Development of a Machine Learning Model for Determining Alignment in Knees Following Total Knee Arthroplasty

Comprehensive Summary

This study aimed to develop a machine learning (ML) model to accurately determine total knee arthroplasty (TKA) alignment from hip to ankle imaging. 550 radiographs from patients who had undergone TKA were collected, 440 of which were used to train the ML model, using landmark features of the leg and knee. Four separate identifiers were measured: mechanical hip-knee-ankle angle (mHKA), lateral distal femoral angle (LDFA), medial proximal tibia angle (MPTA), and joint line obliquity (JLO). 110 separate random radiographs were used to test the model’s accuracy. Mean errors between human measured and ML model measured values were: 0.08° for mHKA, 0.7° for LDFA, 0.4° for MPTA, and 0.7° for JLO. Overall, these results signify a highly accurate ML model, which can measure angles very closely to the capabilities of a human, with the advantage of only taking 0.1 seconds per image.

Outcomes and Implications

This machine learning model is a huge advantage towards streamlining clinical workflows, as it can very rapidly and accurately measure key characteristics after total knee arthroplasty. For physicians, this makes follow-ups much quicker and easier, and additionally this study provides a basis for future research into total knee arthroplasty procedures. Overall, the study points to a future with machine learning models as a primary tool for imaging and outcome assessment.