Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Statistical Analysis of Random Symmetric Positive Definite Matrices Via Eigen-Decomposition

Rooks, Brian (2018) Statistical Analysis of Random Symmetric Positive Definite Matrices Via Eigen-Decomposition. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img]
Preview
PDF
Download (13MB) | Preview

Abstract

The work in this dissertation is motivated by applications in the analysis of imaging data, with an emphasis on diffusion tensor imaging (DTI), a modality of MRI used to non-invasively map the structure of the brain in living subjects. In the DTI model, the local movement of water molecules within a small region of the brain is summarized by a 3-by-3 symmetric positive-definite (SPD) matrix, called a diffusion tensor. Diffusion tensors can be uniquely associated with three-dimensional ellipsoids which, when plotted, provide an image of the brain. We are interested in analyzing diffusion tensor data on the eigen-decomposition space because the eigenvalues and eigenvectors of a diffusion tensor describe the shape and orientation of its corresponding ellipsoid, respectively. One of the major contributions of this dissertation is the creation of the first statistical estimation framework for SPD matrices using the eigen-decomposition-based scaling-rotation (SR) geometric framework from Jung et al (2015). In chapter 3, we define the set of sample scaling-rotation means of a sample of SPD matrices, propose a procedure for approximating the sample SR mean set, provide conditions under which this procedure will provide a unique solution, and provide conditions guaranteeing consistency and a Central Limit Theorem for the sample SR mean set. Our procedure for approximating the sample SR mean can also be extended to compute a weighted SR mean, which can be useful for smoothing DTI data or interpolation to improve image resolution. In chapter 4, we present moment-based hypothesis tests concerning the eigenvalue multiplicity pattern of the mean of a sample of diffusion tensors which can be used to classify the mean as one of four possible shapes: isotropic, prolate, oblate, or triaxial. The derivations of these test procedures lead to the creation of consistent estimators of the eigenvalues of the mean diffusion tensor. In the final chapter, we present a mixture distribution framework which can be used to model the variability of SPD matrices on the eigen-decomposition space, and an accompanying likelihood based estimation procedure which can be used for estimation of parameters of interest or inference via likelihood ratio tests.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Rooks, Brianbtr15@pitt.edubtr15
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairJung, Sungkyusungkyu@pitt.edusungkyu
Committee MemberIyengar, Satishssi@pitt.edussi
Committee MemberRen, Zhaozren@pitt.eduzren
Committee MemberKrafty, Robertrkrafty@pitt.edurkrafty
Date: 28 June 2018
Date Type: Publication
Defense Date: 4 April 2018
Approval Date: 28 June 2018
Submission Date: 9 April 2018
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 112
Institution: University of Pittsburgh
Schools and Programs: Dietrich School of Arts and Sciences > Statistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Symmetric Positive Definite Matrices, Shape Analysis, Non-Euclidean Statistics, Eigen-Decomposition, Diffusion Tensors
Date Deposited: 28 Jun 2018 19:30
Last Modified: 28 Jun 2018 19:30
URI: http://d-scholarship.pitt.edu/id/eprint/34231

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item