Deep Learning-Driven Abbreviated Shoulder MRI Protocols: Diagnostic Accuracy in Clinical Practice

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Deep learning (DL) reconstruction shows potential for reducing MRI scan times without compromising image quality.
• Clinical impact of accelerated DL shoulder MRI protocols on diagnostic accuracy is not well-established.

Purpose of the Study:

• To assess the diagnostic accuracy of 2-fold (DL2) and 4-fold (DL4) DL-accelerated shoulder MRI protocols compared to standard MRI.
• Evaluate the impact of acceleration factors on clinical diagnostic performance.

Main Methods:

• Prospective study of 88 patients undergoing standard, DL2, and DL4 shoulder MRI.
• Four radiologists assessed bone marrow edema, rotator cuff tears, and labral lesions.
• Calculated sensitivity, specificity, interobserver agreement, and diagnostic confidence.

Main Results:

• DL2 and DL4 achieved high diagnostic accuracy for bone marrow edema (100% sensitivity/specificity).
• Rotator cuff tear detection showed high sensitivity and specificity for both DL2 (98-100%/99-100%) and DL4 (95-98%/99-100%).
• Labral tear detection was perfect with DL2 and slightly reduced with DL4 (89-100% sensitivity); interobserver agreement was excellent (Kendall's W = 0.92-0.98).

Conclusions:

• DL-accelerated shoulder MRI, particularly DL2, can significantly reduce acquisition time (by 50%) while maintaining diagnostic reliability.
• DL2 protocols offer an optimal balance of speed and diagnostic confidence, enhancing patient throughput and MRI accessibility.
• DL4 provides acceptable accuracy but may show reduced sensitivity for subtle findings, especially with less experienced readers.