UCSF

Diffusion Tensor MRI (DTI) is an imaging technique that can probe the properties of microstructures in the brain tissue. DTI has proved to be useful in detecting brain pathologies invisible by other techniques and in delineating specific white matter pathways. Analyzing DTI data from a single subject at multiple time points is desirable when the disease of interest presents with large inter-subject variability. However no such analysis framework exists due to the lack of proper tools to accurately estimate uncertainties of DTI derived parameters in a single subject.

In this dissertation, novel DTI resampling statistical analysis frameworks were developed. Bootstrap algorithms were investigated by Monte Carlo simulation in order to better understand the properties of bootstrap and determine the optimal bootstrap algorithm for DTI. Then, DTI bootstrap was used to compare the uncertainties of DTI derived parameters with or without cardiac gating in order to determine the necessity of cardiac gating for DTI with single-shot EPI acquisitions. Bootstrap was also used to perform voxel-wise T-testing between DTI data from two time points of a single subject, a technique called BLADE (Bootstrap-based Longitudinal Analysis of Diffusion Estimates). An alternative analysis framework based on permutation testing, another resampling technique, was developed as well. This framework called PERVADE (Permutation Voxel-wise Analysis of Diffusion Estimates) was designed to overcome some of the limitations of BLADE.

Monte Carlo simulation studies of bootstrap algorithms showed that the residual bootstrap algorithm developed in this work can estimate the uncertainties of DTI derived parameters with the highest accuracy. A cardiac gating scheme with minor increase in the scan time turned out to reduce the bootstrap-estimated uncertainties enough to justify the longer scan time, making cardiac gating necessary with partial Fourier acquisition. BLADE and PERVADE were able to detect FA changes in the normal appearing white matter as well as lesions of patients with traumatic brain injury and multiple sclerosis. Resampling techniques have shown great potential in subject-specific DTI analyses and quantification of artifacts, and are anticipated to play even bigger roles in the next generation diffusion MRI studies.