Deep Learning Superresolution for Simultaneous Multislice Parallel Imaging-Accelerated Knee MRI Using Arthroscopy...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Deep learning (DL) enhances accelerated MRI but requires validation against reference standards.
• Accelerated MRI techniques like simultaneous-multislice (SMSx2) and parallel imaging (PIx2) are crucial for efficient imaging.
• Ensuring the accuracy and robustness of DL-enhanced MRI is vital for clinical adoption.

Purpose of the Study:

• To validate the diagnostic performance of DL superresolution MRI accelerated with SMSx2 and PIx2.
• To compare the DL approach against conventional SMSx2-PIx2-accelerated MRI in knee imaging.
• To assess image quality and diagnostic accuracy using arthroscopy as the reference standard.

Main Methods:

• Prospective enrollment of adult patients with knee conditions.
• Acquisition of fourfold accelerated standard-of-care and DL superresolution MRI at 3 T.
• Independent radiologist evaluation of image quality, meniscus tears, ligament tears, and articular cartilage defects.
• Statistical analysis including interreader agreement and area under the receiver operating characteristic curve (AUC).

Main Results:

• DL superresolution MRI demonstrated superior image quality (median Likert score 5 vs 4, P < .001).
• Diagnostic performance for meniscus and ligament tears was similar between DL and conventional MRI (all P > .96).
• DL superresolution MRI showed improved performance for detecting articular cartilage lesions (AUC 0.78 vs 0.71, P = .002).

Conclusions:

• Fourfold accelerated DL superresolution MRI offers enhanced image quality in knee imaging.
• DL MRI provides comparable diagnostic accuracy for meniscal and ligamentous injuries.
• DL superresolution MRI improves the detection of articular cartilage lesions compared to conventional accelerated MRI.