Subvoxel model-based 3D segmentation using implicit snakes

Abstract

Building accurate representations of the surfaces of bones in 3D medical images is an important task in many computer-assisted surgery procedures. We first present a survey of the existing solutions to this problem. Then, we give the results of experiments on new 2D and 3D deformable models. We propose to maximize the correlation of raw image values along contours parallel to the surface of bones. We also propose new representations of deformable contours : an orientation-based representation for 2D contours, and an implicit representation using 4D splines for 3D contours. We finally present some results of subvoxel segmentation of bone surfaces.