Automatic Segmentation Improved Active Contour Model Based On Ultrasound Images

Because ultrasound is no harm, no pain and real-time, it has been widely applied in clinical diagnosis. With the wide application of ultrasound images in the medical field, automatic segmentation is needed to obtain the anatomical structure or the interested area. Because of the ambiguity, complexity and diversity of ultrasound medical images containing speckle noise, conventional edges detection operators fail to extract continuous boundaries, and traditional image segmentation technology may lead to boundary leak in weak edge and obtain inaccurate segmentation results. Therefore, this thesis based on the traditional active contour algorithm, combines other algorithms to accurately extract the contour of the region of interest for ultrasound images.The main contributions of this paper are as follows:1. A novel active contour models based on Fuzzy C-Means energy minimization is put forward. The proposed method uses the membership values of Fuzzy C-Means clustering and local pixel information of images to drive the evolution of traditional active contour model curve. The proposed model applies a fast algorithm to directly calculate the fuzzy C-means clustering energy minimization, rather than solving the Euler-Largrange equation to make the energy minimization of the active contour models. The fast algorithm solves the problem of the convergence of traditional active contour model in image depression. The proposed algorithm can be used to segment ultrasound images with the feature of intensity inhomogeous, fuzzy boundaries and corrupted by speckle noise. Compared to the LBF algorithm model and algorithm of Krinidis model, the proposed algorithm achieves superior segmentation performance and speed.2. A novel active contour model with the variance energy minimization is proposed. The proposed model is an automated active contour technology, which is based on the traditional active contour models. The proposed method does not need to manually select the seed point on the images and overcomes the initialization process of the traditional active contour model. At the same time, the different variance and mean of the Gaussian distribution model between the target areas and normal areas, based on the image statistical characteristics, is applied to control the traditional active contour model’s internal and external energy to segment the Ultrasound images. Therefore, the proposed method can effectively overcome the interference of ultrasound image’s speckle noise. The experiment results show that the proposed method can overcome the speckle noise impact of the ultrasound image and can quickly and accurately segment the target area on medical ultrasound images.