Advisory Committee Chair
Inmaculada B Aban
Advisory Committee Members
Marguerite R Irvin
Andrew D Smith
Hemant Tiwari
Nengjun Yi
Document Type
Dissertation
Date of Award
2017
Degree Name by School
Doctor of Philosophy (PhD) School of Public Health
Abstract
Stroke is, and will continue to be, a pervasive problem in the both the United States and across the world, and computed tomography (CT) perfusion scanning will continue to be a first-line diagnostic tool to quantify where and how much blood flow occlusion is present in the brain. CT perfusion maps displaying several scalar perfusion parameters for each brain voxel will continue to be used by radiologists and other clinicians. The goal of this dissertation is to examine and improve upon current standards of practice concerning CT perfusion. We first exhibit how software that constructs perfusion maps can be implemented and distributed completely within an open-source environment, alleviating either the cost or accessibility hindrances that often accompany access to such. Although challenging and far from perfect, we prove that such a product can be delivered to researchers wishing to use it. We then seek to answer whether using the entire residue function rather than distilling it down to a few scalar quantities would prove to be a useful and enlightening approach. But before we answered that question, we first determine which modeling framework we need to employ. We then, through an extensive simulation study, go on to show that the Functional Linear Regression That’s Interpretable (FLiRTI) method produced results that have the best predictive accuracy while also not producing too much of a computational burden. Finally, we take on the analysis of 93 participants with validated stroke/no stroke to assess the accuracy of cerebral blood flow, cerebral blood volume, mean transit time, the brain tissue time-attenuation curve, and the residue function to predict the presence of stroke. Logistic regression was used for the three scalar quantities and the FLiRTI method was used for the two functional curves. It was shown that the residue function did, in fact, offer the most in terms of predictive ability.
Recommended Citation
Lirette, Seth Thomas, "A Statistical Approach to Computed Tomography Perfusion" (2017). All ETDs from UAB. 2302.
https://digitalcommons.library.uab.edu/etd-collection/2302
