Automated Artificial Intelligence-Based Assessment of Lower Limb Alignment Validated on Weight-Bearing Pre- and...

Area of Science:

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

Background:

• Standing weight-bearing full-leg radiographs (FLR) are standard for knee alignment assessment.
• Manual measurement of the load-bearing axis from FLR is time-consuming and prone to variability.

Purpose of the Study:

• To develop and validate a novel artificial intelligence (AI) algorithm for automated lower limb alignment assessment.
• To improve the efficiency and objectivity of analyzing full-leg radiographs.

Main Methods:

• A customized mask-RCNN model was trained for anatomical structure and implant detection in FLR.
• UNet-based neural networks were employed to automatically identify anatomical landmarks.
• The algorithm calculated five key lower limb alignment angles using detected landmarks.

Main Results:

• The AI algorithm achieved high detection rates for lower limb angles (92.4%-98.9%).
• AI demonstrated excellent inter-rater reliability compared to human measurements (ICCs 0.80-1.0).
• The algorithm showed high agreement with manual measurements on a validation dataset of 200 FLR.

Conclusions:

• The developed AI algorithm provides an objective and time-saving method for analyzing lower limb alignment from FLR.
• This tool has the potential to streamline the analysis of large datasets and assist physicians in routine practice.
• The AI algorithm's performance was validated externally, showing excellent reliability comparable to human experts.