Radiomics-based MRI model for predicting the severity of rotator cuff tears

Differentiating full thickness from partial thickness rotator cuff tears is crucial for optimal surgical planning and clinical decision making. Conventional MRI assessment relies largely on subjective interpretation and may lead to diagnostic variability.

This study aimed to develop and validate an MRI-based radiomics model capable of accurately distinguishing full thickness from partial thickness rotator cuff tears and to evaluate its diagnostic performance and potential clinical utility.

A total of 120 patients were included (full thickness, n = 60; partial thickness, n = 60). Shoulder MRI, with proton density-weighted fat-suppressed imaging as the core acquisition, was used for region of interest segmentation by two radiologists. The IBSI-compliant radiomic features were extracted, preselected by mRMR, and modelled using LASSO regularization. Patients were stratified by class into a training set (n = 84) and an independent testing set (n = 36). Using fivefold cross-validation and a fixed threshold, we evaluated the AUC, accuracy, sensitivity, and specificity in both cohorts.

The final radiomics model comprising five features achieved an AUC of 0.86 (95% confidence interval, CI 0.78-0.94) in the training set and 0.82 (95% CI 0.68-0.96) in the testing set. On the testing set, sensitivity was 0.83 (95% CI 0.61-0.94), specificity 0.61 (95% CI 0.39-0.80), and accuracy 0.72 (95% CI 0.56-0.84).

An interpretable radiomics model derived from routine shoulder MRI reliably distinguishes full from partial thickness rotator cuff tears and shows promise as a quantitative tool for preoperative stratification and decision support.