The detection of anatomical landmarks is an important prerequisite to analyze medical images fully automatically.
Several machine learning approaches have been proposed to parse 3D CT datasets and to determine the
location of landmarks with associated uncertainty. However, it is a challenging task to incorporate high-level
anatomical knowledge to improve these classification results. We propose a new approach to validate candidates
for vessel bifurcation landmarks which is also applied to systematically search missed and to validate ambiguous
landmarks. A knowledge base is trained providing human-readable geometric information of the vascular system,
mainly vessel lengths, radii and curvature information, for validation of landmarks and to guide the search
process. To analyze the bifurcation area surrounding a vessel landmark of interest, a new approach is proposed
which is based on Fast Marching and incorporates anatomical information from the knowledge base. Using the
proposed algorithms, an anatomical knowledge base has been generated based on 90 manually annotated CT
images containing different parts of the body. To evaluate the landmark validation a set of 50 carotid datasets
has been tested in combination with a state of the art landmark detector with excellent results. Beside the
carotid bifurcation the algorithm is designed to handle a wide range of vascular landmarks, e.g. celiac, superior
mesenteric, renal, aortic, iliac and femoral bifurcation.