Magnetic resonance shoulder imaging using deep learning-based algorithm.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Orthopedic Imaging

Background:

• Conventional MRI techniques for shoulder imaging can be time-consuming.
• Image quality and diagnostic accuracy are critical for effective shoulder joint assessment.
• Deep learning offers potential for optimizing medical imaging protocols.

Purpose of the Study:

• To evaluate the feasibility of deep learning-based MRI (DL-MRI) for shoulder imaging.
• To compare the performance of DL-MRI against conventional non-DL-MRI.
• To assess image quality, scan time, and diagnostic agreement of DL-MRI.

Main Methods:

• Retrospective analysis of 70 patients who underwent both DL-MRI and non-DL-MRI.
• Image quality and lesion diagnosis comparison using radiologist assessments and statistical analysis (Kendall W, Wilcoxon, kappa).
• Satisfaction survey of 400 patients who underwent DL-MRI for image quality assessment.

Main Results:

• DL-MRI reduced scan time by approximately 50% (6 min 1s vs. 11 min 25s).
• DL-MRI demonstrated superior image quality in PDWI and T2WI sequences with higher SNRs and CNRs.
• High interobserver and interstudy agreement for lesion diagnosis (kappa = 0.913–1.000) and excellent radiologist satisfaction scores were observed for DL-MRI.

Conclusions:

• DL-MRI significantly decreases scan time and enhances image quality for shoulder PDWI and T2WI compared to conventional MRI.
• DL-MRI exhibits high diagnostic agreement and radiologist satisfaction, indicating its clinical utility.
• Shoulder 2D DL-MRI is a viable option for clinical routine, potentially improving work efficiency.