Subtyping Autism Spectrum Disorder via Joint Modeling of Clinical and Connectomic Profiles

• DOI: 10.1089/brain.2020.0997
• Publication Date: 2021-06-09T08:06:27Z
• Authors: Alexandra M Reardon; Xiaoping P. Hu; Kaiming Li; Jason Langley
• Keywords: Subtyping

Background: Autism spectrum disorder (ASD) is a highly heterogeneous developmental with diverse clinical manifestations. Neuroimaging studies have explored functional connectivity (FC) of ASD through resting-state magnetic resonance imaging studies; however, the findings remained inconsistent, thus reflecting possibility multiple subtypes. Identification relationship between symptoms and FC measures may help clarify inconsistencies in earlier advance our understanding Methods: Canonical correlation analysis was performed on 210 subjects from Brain Imaging Data Exchange to identify significant linear combinations connectomic profiles ASD. Then, hierarchical clustering defined subtypes based distinct brain-behavior relationships. Finally, support vector machine (SVM) classifier used verify that comprised features. Results: Three were identified. Subtype 1 exhibited increased intra-network FC, Intelligence Quotient (IQ) scores, restricted repetitive behaviors. 2 characterized by decreased whole-brain more severe Diagnostic Interview-Revised Social Responsiveness Scale symptoms. 3 demonstrated mixed low IQ as well social motivation verbal deficits. To subtype assignment, multi-class SVM using yielded an average accuracy 71.3% 65.2% respectively for classification, which significantly higher than chance (33.3%). Conclusion: The present study demonstrates combining behavioral powerful approach disease subtyping suggests there are profiles.