The follow-up of lower limb implants is routinely performed through the analysis of X-ray images to evaluate the characteristics of the implanted prosthesis and the limb alignment. This task is time-consuming and remains challenging for non-specialized physicians and junior surgeons. In this study, we present and evaluate a suite of deep learning algorithms. Our tool can automatically (1) perform quality control of X-rays to ensure their characteristics, (2) measure the prosthesis positioning on the knee close-up image and (3) determine the lower limb alignment angles on long-leg images. We collected a retrospective database of 103,360 X-rays belonging to 19,560 adult patients including both knee close-up and long-leg images. First, we cleaned our database to ensure its quality by defining 8 processing steps and training one convolutional neural network (CNN) for each of these steps. Secondly, we trained different neural networks to measure prosthesis positioning (i.e. angle between the prosthesis components and the bones’ axes) and limb alignment from manually labeled anatomical landmarks.

For the knee close-up, the algorithms obtained a mean error 1.71° (std 1.53°), close to the surgeons’ mean difference of 1.69° (std 1.52°) on a comparison subset. For the long-leg images, our pipeline reached a mean angle error of 2.59° (std 3.94°).

This study shows that innovative technologies can be integrated into the orthopedic surgeon’s routine. In particular artificial intelligence is well suited to assist in medical imaging analysis as it produces accurate and standardized measurements.