Recognition of Autism Spectrum Disorder via Dual-Channel Feature Fusion and Multi-Resolution Brain Region Segmentation

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by disruptions in brain functional networks. Current classification studies using functional magnetic resonance imaging (fMRI) often struggle with issues of model interpretability and incorporating multi-scale features effectively. This study proposes a multi-resolution brain network model that integrates degree-normalized graph neural networks with fully connected attention mechanisms to boost classification accuracy and identify critical brain networks related to ASD. Experimental results show that the proposed model outperforms traditional methods in accuracy, sensitivity, and other evaluation metrics. Furthermore, analysis of brain network correlations reveals functional networks closely associated with ASD, providing valuable insights and a strong reference for clinical diagnosis.