Comprehensive Summary
In this study, Chandrashekar et al. developed a machine learning model to determine alignment in knees following total knee arthroplasty (TKA) by analyzing long leg radiographs (LLRs). The model was trained to identify and label key landmarks on the femur, tibia, and joint space. Using these landmarks, four metrics were calculated: mechanical hip-knee-ankle angle (mHKA), lateral distal femoral angle (LDFA), medial tibial plateau angle (MPTA), and joint line obliquity (JLO). The model was trained on 440 patients and tested on 110 LLRs, achieving a mean error of 0.08º for mHKA, 0.7º for LDFA, 0.4º for MPTA, and 0.7º for JLO. Landmark placement accuracy had a mean root mean square error (RMSE) of 1.53 pixels. The model processes over 10 images per second, offering a significant improvement in efficiency over manual annotation.
Outcomes and Implications
The development of this machine learning model has significant implications for the field of orthopedics, particularly in improving the workflow and productivity of clinics performing total knee arthroplasties. By automating the process of determining implant alignment, the model reduces the need for manual annotation by trained professionals, saving time and resources. This increased efficiency allows for more thorough routine screenings of postoperative implant alignment, ensuring prompt discovery of patients at higher risk for joint deterioration. The model's potential expansion to short leg radiographs (SLRs) could further enhance its applicability in clinical settings.