Abstract
Machine learning (ML), increasingly used for predictive modeling, has seen rapid growth in osteoarthritis (OA) research over the past decade. This review highlights recent advances in ML model development across four OA outcome domains: clinical, structural (radiographic and MRI-based), and surgical endpoints, each addressing different but interrelated aspects of the disease. For clinical outcomes, ML studies have focused on predicting changes in patient-reported clinical measures (e.g., pain and function). Radiographic OA has been characterized using deep learning (DL) models, and ML approaches have also been used to predict progression of Kellgren Lawrence grades and joint space narrowing. For MRI-based features, DL-based tools have been developed for automatic quantification of cartilage, bone marrow lesions, and subcutaneous fat; they have improved scalability and supported development of ML prediction models with cartilage loss outcomes. For total knee replacement outcomes, ML models have demonstrated strong performance, offering the potential for both early intervention and surgical planning. This review also discusses emerging directions for ML in OA research, including the integration of multimodal data sources, the development of interpretable and explainable ML models, and the use of automated ML to streamline model development. Future approaches may include OA subtype-specific prediction models, alignment of ML approaches with clinical workflows, and enhanced external validation to ensure generalizability. These evolving strategies underscore the growing potential of ML to improve the detection of early OA, individualized risk stratification, and personalized interventions in OA clinical care.