Automatic Generation of Shape Models Using Nonrigid Registration with a Single Segmented Template Mesh

Statistical shape modeling using point distribution models (PDMs) has been studied extensively for segmentation and other image analysis tasks. Methods investigated in the literature begin with a set of segmented training images and attempt to find point correspondences between the segmented shapes before performing the statistical analysis. This requires a time-consuming preprocessing stage where each shape must be manually or semi-automatically segmented by an expert. In this paper we present a method for PDM generation requiring only one shape to be segmented prior to the training phase. The mesh representation generated from the single template shape is then propagated to the other training shapes using a nonrigid registration process. This automatically produces a set of meshes with correspondences between them. The resulting meshes are combined through Procrustes analysis and principal component analysis into a statistical model. A model of the C7 vertebra was created and evalua...

• Type: Conference
• Year: 2004
• Where: VMV
• Authors: Geremy Heitz, Torsten Rohlfing, Calvin R. Maurer Jr.
• Keywords: Segmented Shapes, Segmented Training Images, Statistical Shape Modeling, VMV 2004, VMV 2008