Graph to graph matching: Facing clinical challenges

State of the art anatomical tree matching algorithms find correspondences between trees that contain topological differences. However there are still open problems that were not considered until now. For example, when the liver vas-culature is segmented, portal and hepatic vein are not separated due to segmentation errors. Because of this reason the resulting structure is not a tree but a graph. On the other hand, inaccuracies in the generation of the graph, as well as artifacts or inhomogeneities in the contrast medium result in graphs containing gaps. In this work, we present a novel graph to graph matching algorithm. It solves the aforementioned problems by taking the whole graph structure into account and does not depend on separated trees. In addition to this it is robust against gaps in the graph. We developed our algorithm so that it does not depend on the root of the graph which is often assumed to be known. The algorithm was evaluated on real clinical data of the liver.