Research On Statistical Segmentation Of Vascular Images

Vascular diseases are the top of the major sources of deaths in the world. Angiographic imaging is commonly used world-wide for the diagnosis of cardiovascular, cerebrovascular, and peripheral vascular disease.Magnetic resonance angiography (MRA) is a noninvasive MRI-based flow imaging technique. Its ability to provide detailed images of blood vessels enabled its use in the diagnosis and surgical planning of the blood vessels diseases. It is widely used clinically.There are two techniques commonly used in performing MRA: time-of-flight (TOF) angiography, phase contrast angiography (PCA).Because of its fast and high contrast, TOF is widely used clinically and is the main motivation behind our work. However, it is the big challenge for the nowdays image processing technique as the limitation of MRA imaging and the complex vasculature structure.To implement the magnetic resonance angiography image-based patient-specific three dimension reconstruction, this dissertation focuses on the research of statistical model based blood vessels extraction.The major works are as follows: 1. A deep research on statistical models based segmentation method was focused, after making a comprehensive research on major blood vessel segmentation techniques proposed in the up-to-date literature. The main research content includes: Bayes segmentation framework, the selection of finite mixture model and parameters estimation techniques; the difference between different Markov random field models and energy minimization algorithms.2. In TOF MRA, intensity values are often not sufficiently high in the low flow regions where the vascular signal approximates that of the background. A drawback of this kind of methods is with poor capability in capturing distal blood vessels, while in some applications this finer detail may be required.A multi-feature incorporated bayes segmentation framework was proposed. Different from most of the existing methods, we have paid same attention to the shape feature of blood vessels and intensity feature. Both high order multiscale feature and intensity feature are incorporated into a Bayesian segmentation framework. Maximum a posterior (MAP) method is further used to estimate the posterior probabilities of vessel and background for classification. The results present clearly that the segmentation algorithm does capture many of the distal arteries.3. An adaptive statistical approach to extracting whole cerebrovascular tree in time-of-flight magnetic resonance angiography (TOF-MRA) is proposed. Firstly, in order to get a more accurate segmentation result, a localized observation model is proposed instead of defining the observation model over the entire dataset. Secondly, for the binary segmentation, an improved Iterative Conditional Model (ICM) algorithm is presented to accelerate the segmentation process. The experimental results showed that the proposed algorithm can obtain more satisfactory segmentation results and save more processing time than conventional approaches, simultaneously.