Background

Chemical-shift-based proton density fat fraction (CS-PDFF) maps have become a staple for evaluation of liver fat content given their ability to survey the entire organ, thus allowing for not only characterization of overall fat content, but also the recognition of any localized increases in fat fraction. However, along with these benefits, CS-PDFF also presents challenges related to how to optimize ROI-based assessments. This article summarizes the considerations that our facility faced when performing PDFF assessments while using a 3D Gradient Echo Fat-Fraction Quantification technique. Data was acquired on a 3T Canon Vantage Galan MR system (Canon Medical Systems, Japan) and collected from a cohort of 51 subjects who were free from liver disease or complications (26 men, age 47±14 years, BMI 27±6 kg/m²) .

Teaching Points

ROI style

Common approaches when measuring PDFF i­nclude drawing ROIs of small area (~2cm²), large area (~4cm²), and best fit per Couinaud segment, as well tracing the footprint of the entire liver or target lobe (“contour”). The topic of optimal ROI size as it relates to accuracy, reproducibility, and workflow is still being examined in our community ^1,2. In our local cohort, there was minimal difference when comparing measurements from multiple small or large ROIs (Figure 1), assuming that these ROIs were placed by the same skilled reviewer and in similar hepatic segments.

However, optimal ROI style must take into consideration staff experience and ability to avoid confounding factors such as:

1. Replication
2. Unfolding errors
3. Partial volume averaging
4. Noise

Since these factors affect pixel intensity, their inclusion in ROIs can impact reported PDFF. To judge the severity of the risk, we triggered each factor on a volunteer and targeted measurements to the affected areas using various ROI styles. ROIs ≥ ~4cm² appeared relatively insensitive to localized artifacts while simultaneously providing robust correlation with the subject’s mean PDFF (Figure 2).

Note that Water-fat “swaps”, while not listed among the above factors, will completely invalidate (i.e.: not simply “invert”) measurements due to failed modeling of the lipid peak. As such, measurement of swapped tissue must be avoided at all costs. Luckily, swaps in the liver are rare and were unobserved in our cohort.

Impact of ROI location

Measurements from the most caudal and cranial sections of the liver may overestimate PDFF^1,2. It has been suggested that this is a result of partial volume averaging² or inadequate shimming (especially in the dome, which neighbors the lung-fields)¹. However, when we assessed the dome on a volunteer after no shimming, bad shimming, and optimal shimming, no significant difference was seen in resulting PDFF (Figure 3). Of additional note, while elevated PDFF in cranial slices was substantiated by our local cohort, there was <1% difference when compared with measurements from the right lobe (Figure 3).

Documented differences between the right (RLL) and left liver lobes (LLL) ^{3, 4} were also supported by analysis of our cohort (Figure 4). We hypothesized that these differences might be exaggerated due to intra-voxel dephasing caused by cardiac motion or dielectric effect. To test the first hypothesis, we compared LLL measurements between inspiration and expiration, but found that the additional space that inspiration places between heart and liver did not lead to increased PDFF values (Figure 4). In a second experiment, we compared PDFF measurements acquired in both supine and decubitus positions. Since dielectric effect manifests based on tissue geometry, we expected to see variations in measurements if this phenomenon were in effect; surprisingly, average measurements from the decubitus position did increase, suggesting that the impact of dielectric effect on LLL measurements should be further explored.

Capturing Heterogeneity

Hepatic steatosis is typically diffuse in nature, but heterogenous manifestations are not uncommon⁴, with an incidence of approximately 14% noted in our cohort (Figure 5). These cases present a challenge, as ROIs that are carelessly placed, or too few in number, may underestimate maximum PDFF. Since disease modeling has shown that heterogenic manifestations of steatosis correlate with progression to severe NAFLD⁵, recognizing patterns of spatially elevated fat content may be critical in reversing early-stage disease. Additionally, heterogeneity is important during longitudinal assessment, in which alterations of PDFF are often noted in some segments but not others⁶.

Heterogeneity is not always obvious to all staff on standard output images and preliminary experiments at our facility showed that higher PDFF values were captured when referencing color maps (Figure 5). This is in-line with color-map experiments applied to standard signal intensity techniques in the past⁷.

Conclusion

CS-PDFF is an indispensable method of assessing steatosis. However, each site must make informed decisions regarding how to characterize PDFF within their clinical population. Taken out of context, measurements targeted to certain areas of the liver may misestimate average PDFF. Confounding factors can also impact measurements, especially with small ROIs.

Experiments at our facility suggest an ROI ≥ 4cm² placed near the center of the RLL is useful for characterizing average hepatic PDFF, without excessive influence from confounding factors, and without introducing measurements from regions prone to aberrant values. ROI’s of ~13cm² were specifically closest to known mean and had the lowest standard deviation. If characterization of maximum PDFF is considered important, referencing color maps may provide additional utility.

Acknowledgements

The authors would like to acknowledge Yurian Falls, CMSC, for his invaluable help with the figures in this abstract.