Related Experiment Video
Updated: Aug 13, 2026
05:37
A 3D Quantification Technique for Liver Fat Fraction Distribution Analysis Using Dixon Magnetic Resonance Imaging
Published on: October 20, 2023
A deep learning pipeline for liver macromolecular proton fraction quantification without subject-specific B1
Hongjian Kang1,2, Vincent W S Wong3, Jiabo Xu1,2
1Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Hong Kong, China.
Quantitative Imaging in Medicine and Surgery
|August 12, 2026
Summary
This study introduces a deep learning pipeline for accurate liver fibrosis staging using macromolecular proton fraction (MPF) quantification. The automated method eliminates the need for B1-inhomogeneity acquisition, improving reproducibility.
Area of Science:
- Medical Imaging
- Biomarker Discovery
- Artificial Intelligence in Medicine
Background:
- Macromolecular proton fraction (MPF) is a key noninvasive biomarker for liver fibrosis staging.
- Current MPF quantification methods require B1-inhomogeneity acquisition and manual region of interest (ROI) selection, leading to subjectivity and variability.
- A novel deep learning approach is proposed to overcome these limitations.
Purpose of the Study:
- To develop and validate a deep learning pipeline for automated liver MPF quantification.
- To eliminate the need for subject-specific B1 acquisition in MPF measurements.
- To enhance the objectivity and reproducibility of liver fibrosis staging.
Main Methods:
- A retrospective study involving 44 patients was conducted.
- A three-model deep learning pipeline was developed: segmentation, registration, and quantification networks.
- An uncertainty-guided strategy was employed for automatic ROI selection, and accuracy/reproducibility were assessed using MAE, SSIM, PSNR, ICC, and Bland-Altman analysis.
Main Results:
- The pipeline achieved high accuracy in MPF quantification with low Mean Absolute Error (MAE).
- Excellent reproducibility was demonstrated, with high Intraclass Correlation Coefficients (ICC) between automated analysis and expert manual selections.
- Bland-Altman analysis showed minimal bias and narrow limits of agreement, indicating strong concordance.
Conclusions:
- The proposed deep learning pipeline enables accurate and reproducible liver MPF quantification without subject-specific B1 acquisition.
- The automated approach, utilizing an atlas-based B1 substitution strategy, offers a significant advancement for liver fibrosis staging.
- This method holds promise for more objective and efficient clinical assessment of liver fibrosis.
