Validation and performance of a machine-learning derived prediction guide for total knee arthroplasty component...

Area of Science:

• Orthopedic Surgery
• Biomedical Engineering
• Data Science

Background:

• Accurate prediction of patient-specific component sizes before total knee arthroplasty (TKA) is crucial for optimizing surgical outcomes and minimizing costs.
• Current methods like templating are inconsistent and time-consuming, leading to potential sizing errors.
• Machine learning (ML) offers a promising alternative for real-time, accurate TKA component size prediction.

Purpose of the Study:

• To evaluate the performance of ML algorithms in predicting TKA component sizes.
• To identify key demographic factors influencing TKA component size prediction.
• To develop a practical tool for real-time TKA size prediction.

Main Methods:

• Trained five ML algorithms on demographic data (age, height, weight, BMI, sex) from 17,283 primary TKA patients (2012-2020).
• Validated algorithms internally on 20% of the patient cohort.
• Assessed performance using accuracy, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE).

Main Results:

• The SGB model achieved 95.0% accuracy for femoral component size and 97.8% for tibial component size.
• Predictive accuracy for femoral anteroposterior diameter was 83.6% (±4 mm) and for tibial medial/lateral diameter was 83.0% (±4 mm).
• Patient sex was identified as the most influential feature for both femoral and tibial component size prediction.

Conclusions:

• Novel ML algorithms demonstrate high accuracy in predicting TKA component sizes.
• Patient sex plays a significant role in TKA size prediction.
• A web-based application for real-time TKA size prediction was developed, requiring external validation before clinical implementation.