|
Abstract:
|
We propose a multilevel functional principal components model for multivariate spatial objects with one and two-dimensional functional arguments. As a motivating example, we consider three-dimensional CT image data of rib cages from normative pediatric subjects and those with skeletal deformities. Variation and symmetry in the shape of individual ribs and of the exterior surface of each side of the rib cage are of interest. Existing analysis methods for thoracic shape typically focus on a priori defined simple geometric summaries. The data have a multilevel structure with multiple functional objects (ribs, sides) observed within each subject's image, and variation is modeled at the levels of subject, rib, and side. We propose a multilevel functional principal components model for shape of ribs and the ribcage surface, using a spherical harmonic basis to represent the functional objects. Derived components are used to assess lateral symmetry within each subject, to model the relationship between shape and covariates of interest, and to identify latent clinical subpopulations in patients with thoracic deformities.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.