Meeting Banner
Abstract #1868

Comparisons of Distance Function Based Permutation Testing in Diffusion Tensor-MRI with Multiple Sclerosis Induced Microstructural Variations

Lingchih Lin1, Jianhui Zhong1, 2, Walter G. O'Dell3

1Department of Physics and Astronomy, University of Rochester, Rochester, NY, United States; 2Department of Imaging Sciences, University of Rochester, Rochester, NY, United States; 3Department of Radiation Oncology, University of Florida, Gainesville, FL, United States

The distributions of inter- and intra- distances within and between groups of subjects are compared by Euclidean, Squared Euclidean, and Log-Euclidean based distance functions for DT-MRI to calculate the exact p-values in the permutation testing. A novel approach by approximating the tail distributions of distance functions as the test statistics from a generalized Pareto model (GPD) is proposed to increase the sensitivity of detection and computation efficiency at the same time. Higher ratios of deterioration induced by multiple sclerosis are detected from Log-Euclidean based distance functions compared to mean and median based distance functions in a larger portion of regions of corpus callosum and corona radiata. Densities of distribution function are quantified by parameters estimated from maximizing likelihood, and compared with methods of moments, and probability weighted moments.