Automatic labeling of the coronary tree using a three dimensional reference prior model

This paper proposes an automatic labeling method of an unknown single 2D X-ray coronary angiographic view. It uses an anatomical coronary tree model in a graph structure. Our 3D topological and semantic model is built from prior anatomical knowledge. It must be general enough to fit the various patient anatomies. The 3D model graph is projected under different angles of view to give the reference graphs for the labeling. The vessel skeleton extracted from the 2D angiograms is also represented with a graph structure, firstly processed to delete loops or crossings. We perform then a coarse matching based on graph shape features in order to select the reference graph which is the closest to the data. A fine matching based on structural and geometrical features allows to couple corresponding vessels in both graphs. Finally, we label unknown vessels with the anatomical names, stored in the reference graph. We obtained good results on pig and patients coronary angiograms.