Identification of landmarks on lower limb joint from CT images for kinematics studies. A totally semi-automatic procedure

The identification of an accurate, reliable and patient specific coordinate system for a bone is fundamental to analyze the kinematics of a human joint. The accuracy in the localization of anatomical landmarks of joint surfaces is extremely important because even a small variation in their positions could induce a high variation in the definition of anatomical axes and further on the kinematics output. The aim of this study was to develop and validate a semi-automatic, accurate, and reproducible routine able to identify the position of anatomical landmarks on joint surfaces. This routine, starting form a CT of a femoral bone, used as input, is able to identify semi-automatically the femoral head and the medial and lateral distal femoral condyles. Moreover it allows the identification of the following anatomical landmarks: the Femoral Hip Center (FHC), the Femoral Medial Epicondyle (FME) and the Femoral Lateral Epicondyle (FLE). From these points a standard coordinate system of the femur is univocally determined according to previous literature works. Compared to other commercial process, extensive used in this field, one peculiarity of this routine is that it is not necessary to generate the 3D model of the joint in order to define the anatomical landmarks. Usually, to generate a 3D lower limb model, with the commercial process, 4 to 5 hours are needed, with this approach we can significantly reduce this time. To validate the routine we analyze ten different CT of lower limbs. Two different tests were performed. The first test was performed to verify and check the output geometry of the model; the second test was aimed to estimate the repeatability and reproducibility of the procedure. For such task five different operators identified for each model the three anatomical landmarks, three times each. The Intra-Class Correlation coefficient (ICC) values (intra and inter) obtained for the landmarks were always higher than 0.996. Comparing the results obtained with this routine with the results obtained using largely used commercial software we found a significant reduction of the error as regards the evaluation of landmarks in terms of inter and intra-observer variability. For example, in the worst condition, on the identification of the femoral lateral condyle point (FLE), the same operator found an average and maximum distance between the real point and the landmark found of respectively 3.5 and 8.8 mm with the use of the commercial software and of respectively 0.8 and 0.9 mm with the use of our routine.