Segmentation Of Angiography Images

Vascular tree angiograms are important tools for computer-aided diagnosing the presence of vascular disease, interactive therapy, surgical navigation and tracing the progress of diseases. In the vascular analysis system, segmentation of angiogram is the basis for description of vascular shape as well as scale, and plays a key role for the objective and precise quantitative analysis of vascular tree. The segmentation is also the basis of registration and 3-D reconstruction. Vascular tree angiograms suffer from noise and they are fuzzy, low-contrast with the vessel disturbed by background. Thus, segmentation is a difficult task.In particular, allowing for the special property of angiograms, for the angiograms under x-ray, this paper mainly addresses two kinds of segmentation methods: data-driven method using fuzzy morphology technology, and model-driven method that using point process under stochastic geometry framework.The paper reviews the main segmentation methods for angiogram. It has analyzed the basic principle of different methods. Based on Bayer's theory, the paper unifies the different methods in a uniform segmentation object and relates them in a consistent segmentation process.As a preparation for segmentation, the paper addresses two types of enhancement methods, i.e., contrast enhancement based on background estimation and nonlinear diffusion enhancement based on morphology measure. To increase vessel contrast and attenuate background for vessel detection, the multiscale morphology opening, with variable size structuring elements, is employed to estimate the slowly varying background of vessels. Subtracting the estimated background from the original angiogram is adopted to enhance the angiogram. The scale used by opening operator for each pixel is detected by multiple scales linear filters under multiple orientations.Besides, the paper has analyzed the morphology difference property of angiogram and obtained a measure for angiogram: maximum morphology difference. As a measure for the vessel structure, more important, as a measure to distinguish vessel structure from background, the morphology differential property is incorporated into a nonlinear diffusion process, a general version of P-M diffusion equation to achieve smoothing enhancement of angiogram.Under the assumption that vessels are bright piecewise connected and local linear pattern in a noisy environment, fuzzy morphology methods are used to extract vessel. The algorithm starts with a multiscale enhancement process presented previous. Subsequently, fuzzy morphology method simplifies the enhanced angiogram with a combination of the mathematical morphology operation, fuzzy opening, with a linear rotating structuring element, and its dual operation, aiming to fit the vessel patterns and strengthen them. The fuzzy filter process achieves smoothing within object region while keeping the contrast between vessel and background. These filtering processes are then followed by threshold and thinning to produce the approximate centerline of vessel. The final vessel boundaries are detected using watershed techniques with the obtained vessel centerline as prior marker for vessel region.To make use of the high-level shape knowledge to achieve robustness in segmentation, an algorithm is presented under the stochastic geometry framework to perform automatic extraction of vessel tree on angiogram. The approximate vessel centerline is modeled as marked point process with each point denoting a line segment. A Double Area prior model is proposed to incorporate the geometrical and topological constraints of segments through potentials on the interaction and the type of segments, which favors the consistency of segment orientation and connectivity of segments. Data likelihood allows for the contrast between vessel part associated to a segment on centerline and the background. Optimization is realized by simulated annealing scheme using a Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm. Moreover, it utilizes Data-Driven techniques to compute important proposal probabilities, which effectively drive the Markov chain dynamics and accelerate the speed of convergence. Finally, the watershed algorithm is used to explore the accurate vessel edges with the approximate vessel centerline as marker.