Deep learning-reconstructed hepatobiliary MRI: Enhancing biliary delineation, lesion margin sharpness, and small...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Hepatobiliary Imaging

Background:

• Conventional T1-weighted volumetric interpolated breath-hold examination (VIBE) MRI is crucial for hepatobiliary imaging.
• Deep learning (DL) reconstruction techniques offer potential for improving MRI image quality and diagnostic performance.

Purpose of the Study:

• To compare the performance of DL-VIBE against conventional VIBE for biliary tree visualization.
• To evaluate the efficacy of DL-VIBE in hepatic lesion detection and characterization.
• To assess the impact of DL-VIBE on image artifact suppression and overall image quality.

Main Methods:

• Retrospective analysis of 94 patients with suspected hepatic lesions who underwent gadoxetic acid-enhanced MRI.
• Comparison of DL-VIBE (1.5 mm slice thickness) with conventional VIBE (3.0 mm slice thickness) at the hepatobiliary phase.
• Quantitative and qualitative assessments of biliary anatomy, lesion margins, image noise, artifacts, and contrast-to-noise ratio (CNR).

Main Results:

• DL-VIBE showed significantly improved tertiary bile duct detection (163% increase) and continuity.
• Lesion margin sharpness was enhanced (ERD 0.93 vs. 1.51 mm) with higher overall (97.2% vs. 91.0%) and micro-lesion detection rates (96.3% vs. 87.6%).
• DL-VIBE reduced image noise and suppressed artifacts while maintaining bile duct CNR, despite slightly lower texture naturalness.

Conclusions:

• DL-reconstructed 1.5-mm HBP VIBE is a superior alternative to conventional 3-mm VIBE for hepatobiliary MRI.
• DL-VIBE enhances delineation of biliary microanatomy, lesion margin resolution, and detection of small hepatic lesions.
• This technique holds clinical utility for precision imaging and improved decision-making in standard breath-hold durations.