Shoulder measurements on MRI: Statistical analysis of patients without and with rotator cuff tears and predictive modeling.

This study aims to determine significant differences between various measurements of shoulder structures and relationships in patients with and without rotator cuff tears and identify a subset of these measurements that may merit further investigation. Shoulder MRIs of 120 patients with rotator cuff tears (RCT group) and 80 patients with normal examinations (No RCT group) were retrospectively identified. Numerous measurements of shoulder anatomy and relationships and pathology characterization were performed for each study. The mean and ranges of measurements for each group were identified, and P-values were calculated to assess differences between the two groups. Classification and Regression Tree (CART) identified prediction models for separating the two groups based on these shoulder MRI measurements. Statistically significant differences were observed for 9 of 18 of the shoulder measurements between the two groups (P < 0.05). Using long head of biceps tendon tear, tendinosis, and subacromial-subdeltoid bursal fluid, the prediction model from CART demonstrated 99.5% accuracy in separating the two groups. Subacromial distance, subacromial spur size, and acromioclavicular osteophyte could also separate the two groups with 97.5% accuracy. Other measures less commonly associated with impingement and rotator cuff tears could also be used to separate the two groups with up to 81.5% accuracy. This study introduces new measures and clarifies ranges for existing measurements on shoulder MRI. Abnormalities of some of these measurements may be associated with subacromial impingement, and combinations of these parameters may be useful for separating patients with or without rotator cuff tears. Clin. Anat. 33:173-186, 2020. © 2019 Wiley Periodicals, Inc.