Deep-Learning-Based Automated Rotator Cuff Tear Screening in Three Planes of Shoulder MRI.

Area of Science:

• Radiology
• Artificial Intelligence
• Orthopedics

Background:

• Rotator cuff tears are common injuries requiring accurate diagnosis.
• Shoulder MRI is the standard imaging modality for rotator cuff assessment.
• Current diagnostic methods can be time-consuming and may vary in accuracy.

Purpose of the Study:

• To develop and evaluate a deep neural network model for automated rotator cuff tear detection.
• To assess the model's performance using all three standard MRI planes (axial, coronal, sagittal) simultaneously and individually.
• To compare the deep learning model's performance against radiologist assessments.

Main Methods:

• Utilized 794 shoulder MRI scans (374 male, 420 female; mean age 59 ± 11 years).
• Rotator cuff tears were labeled by three musculoskeletal radiologists.
• Trained a YOLO v8 deep learning model using axial, coronal, and sagittal MRI planes, both separately and combined.
• Evaluated model performance using receiver operating characteristic curves and area under the curve (AUC).

Main Results:

• The model trained on all three imaging planes simultaneously achieved the highest Area Under the Curve (AUC) of 0.94 (p < 0.05).
• Single-plane performance was best with the axial plane (AUC: 0.71), followed by sagittal (0.70) and coronal (0.68).
• The all-plane training model demonstrated superior sensitivity (0.98) and accuracy (0.96).

Conclusions:

• Deep learning-based rotator cuff tear detection using multi-planar MRI is highly effective.
• Simultaneous analysis of all three MRI planes significantly enhances diagnostic performance.
• This automated approach shows promise for improving the efficiency and accuracy of rotator cuff tear diagnosis.