Deep learning-driven abbreviated knee MRI protocols: diagnostic accuracy in clinical practice.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Deep learning (DL) reconstruction in MRI offers reduced acquisition times but its effect on knee MRI diagnostic accuracy is not well-established.
• Assessing the impact of varying acceleration factors is crucial for clinical implementation.

Purpose of the Study:

• To evaluate the diagnostic performance of twofold, fourfold, and sixfold DL-accelerated knee MRI protocols compared to standard protocols.
• To determine the optimal DL acceleration factor for knee MRI.

Main Methods:

• A prospective study involving 71 patients undergoing knee MRI with standard, DL2, DL4, and DL6 accelerated protocols.
• Four radiologists assessed ligament tears, meniscal lesions, bone marrow edema, chondropathy, and extensor abnormalities.
• Diagnostic accuracy metrics including sensitivity, specificity, and interobserver agreement were calculated.

Main Results:

• DL2 and DL4 protocols demonstrated high diagnostic accuracy for most pathologies, comparable to standard protocols.
• DL6 showed reduced sensitivity for subtle lesions, particularly for ACL tears and meniscal evaluation.
• Bone marrow edema detection remained excellent across all DL acceleration factors; interobserver agreement was excellent for DL2/DL4 and good for DL6.

Conclusions:

• DL2 protocols closely match standard MRI performance, while DL4 offers acceptable diagnostic accuracy with significant time savings (up to 50%).
• DL6 acceleration may compromise diagnostic confidence for subtle abnormalities, especially for less experienced readers.
• DL2 and DL4 represent a favorable trade-off between reduced MRI acquisition time (50-75%) and diagnostic performance in knee imaging.