Deep learning reconstruction for accelerated high-resolution upper abdominal MRI improves lesion detection without...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Conventional T1-weighted volumetric interpolated breath-hold examination (VIBE) sequences are standard for upper abdominal MRI.
• Accelerated MRI techniques are crucial for improving patient comfort and workflow efficiency.
• Deep learning (DL) reconstruction offers potential for enhancing image quality in accelerated MRI.

Purpose of the Study:

• To compare a conventional VIBE sequence with a DL-reconstructed accelerated HR-VIBE_DL sequence.
• To evaluate differences in image quality, lesion conspicuity, and lesion detection rates.
• To assess the performance of HR-VIBE_DL in upper abdominal MRI.

Main Methods:

• Prospective enrollment of patients undergoing upper abdominal MRI at 1.5 T.
• Acquisition of both conventional VIBE (fourfold acceleration) and HR-VIBE_DL (sixfold acceleration) sequences with identical 16s acquisition time.
• Blinded, randomized image analysis by three readers focusing on image quality, lesion conspicuity, and detection in liver, pancreas, spleen, lymph nodes, and adrenal glands.

Main Results:

• HR-VIBE_DL demonstrated superior signal-to-noise ratio and edge sharpness compared to conventional VIBE (P < 0.001).
• Lesion conspicuity was significantly higher on HR-VIBE_DL images (P = 0.005).
• HR-VIBE_DL achieved a higher lesion detection rate (97.5%) versus conventional VIBE (93.2%) (P = 0.002).

Conclusions:

• DL-reconstructed HR-VIBE_DL provides superior image quality in upper abdominal MRI.
• The technique enhances lesion conspicuity and detection rates compared to conventional VIBE.
• HR-VIBE_DL offers these benefits without any increase in examination time.