Medical Image Segmentation Under The Framework Of The Level Set

With the development of imaging technology, the use of computer technology for biomedical image analysis has become an important means of the clinical medicine and auxiliary diagnosis.However, under the influence of imaging mechanism and other factors, the majority of medical images often contain noise, weak boundary, intensity inhomogeneity (bias field), as well as complex background,etc. These make the traditional image segmentation method difficult to get a better segmentation results. Due to the support of rigorous mathematical theory, the variational model based on the level set theory can automatically handle topological transformation, and integrate the advantage of a variety of characteristics and the method easily. It gradually become the focus of many scholars in the field of biomedical image segmentation.In this context, this paper puts attention on variational method under complex scenes, mainly including the uneven grayscale, strong noise, complex topological relations scene, which is easy to fall into local optimal, sensitive to initial contour curve, less accurate and computational slowly. We carry out a theoretical framework of the level set method.The main work of the paper as follows:1. To address the interference of intensity inhomogeneity and achieve accurate segmentation, we combine local statistics with bias estimation, and propose a bias recovery coupled model based on local information to overcome intensity inhomogeneity and restore the bias field. In addition, to avoid the local optimum phenomenon and the impact of the initial level set, we carry out global convex optimization studies under l1norm to solve the model.2. Aiming at the lower computation rate, we propose an algorithm based on the Split-Bregman method to solve the coupled model. Firstly, we simplify the equation and reconfigurate our model, then construct global convex optimization model under l1norm, and finally get fast algorithm based on the Split-Bregman method for solving our model.3. In order to overcome the shortcoming that traditional methods are susceptible to noise or redundant details, we carry out the research of level set method under the framework of the local entropy. Local entropy describe the relationship between spatial region by using the degree of dispersion of pixels and noise condition, and can reflect the richness of the information. Therefore, the paper designs a bias recovery coupled segmentation model driven by local entropy, aiming at enhancing anti-interference ability of our model to the noise or detail and improving the accuracy of segmentation.4. GAC model efficiency is low, applicable scope is small. Thence, we propose a novel GAC model based on the local entropy to overcome the influence of complex background by extract and analyse the local entropy information of image. Instead of the edge stopping function, we use the local entropy to guide the curve evolve toward objective boundary, in order to overcome the complex background and extract the image objective fast and accurately.