Medical Image Segmentation Based On Vese-Chan Multiphase Level Set Method

Medical image segmentation is a traditional and challenging research point. Due to theintrinsic features of medical image, traditional single level set method is difficult toaccurately solve its segmentation and it is very imperative to further the research onmultiphase level set method. A deep study is made in this thesis and it covers four partswhich is organized as follows:Firstly, the basic theory on level set method and Mumford-Shah Model issystematically researched and their application in image segmentation field is discussed.And the drawbacks of single level set method in segmenting multiphase image is pointedout through experiments which lays a solid foundation for the following research.Secondly, several representative multiphase level set methods, especially Vese-Chanmultiphase level set method, are presented. Vese-Chan multiphase level set method has anideal scheme of region partition and is able to segment piecewise constant and piecewisesmooth image and naturally avoid overlap and vacumm. However, this method is difficultto segment multiphase image whose noise complies with complicated probabilitydistribution model under the assumption that image's noise complies piecewise constantdistribution. In this thesis, an improved Vese-Chan multiphase level set method is proposedand respectively used to segment Ultrasound image and MRI image whose noise abides byRayleigh and Gauss Probability Distribution Model respectively.Thirdly, the simulation process of Vese-Chan multiphase level Set method and theproposed Vese-Chan multiphase level set method is implemented. Then the programs areapplied to real medical image segmentation experiment and the experiment resultsdemonstrate that the improved Vese-Chan multiphase level set method achievesprospective effect in Segmenting Ultrasound image and MRI image after comparing andanalyzing the results with Vese-Chan multiphase level set method.Fourthly, the future research direction is put forward.