Coronary tree extraction from X-ray angiograms using marked point processes

In this paper, we use marked point processes to perform an unsupervised extraction of the coronary tree from 2D X-ray angiography. These processes provide a rigorous framework based on measure theory to describe a scene by an unordered set of objects. Firstly, the thick branches are detected at low resolution using a segment process. Secondly, a polygon tree is derived from this first result at high resolution to represent the main part of the coronary tree. Finally, new branches are extracted using a recursive algorithm based on the modeling of the descendants of a given branch by a polyline process in the neighborhood of this branch. Process optimization is done via simulated annealing using a reversible jump Markov chain Monte Carlo algorithm. Experimental results show the relevance of the object process models