Abstract
In this paper, we present a method for extracting center axis representations of tubular shapes obtained from medical images. The method extracts centerlines by computing minimum-cost paths from a graph based optimization algorithm which minimizes responses obtained from multi-scale medialness filters. These filters are designed from the assumption that tubular structures, in general, has circular shapes in cross-sectional views. The proposed algorithm can produce a local centerline segment between two points as well as a centerline tree from a single seed, which can be detected automatically. The main contribution of the method is its ability to extract center axis of tubular shapes which may contain boundary noise, holes, gaps and attachment of nearby structures or pathologies due to the errors from segmentation algorithms.