Simultaneous Segmentation of Multiple Closed Surfaces Using Optimal Graph Searching
Abstract
This paper presents a general graph-theoretic technique for simultaneously segmenting multiple closed surfaces in volumetric images, which employs a novel graph-construction scheme based on triangulated surface meshes obtained from a topological presegmentation. The method utilizes an efficient graph-cut algorithm that guarantees global optimality of the solution under given cost functions and geometric constraints. The method’s applicability to difficult biomedical image analysis problems was demonstrated in a case study of co-segmenting the bone and cartilage surfaces in 3-D magnetic resonance (MR) images of human ankles. The results of our automated segmentation were validated against manual tracings in 55 randomly selected image slices. Highly accurate segmentation results were obtained, with signed surface positioning errors for the bone and cartilage surfaces being 0.02±0.11mm and 0.17±0.12mm, respectively.
BibTeX
@conference{Li-2005-9242,author = {Kang Li and Steven Millington and Xiaodong Wu and Danny Z. Chen and Milan Sonka},
title = {Simultaneous Segmentation of Multiple Closed Surfaces Using Optimal Graph Searching},
booktitle = {Proceedings of Biennial International Conference on Information Processing in Medical Imaging (IPMI '05)},
year = {2005},
month = {July},
pages = {406 - 417},
}