Vascular Segmentation Based On Adaptive Region Growing And4d-curves In CT Image

Vascular disease is attracting more and more people’s attention. With thedevelopment of modern science and technology, the detection methods of thesediseases have become more mature. The emergence of medical imaging technology，which is a milestone in the history of disease detection，provides people with anefficient detection method. In order to be able to further improve the efficiency ofthis method, many researchers put a lot of efforts in this regard and utilize currentcomputer technology, so that many vessel segmentation algorithms appear.Vessel segmentation in CT images is different from that in general image, andthe development is very challenging because of the characteristics of CT image andthe complexity of blood vessels. In order to improve the practicality of vesselsegmentation algorithms in CT images, this thesis focuses on research on thefollowing two aspects:(1) Finding the deficiencies of the existing algorithms andanalyzing the reasons of their problems.(2) Proposing improvements to existingalgorithms, solving problems for engineering applications, designing rational datastructures to enable finishing improvements, dealing the problem of large amount ofdata, reducing time and space consuming, achieving better human-computerinteraction, reducing the effects of human factors, meeting the requirements ofactual clinical applications. The details are：1. Analysis of existing algorithms, including3D locally adaptive regiongrowing algorithm,4D Curves based on active contour and competitive regionalgrowing algorithm. These three methods are all based on statistical thesis. Byevaluating the results of vessel segmentation on legs, neck and abdominal CTimages, we find out all these algorithms have good anti-noise ability. Besides, weidentify the strengths and weaknesses of these algorithms according to their results.2. The3D locally adaptive region growing algorithm can detect branch invessel segmentation but it needs several parameters. This thesis provides animproved3D locally region growing algorithm, which changes the rule of cubemoving and the way to find seed regions to reduce the number of parameter andcomputation. For4D Curves algorithm, this thesis provides two improvements,which are both based on narrowing the sphere model and can reduce the amount ofcomputation.