Simplifying Knee OA Prognosis: A Deep Learning Approach Using Radiographs and Minimal Clinical Inputs

Abstract

Objectives: To predict the progression of knee osteoarthritis (OA), a deep convolutional neural network model was developed and applied to basic images and clinical data. Design: A vision transformer-based model was trained using 5565 knee radiographs as baseline images from the osteoarthritis initiative (OAI), including 578 testing images. Each knee had a corresponding Kellgren and Lawrence (KL) stage after 48 months of follow-up. Another 274 cases from the Far Eastern Memorial Hospital were used for external validation. The data included a combination of single/pairing images and full/essential clinical factors. Area under the receiver operating characteristics (AUROC), accuracy, sensitivity, specificity, odds ratio, and ability to discriminate surgical candidates were applied to evaluate model performance. Results: In cases with OA progression, the AUROC for identifying surgical candidates was 0.844, 0.804, 0.766, and 0.718 in the combination of a single image with essential factors, single image with full factors, pairing images with essential factors, and pairing images with full factors, respectively. In OAI testing using the simplest input, AUROC of identifying OA progression was 0.808, with 74.1% accuracy, 91.8% sensitivity, and 71% specificity. In external validation, AUROC of identifying OA progression was 0.709, with 71.2% accuracy, 72.2% sensitivity, and 70.3% specificity. Positive model prediction had an odds ratio of 23.87 (CI: 11.24~50.67) in OAI and 5.92 (CI: 3.50~10.03) in external validation. Conclusions: Our model provides reliable prediction results for knee OA cases with the advantages of simplicity and flexibility. The model performance was excellent in progression cases, potentially making early intervention in OA patients more efficient.