The quality of cardiac images acquired with multi-detector CT scanners has improved significantly, to the point where minimally invasive examination of the coronary arteries became reality. The interpretation of such images requires efficient post-processing tools to isolate the vessels from other structures, such that they can be properly analyzed quantitatively or visually. In this paper we evaluate a method for semiautomated extraction of the central axis of coronary arteries in these images. First the vessels are enhanced with a local filter that analyzes the main modes of second-order variation in image intensity to determine the type of local structure. Secondly, the extremities of the axis are indicated by the user. Finally, a connected path between the given points is automatically determined with a minimum cost path algorithm, where the cost corresponds to the reciprocal of the enhanced image. The results obtained with different vessel enhancement filters are compared with man...