Automatic detection of landmarks and axes at the human knee joint

Abstract

Motivation: Anatomical landmarks are of great importance for many medical fields. Especially at the knee joint, landmarks are used for orientation during surgery or to define axes. Consequently, knowledge of their exact location is crucial. Previous methods to detect these points are very time consuming or highly complex. In 2009, a fully automatic approach to detect peak landmarks based on surface curvatures was presented by Subburaj et al. This method was implemented and tested. In addition, the method was extended to detect different geometric structures and axes. Material and methods: As the proposed method is based on curvature values of surface meshes, segmented computer tomography scans were used to obtain triangular meshes of 15 knees. A semi-automatic multi-step process was implemented to detect bony landmarks with the shape of a peak, based on curvature values and adjacency relationships. The implemented algorithm was extended to detect further geometrical shapes of the surface such as edges and valleys. This enabled the computation of the farthest points of the tibial plateau and the patella as well as the cylindrical axis and the path of the trochlear groove. In addition, the shaft axes of the femur and tibia were calculated. This algorithm was applied to all 15 specimens. The results obtained for the bony landmarks were compared with the landmarks labeled by an experienced orthopedic surgeon on 7 of those 15 specimens. Results: Manual intervention was necessary for all specimens to allow the algorithm to detect bony landmarks on the surface. However, the algorithm was only able to detect all landmarks of 5 femura and 6 tibias. The variability in location of bony landmarks of the femur and tibia, compared to the landmarks labeled by the surgeon, were found to be in range of 1.90 to 6.96 mm. The computation of the farthest points of the patella as well as tibial plateau, the trochlear groove, and the cylindrical axis delivered adequate results, required manual interventions for roughly half of all specimens. The computation of the shaft axis of the femur and tibia did not require any interventions to get results. Discussion: The methods for detecting landmarks or axes which were based on larger regions or an overall contour of the bone were relatively stable. In contrast, using curvature values for locating small peak landmarks required partly manual interventions and this strategy was not successful for all specimens. The landmarks found were in good agreement with the literature. In conclusion, semi-automatic detecting and labeling of anatomical landmarks and axes could be achieved with the implemented algorithms.