Multimodal Imaging Features Associated with Autism Spectrum Disorder
Objectives: We aimed to (1) identify multimodal components representing the inter-relationship between different in vivo MRI measures of grey matter volumes, cortical morphology, and white matter diffusion metrics; and (2) relate these components to the ASD phenotype.
Methods: 98 adults with ASD (49 males and 49 females; diagnosed using the ADI-R and ADOS) and 98 matched typically developing controls (51 males and 47 females) aged 18-42 years received structural and diffusion MRI scans at the Institute of Psychiatry, Psychology and Neuroscience, London, and the Autism Research Centre, Cambridge. For each participant; voxel based morphometry (VBM) was used to segment whole brain grey matter volumes, FreeSurfer software was used to estimate five morphometric features; i.e. cortical thickness, surface area, sulcal depth, local gyrification index, and grey to white matter signal intensity ratios; and track based spatial statistics (TBSS) skeletons of white matter fractional anisotropy (FA) and mean diffusivity (MD) measures were calculated. A multimodal fusion technique ‘linked independent components analysis’ (linked ICA) (Groves et al. 2011, 2012), was used to identify components representing shared inter-subject variation between the different measures. Relationships between individual multi-modal components and ASD diagnosis and Autism Spectrum Quotient (AQ) scores were assessed through correlation analysis.
Results: We identified one component that had a significant negative correlation with diagnosis of ASD (p=0.002) and AQ scores (p=0.0019). This component represented a spatially distributed pattern of grey matter volumetric increases spanning the cerebellum, temporal and parietal regions, as well as a pattern of decreased volumes in the frontal lobe in addition to increased gyrification in the cingulate gyrus and temporal lobes. A second component representing a similar diffuse pattern of grey matter volumetric differences as well as increased signal contrast, cortical thickness, and gyrification across temporal regions, was significantly negatively correlated with AQ scores (p=5.1e-4).
Conclusions: We found significant correlations between a diagnosis of ASD and multimodal components representing spatially-distributed patterns of volumetric differences as well as decreased grey-white matter boundary integrity, cortical thickness and gyrification. Our findings enrich our understanding of the relationship between different imaging features, and may aid the identification of neurobiological pathways that contribute to the cross-modal pattern of atypical brain structure observed in ASD.