enable deeper GCN architectures, enhancing feature extraction by facilitating deeper message diffusion among brain nodes.•Our Graph Sparse Fitting module efficiently reduces graph dimensionality by evaluating inductive node scores, enabling our model to accurately highlight salient brain regions.•RGTNet identifies ASD biomarkers aligned with referenced medical research, offering unique insights into clinical diagnosis.