Deep Learning-Accelerated Non-Contrast Abbreviated Liver MRI for Detecting Malignant Focal Hepatic Lesions:...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Standard abbreviated MRI (AMRI_STD) protocols are used for liver imaging.
• Deep learning (DL) acceleration offers potential for improving MRI efficiency and quality.

Purpose of the Study:

• To compare a deep learning-accelerated non-enhanced abbreviated MRI (AMRI_DL) protocol against standard AMRI (AMRI_STD).
• To evaluate differences in image quality and malignant focal lesion detection in the liver.

Main Methods:

• Retrospective study of 155 patients undergoing both AMRI_STD and AMRI_DL (including DL-accelerated SSFSE_DL and DWI_DL).
• Phantom experiments assessed SNR, CNR, and ADC values.
• Three reviewers evaluated image quality (Likert scale) and detected malignant lesions; statistical comparison using Wilcoxon signed-rank test and logistic regression.

Main Results:

• AMRI_DL showed comparable per-lesion sensitivity (97.6%) to AMRI_STD for malignant lesions.
• AMRI_DL demonstrated superior image quality (sharpness, fewer artifacts, less noise, P < 0.001) and reduced scan time by ~50%.
• For HCC detection in chronic liver disease patients, AMRI_DL sensitivity was 96.6% vs. 96.5% for AMRI_STD.

Conclusions:

• The AMRI_DL protocol provides non-inferior sensitivity for detecting malignant liver lesions, including HCC.
• AMRI_DL significantly enhances image quality and reduces examination time by approximately 50% compared to AMRI_STD.