Integrating Multimodal Neuroimaging and Physical-Health Markers for Autism Spectrum Disorder in the ABCD Study.
Authors
Affiliations (2)
Affiliations (2)
- Department of Radiology, Columbia University Irving Medical Center, New York Presbyterian Hospital, New York, NY 10032, USA.
- Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT 06520, USA.
Abstract
Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by diverse presentations, which complicates the identification of consistent biological markers. This study examined whether integrating multimodal neuroimaging and physical-health measures from a population-based cohort can improve ASD classification and reveal interpretable markers that reflect both clinical and community variation. Data were drawn from the Adolescent Brain Cognitive Development (ABCD) Study, a large community-based cohort of adolescents recruited from the general population. Participants with and without ASD were selected from this cohort, allowing contrasts that reflect natural variability across individuals. Structural, diffusion, and resting-state functional magnetic resonance imaging (MRI) data were integrated with physical-health markers, including sleep, growth, and early development. Propensity-score matching created demographically balanced groups, and multimodal machine learning models were evaluated through stratified cross-validation. The multimodal integration of brain and physical-health markers outperformed single-modality models (area under the receiver operating characteristic curve [AUC-ROC] = 0.68, 95% confidence interval [CI]: 0.62-0.73; area under the precision-recall curve [AUC-PR] = 0.66, 95% CI: 0.60-0.73). Among physical-health markers, sleep function contributed most strongly to ASD classification, while neuroimaging predictors included cortical thickness in the right superior temporal gyrus and connectivity between the cingulo-opercular and default mode networks. These findings indicate that integrating modalities capturing both neural and physiological systems provides complementary information for identifying ASD-related differences within a population-based framework. This study provides a proof of concept that combining multimodal MRI and physical-health data within a large, demographically representative cohort can enhance ASD classification and yield biologically interpretable features. The population-based design situates these findings within a community context and offers a preliminary framework for integrating neural and physiological measures in future large-scale studies of neurodevelopmental diversity.