Research On Medical Image Segmentation Method Based On Active Contour Model

In recent years, active contour model (ACM) based on partial differential equation has become an important method of image segmentation, but the defect of ACM limits its application in medical image segmentation. In this paper, firstly we introduce the two classification of ACM, and design some experiments to verify its superiority in medical image segmentation. We deeply study the classical representative of parametric active contour model which is the traditional Snake model, and two improved models based on the traditional method:Balloon Snake model and the gradient vector flow (GVF) model. We design different experiments with the three methods to compare their performance on the segmentation of human lung CT images. Then we analyze the advantages and shortcomings of each method, and improve the following defects in this paper:(1) Since the ACM is easy to converge to the false edge points of which gray value changes larger than others, in this paper, an improved model was used to segment medical image, by introducing gray control information to the traditional model, which makes the evolution curve obtain a grey binding in the large gradient of false edge points and stops the curve motion to the wrong edge. A large number of CT slices of human lung organ were segmented with the improved algorithm, and the experimental results show that, even if the initial contour is deviated from the true contour of the target, or there is a large interference near the true contour, the improved algorithm can still segment the human lung organ accurately. It overcomes the shortcomings of the traditional Snake model which is more sensitive to the position of the initial contour line.(2) In addition, ACM need to manually or semi-automatically initialize contour, so in this paper we propose an improved method by combining mean shape point cloud model with the ACM for segmentation. Firstly, multiple organs were manually segmented to gather the training samples, then the corresponding point cloud data of these samples were obtained according to the mesh generation method. After that, an average processing of these corresponding points was taken to get a mean shape point cloud as the mean model. By calculating the correspondence of characteristics point, the mean model is mapped to the organs atlas which is to be segmented and the initial outline is obtained, at last the organ edge of each slice is extracted with the Snake algorithm. A large number of CT slices of mouse kidney organ were segmented with the improved algorithm, and the experimental results show that, the method can automatically segment the interested organ, and the segmentation accuracy was higher than the traditional method.