Diseases of the cardiovascular system are one of the main causes of death in the Western world. Especially the
aorta and its main descending vessels are of high importance for diagnosis and treatment.
Today, minimally invasive interventions are becoming increasingly popular due to their advantages like cost
effectiveness and minimized risk for the patient. The training of such interventions, which require much of
coordination skills, can be trained by task training systems, which are operation simualtion units. These systems
require a data model that can be reconstructed from given patient data sets. In this paper, we present a
method that allows to segment and classify aorta, carotides, and ostium (including coronary arteries) in one
run, fully automatic and highly robust. The system tolerates changes in topology, streak artifacts in CT caused
by calcification and inhomogeneous distribution of contrast agent. Both CT and MRI-Images can be processed.
The underlying algorithm is based on a combination of Vesselness Enhancement Diffusion, Region Growing, and
the Level Set Method. The system showed good results on all 15 real patient data sets whereby the deviation
was smaller than two voxels.