MRI-based predictors of MRI-detected ramp lesions in ACL-injured knees: an exploratory logistic regression and machine learning model

To identify clinical, demographic, and MRI-based morphological factors associated with MRI-detected medial meniscus ramp lesions in patients with anterior cruciate ligament (ACL) injuries and to develop an exploratory prediction model, comparing logistic regression with several machine learning (ML) algorithms. One hundred thirty patients with MRI-confirmed ACL rupture were retrospectively analyzed. Ramp lesions were defined on MRI. Clinical variables and quantitative MRI parameters (medial meniscus area, medial tibial plateau area, and medial posterior tibial slope angle) were independently assessed by two observers who were blinded to clinical information and to each other’s readings. Between-group differences were analyzed using chi-square and independent samples t-tests. Independent associations were identified by multivariate logistic regression. Five ML algorithms were compared using 10-fold cross-validation. Ramp lesions were identified in 31.5% of patients. In multivariable analysis, younger age (OR 0.92, 95% CI 0.86–0.98), larger medial meniscus area (OR 2.26, 95% CI 1.08–4.73), and greater posterior tibial slope angle (OR 1.36, 95% CI 1.08–1.70) were independently associated with ramp lesions, whereas female sex was associated with lower adjusted odds (OR 0.16, 95% CI 0.03–0.82). Logistic regression achieved the highest discrimination (cross-validated AUC = 0.83 +/- 0.15; pooled ROC AUC = 0.82). Calibration analysis indicated adequate model fit (Brier score: 0.159), and decision curve analysis suggested net clinical benefit across the 12–60% threshold range. Larger medial meniscus area and greater posterior tibial slope angle, were independently associated with MRI-detected ramp lesions. The exploratory logistic regression model showed moderate discrimination but limited sensitivity. This model may, after external validation, help raise preoperative suspicion of ramp lesions.