Anatomical registration and segmentation by warping template finite element models

Abstract

Image segmentation and anatomical registration play an important role in subject-specific computational modeling and image analysis. Often a three-dimensional (3D) segmentation is available for a canonical template image dataset of a single subject. The goal of the present work is to apply this a priori knowledge to facilitate segmentation of anatomical structures in other subjects. A 'Warping' method was developed to deform the template anatomy and register it with specific target anatomies. This was achieved by direct incorporation of image data into a nonlinear finite element (FE) analysis program. The algorithm searches all admissible material configurations for the one which minimizes the difference between the target and the deformed template. FE models of specific anatomical structures were generated from the anatomy of one specific template subject. The FE model deforms under the laws of nonlinear continuum mechanics such that one-to-one correspondence of differential lines, areas, and volumes is guaranteed. The method has been successfully applied to 2D and 3D segmentation, registration, and geometrical model construction. Example results are provided for segmentation of the distal femur using X-ray computed tomography (CT) data, and registration of neuroanatomical structures using optical cryosection image data.