Jitka Starekova1, Ruiyang Zhao1,2, Timothy J Colgan1, Kevin M Johnson1,2, Jennifer L Rehm3, Scott B Reeder1,2,4,5,6, and Diego Hernando1,2
1Department of Radiology, University of Wisconsin, Madison, Madison, WI, United States, 2Department of Medical Physics, University of Wisconsin, Madison, Madison, WI, United States, 3Department of Pediatrics, University of Wisconsin, Madison, Madison, WI, United States, 4Department of Biomedical Engineering, University of Wisconsin, Madison, Madison, WI, United States, 5Department of Medicine, University of Wisconsin, Madison, Madison, WI, United States, 6Department of Emergency Medicine, University of Wisconsin, Madison, Madison, WI, United States
Synopsis
Chemical shift-encoded (CSE)-MRI enables
accurate and precise quantification of proton density
fat-fraction (PDFF) in the liver. Widely used 3D multi-echo spoiled gradient
echo (SGRE) CSE-MRI requires reliable breath-holding to avoid motion-related
artifacts. This is a major limitation for children, the elderly, and sick patients.
Free-breathing 2D sequential CSE-MRI is motion-robust, however, suffers from
low signal-to-noise-ratio (SNR). To overcome these limitations, we combined variable
flip angle (VFA) 2D acquisitions and nonlocal means (NLM) motion-corrected
averaging. In this prospective study, free-breathing multi-repetition VFA-NLM demonstrated
high SNR and reduced artifacts compared to the conventional 3D-SGRE, while
preserving accuracy of PDFF quantification.
INTRODUCTION
Chemical shift encoded (CSE)-MRI methods
enable mapping of proton density fat-fraction (PDFF), a biomarker
of triglyceride concentration1 with applications in
the assessment of non-alcoholic fatty liver disease (NAFLD)1. Current
CSE-MRI methods are typically based on 3D multi-echo spoiled gradient echo (SGRE)
acquisitions2. Unfortunately, reliable breath-holding is necessary
to avoid motion-related artifacts. This
is problematic in children and many patients, where
breath-holding may be limited or not possible. 2D sequential CSE-MRI is inherently
motion-robust due to its very short temporal footprint and can be reliably
acquired during free-breathing3. However, the use of low flip angles
with linear phase encoding to avoid T1-related bias, leads to low signal-to-noise-ratio
(SNR) with 2D sequential techniques.
A recently introduced centric
encoding variable flip angle (VFA) acquisition approach improves SNR,
avoiding T1 bias through a novel variable flip angle schedule4. In order to further improve SNR, several repetitions may
be acquired during free-breathing and averaged. However, this averaging will
result in blurring and bias due to the presence of motion between repetitions.
Recently, nonlocal means (NLM) methods have been shown
to enable motion-corrected averaging of multi-repetition 2D CSE acquisitions,
enabling additional SNR gains while minimizing blurring5. However,
it is unknown whether NLM can be synergistically combined with VFA-based 2D CSE-MRI.
Thus, the purpose of this work is to demonstrate the feasibility
of a VFA CSE-MRI strategy, combined with NLM-based motion corrected averaging to
obtain free-breathing, high-quality, high SNR liver PDFF quantification in
children and adults. METHODS
Technique: This study evaluated a combination
of VFA and NLM techniques. A novel 2D sequential CSE-MRI technique with centric
encoding and VFA was recently developed and validated4, which
provides optimized SNR, low T1 bias, and low blurring in liver fat
quantification. NLM-based motion-corrected averaging was applied across multiple
repetitions of the acquired multi-echo complex CSE-MRI data. This process was
performed jointly for all acquired echo times. The algorithm was designed to
find the most similar pixel in each of the different slices and repetitions
based on a neighborhood similarity metric and perform weighted averaging across
repetitions based on this metric.
Subjects: 15 subjects (mean age 23 years, age
range 11-48 years; mean BMI 27.1, range 17.3-35.3) were imaged on a clinical
3T MRI system (MR750, GE Healthcare Waukesha, WI) using a 32-channel
phased-array torso coil. In this prospective, IRB approved study all
participants provided informed assent and parents provided written informed
consent.
MRI Acquisition: Multi-echo 3D and 2D CSE-MRI data
were acquired to obtain PDFF maps. Three subsets of a single 2D acquisition
with 10 repetitions were explored, VFA with a single repetition (VFA), VFA-NLM
with 5 repetitions (VFA-NLM 5 REP) and VFA-NLM with all 10 repetitions (VFA-NLM
10 REP). Detailed acquisition parameters are listed in Table 1.
Reconstruction and Statistical
Analysis: Source
CSE-MRI data with multiple repetitions were processed using an NLM-based motion
corrected averaging algorithm. The resulting echo images were subsequently
processed using a confounder-corrected PDFF mapping algorithm6. All
reconstruction and processing was implemented using MATLAB (MathWorks, Natick,
MA). A region of interest (ROI) was placed in each of the nine liver Couinaud segments
to measure mean and standard deviation of PDFF using OsiriX (OsiriX Lite). For
each subject, ROI measurements were averaged across all segments to obtain an
overall measure of liver PDFF, and a measure of standard deviation. Linear
regression and Bland-Altman analysis were performed to compare PDFF from different
techniques (VFA, VFA-NLM 5 REP, VFA-NLM 10 REP) to the reference (3D CSE-MRI). Further,
in order to assess SNR performance, PDFF standard deviation distribution was
plotted across different techniques and two-sided Student’s t-test was
performed to compare each technique with the reference technique. All
statistical analysis was performed using Python. RESULTS
As shown in Figure 1, free-breathing VFA,
VFA-NLM-5 REP and VFA-NLM-10 REP showed similar liver PDFF measurements, with high
correlation and narrow limits of agreement compared to the 3D CSE-MRI reference.
VFA showed similar SNR to 3D CSE-MRI (p=0.88). Importantly, VFA-NLM was
observed to improve SNR with increasing number of repetitions (p<0.05 for
VFA-NLM 5 REP, p<0.01 VFA-NLM 5 REP) in comparison to 3D CSE-MRI (Figure 2 and 3). In addition to the excellent noise reduction, the
proposed VFA-NLM approach avoided motion artifacts and preserved image
sharpness at tissue and vessel boundaries. Using VFA-NLM 10 REP, a striking reduction
of motion-related
artifacts was
observed (Figure 3 and 4) in comparison to 3D CSE-MRI. Additionally,
VFA-NLM 10 REP demonstrated improved visibility of organ boundaries, intrahepatic
vessels (Figure 4).DISCUSSION AND CONCLUSION
This study reports on a novel free-breathing
PDFF mapping strategy evaluated in 15 subjects. The combination of multi-repetition
VFA with NLM-based motion-corrected averaging enabled free-breathing fat
quantification that is accurate, has high SNR performance and minimal artifacts,
overcoming the limitations of standard
breath-held CSE-MRI methods. VFA-NLM is a promising method for fat
quantification especially in children and may have broad applications in patients
who are incapable of sustaining a prolonged breath-hold.
Acknowledgements
The authors wish to acknowledge
support from the NIH (UL1TR00427, R01-DK117354, R01-DK083380, R01-DK088925,
R01-DK100651, K24-DK102595). Further, we wish to acknowledge GE Healthcare and
Bracco Diagnostics, who provides research support to the University of
Wisconsin. Finally, 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. Caussy C, Reeder
SB, Sirlin CB, Loomba R. Noninvasive, Quantitative Assessment of Liver Fat by
MRI-PDFF as an Endpoint in NASH Trials. Hepatology. Wiley-Blackwell; 2018
Aug;68(2):763–72.
2. Yokoo
T, Bydder M, Hamilton G, Middleton MS, Gamst AC, Wolfson T, et al. Nonalcoholic
fatty liver disease: diagnostic and fat-grading accuracy of low-flip-angle multiecho
gradient-recalled-echo MR imaging at 1.5 T. Radiology. Radiological Society of
North America; 2009 Apr;251(1):67–76.
3. Pooler BD, Hernando D, Ruby JA, Ishii
H, Shimakawa A, Reeder SB. Validation of a motion-robust 2D sequential
technique for quantification of hepatic proton density fat fraction during free
breathing. J Magn Reson Imaging. John Wiley & Sons, Ltd; 2018
Dec;48(6):1578–85.
4. Zhao R, Zhang Y, Wang X, Johnson KM,
Reeder SB, Hernando D. Motion-Robust, High-SNR Fat Quantification using a Variable Flip Angle Approach. 27th ISMRM Annual Meeting 2019; pp. 1–2. https://www.ismrm.org/19/program files/O23.htm.
5. Luo, Huiwen,
Curtis Wiens, Ann Shimakawa, Scott B Reeder, Kevin M Johnson, and Diego
Hernando. “A Novel Fat and Iron Quantification
Technique with Non-Rigid Motion-Corrected Averaging Based on Non-
Local Means,” Proc. Intl. Soc. Mag. Reson. Med. 25
(2017): 1019.
6. Zhang
X, Hou G, Ma J, et al. Denoising MR images using nonlocal means filter with
combined patch and pixel similarity. PLoS One. 2014;9(6):e100240. Published 2014 Jun 16. doi:10.1371/journal.pone.0100240
7. Hernando D, Kellman P, Halder JP, Liang ZP. Robust water/fat
separation in the presence of large field inhomogeneities using a graph cut
algorithm. Magnetic resonance in medicine: official journal of the Society of
Magnetic Resonance in Medicine/Society of Magnetic Resonance in Medicine.2010;
63(1):79-90