Predicting Knee osteoarthritis progression using explainable machine learning and clinical imaging data
Rayan Harari1,2, Stacy E Smith1,3, Sara M Bahouth1
1Department of Radiology, Mass General Brigham, Harvard Medical School, Boston, MA, USA.
Osteoarthritis and Cartilage Open
|June 29, 2026
Summary
Explainable AI accurately predicts knee osteoarthritis radiographic progression using MRI and clinical data. Key factors include cartilage loss and bone marrow lesions, offering insights into disease development.
Area of Science:
- Orthopedics
- Radiology
- Artificial Intelligence
Background:
- Knee osteoarthritis (KOA) poses a significant health burden.
- Predicting KOA progression is crucial for timely intervention.
- Quantitative MRI and clinical data offer rich insights into KOA.
Purpose of the Study:
- To evaluate explainable AI models for KOA progression prediction.
- To assess models for both composite (radiographic + pain) and radiographic-only endpoints.
- To identify key predictive features using explainability methods.
Main Methods:
- Analysis of 600 participants from the FNIH Osteoarthritis Biomarkers Consortium (OAI).
- Utilized demographic, clinical, and quantitative MRI features (cartilage, bone marrow lesions, osteophytes, effusion-synovitis).
- Employed five classifiers (e.g., random forest, XGBoost) with explainability techniques (SHAP, Gini importance).
Main Results:
- Radiographic progression was accurately predicted (AUC=0.87) using longitudinal MRI changes.
- Baseline features also showed strong predictive performance (AUC=0.80).
- Composite progression prediction was less accurate (AUCs=0.66-0.70).
Conclusions:
- Explainable AI with quantitative MRI enables interpretable KOA progression prediction.
- This study integrates longitudinal MRI features and model-agnostic explanations in the FNIH/OAI cohort.
- Identified medial femoral cartilage loss, bone marrow lesions, osteophytes, and effusion-synovitis as key predictors.
