Date on Master's Thesis/Doctoral Dissertation
Interdisciplinary and Graduate Studies
Interdisciplinary Studies (Individualized Degree), PhD
Committee Co-Chair (if applicable)
MRI; fMRI; rs-fMRI; DTI; ASD; CAD; ML; feature engineering
There is approximately 1 in every 44 children in the United States suffers from autism spectrum disorder (ASD), a disorder characterized by social and behavioral impairments. Communication difficulties, interpersonal difficulties, and behavioral difficulties are the top common symptoms. Even though symptoms can begin as early as infancy, it may take multiple visits to a pediatric specialist before an accurate diagnosis can be made. In addition, the diagnosis can be subjective, and different specialists may give different scores. There is a growing body of research suggesting differences in brain development and/or environmental and/or genetic factors contribute to autism development, but scientists have yet to identify exactly the pathology of this disorder. ASD can currently be diagnosed by a set of diagnostic evaluations, regarded as the gold standard, such as the Autism Diagnostic Observation Schedule (ADOS) or the Autism Diagnostic Interview-Revised (ADI-R). A team of qualified clinicians is needed for performing the behavioral and communication tests as well as clinical history information, hence a considerable amount of time, effort, and subjective judgment is involved in using these gold-standard diagnostic instruments. In addition to standard observational assessment, recent advancements in neuroimaging and machine learning suggest a rapid and objective alternative, using brain imaging. An investigation of the employment of different imaging modalities, namely Diffusion Tensor Imaging (DTI), and resting state functional MRI (rs-fMRI) for autism diagnosis is presented in this comprehensive work. A detailed study of the implementation of feature engineering tools to find discriminant insights from different brain imaging modalities, including the use of novel feature representations, and the use of a machine learning framework to assist in the accurate classification of autistic individuals is introduced in this dissertation. Based on three large publicly available datasets, this extensive research highlights different decisions along the pipeline and their impact on diagnostic accuracy. It also identifies potentially impacted brain regions that contribute to an autism diagnosis. Achieving high global state-of-the-art cross-validated accuracy confirms the benefits of feature representation and feature engineering in extracting useful information, as well as the potential benefits of utilizing neuroimaging in the diagnosis of autism. This should enable an early, automated, and more objective personalized diagnosis.
ElNakieb, Yaser, "The role of MRI in diagnosing autism: a machine learning perspective." (2022). Electronic Theses and Dissertations. Paper 4003.