A radiographic artificial intelligence tool to identify candidates suitable for partial knee arthroplasty

Area of Science:

• Orthopaedic Surgery
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Knee osteoarthritis is a widespread condition requiring significant knee replacement surgeries, with increasing demand.
• Partial knee arthroplasty (PKA) presents benefits over total knee arthroplasty (TKA) but is underutilized, partly due to the limited use of radiographic decision aids.
• Clinician time constraints hinder the adoption of existing decision support tools for PKA candidacy.

Purpose of the Study:

• To develop and evaluate a novel radiographic artificial intelligence (AI) tool for identifying patients suitable for PKA.
• To address the underutilization of PKA by providing an efficient decision support mechanism.

Main Methods:

• Trained six AI models on a dataset of 1241 knee radiograph series, assessed by expert orthopaedic surgeons.
• Utilized EfficientNet-ES architecture for model development to predict PKA candidacy.

Main Results:

• AI models achieved statistically significant accuracies exceeding random assignment.
• The EfficientNet-ES model demonstrated high performance with an AUC of 95%, F1 score of 83%, and accuracy of 80% in identifying PKA candidates.

Conclusions:

• The developed AI tool shows significant potential for identifying PKA candidates, potentially increasing the utilization of this beneficial procedure.
• Integration of this AI tool into clinical workflows could enhance shared decision-making processes between clinicians and patients.
• Further research is necessary to validate the AI tool's real-world effectiveness and impact on patient outcomes.