Level Set Theory And Its Application In Medical Image Segmentation

Leap progress of information technology and computer science research has brought a revolutionary development for clinical medical imaging field. Medical image segmentation is the understanding of medical image processing and analysis. However, there are some difficulties in the existing segmentation methods for extraction of ROI (Region of Interest), for complex imaging mechanism, ambiguous evaluation etc. Level set method, shortly called LSM, which is based on geometric deformation model, describes the process of interface dynamics, by implicitly representing mobile and deformable curves or surfaces as zero level set function of surfaces on higher dimension space, problem of curve’s dynamic evolution over time turns into numerical solving of partial differential equations. Now this method has been widely used in image segmentation, and is a representative numerical analysis tool for closed interface evolution research.In view of the specialty and challenging of the medical image processing field, this paper will be to the level set theory and its application in medical image segmentation. The main study work is as follows.First, this thesis summarizes level set theoretical basis and its research status applied to medical image segmentation, especially the problem of re-initialization, existence and uniqueness of numerical solution.Then, on the basis of summary and analysis to classical variation level set method, to effectively perform imaging segmentation in the case of intensity in homogeneity, the paper proposes a novel level set method combing global and local information (GLCCV). Based on the idea of neighborhood similarity and gray level gradient, we make kernel function to extract local information, which is incorporated into the classical Chan-Vese (C-V) model. According to the proportion the local intensity fitting term accounts for the global term, this paper presents a self-adaptive indicator function to balance the global and local effect. Add penalty term to avoid re-initialization and speed up the evolution. Contrast experiments show that the proposed method is efficient to segment medical images in variety case of intensity inhomogeneity scenes, and is better than conventional methods in noise resistance and accuracy.Next, in response to phenomenon like general level set methods’ sensitivity to initialization and noise, medical imaging field effect, weak boundary etc, we researched level set algorithm based on edge detection with scale transformation combined kernel fuzzy C-means with special information,shortly called KFCM_S, to segment medical images accurately and fast.In the algorithm, special neighborhood information is added to the existing method KFCM, which generates initial contour. Then, do edge detection with scale transformation to region scalable fitting model RSF introducing bias field, we can get accurate segmenting result for intensity in homogeneity and extend it to multi phases conveniently when prevent boundary leakage. This method can avoid the dependence of the level set method to initial contour, run fast, and get more accurate result.Last, the thesis summarizes all work it has done and points out future research direction.