MVMS-GCN: A Multi-view Multi-source data fusion graph convolution neural network for predicting autism spectrum disorder with fMRI

Autism Spectrum Disorder (ASD) is a neurological disorder diagnosed based on symptoms, patient descriptions, and assessment scales such as the Hamilton Scale. Although Functional Magnetic Resonance Imaging (fMRI) and brain functional network analysis are used for the diagnosis and classification of neurological disorders, challenges remain due to the heterogeneity of the brain and noise connections. To overcome this, Graph Neural Networks (GNNs) have gained attention in analyzing unstructured fMRI data. Our research aims to propose a novel fMRI-based auxiliary diagnosis method for ASD. We have developed a new method, MVMS-GCN for fMRI analysis that includes graph structure learning, multi-view methodology, brain atlas-based graph clustering, and multi-source data fusion. Our approach facilitates the discovery of functional connectivity patterns in the brain network via an end-to-end methodology. Furthermore, by integrating non-imaging data, our model improves subject similarities, leading to a more precise classification performance. Our approach, MVMS-GCN, was evaluated on the ABIDE data-set for autism analysis and the REST-meta-MDD data-set for depression analysis. In comparison to other methods, we found that MVMS-GCN achieved higher accuracy (71.9% on ABIDE, an increase of approximately 3.79% compared to other methods) and exhibited superior classification performance on the REST-meta-MDD data-set. These findings underscore the potential of our approach to enhance auxiliary diagnosis and classification of neurological disorders.