Automated Femur Segmentation and Joint Axis Extraction

3D Slicer 4.13 (Fedorov et al., 2012 (link)) was used to segment MRI images. Each femur was split into five parts similar to previous studies (Kainz et al., 2020 (link))—the proximal trabecular bone, the growth plate, the cortical bone of the shaft, the bone marrow and the distal trabecular bone. STL-files of all parts and additionally a file containing the full femur were exported. The STAPLE-Toolbox of Modenese and Renault (2021) (link) was used to identify the femoral head and the epicondyles representing the hip and the knee joint axis using the “GIBOC-Femur” and “GIBOC-Cylinder” algorithms, respectively. If “GIBOC-Cylinder” algorithm failed to fit a cylinder through both epicondyles, “GIBOC-Ellipsoids” algorithm was used to fit ellipsoids through medial and lateral epicondyles. The hip joint center and knee joint axis were required to transform the femur into the OpenSim coordinate system.
The diaphysis of the femur was defined by removing 20% off the top and bottom of the femur. Then, the principal inertia axis of the remaining part was calculated to identify the shaft axis. The neck axis was defined by fitting a least-squares cylinder through surface nodes of the femoral neck. The longitudinal axis of this cylinder was constrained to pass through the femoral head center. The AVA was calculated as the angle between the neck axis and the medial-lateral knee axis obtained from STAPLE-Toolbox (Modenese and Renault, 2021 (link)) in the transverse plane. The NSA was computed as the angle between the neck axis and shaft axis in 3D space.