Background: In recent years, CT-derived liver fat quantification methods named multi-material decomposition (MMD) is playing an increasingly important role as imaging biomarker of hepatic steatosis. However, there were various measurement ways with various results among different researches, and the impact of measurement methods on the results is unknown.

Purpose: To evaluate the reproducibility of liver fat volume fraction (FVF) using MMD algorithm in nonalcoholic fatty liver disease (NAFLD) patients when taking vascular, location and iodine contrast into account during measurement.

Methods: This retrospective study was approved by the institutional ethics committee, and the requirement for informed consent was waived because of the retrospective nature of the study. 101 patients with nonalcoholic fatty liver disease (NAFLD) were enrolled in this study. All participants underwent non-contrast phase (NCP) and two-phase enhanced abdomen CT scanning (late arterial phase (LAP) and portal vein phase (PVP)) with GSI mode on a DECT scanner (Revolution CT, GE Healthcare, Chicago, IL). The Fat volume fractions (FVF) of region of interest (ROI) were obtained through the software named ‘GSI liver fat’ on the Advanced Workstation (AW4.7; GE Healthcare). ROIs with size of ~400 mm² were placed at right posterior lobe (RPL), right anterior lobe (RAL) and left lateral lobe (LLL) to obtain FVF values on liver fat images without and with the reference of enhanced CT images (FVF_without, FVF_with). The differences of FVF values under NCP obtained without and with the reference of enhanced CT images (FVF_without vs. FVF_with) were compared using Wilcoxon matched-pairs rank test. The difference of FVF values obtained from NCP images with the reference of enhanced CT images between each two lobes (FVFwith: RPL vs. RAL, RPL vs. LLL, RAL vs. LLL) were also compared. Friedman test was used to compare FVF values among three phases for each lobe, while the consistency of FVF values were assessed between each two phases using Bland-Altman analysis.

Results: Significant difference was found between FVF values obtained without and with the reference of enhanced CT images (Table 1, all p values<0.05). And there was no significant difference about FVF values obtained from NCP images with the reference of enhanced CT images between any two lobes or among three lobes (RPL:15.15 (11.57, 20.49), RAL:15.76 (10.91, 20.62), LLL: 14.94 (10.99, 20.57)). The FVF value increased after the contrast injection. FVF value under PVP was highest, while under NCP was lowest (Table 2), and there were significant differences in the FVF values among three scanning phases (Fig.1). Poor consistencies of FVF values between each two phases were found in each lobe by Bland-Altman analysis.

Discussion: In our study, we found that FVF values obtained with the reference of enhanced CT images were significantly higher than those obtained without the reference of enhanced CT images, which indicated that visible vessels should be avoided during ROI measurement. However, abdomen CT scanning without contrast was most frequently used for the liver fat evaluation of NAFLD patients in clinical. Therefore, it was difficult to avoid vessels, especially in patients with mild or moderate liver fat, just according to non-contrast images, which may be the potential limitation of MMD algorithm when we apply it to liver fat quantification.   Besides, with the reference of enhanced CT images, no significant difference between the FVF values obtained from NCP images of any two lobes or among all three lobes were found. Therefore, with the reference of enhanced CT images, the location of ROI might have no influence on the measurement of liver FVF. Although some previous studies resulted that contrast agent had no impact on FVF result, the sample size in each study was very small. In theory, virtual unenhanced attenuation would be under-estimated when using rsDECT, which resulted in over-estimated FVF values. In our study, the FVF values after contrast injection were indeed significantly higher than those on unenhanced images, which was in lined with the theory.

Conclusion: MMD algorithm quantifying hepatic fat might be influenced by obvious vascular and iodine contrast.