Machine learning-based prediction of rotator cuff tears using anatomical parameters: a retrospective cohort study.

BACKGROUND: To apply machine learning to improve risk assessment for rotator cuff tears by developing an integrated predictive model that combines acromial morphology (structural factors) and patient characteristics (systemic factors). METHODS: A total of 342 patients who underwent shoulder radiography and MRI between 2023 and 2025 were included (180 tears, 162 intact). Demographic and radiographic parameters—including the novel acromial angle (NAA), critical shoulder angle, acromion index (AI), and acromial tilt—were collected. Six machine learning classifiers (logistic regression, decision tree, random forest, k-nearest neighbors, support vector machine, and XGBoost) were trained using 10-fold cross-validation after LASSO-based feature selection. Performance was evaluated on a hold-out test set using area under the curve (AUC) and secondary metrics. SHapley Additive exPlanations (SHAP) values were used to interpret feature contributions in the best model. RESULTS: The tear group was older and showed a significantly smaller NAA (142.8° vs. 147.4°) and higher AI (0.80 vs. 0.77) than the intact group (P < 0.001). Diabetes was more common and BMI lower among patients with tears (P < 0.01). XGBoost achieved the best performance (AUC 0.87 in training, 0.74 in testing), with higher balanced accuracy and F1-score than other classifiers. SHAP analysis indicated that NAA, age, AI, and dominant-arm involvement were the strongest predictors. CONCLUSIONS: A machine-learning model integrating clinical and acromial morphological factors can effectively predict rotator cuff tear risk. The novel acromial angle is a key determinant. This interpretable model may support early identification of high-risk individuals and inform preventive strategies, but prospective validation is needed.