A Deep Learning Method for Autism Spectrum Disorder Classification Based on Multimodal Neuroimaging Data

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social interaction and communication skills. Accurate, early-stage differentiation of individuals with ASD from typically developing controls (TC) is essential for timely intervention and treatment. In this paper, we propose a predictive model based on multimodal feature fusion, using both functional magnetic resonance imaging (fMRI) and structural magnetic resonance imaging (sMRI) data to improve the classification of ASD. By integrating complementary information from these two modalities, our method constructs a more comprehensive feature space, capturing complex neuropathological signatures that a single modality cannot provide. We evaluated the proposed approach using imaging data from the ABIDE NYU site under a five-fold cross-validation scheme. The experimental results show that the proposed method achieved an average accuracy of 82.63%, an area under the receiver operating characteristic curve (AUC) of 89.31%, a sensitivity of 81.45%, and a specificity of 82.86%. These findings suggest that the proposed multimodal feature fusion strategy significantly enhances ASD identification, offering a promising approach to the precise diagnosis of brain disorders.Clinical Relevance- We proposed a learning framework that integrates multi-modality neuroimaging data, addressing the heterogeneity of ASD-related brain features and the challenges posed by limited training data. This framework contributes to improving diagnostic accuracy and supports early clinical decision-making for ASD, thereby facilitating timely intervention and the development of personalized treatment strategies in clinical practice.