Determining the Mechanical Axis of the Femur From a Standard Antero-Posterior Knee Radiograph With Deep Learning

Area of Science:

• Orthopedic surgery
• Medical imaging analysis
• Artificial intelligence in healthcare

Background:

• Standard antero-posterior (AP) knee views lack femoral head visibility, complicating alignment assessment.
• Surgeons currently rely on incomplete measurements, less precise anatomic axes, or approximations for knee alignment.
• Accurate assessment of femoral alignment is crucial for surgical planning and outcomes.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting the correction factor between the anatomical and mechanical axes of the femur.
• To enable accurate femoral alignment assessment using only standard AP knee radiographs.

Main Methods:

• Utilized a dataset of bilateral AP knee and long-leg radiographs taken within 90 days.
• Employed validated DL algorithms to measure mechanical and anatomical axes.
• Trained a DL regression model to predict the correction factor from AP images alone.
• Compared DL model performance against linear regression and a 6-degree varus addition method.

Main Results:

• The DL model achieved a mean absolute error (MAE) of 1.02° (SD: 0.94°) for predicting the correction factor.
• Linear regression yielded an MAE of 1.34° (SD: 1.10°).
• The 6-degree varus addition method resulted in an MAE of 2.00° (SD: 1.47°).

Conclusions:

• A DL model was successfully developed to predict the femoral mechanical axis from standard AP knee radiographs.
• The DL model demonstrated clinically relevant accuracy within 1°, outperforming existing methods.
• This automated tool offers significant potential benefits for clinical practice and research in orthopedic alignment assessment.