|
Activity Number:
|
475
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #305692 |
|
Title:
|
Validity Diagnostics for DTI Heterogeneity Models
|
|
Author(s):
|
Meagan E. Clement*+ and Keith E. Muller and Guido Gerig and Matthew Gribbin and Joseph Piven
|
|
Companies:
|
Rho, Inc. and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
|
|
Address:
|
1010 Goldmist Lane, Durham, NC, 27713,
|
|
Keywords:
|
diffusion tensor imaging ; heterogeneity ; bimodal mixture distribution
|
|
Abstract:
|
Widely used summary measures from diffusion tensor imaging (DTI) can be interpreted as statistical estimators of population properties for Gaussian stochastic processes. Despite concerns about an underlying assumption of homogeneity, one-to-one transformations of the measures have observed distributions accurately represented in terms of only two estimated parameters each. Furthermore, standard statistical methods provide diagnostic tools for checking the homogeneity assumption, as should be done in every analysis. DTI data from small regions of the brain illustrate the process. Inadvertently including both white and grey matter in the brain due to region definition generates bimodal mixture distributions. Kernel density estimation and related histogram tools allow detecting the problem.
|
