Medical Image Segmentation Research And Application In Heart Segmentation

Medical image segmentation is a fundamental research topic which is quite popular in the recent years. Many different approaches have been proposed dealing with different kinds of cases. Generally speaking, all segmentation methods can be divided into two major categories. One is boundaries segmentation (pixel differences). And the other one is coherent regions segmentation (pixel similarities).According to the world health report of World Health Organization (WHO), cardiovascular disease has become the leading direct or contributing cause of non-accidental deaths around the world. Consequently, it is really important to pay more attention to early diagnosis and therapy plan of cardiovascular diseases, in which heart segmentation plays a key role. Heart segmentation is a difficult task for its similarity in gray level with neighboring organs coursed by tissue conglutination and the complexity of anatomical configuration. Some earlier methods focused on the segmentation of parts of heart, e.g. ventricles, which is much easier even though they are time consuming and do not converge in some cases. Some early researches focus on part of heart, e.g. left heart ventricle, rather than whole heart, which are easier but not suitable in some diagnosis and therapy cases.In our method, heart segmentation consists of two steps. One step mainly deals with segmentation of the interior anatomic configuration of heart. The basic idea is to use a classification method based on the EM (Expectation Maximization) algorithm and Gaussian Mixture Model, combined with optimizing processing. The other step focuses on extracting the exterior boundary of whole heart from the neighboring organs in medical images. This step is mainly based on an optimized version of Level Set Method guided with a prior shape model.The main works and innovations are described as follows:1. Overview of the popular medical image segmentation methods, which emphasis on Expectation Method and Level Set Methods.2. Propose some optimizing methods to the original EM algorithm according to the characteristic of heart segmentation problem. Also we will compare the results of them to verify that our optimizing method will suit heart segmentation problem better than before.3. To deal with the whole heart with configuration segmentation problem, we propose a hybrid model consists of optimized EM algorithm and Level Set Method with the guidance of prior shape model. The hybrid combines the advantages of both the methods, and can produce better result of whole heart methods, together with the heart internal configuration in information.4. Using our hybrid model based on optimized EM algorithms and Level Set Method with prior shape model, we implemented several sets of experiments on CT and MRI images. Then we compared and verified our model with golden standards and other methods. And we discussed some applications of our research results.