Gray matter subtyping with SuStaIn reveals neuroanatomical patterns in autism spectrum disorder.
Authors
Affiliations (10)
Affiliations (10)
- College of Computer Science, Chengdu University, Chengdu, 610106, P. R. China.
- Department of Computer and Software, Chengdu Jincheng College, Chengdu, P. R. China.
- The Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, University of Electronic Science and Technology of China, Chengdu, 611731, P. R. China.
- China-Cuba Belt and Road Joint Laboratory on Neurotechnology and Brain-Apparatus Communication, University of Electronic Science and Technology of China, Chengdu, 611731, P. R. China.
- Stirling College, Chengdu University, Chengdu, 610106, P. R. China.
- Cuban Neuroscience Center, La Habana, Cuba.
- Affiliated Hospital of Chengdu University, Chengdu, 610106, P. R. China.
- The Clinical Hospital of Chengdu Brain Science Institute, School of Life Science and Technology, University of Electronic Science and Technology of China, Chengdu, 611731, P. R. China. [email protected].
- China-Cuba Belt and Road Joint Laboratory on Neurotechnology and Brain-Apparatus Communication, University of Electronic Science and Technology of China, Chengdu, 611731, P. R. China. [email protected].
- College of Computer Science, Chengdu University, Chengdu, 610106, P. R. China. [email protected].
Abstract
Autism spectrum disorder (ASD) is highly heterogeneous, and structural magnetic resonance imaging (MRI) studies have reported both increased and decreased gray matter volume relative to typically developing controls. This variability may obscure distinct neuroanatomical patterns in conventional group-level analyses. It therefore motivates subtyping approaches that can characterize neuroanatomical heterogeneity beyond conventional group-level analyses. We applied Subtype and Stage Inference (SuStaIn) to structural MRI data from the Autism Brain Imaging Data Exchange (ABIDE) to identify gray matter subtypes and stage-like variation in ASD. Original and sign-reversed z-score representations were modeled to capture gray matter deviations above and below the control reference. We then evaluated whether the organization of subtype-related brain regions was consistent across deviation directions. Cross-direction consistency, within-ABIDE cross-cohort separability, and structural covariance network (SCN) differences between stage groups were assessed using subtype alignment, Dice coefficients, cross-cohort machine-learning classification, and structural covariance analyses. We identified two gray matter subtype patterns in ASD that were consistent across deviation directions. Subtype 1 was centered on cerebellar-limbic-striatal regions, whereas Subtype 2 involved anterior fronto-cingulo-insular regions. In exploratory covariate-adjusted analyses, higher stage in sign-reversed-model S1 was associated with lower verbal intelligence quotient (IQ) scores, whereas S2 showed modestly higher Autism Diagnostic Observation Schedule (ADOS) scores. The two subtypes showed regionally distinct volume profiles. The two-subtype solution showed strong cross-direction agreement, and within-ABIDE cross-cohort classification supported structural separability of the aligned subtypes. In the sign-reversed model, SCN analyses further identified stage-group structural covariance differences within each subtype, with little evidence for a stable covariance difference between subtypes. This study identifies two gray matter-based ASD subtype patterns that were consistent across the original and sign-reversed z-score representations, with distinct regional volume profiles, selective clinical associations, and stage-group structural covariance differences. The findings provide a neuroanatomical framework for organizing ASD heterogeneity and for examining brain-behavior variation in future studies.