Unsupervised Machine Learning Drives Dynamic Alignment Classification in Navigated Total Knee Arthroplasty

Area of Science:

• Orthopaedic Surgery
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Current total knee arthroplasty (TKA) classifications are static and do not reflect dynamic limb behavior during gait.
• A novel intraoperative method using an intra-articular device and computer-assisted orthopaedic surgery (CAOS) system measures dynamic hip-knee angle (dHKA).

Purpose of the Study:

• To develop a machine learning (ML) model for classifying patient-specific dHKA profiles intraoperatively.
• To analyze the stability of these dHKA profiles before and after the femoral cut in a tibia-first TKA workflow.
• To assess the correlation between dHKA profile preservation and early functional outcomes.

Main Methods:

• Reviewed 1,890 tibia-first TKA cases across 11 surgeons, recording dHKA at 12 flexion angles pre- and post-femoral cut.
• Utilized a K-means clustering model trained on pre-cut data to identify alignment profiles, then applied it to post-cut data.
• Analyzed a subset of 141 cases for the association between cluster preservation and one-year Knee Injury and Osteoarthritis Outcome Score for Joint Replacement (KOOS Jr.) scores.

Main Results:

• Identified four distinct dHKA clusters (valgus/neutral, neutral, low-to-moderate varus, moderate-to-high varus).
• Overall, 69.4% of cases maintained their pre-cut cluster after the femoral cut, with surgeon variation (61-88%).
• In the outcomes subset, 72.3% preserved their cluster, which was associated with improved KOOS Jr. scores, particularly in low-to-moderate varus knees.

Conclusions:

• Demonstrated the first unsupervised ML application for classifying intraoperative dHKA profiles using a force-controlled device and CAOS system.
• This method provides real-time feedback for surgical alignment.
• Establishes a foundation for automated alignment classification guidance in personalized TKA.