Deep Graph Matching On Liver Blood Vessels

Liver cancer is one of the most common cancer-related causes of death [38]. To evaluate the results of surgical interventions treating liver cancer pre intervention and post intervention images are usually registrated. Automatic registration using landmark based registration improves the consistency and quality of the registration compared to manual registration. One type of landmarks are correspondences of graph structures. To find these correspondences is the main task of this work. In this master thesis ’deep graph matching on liver blood vessels’ a method to acquire corresponding pairs between graphs representing liver blood vessels is proposed. The deep graph matching pipeline based on DGMC [16] is trained. The data used for the supervised learning is synthetically generated based on extracted vascular graphs of the initial real world data.