Purpose: To determine interobserver and intraobserver variation in the interpretation of magnetic resonance (MR) images in rotator cuff disorders.
Materials and methods: MR images of the shoulder in 97 patients were retrospectively reviewed twice, with a 3-week interval. Surgical findings indicated a full-thickness tear in 29 patients, grade 1 impingement in 19 (tendinitis), and grade 2 impingement (partial tear) in 26. The control population comprised 23 asymptomatic volunteers or patients.
Results: All observers were accurate in the diagnosis of a full-thickness tear (89%-98%), with good intraobserver (kappa = 0.67-0.84) and interobserver agreement (kappa = 0.74-0.92). In diagnoses of tendinitis, partial tear, and normal cuff, there were wide ranges of sensitivity (13%-74%) and specificity (72%-93%), as well as poor interobserver (kappa = 0.12-0.60) and intraobserver agreement (kappa = 0.35-0.78).
Conclusion: Full-thickness tears of the rotator cuff can be accurately identified at MR imaging with little observer variation. Consistent differentiation of normal rotator cuff, tendinitis, and partial thickness tears is more difficult.