A Research Of Medical Image Segmentation Using Level Set Method

In the past decades, medical imaging technology is developing so rapidly that medical image research has become a hot topic. Medical image segmentation is the key technology of medical image processing, analysis and understanding, and it is also the prerequisite for image registration, three-dimensional reconstruction and computer-aided diagnosis. Thus, the medical image segmentation is of great significance in the medical image research. So far, researchers have proposed kinds of theories and methods to solve image segmentation problem, of which the level set segmentation model is a desirable one. This model imports high-level information according to an energy function, which corresponds with the general process of people’s visual perception. Existing level set segmentation methods can be categorized into two major classed:edge-based models and region-based models, depending on the construction of the energy functions. The edge-based models commonly drive the active contours to approximate object boundaries using an edge-detector based on image gradient, while the region-based models control the evolution of the active contours according to the similarity of image feathers in the regions. But in most of the time, these models are not able to meet the specific requirements of segmentation for medical images. The reasons include the medical images are usually fuzzy and uneven due to the influence of noise, field shift effect, local body effect and tissue movement. In addition, the anatomical structure and shape of the human body is complex, and meanwhile there are considerable differences between persons.At present, due to the complexity and diversity of the medical images, we still do not find a general method, which can be applied to a variety of segmentation tasks. Although there are already many medical image segmentation methods, they are mostly targeted to a particular task, or the segmentation errors are often great. As a result, it is prevalent to combine edge-based methods with region-based methods. Studying the integrated technology of edge information and region information, these new methods tend to be more robust and more applicable. For the medical images, what way should be taken to combine, and how to combine to give full play to their strengths so as to obtain desirable segmentation effect is the focus of our study.Based on this, the thesis proposes a novel level set energy function model, which combines edge information with region information dynamically, and will make the initial contour evolve towards the desirable boundaries while not leak at weak or discrete edge positions. In addition, in order to avoid the periodic re-initialization of the evolving level set function, we introduce a new simple regularization term, which can eliminate radical changes of the level set function (LSF) far away from the contour, and make the LSF prone to be a signed distance function around the contour as well. Finally, experimental results demonstrate that the proposed method can segment given organs or tissues in medical images exactly, especially for tumors in CT images, which proves the rationality and effectiveness.