An Improved Chan-Vese Model for Medical Image Segmentation

Abstract

Chan-Vese model, based on Mumford-Shan segmentation techniques and the level set method, is one of classical active contour models. It is improved by introducing gradient of images to it in this paper, because gradient of images can reflect the characteristic of all contours in images. This new model can detect objects whose boundaries are interior contours. Bones always appear to be the brightest tissue in CT medical images, while its boundaries always are interior contours which can not be detected by classical C-V model or other existing models. Meanwhile special surgery instruments in CT images for minimal invasive spinal surgery can not be detected by them too. But by this new model, they can be detected exactly, which can help doctors or surgical robot to finish their surgery better. This model has been applied on both synthetic images and CT medical images with promising results.