Automatic segmentation in image stacks based on multi-constraint level-set evolution

Contour extraction of image stacks is a basic task in medical modeling. The existing level-set methods usually suffer from some problems (e.g. serious errors around sharp features, incorrect split of topology and contour occlusions). This paper proposes a novel method of multi-constraint level-set evolution to avoid above-mentioned problems. Interpolation constraint and deviation constraint are added to evolution processin addition to existing constraints (such as edge points and areas). In order to prevent occlusions, it proposes a method of three-phase level-set evolution. The first phase obtains a rough contour according to selected edge points. The second phase applies an expanding LSE (level-set evolution). Missing edge points in the first phase are added when occlusions probably appear. In the third phase, occlusions are deleted and a refining evolution is implemented. As proved by final experiments, our method can steadily extract contours slice by slice when the shapes of previous contours (contours in the previous slice) are similar to current contours (contours in the current slice). Furthermore, there is no error propagation during the process of contour extraction.