A novel cross-modal contrastive learning-based framework for autism spectrum disorder diagnosis

Given the escalating prevalence of Autism Spectrum Disorder (ASD) across international populations, developing an efficient and accurate computer-aided diagnosis method has become an urgent necessity. Population-based method serves as an effective tool for large-scale ASD analysis. However, existing methods typically neglect data heterogeneity between imaging and non-imaging features while underutilizing non-imaging data. To address this, we propose a novel cross-modal contrastive learning-based graph convolutional framework (CMCL-GCN) that innovatively employs cross-modal feature alignment (CMFA) for mulimodal feature alignment. Through pre-trained encoders, our method efficiently extracts multimodal features to construct population graphs. The framework performance is tested on ABIDE dataset for ASD prediction. Experimental results show that our method can effectively improve the diagnostic accuracy of ASD.