Research On Methods Of Medical Image Segmentation And Registration

As the basis of medical image processing and analysis, image registration and segmentation is the most basic problem in the medical image processing and analysis, it is the hot point of the research issues, it is also the urgency hard and key issue to be solved demoned by clinic diagnosis, virtual surgery and human body dynamic simulation, and so on.This dissertation focuses on the registration and segmentation methods for the application of medical image analysis. It surveys deformable models, researches their theory foundations, development, mathematical implementation. Image segmentation based on deformable model is discussed in detail, and some new ideas and improved algorithm are proposed in this thesis. This thesis researches on the registration method based maximization of mutual information. As it is computational expensive, we use the moment and principal axes method for a coarse registration and then operate accurate registration, so as to reduce the hole computing cost.The major works are as follows:1.We proposed a new image segmentation method base on the Snake model. In the beginning, we divided the pixels of the image into three classes. Refine the strong edge pixels, and then we get an edge curve through post threshold selection. Take it as the initial curve of the Snake model, Adjust the curve step by step. Finally, we can get the results.2.We also made some improvement to the Level Set Method. First, we divide the pixels into two classes, then define different evolve speeds to enhance the performance and the efficiency. The results of the experiments showed us the advantages of the new method.3.Given the characteristic of human body CT image, we present a new registration strategy, which include a coarse registration and accurate registration. Firstly, we attain coarse registration by matching the obtained segmentation result of two images, and get the inatial parameters for the accurate. Then we adopt MI (Mutual Information) as cost function, the POWELL optimization algorithm as our optimization algorithm to search the best registration parameters. Theories and experiments indicated that this method not only speeds up the computing process but also has the advantages of high precision, and good ability of resisting noise.