Dissertation/Thesis Abstract

Clustering Methods for Discovering Patterns in the Alzheimer's Disease Neuroimaging Initiative (ADNI) Data
by Fraundorf, Scott, M.S., Southern Illinois University at Edwardsville, 2019, 213; 13877627
Abstract (Summary)

Classification is the means by which information is categorized. Clustering algorithms accomplish classification and are a type of machine learning. One clustering algorithm, ITERATE, makes use of category utility, and accounts for order bias. ITERATE was implemented and applied to patient data provided by the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The time complexity of the algorithm was analyzed as well. Because of a tendency for high polynomial time complexity, the algorithm was parallelized using thread pools, and run on the Pittsburgh Supercomputing Center. For the purpose of comparison, k-means and k-means ++ were also implemented and applied to the same data. ITERATE was found to perform in a reasonable time on datasets in the thousands of tuples such as those found in ADNI’s data. However, for some tables, it was discovered that parallelization and high-performance computing might be required. ITERATE would most likely not perform a reasonable amount of time on very large datasets such as entire genomes. During this study, it was found that the algorithm has a time complexity that varies between θ(n3) and θ(n5). Because of that it could be prohibitive for data as big as genomes, and if not, it might require tremendous computing resources. ITERATE, unlike k-means and k-means ++ does not partition the data between clusters along every dimension evenly but was observed not to cluster along dimensions where little predictiveness is gained. One recurrent observation in ADNI’s data was an inverse ranking of clusters between glucose level in the hippocampus, and age (i.e. when age goes up, glucose level goes down). There is also an indication of a positive correlation between the rankings of clusters when the ITERATE algorithm was applied to MMSE scores and education, that corresponds to results in another study. [1]

Indexing (document details)
Advisor: Weinberg, Jerry
Commitee: Sandoval, Karin, Yu, William
School: Southern Illinois University at Edwardsville
Department: Computer Science
School Location: United States -- Illinois
Source: MAI 58/06M(E), Masters Abstracts International
Subjects: Computer science
Publication Number: 13877627
ISBN: 978-1-392-25969-6
Copyright © 2020 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy