Meeting Banner
Abstract #3207

A Novel Method for Determining the Reliability of Diffusion-Weighted Imaging Data

Karl G. Helmer1, Ming-Chung Chou2, Ronny I. Preciado, Allen Song3, Jessica Turner4, Barjor Gimi5, Susumu Mori6, 7

1Radiology, Massachusetts General Hospital, Charlestown, MA, United States; 2Department of Medical Imaging and Radiological Sciences, Kaohsiung Medical University, Kaohsiung, Taiwan; 3Radiology, Duke University, Durham, NC, United States; 4Translational Neuroscience, The Mind Research Network, Albuquerque, NM, United States; 5Radiology and Medicine, Dartmouth-Hitchcock Medical Center, Lebanon, NH, United States; 6Radiology, Johns Hopkins School of Medicine, Baltimore, MD, United States; 7Kennedy Kreiger Institute, Baltimore, MD, United States

We report on a method using histogram-similarity measures to establish normative data which can be used in the calibration of diffusion-weighted imaging data collected at a single or multiple sites. This method can be used to quantitatively determine the quality of new data acquired over time by comparing it to the normative data. We have calculated histogram similarity using different metric types, both within and between sites and have shown that these metrics are sensitive measures for both fractional anisotropy (FA) and mean diffusivity (MD). Statistical significance of the results was determined using simulations of two different histogram distributions.

Abstract PDF Presentation Video

Keywords

addition adequately applicable assess audience bays become bins brain calculation characterize commonly comparing computational concatenating consistent constraints correlations coverage curve curves cutoff decreasing defined described determine determined determining develop deviations dictated diffusion diffusivity distance distances distinguish distribution distributions done duke elements example experiment experimental families family field final five fractional full gave general generally generated giving goal gold gradient hard harmonized histogram histograms hospital human identify include includes increasing investigated johns knowing largest little locally look made manufacturers matrix medical medicine metric metrics mind ming minimizing much muller needed network novel original others parallel percentile performance plotted process protocol quality quantitative quantitatively radiological radiology random recalculated reliability remove reproducibility samples scanned scanner school sciences sense separate sequentially serial sets several shuffle significance simulated simulations site sites slices soft song space statistical strength strengths studies subject subjects target tensors track translational turner unique unit variability variance vary vendor vendors ware whole wide