An Automated, Deep Learning Approach to Systematically & Sequentially Derive Three-Dimensional Knee Kinematics Directly from Two-Dimensional Fluoroscopic Video

Total knee arthroplasty (TKA), also known as total knee replacement, is a surgical procedure to replace damaged parts of the knee joint with artificial components. It aims to relieve pain and improve knee function. TKA can improve knee kinematics and reduce pain, but it may also cause altered joint mechanics and complications. Proper patient selection, implant design, and surgical technique are important for successful outcomes. Kinematics analysis plays a vital role in TKA by evaluating knee joint movement and mechanics. It helps assess surgery success, guides implant and technique selection, informs implant design improvements, detects problems early, and improves patient outcomes. However, evaluating the kinematics of patients using conventional approaches presents significant challenges. The reliance on 3D CAD models limits applicability, as not all patients have access to such models. Moreover, the manual and time-consuming nature of the process makes it impractical for timely evaluations. Furthermore, the evaluation is confined to laboratory settings, limiting its feasibility in various locations.

This study aims to address these limitations by introducing a new methodology for analyzing in vivo 3D kinematics using an automated deep learning approach. The proposed methodology involves several steps, starting with image segmentation of the femur and tibia using a robust deep learning approach. Subsequently, 3D reconstruction of the implants is performed, followed by automated registration. Finally, efficient knee kinematics modeling is conducted. The final kinematics results showed potential for reducing workload and increasing efficiency. The algorithms demonstrated high speed and accuracy, which could enable real-time TKA kinematics analysis in the operating room or clinical settings. Unlike previous studies that relied on sponsorships and limited patient samples, this algorithm allows the analysis of any patient, anywhere, and at any time, accommodating larger subject populations and complete fluoroscopic sequences. Although further improvements can be made, the study showcases the potential of machine learning to expand access to TKA analysis tools and advance biomedical engineering applications.