Purpose: To see if 3D imaging could improve our understanding of rotator cuff tendon tear shapes on MRI.
Methods: We performed a retrospective review of 1.5T/3T MR examinations, conducted over an 18-month period, of patients with arthroscopically proven full- thickness rotator cuff tears. Two orthopaedic surgeons reviewed the operative reports/arthroscopic photos for each patient, and characterized, in consensus, the shape of the tear based on a comparison of the tear’s width (size) and length (retraction), and involvement of the rotator interval without measurements into the following categories: crescent, longitudinal, U or L-shaped longitudinal, and massive-type. Two musculoskeletal radiologists reviewed the pre-operative MR examination for each patient independently/blind to the arthroscopic findings. Initially, the readers characterized the shape of the tendon tears by reviewing the standard 2DMR sequences and using the same criteria as the surgeons used during their review of the scope images. Next, the readers measured and documented the width and length of each tendon tear using the 2D images. The shape of the tear was then classified based on a previously published MR-based system as either crescent, longitudinal, U or L-shaped, or massive. Four weeks after the initial imaging evaluation, 3DMR reconstructions of each tear were reviewed and the shape documented by each radiologist independently/blind to the arthroscopic results using the same system used by our orthopaedic colleagues. These results were then compared to the 2D imaging evaluations and arthroscopic findings. Statistical analysis included 95% confidence intervals, McNemar test, and intra-class correlation coefficients.
Results: A total of 34 patients were included in the study; 21 had crescent shaped tears and 13 had longitudinal tears during arthroscopy. Of the 13 longitudinal tears, 8 were subtyped as U-shaped, while 5 were described as L-shaped. 6 of the 13 longitudinal tears were additionally classified as massive-type. There was no significant difference when comparing the accuracy of the tear shape characterizations made on 2DMR imaging without measurements (pre) and with measurements (post). The accuracy for differentiating between crescent shaped, longitudinal, and massive tears was the same for reader 1, 70.6% (24/34; p=1) and more accurate using the post 2D data for reader 2 (67.6%(post) vs. 61.8%(pre), p=0.5). The accuracy for tear shape characterization between crescent and longitudinal using the 3D reconstructions for reader #1 was 97.1% (33/34) and 88.2% (30/34) for reader #2. When this characterization included subclassifying the longitudinal tears into U or-L shaped, the accuracy for reader #1 was 97.1% and 82.4% for reader #2. When further characterizing the longitudinal tears as massive or not, both readers had an accuracy of 76.9% (10/13). The overall accuracy of the 3D reconstructions was 82.4% (56/68), significantly different (p=0.021) from the post 2D accuracy (64.7%) and pre 2D accuracy (60.3%, p=0.001). The intraclass correlation coefficient for the 2D measurements of width and length were 0.81, moderate agreement, for width and 0.95, strong agreement, for length.
Conclusion: Our study has demonstrated that 3DMR reconstructions of the rotator cuff improve the accuracy of characterizing rotator cuff tear shapes compared to the current 2DMR imaging based techniques.