Automatic Landmark Detection on Knee Bony Surfaces in PSI-Based Total Knee Arthroplasty Using AI-Enhanced Statistical Shape Models

Total knee arthroplasty (TKA) is surgical a procedure involving the reconstruction of the entire knee joint. It requires the precise identification of specific landmarks (i.e., reference points) which are crucial for correct surgical planning. This study proposes a novel approach for the identification of such points by means of statistical shape models (SSMs) applied to 3D femur volumes. Given a 3D surface as input, the proposed method reconstructs it by deforming an average shape derived from the dataset. This average shape serves as a template onto which landmarks are manually positioned. As the deformation occurs, the landmarks shift from their original positions to new locations on the reconstructed surfaces. The construction of SSM necessitates employing registration algorithms between surfaces, with parameters selected using an intelligent method aimed at minimizing the RMSE between a chosen target shape and a reference one. The effectiveness of this approach was assessed through performance metrics including Hausdorff distance (HD) and RMSE for surface reconstruction, as well as the distance between target and predicted points. The results of this work emphasize the importance of stratification, high resolution, and specific preprocessing steps in model creation.