Abstract: Cardiac multi-slice computer tomography (MSCT) as a widely used non-invasive imaging modality for cardiovascular diseases diagnosis can image anatomical structures of the heart including cavities and coronary tree. In this paper, we present a new centerline extraction algorithm of coronary arteries in MSCT data sets. Based on a local vessel model, the algorithm uses the 3D geometrical moment operator, multi-scale Hessian matrix analysis and vessel filter to estimate the centroid coordinates, diameter and direction of the local vessel with the optimal scale. A tracing-based strategy is adopted to improve the algorithm’s efficiency. Beginning with interactively initialized seed points, the extraction follows the local estimated direction to search the points of centerline consecutively. A bifurcation detection operator can locate the potential bifurcations and extract seed points of subordinate branches of a single centerline. Promising experimental results are presented finally by illustrating a hierarchical coronary centerline model combined with estimated local diameters. The extraction process can be achieved in few minutes for one volume on a standard PC platform.

Key words: coronary artery, Hessian matrix
, centerline extraction, multi-slice computer tomography (MSCT), geometrical moment, multi-scale