PhD Defense: Tong Zhu
Towards Optimal Human Diffusion Tensor Imaging (DTI) Protocols with Wild Bootstrap Analysis
Diffusion Tensor Imaging (DTI) depicts tissue morphology via unique patterns of random molecular motions of water inside tissues. DTI-derived parameters have been explored as surrogate biomarkers in a variety of neurological and clinical applications to non-invasively infer the underlying anatomical architectures as well as their alterations due to pathological processes of diseases. However, when the DTI technique is applied, imprecision due to measurement uncertainties decreases the sensitivity and specificity of these DTI-derived parameters as surrogate biomarkers for various applications.
The main goal of this dissertation is to apply an optimized wild bootstrap analysis, which is a nonparametric and empirical statistical method, to estimate measurement uncertainties of DTI-derived parameters within each voxel of DTI data of human brain. In contrast to previous analytical approaches, this method does not impose any assumptions about underlying noise distributions and is therefore capable of depicting variations in acquired DTI data containing sources of complex uncertainties in real DTI acquisitions.
In this study, evidence collected from real human DTI data of a group of 13 volunteers with an optimized wild bootstrap analysis provides, for the first time, criteria for optimizing DTI acquisition protocols with minimal measurement variations within clinically feasible acquisition time. Empirical distributions generated with the wild bootstrap method also enable statistical inferences between longitudinal DTI data of the same subject to detect subject-specific alternation patterns of diffusion characteristics with mild traumatic brain injuries due to sports-related concussions.