ASSESSING DIAGNOSTIC ACCURACY OF MRI IN ROTATOR CUFF TEAR DETECTION

Abstract

Rotator cuff tears are a leading cause of shoulder pain and disability, significantly affecting quality of life and functional outcomes. Accurate diagnosis is essential to guide treatment and prevent progression from partial- to full-thickness tears. Magnetic resonance imaging is widely recognized as the preferred modality for tendon evaluation due to its ability to provide detailed soft tissue visualization. Despite its strengths, diagnostic variability persists, particularly in detecting partial-thickness tears, which often present with subtle features. This study evaluated the diagnostic performance of magnetic resonance imaging in detecting normal tendons, partial-thickness tears, and full-thickness tears, while also assessing the impact of dataset imbalance. A secondary analysis was performed using a publicly available dataset of 2,447 cases, with 242 cases forming the test set (160 normal, 16 partial-thickness, 66 full-thickness). Classification performance was measured using sensitivity, specificity, predictive values, and overall accuracy. Results showed that overall accuracy was 88%. Normal tendons were detected with sensitivity of 92%, specificity of 94%, and precision of 97%, while full-thickness tears achieved sensitivity of 89%, specificity of 94%, and precision of 86%. In contrast, partial-thickness tears showed weaker results, with sensitivity of 50% and precision of 38%, largely due to their low prevalence of 6.6% in the dataset. In conclusion, magnetic resonance imaging remains a robust diagnostic tool for normal tendons and full-thickness tears. However, improving recognition of partial-thickness tears and addressing dataset imbalance are critical for reducing misclassification and optimizing clinical management.