This paper describes the construction of 3D dynamic statistical deformable models for complex topological shapes. It significantly extents the existing framework in that surfaces with higher genus can be effectively modeled. Criteria based on surface conformality and minimum description length is used to simultaneously identify the intrinsic global correspondence of the training data. The proposed method requires neither surface partitioning nor artificial grids on the parameterization manifold. The strength of the method is demonstrated by building a statistical model of the complex anatomical structure of the left side of human heart that includes the left ventricle, left atrium, aortic outflow tract, and pulmonary veins. The analysis of variance and leave-one-out-cross-validation indicate that the derived model not only captures physiologically plausible modes of variation but also is robust and concise, thus greatly enhancing its potential clinical value.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.