Predicting anatomical landmarks and bone morphology of the femur using local region matching.

Purpose: Anatomical landmarks and bony features are frequently used in biomechanical and surgical applications. The purpose of this study was to develop a local region matching-based anatomical landmark prediction method.

Methods: A reference femur model with anatomical landmarks and a surface division map was prepared. Initial registration between the reference femur model and a target femur model was performed in three-dimensional Cartesian space, and closest point pairs were determined by the initial surface correspondence. The models were mapped to unit spheres through spherical parameterization. Spherical registration using the closest point pairs in the spherical parametric space enabled the application of a division map from the reference model to the target model. The reference and target models were divided into local regions defined in the division map, and the corresponding regions were again registered in Cartesian space. Anatomical landmarks in the local regions were identified in the target model.

Results: The accuracy of the proposed method was tested for anatomical landmarks marked by a clinician on 35 femoral models. The effectiveness of local region matching was demonstrated by automatic measurements of the femoral neck-shaft angle. The average prediction error for all eight anatomical landmarks of the femur was 2.74 (±1.78) mm. The average of the predicted neck-shaft angle for our Korean subjects was 126.41° (±4.92°), which was comparable to previous studies in Japanese and Chinese populations.

Conclusion: Anatomical landmarks and features could be accurately predicted using the proposed local region matching method. This method offers robustness and accuracy in determining anatomical bony landmarks and bone morphology for clinical and biomechanical applications.