Robust vessel tree modeling.

In this paper, we present a novel method for extracting center axis representations (centerlines) of blood vessels in contrast enhanced (CE)-CTA/MRA, robustly and accurately. This graph-based optimization algorithm which employs multi-scale medialness filters extracts vessel centerlines by computing the minimum-cost paths. Specifically, first, new medialness filters are designed from the assumption of circular/elliptic vessel cross-sections. These filters produce contrast and scale independent responses even the presence of nearby structures. Second, they are incorporated to the minimum-cost path detection algorithm in a novel way for the computational efficiency and accuracy. Third, the full vessel centerline tree is constructed from this optimization technique by assigning a saliency measure for each centerline from their length and radius information. The proposed method is computationally efficient and produces results that are comparable in quality to the ones created by experts. It has been tested on more than 100 coronary artery data set where the full coronary artery trees are extracted in 21 seconds in average on a 3.2 GHz PC.