Study on spatio-temporal patterns of gray matter and white matter functional connectivity in autism spectrum disorder based on Mamba-CNN framework

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that widely affects children and adults. Early and accurate diagnosis is crucial for improving patient outcomes, but traditional approaches face challenges in capturing the complex spatio-temporal dynamics of brain networks. This study included multi-center data from ABIDE I and ABIDE II, comprising resting-state functional magnetic resonance imaging and structural magnetic resonance imaging data of 1175 participants. To address the technical challenges in analyzing neuroimaging data, we proposed an innovative deep learning framework based on the Mamba-CNN architecture, which leverages selective state space mechanisms to capture dynamic interaction features and long-range dependencies. Additionally, a Hierarchical Multi-scale Fusion module was introduced to integrate brain network features across different spatiotemporal scales. We systematically investigated the specific features of dynamic network connectivity between gray matter and white matter in ASD populations for the first time, identifying unique abnormal patterns such as enhanced connectivity between the superior longitudinal fasciculus and dorsal anterior cingulate cortex, and reduced connectivity between the corpus callosum and somatosensory association cortex. Experimental results demonstrated that this framework achieved an 87.49% accuracy, 87.31% sensitivity, 87.68% specificity in ASD diagnosis, outperforming other ASD classification methods. This study not only enriches the understanding of ASD’s neural mechanisms from the perspective of gray matter and white matter dynamic connectivity but also provides a new technical solution for neuroimaging data analysis.