Abstract
In this work, we adapt a method based on multiple hypothesis tracking (MHT)that has been shown to give state-of-the-art vessel segmentation results ininteractive settings, for the purpose of extracting trees. Regularly spacedtubular templates are fit to image data forming local hypotheses. These localhypotheses are used to construct the MHT tree, which is then traversed to makesegmentation decisions. However, some critical parameters in this method arescale-dependent and have an adverse effect when tracking structures of varyingdimensions. We propose to use statistical ranking of local hypotheses inconstructing the MHT tree, which yields a probabilistic interpretation ofscores across scales and helps alleviate the scale-dependence of MHTparameters. This enables our method to track trees starting from a single seedpoint. Our method is evaluated on chest CT data to extract airway trees andcoronary arteries. In both cases, we show that our method performssignificantly better than the original MHT method.