Related Experiment Video
Updated: Apr 7, 2026
Fat-Water Phantoms for Magnetic Resonance Imaging Validation: A Flexible and Scalable Protocol
Published on: September 7, 2018
Multi-vendor, multi-field strength validation of MRI-Proton density fat fraction for fat quantification: A
Kuang-Chen Yen1, Po-Chin Liang2, Yueh-Hsun Lu3
1Department of Medical Imaging, National Taiwan University Cancer Center, Taipei, Taiwan; Department of Radiology, National Taiwan University, Taipei, Taiwan.
Background/Purpose:
This study aimed to validate the accuracy and consistency of Magnetic Resonance Imaging Proton Density Fat Fraction (MRI-PDFF) across systems from three major vendors at two field strengths.
Methods:
A PDFF phantom with five calibrated values (0%, 10%, 20%, 30%, and 40%) was scanned at six hospitals. The platforms included GE, Siemens, and Philips systems at both 1.5T and 3.0T, utilizing corresponding commercial PDFF sequences. Correlation with reference values was assessed using Pearson's correlation coefficient (r). Overall agreement and systematic bias were evaluated using the Concordance Correlation Coefficient (CCC) and Bland-Altman analysis, respectively. Finally, linearly weighted Kappa (κ) statistics were used to evaluate agreement for clinical steatosis grading.
Results:
All evaluated systems exhibited an excellent correlation with reference values, yielding a Pearson correlation coefficient (r) > 0.99 (p < 0.0001). The overall agreement measured by CCC was also excellent, ranging from 0.9797 to 0.9964. Bland-Altman analysis revealed no statistically significant bias for Siemens platforms (p > 0.05). However, a small but statistically significant systematic underestimation was observed for GE (mean bias: 1.7 to -1.9%) and Philips (mean bias: 2.1 to -2.7%) platforms (p < 0.0001). Weighted Kappa analysis showed excellent to perfect agreement for steatosis grading (κ = 0.83 to 1.00).
Conclusion:
MRI-PDFF is an accurate, precise, and consistent biomarker for liver fat quantification across different vendors and field strengths. The minor, systematic inter-vendor biases observed are likely due to proprietary algorithmic differences and, while quantitatively small, highlight the need for harmonization in multi-center clinical trials.
