<title>Vessel tracking by template string in angiography</title>

Abstract

ABSTRACT The motion of coronary arteries has attracted more and more attention because of its possible effects on the development of atherosclerosis and potential clinical application for diagnosis of cardiovascular disease. Angiography is the best clinical modality so far to extract this information spatially and temporally. In this paper, a new method, which combines “snakes” and a template matching technique, is proposed to track vessel segments in angiographic image sequences. By considering both global and local motion, the vessel can be tracked well in a cardiac cycle. Keywords: Vessel tracking, Angiography, Active contour model, Template matching 1. INTRODUCTION The dynamic information from coronary arteries has attracted increasing attention in cardiac research in recent years. On one hand, it can be used for evaluation of regional myocardial performance and exploring cardiac motion. On the other hand, the artery motion might play an important role in the pathogenesis of coronary atherosclerosis by modulating the vessel wall mechanics and luminal fluid dynamics, which have been widely recognized to be involved in the initiation and development of atherosclerosis. Among the several clinical cardiac imaging modalities, angiography is the best one so far to extract this information spatially and temporally. Early image tracking methods were mainly based on template matching [1]. In such methods, a small square or rectangular window that contains the object to be tracked and fits the size of the object is set up in the first image. Then th e object is tracked by matching this window image (template) in the second image. The matching process can be mainly realized by three algorithms: cross correlation, MAD (Minimum Absolute Difference) and MSD (Minimum Square Difference). Because of the shape and the deformable motion of the vessel, it is hard to directly apply the basic template matching technique to vessel tracking. Stevenson et al. proposed tracking arterial bifurcation points by a template matching technique, followed by a centerline-tracing algorithm to join them [2]. There are two main problems in using template matching techniques to track a vessel. First, it is not good to set a template to contain the whole vessel segment, because the vessel will occupy only a very small part of the rectangular window. Then the template can hardly represent the vessel. Second, if using some small templates, instead of a big template, to track some feature points of the vessel, as in [2], since there is no constraint among those small templates, each small template must track correctly so that the right result can be obtained. Since late 1980’s, active contour models, also called “snakes”, have become popular in tracking deformable objects [3]. Hyche et al. applied this method to track the coronary artery of a pig’s heart in angiography [4]. The original idea of “snakes ” was to locate some image curve features, such as lines and edges, from an initial guess. When it is used for tracking objects, it is actually repeating the localization process. If the contour being tracked is closed, “snakes” work well. But when the contour is an open one, such as in vessel tracking, the contour may shrink or shift along the object, unless some external forces are applied to the two ends of the contour. Considering the advantages and disadvantages of both template matching and “snakes”, we present a new method for tracking vessels in angiography. Our new method is based on “snakes”, but involves the template matching technique in constructing the external energy term of the model. In addition, a global motion model is incorporated so that fairly large movement can be tracked. In Section 2, the tracking method based on “snakes” will be briefly described and template matching techniques will be adopted to improve the external forces. A 2-D affine transformation will be introduced to model the global motion of the vessel. Some experimental results and discussion will be presented in Section 3, and Section 4 is the conclusion.