Abstract / Summary
Abstract Background Understanding the diagnostic agreement between advanced imaging and arthroscopy is essential for accurate preoperative planning and tailored therapy. This study aimed to evaluate the diagnostic concordance between preoperative magnetic resonance imaging (MRI) and direct arthroscopic examination for principal ankle pathologies and identify key predictors of arthroscopically confirmed osteochondral lesions (OLs) using machine learning (ML) predictive algorithms. Materials and Methods This retrospective study included 54 adults with chronic or recurrent ankle symptoms, who underwent preoperative imaging and subsequent ankle arthroscopy. Arthroscopy served as the reference standard. Sensitivity, specificity, positive and negative predictive values, accuracy, and Cohen κ were calculated with 95% confidence intervals. Exploratory ML models predicting arthroscopically confirmed OLs were evaluated using leave-one-out cross-validation (LOOCV), followed by feature importance and ablation analyses. Results For OLs, MRI demonstrated 82.1% sensitivity, 80.8% specificity, 81.5% accuracy, and substantial agreement with arthroscopy (κ = 0.629). For synovitis, MRI showed 94.3% sensitivity and 100% specificity, although the specificity was based on only one arthroscopy-negative case. For loose body detection, MRI demonstrated 50.0% sensitivity and 90.5% specificity. For ligamentous injury, the sensitivity and specificity were 100% and 25.0%, respectively, with directional discordance between MRI and arthroscopy (exact McNemar P = 0.004). In the exploratory ML analysis, linear Support Vector Machine and Logistic Regression achieved LOOCV accuracies of 81.5% and 79.6%, respectively. Direct MRI visualization of the OL had the largest Logistic Regression feature weight (56.5%), and removing this feature reduced the Logistic Regression Accuracy to 50.0%. Conclusions MRI diagnostic performance varied according to lesion type, with balanced performance for OLs and high sensitivity for synovitis and ligamentous injury. Exploratory ML analysis indicated that direct MRI visualization was the predominant predictor of arthroscopically confirmed OLs, with smaller contributions from coexisting loose bodies and symptom duration. Preoperative MRI provides excellent accuracy and unbiased diagnostic agreement for OLs and synovitis and serves as an indispensable tool for surgical planning. However, clinicians should recognize the potential for discordance between MRI and arthroscopic findings in the assessment of ligamentous injury as well as the limited sensitivity of MRI for detecting small intra-articular loose bodies. Level of Evidence: Level 3