Abdominal Aorta Segmentation Based On CT Image

The vascular disease is the killer of human health, early detection and treatment can greatly reduce the mortality rate. Currently, the world recognized for assessing vascular disease is the using of CT imaging to observe the morphology of the vessel and confirm it. Therefore, the vessel segmentation based on the CT image is an important means of computer-aided diagnosis and treatment; the use of digital image processing and artificial intelligence techniques to extract abdominal aorta and its branches improve the segmentation accuracy and the vascular disease detection rate for clinical preoperative diagnosis and treatment, is a very significance research.Considered to the complexity and individuality of human blood vessel, combined with by the huge information of image by CT machine, the segmentation of abdominal aorta is a big challenge. Combine to the gray, morphology and structure of medical knowledge, we propose a 3D segmentation algorithm based on symmetric region growing and hybrid model to extract the 3D vessel tree.First of all, we use the Guassian Mixture model and LBF model, considering the constraints of the 3D space of the target region to obtain the interest regions of 2D direction. Then, according the characteristics of 2D segmentation results, we alter the grown and combination rule of symmetric region growing to fit the algorithms to extract the vessel on 3D direction. The algorithm combines the characteristics of region segmentation and edge detection, considering the continuity of 3D image in the interest areas not only keep the vascular wall smooth, but also get the high accuracy rate of the tiny blood vessels in the segmentation. According to the complex environment of intraperitoneal and the blood branches, we simplify the process of symmetric region growing to reduce the complexity of calculation, by using the 2D result of segmentation, we restore the true abdominal aorta and its branches.The experiment on 9 cases provided by Hubei Cancer Hospital. We proposed present the advantages of clear visualization and the accuracy of the segmentation, especially to retain the integrity of the main vessel wall and the tiny blood vessels. Results based on Overlapping rate, the maximum distance, Sensitivity and Kappa to analysis and compare with the existing algorithms. In the segmentation performance, the proposed algorithm has a relatively high accuracy and stability, which provide a solid theoretical foundation of human aortic vascular segmentation for the clinical application of computer-aided diagnosis system.