Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.



Degree Program

Interdisciplinary Studies (Individualized Degree), PhD

Committee Chair

El-Baz, Ayman

Committee Co-Chair (if applicable)

Barnes, Gregory Neal

Committee Member

Barnes, Gregory Neal

Committee Member

Roussel, Tommy

Committee Member

Mohamed, Tamer M. A.

Committee Member

Frieboes, Hermann

Author's Keywords

Autism; structural MRI; aritificial inteligence; machine learning; feature engineering; social responsive scale


Autism Spectrum Disorder (ASD) is a condition characterized by social and behavioral impairments, affecting approximately 1 in every 44 children in the United States. Common symptoms include difficulties in communication, interpersonal interactions, and behavior. While symptoms can manifest as early as infancy, obtaining an accurate diagnosis may require multiple visits to a pediatric specialist due to the subjective nature of the assessment, which may yield varying scores from different specialists. Despite growing evidence of the role of differences in brain development and/or environmental and/or genetic factors in autism development, the exact pathology of this disorder has yet to be fully elucidated by scientists. At present, the diagnosis of ASD typically involves a set of gold-standard diagnostic evaluations, such as the Autism Diagnostic Observation Schedule (ADOS), the Autism Diagnostic Interview-Revised (ADI-R), and the more cost-effective Social Responsive Scale (SRS). Administering these diagnostic tests, which involve assessing communication and behavioral patterns, along with obtaining a clinical history, requires the expertise of a team of qualified clinicians. This process is time-consuming, effortful, and involves a degree of subjectivity due to the reliance on clinical judgment. Aside from conventional observational assessments, recent developments in neuroimaging and machine learning offer a fast and objective alternative for diagnosing ASD using brain imaging. This comprehensive work explores the use of different imaging modalvities, namely structural MRI (sMRI) and resting-state functional MRI (rs-fMRI), to investigate their potential for autism diagnosis. The proposed study aims to offer a new approach and perspective in comprehending ASD as a multidimensional problem, within a behavioral space that is defined by one of the available ASD diagnostic tools. This dissertation introduces a thorough investigation of the utilization of feature engineering tools to extract distinctive insights from various brain imaging modalities, including the application of novel feature representations. Additionally, the use of a machine learning framework to aid in the precise classification of individuals with autism is also explored in detail. This extensive research, which draws upon large publicly available datasets, sheds light on the influence of various decisions made throughout the pipeline on diagnostic accuracy. Furthermore, it identifies brain regions that may be impacted and contribute to an autism diagnosis. The attainment of high global state-of-the-art cross-validated, and hold-out set accuracy validates the advantages of feature representation and engineering in extracting valuable information, as well as the potential benefits of employing neuroimaging for autism diagnosis. Furthermore, a suggested diagnostic report has been put forth to assist physicians in mapping diagnoses to underlying neuroimaging markers. This approach could enable an earlier, automated, and more objective personalized diagnosis.