Date of Award
6-2022
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2022.
Departments
Computer Science and Engineering
Abstract
Autism Spectrum Disorder (ASD) is typically understood by observing patient behaviors. Advances in neuroimaging research and data collection have made it possible to investigate the neurobiological basis for ASD, and differences between subjects with ASD and healthy controls (HC) have been found using a variety of methods. Functional magnetic resonance imaging (fMRI) in particular has provided evidence for and against certain cognitive models of ASD based on measuring the functional connectivity of subjects’ brains. Various techniques have been explored for applying machine learning (ML) to fMRI data in order to find unique patterns and differences between ASD and HC subjects. In this study, I expand on this search by simplifying the input data given to the machine learning model by thresholding brain connectivity to focus on the largest correlations and anticorrelations between brain regions. This results in a model performing 95.5% accurate classification under extremely favorable conditions, and 68% accurate under more challenging conditions. This demonstrates the thresholding technique’s efficacy, but points to a future challenge of producing a model that can perform well on extremely heterogeneous samples.
Recommended Citation
Luckenbach, Nicholas, "Machine Learning fMRI Autism Classifier" (2022). Interdisciplinary Design Senior Theses. 83.
https://scholarcommons.scu.edu/idp_senior/83