CORE -- IMAGE STATISTICS CORE

University Of Michigan At Ann Arbor, Ann Arbor MI

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The Image Statistics Core will play a crucial role complementing the Computing Core and Biostatistics Core. The Central concerns of the Image Statistics Core will be specialized statistical methods for handling variances and covariances of nonrigid registration rules and the implications of these uncertainties for the applications projects. Services supplied by the Image Statistics Core will be centered around a collection of existing multivariate statistical and morphometrical tools, with one novel addition. The existing tools are the Core director's program package Edgewarp, a navigation system through 3D medical images; the methodology of Procrustes-based biometric statistical analysis for data sets of landmark points and curves (means, principal components, regressions, permutation tests); and the Partial Least Squares toolkit for analysis of relations between shape variations and their causes or effects, an approach based on the sophisticated re-use of to singular-value decomposition of a cross-covariance matrix. In addition to these relatively familiar devices, the Core will fill a serious gap in the literature of nonrigid registration, by supplying a kriging-style estimator for the uncertainty of a single pointto- point correspondence inside a domain that has been subjected to a generalized spline registration. The connection between the thin-plate spline and the general technique of kriging has been known for some time, but the extension to actual calibration of "prediction" error does not seem to have been developed yet. This extension will be exploited by applied projects 1, 2, and 3 as one term in an analysis of systematic error in estimating the quantities of ultimate interest.