Using open-source data repository initiatives to investigate impaired white matter connectivity in autism
Loading...
Authors
Otto, Stephanie Rene
Issue Date
2022
Type
Dissertation
Language
Keywords
autism , biomarkers , connectivity , diffusion MRI , machine learning , white matter
Alternative Title
Abstract
Autism is an early appearing, highly heterogeneous neurodevelopmental condition that is primarily characterized by deficits in social communication and repetitive sensory and motor behaviors. While researchers and clinicians have made enormous progress in identifying the unique behavioral features of autism, the underlying pathophysiology is not well understood, and diagnosis is entirely reliant on behavioral observation. Emerging evidence suggests that an important feature of autism pathology involves disruptions to the connectivity between both proximal and distant brain regions. Much of the existing literature surrounding white matter connectivity in autism has primarily focused on children, while little is known about the structural and behavioral profile across the lifespan into adulthood. Diffusion MRI studies have helped to elucidate the potential relationship between autism etiology and brain connectivity, and these features have been implicated as a promising candidate in the development of sensitive and reliable biomarkers. The purpose of the studies presented in this dissertation was to characterize the extent of white matter diffusion abnormalities in a large sample of individuals with autism. Secondly, we evaluated the predictive power these differences have in identifying such unique features of connectivity in autism that may contribute to the future development of stable biomarkers. For Experiment I, we aggregated dMRI scans from open-sourced data sharing initiatives to investigate the degree to which WM diffusion is abnormal in autism compared to age-matched neurotypical controls (NT). dMRI data was acquired from the ABIDE II repository, Carnegie Mellon University, and the University of Pittsburgh. We analyzed a total of 336 subjects (187 ASD, 149 NT) using Tract Based Spatial Statistics (TBSS). Due to the nested nature of the data, robust linear mixed-effects modeling was used to examine if group differences in diffusion measures were indicated while controlling for several covariates. Compared to NT controls, individuals with ASD showed significantly decreased fractional anisotropy (FA), increased mean diffusivity (MD) and increased radial diffusivity (RD). These results were evident across the entire brain. We did not find any significant relationships between diffusion measures and ADOS or SRS scores. For Experiment II, we focused on a subset of adolescent and adult participants to perform whole-brain automated tractography of 42 major white matter pathways across 103 participants (ASD = 52, NT = 51) using FSL’s XTRACT. We trained a single-layer Perceptron classification model to evaluate whether we could predict autism diagnosis based on these tract features. Our results indicate significant model convergence with a training prediction accuracy of 77% and a testing prediction accuracy of 73%. Our results showed that the mean fractional anisotropy of the left Superior Thalamic Radiation (STR) was the only significant predictor of autism at p < .001. Taken altogether, these results suggest that white matter compromise begins early in autism and persists throughout adulthood. This highlights the need for longitudinal studies to better understand how age-related changes in white matter diffusion properties may relate to the behavioral profile often seen in autism. Understanding the complex relationship between tissue architecture, diffusion measurements and behavioral profiles across the lifespan in autism is at the foundation of developing promising neuroimaging markers.