Medical Image Segmentation Based On Variational Level Set

Image segmentation is an extremely significant part of image processing. Its main purpose is to divide and then extract the region of interest of the image. Image segmentation also lays the foundation for image understanding and recognition. Medical image segmentation, which provides an important reference for the next diagnosis and treatment, is the application of image segmentation in medical field.Among those proposed medical image segmentation methods, the level set method is widely used in medical image segmentation field for it has good topological transformation properties, more accurate calculations, and it is easier to implement in high dimension. The basic research content of this thesis is based on level set method of medical image segmentation. The efforts and results are as follows:1. First, I give a brief introduction of the current research status, the latest research results and the future research direction in medical image segmentation field at home and abroad. Then I simply introduce some classical image segmentation methods which are mainly based on region information, edge information, and texture analysis.2. Elaborate the theoretical knowledge about geometrical evolution of curve and surface, mainly including curve evolution problem, level set method, and variational level set method. Then deduce and analyze Mumford-Shad model, Chan-Vese model and Li model involved in the paper, and compare their strengths and weaknesses.3. To solve problems of edge leakage and incomplete segmentation in medical image segmentation, this paper proposed a level set method based on regional information. By replacing the traditional model Li stop function with signed pressure function, the problem that curve is sensitive to initial position can be effectively solved. Meanwhile, curve overcomes the shortcoming of single direction evolution. It can easily choose the evolution direction according to the regional information. The new speed function can effectively control the speed of curve evolution. When curve gets close to the edge of target area, evolution speed becomes slow. That can avoid edge leakage. When the curve is away from the target area, evolution speed becomes high. That improves the evolution efficiency.4. When segmenting medical images with complex structures, the level set method has the shortcomings, such as large amount of iteration, long iteration time, incomplete segmentation and so on. This paper thus presents a new image segmentation algorithm, which combines marked watershed with improved Li model. In step one, the original image that is segmented for the first time by marked watershed algorithm, rapidly and accurately positions the target area information. In step two, use improved Li model to segment the image accurately for the second time. This combination algorithm is time-saving with higher precision.5. The tradition C-V model can divide the image into two sections-target and background. But it can not succeed in segmenting the multiple objects at the same time. Multiphase C-V model can achieve the goal of multi-object image segmentation with much more iteration and calculation. In order to solve those problems, this paper proposes a double level set segmentation algorithm based on image layer. This algorithm forms a multilevel image by introducing the background filling technology to change the image background continuously. As a result, the double level set can segment the object in the new image layer until all objects are segmented. This double level set segmentation algorithm needs less number of iterations with strong anti-interference ability and fast convergence speed.