Timothy J Colgan1, Diego Hernando1,2, and Scott B Reeder1,2,3,4,5
1Radiology, University of Wisconsin - Madison, Madison, WI, United States, 2Medical Physics, University of Wisconsin - Madison, Madison, WI, United States, 3Biomedical Engineering, University of Wisconsin - Madison, Madison, WI, United States, 4Medicine, University of Wisconsin - Madison, Madison, WI, United States, 5Emergency Medicine, University of Wisconsin - Madison, Madison, WI, United States
Synopsis
Quantitative
confounder-corrected chemical shift encoded MRI estimates of proton density fat
fraction (PDFF) are corrected for R2* signal decay. However, at very high
levels of iron overload, the observed signal in CSE-MRI decays rapidly (high
R2*), complicating the separation of water and fat signals, and therefore the
quantification of PDFF. This work characterized the degree of iron
overload above which the quantification of PDFF is limited and developed
practical clinical and research guidelines for PDFF estimation protocol design
for 1.5T and 3.0T.
Introduction:
Intracellular
accumulation of triglycerides within hepatocytes is the earliest and hallmark
feature of nonalcoholic fatty liver disease (NAFLD). Iron overload is also common
with chronic liver disease and often coexists with hepatic steatosis in
patients with NAFLD1. Quantitative chemical shift
encoded MRI (CSE-MRI) measures liver fat based on estimates of the proton
density fat fraction (PDFF) corrected for all relevant confounding factors,
including R2* signal decay. However, at very high levels of iron overload, the
observed signal in CSE-MRI decays rapidly (high R2*), complicating the
separation of water and fat signals, and therefore the quantification of PDFF2. Therefore, the purpose of this
work is to characterize the degree of iron overload above which the
quantification of PDFF is limited and to develop practical clinical and
research guidelines for PDFF estimation protocol design for 1.5T and 3.0T.Theory:
As
shown in Figure 1, dramatic differences in the CSE signal evolution and
spectroscopy experiments can be observed depending on the R2* signal decay. At
severely elevated R2* levels the water and fat peaks coalesce, severely
limiting the ability of CSE-MRI to estimate PDFF as seen in the representative
maps in Figure 1 C-F.Methods:
Cramér–Rao lower bound (CRLB) Simulations
CRLB
analysis for different field strengths, fat fractions, and echo time
combinations were used to identify R2* values above which PDFF estimation
becomes unreliable. This R2* threshold is
based on an effective number of signal averages (NSA) that produces a root mean
square error (RMSE) equal to the coefficient of repeatability of 4.7% from a
recent meta-analysis3.
Monte
Carlo Simulations
Monte
Carlo simulations with PDFF values of 0,10, and 20%, and R2* values ranging from
0-1000s-1 were compared to the CRLB. Two commonly used clinically
feasible acquisition protocols were investigated, one with an early first echo
and short echo spacing, and another with a later first echo and longer echo
spacing (Table 1).
In
Vivo Experiments
A
prospective study was performed in patients with known or suspected iron
overload after obtaining approval from our local Institutional Review Board
(IRB), and informed written consent. Subjects were imaged on a clinical 1.5T
MRI system (HDxt, GE Healthcare, Waukesha, WI) using an 8-channel torso coil
(USA Instruments, Cleveland, OH) and a clinical 3.0T MRI system (Discovery
MR750, GE Healthcare, Waukesha, WI) using a 32-channel torso coil (NeoCoil,
Pewaukee, WI). The two acquisition protocols (Table 1) were used with a
quantitative complex confounder-corrected CSE-MRI method4. Further, a Single Voxel
Multi-Echo Stimulated Echo Acquisition Mode (STEAM) MR spectroscopy (MRS) method5 was used as the reference standard
for PDFF.
CSE-MRI
acquisitions were reconstructed offline using a non-linear maximum likelihood
estimator algorithm4. MRS acquisitions were analyzed
through a fully automated processing of the multi-echo STEAM sequence6. Reconstructed PDFF and R2* maps
were analyzed by placing a single region of interest (ROI) in the right lobe of
the liver. The absolute error from each PDFF map was calculated as the
difference between the average value in the ROI and the MRS estimate.Results:
Cramér–Rao lower bound (CRLB) Simulations
The
minimum variance of the CRLB predicts that an NSA of approximately 0.2 produces
a RMSE of the pixel by pixel estimate below the coefficient of repeatability
from Yokoo et al3. Figure 2 summarizes the CRLB
analysis, where the maximum R2* for each $$$TE_1$$$ and $$$\Delta{TE}$$$ pair that leads to an NSA above 0.2 is shown for a range of clinically relevant
protocols.
Based off these CRLB
simulations, expressions for the maximum R2* for a given $$$TE_1$$$, $$$\Delta{TE}$$$ pair where PDFF can be reliably reconstructed
for any fat fraction were determined by fitting a second order polynomial to
all echo time pairs and across fat fraction from 0-30%.
$$$1.5T: Max
R2^*(TE_1,\Delta{TE})=740.5–345.1TE_1+28\Delta{TE}+20.7TE_1^2+94TE_1\Delta{TE}-84.1\Delta{TE}^2$$$
$$$3.0T:
Max R2^*(TE_1,\Delta{TE})=1397–808.8TE_1–382.4\Delta{TE}+132.2TE_1^2+252.6TE1\Delta{TE}-89.1\Delta{TE}^2$$$
For
example, the two protocols used at 1.5 T would have a maximum R2* of 522s-1
and 308s-1 (Protocol 1 and Protocol 2, respectively), above which
PDFF estimates are not reliable. At 3.0T the two protocols would have a max R2*
of 771s-1 and 435s-1 (Protocol 1 and Protocol 2,
respectively).
Monte
Carlo Simulations
The
agreement between CLRB and Monte Carlo simulations is shown in Figure 3. The
choice of echo times leads to a large difference in the variance of PDFF
estimates as a function of R2*.
In
Vivo Experiments
A
total of 26 patients were successfully recruited, with average age of 42 (11-76
range) and 17:9 men:women. There was a wide range of iron overload, with
average R2* of 382s-1 (39-640s-1 range) at 1.5T and many
of these patients had hepatic steatosis with average PDFF of 10% (0.8-29% range).
Figure 4 plots the absolute
error in PDFF as a function of R2* for the different protocols. The acquisition
with a shorter first echo and shorter echo spacing has lower PDFF errors as a
function of R2*.Discussion & Conclusions
In
this work we have investigated and identified R2* thresholds above which PDFF
estimation is unreliable. Practical equations were developed to identify these
thresholds, based on acquisition parameters at both 1.5T and 3T. These results
may provide practical clinical and research guidelines for PDFF estimation in
patients who have concomitant iron overload. Acknowledgements
The
authors wish to acknowledge support from the NIH R01 DK100651, K24 DK102595, R0
DK088925, R41-EB025729, R01DK117354), the Wisconsin Alumni Research Foundation
(WARF) Accelerator Program, as well as GE Healthcare who provides research
support to UW-Madison. Further, Dr. Reeder is a Romnes Faculty Fellow, and has
received an award provided by the University of Wisconsin-Madison Office of the
Vice Chancellor for Research and Graduate Education with funding from the
Wisconsin Alumni Research Foundation.References
1. Britton, L. J., Subramaniam, V. N.
& Crawford, D. H. Iron and non-alcoholic fatty liver disease. World J.
Gastroenterol. 22, 8112–8122 (2016).
2. Hernando, D., Sharma, S. D. &
Reeder, S. B. Limits of liver fat quantification in the presence of severe iron
overload. in Proceedings of the 22rd Annual Meeting of ISMRM (2014).
3. Yokoo, T. et al. Linearity,
Bias, and Precision of Hepatic Proton Density Fat Fraction Measurements by
Using MR Imaging: A Meta-Analysis. Radiology 286, 486–498 (2017).
4. Hernando, D., Kellman, P., Haldar, J.
P. & Liang, Z. P. Robust water/fat separation in the presence of large
field inhomogeneities using a graph cut algorithm. Magn Reson Med 63,
79–90 (2010).
5. Hamilton, G. et al. In vivo
breath-hold (1) H MRS simultaneous estimation of liver proton density fat
fraction, and T1 and T2 of water and fat, with a multi-TR, multi-TE sequence. J
Magn Reson Imaging 42, 1538–43 (2015).
6. Hernando, D., Artz, N. S., Hamilton,
G., Roldan, A. & Reeder, S. B. Fully automated processing of multi-echo
spectroscopy data for liver fat quantification. in Proceedings of the 22nd
Annual Meeting of the International Society of Magnetic Resonance in Medicine
(2014).