To assess the sensitivity and specificity of MR arthrography in detecting cartilage lesions in the shoulder joint as compared to surgical arthroscopy.

85 patients with clinical suspicion for labral or rotator cuff tear underwent shoulder MR arthrography at 1.5 T followed by arthroscopy. MR acquisitions included T1 weighted fat sat SE, PD weighted FSE, T2 weighted fat sat FSE and STIR acquisitions. Images were obtained in axial, oblique sagittal, and oblique coronal planes. The frequency and severity of cartilage changes in the glenoid (anterior, posterior, superior and inferior) and humeral head (anterior, posterior, central, medial, and lateral) were graded on a six point scale (0 = normal, 1 = signal heterogeneity, 2 = fissuring, 3 = thinning <50%, 4 = thinning >50%, 5 = full thickness cartilage loss, 9 = not evaluable). Arthroscopic findings were then correlated and sensitivity, specificity, accuracy, and negative (NPV) and positive (PPV) predictive values of MR in detecting cartilage lesions were calculated.

Arthroscopic correlation was evaluable in 85 patients for labrum and rotator cuff, and 84 patients for cartilage lesions. For the glenoid cartilage, sensitivity and specificity were 82 and 72%; NPV, PPV, and accuracy were 92, 51, 73%, respectively. For the humeral cartilage, sensitivity and specificity were 77 and 74%; NPV, PPV, and accuracy were 90, 51.5, and 73%, respectively. The highest correlation occurred when the cartilage was normal or near - normal or severely diseased. The majority of false negative and false positive findings occurred in early and intermediate stages of cartilage loss.

Cartilage lesions were correctly identified in over 70% of the analyzed cases on pre-operative MR arthrography when compared to arthroscopy. Assessment of the articular cartilage of the shoulder joint is feasible using MR arthrography with high sensitivity and specificity.