Research On Algorithm For Medical Image Segmentation Based On Level Set Method And Fuzzy Cluster

Medical image segmentation has a wide applications and great research values on medical research, clinical diagnosis, medical image 3-D reconstruction and visualization, computer-assisted surgey, etc. The definition of the so-called medical image segmentation is that separating the same region of interest, such as organ and tissue, tumor area, blood vessel, etc. from medical image background.Based on the researches about many medical image segmentation methods and considering the smooth and continuous segmentation boundaries in the segmentation results of active contour model, this paper proposes a fast image segmentation model combined with Fuzzy C-means(FCM) method and curve evolution to solve the problem of FCM in sensitiveness to image noise and unclosed boundaries of segmentation result. Firstly, a pseudo level set and its evolution curves are defined on membership matrixes of FCM. Then, to get the smooth and closed segmentation object boundaries, the Gaussian filter is performed on the pseudo level sets to approximate the function of the curve length regularization term. Based on a new introduced edge-stop function and the mapping relationship between the gray value and membership degree, the gray values of the noisy points are corrected to reduce the influence of the Gaussian filter on the result of FCM. The FCM and the smoothing object boundary stage are performed alternately,which improves the robustness of this model. The experimental results show that the proposed- model can overcome the influence of noise and get better segmentation results.Based on two assumptions on image intensity inhomogeneity, this paper devises a level set model to image segmentation based on direction gradient to against the phenomenon of the intensity inhomogeneity in medical images. Through a new introduced image intensity linear attenuation model in this paper, the relationships between direction gradients of pixel and segmentation objects are presented. A pixel classification method based on the spatial distance weighting is introduced, according to the rate of intensity change in the direction of the locations of the minimum and maximum of pixel intensity respectively in a local image region. To implement the model, variational level set method and spatial neighborhood information of image are used,which enhance the anti-noise capacity of the proposed model with the image gradient information. From the segmentation results of experimental images including synthetic images, MRI and real images,the proposed model can deal with the intensity inhomogeneity and get the comparatively ideal segmentation results properly and efficiently.