Eric Barnhill1, Jing Guo2, Florian Dittmann2, Sebastian Hirsch2, Michael Perrins3, Lucy Hiscox3, Tim Herrmann4, Johannes Bernarding4, Neil Roberts3, Jürgen Braun1, and Ingolf Sack2
1Institute of Medical Informatics, Charité Universitätsmedizin Berlin, Berlin, Germany, 2Department of Radiology, Charité Universitätsmedizin Berlin, Berlin, Germany, 3Clinical Research Imaging Centre, The University of Edinburgh, Edinburgh, United Kingdom, 4Institute for Biometrics and Medical Informatics, Otto von Guericke University Magdeburg, Magdeburg, Germany
Synopsis
We investigated the
impact of field strength and image resolution on brain MRE stiffness
results.
A cohort of 18
healthy volunteer subjects was scanned at 1.5T (2mm isotropic voxels), 3T (2mm) and 7T (1mm), with a fourth set downsampling the 7T to 2mm.
Means were 1634 Pa (+/-613) for 1.5T, 1743
Pa (+/-811) for 3T, 1786 (+/-634) Pa for 7T 2mm, and 927 (+/-364) Pa for 7T 1mm. In the paired
sign-rank tests, there were no significant effects for field strength. Examination of
histograms of example slices suggests that a different distribution
of features is being captured at the higher resolution.Introduction
Recent work has
produced high-resolution Magnetic Resonance Elastography (MRE)
viscoelastic parameter maps, including brain MRE at different field
strengths, vibration frequencies and image resolutions [1-5]. This
variability in methods caused considerable variation of stiffness
estimates ranging from 1.08 kPa (at 40,50,60 Hz) [4] to 3.7 (at 45
Hz) [3] for in vivo white matter. Since brain tissue features spatial
heterogeneities at multiple scales, an improved image resolution, as
is attainable at ultrahigh fields, would add detail within broader
shear wavelengths. We hypothesize that fine features increase wave
curvature which is expected to lower wavelength-based stiffness
estimates.
Purpose
Here we investigated
the impact of field strength and image resolution on brain MRE
stiffness results obtained by multifrequency dual elasto visco (MDEV)
inversion [4].
Methods
A cohort of 18
healthy volunteer subjects was scanned at 1.5T, 3T and 7T using the
protocol described in [4]. We used three drive frequencies of 30, 40
and 50 Hz. The 1.5T and 3T scans had isotropic voxel resolution of
2mm (20 contiguous transversal slices). The 7T scan had isotropic
voxel resolution of 1mm (42 contiguous transversal slices). To
investigate the impact of feature fineness apart from field strength,
a fourth data set was evaluated in which the 7T set was downsampled
to 2mm isotropic voxels using the Matlab function 'imresize' (MathWorks, Natick, MA, USA).
Images were phase
unwrapped using the Laplacian based technique in [7] and denoised
using divergence-free wavelets [8] which were thresholded with SureShrink using
median absolute deviation noise estimates, and complex dualtree
wavelets [9] thresholded with Overlapping Group Sparsity [10] using a
Principal Component Analysis-based blind noise estimation technique
[11].
For measurements,
the cortex was thresholded by measuring the mean and standard
deviation (SD) of a pure noise region in each image, then masking all
values below two SD's above the mean noise level. Mean and SD of each
masked volume were measured. Example slices and histograms of these images were also examined. Each of the four cohorts was tested for
significance of field strength effect using the Wilcoxon signed rank
test.
Results
Plots of mean and SD
for each subject at each field strength are shown in Figure 1. Means
across the cohort for 1.5T, 3T and 7T_2mm resampling were 1634 Pa
(+/-613), 1743 Pa (+/-811) and 1786 (+/-634) Pa respectively, while
mean for the original 7T 1mm resolution was 927 (+/-364) Pa. In the
paired sign-rank tests, there were no significant effects for field
strength between any of the three 2mm acquisitions (p values of 0.13
for 1.5T vs 3T, 0.84 for 1.5T vs. 7T_2mm, and 0.38 for 3T vs. 7T_2mm)
while the impact of 1mm vs. 2mm voxel edge lengths was significant
with p < 0.001 in all cases.
An example slice from
each field strength for a single subject chosen at random is in
Figure 2. Below it are the histograms for the masked brain region in
each case. It can be seen that while the three 2mm voxel edge length histograms are roughly symmetrical, the 7T 1mm histogram is
qualitatively different with heavy right skew.
Discussion
In agreement to
previous work, our study shows no significant impact of field
strength on stiffness estimation at identical voxel size [6]. The
impact of a 1mm isotropic resolution on stiffness values, however, is
highly significant. Examination of histograms of example slices
suggests that a different distribution of features is being captured
at the higher resolution, supporting the hypothesis that the
additional fine viscoelastic features detected at the higher field
strength create more tortuosity in the wave, lowering a
wavelength-based stiffness estimate.
Conclusion
In conclusion, the
MDEV inversion protocol appears robust to field strength, with lower
elastogram values in higher-resolution elastograms likely a function
of finer detail in the image creating more curvature in the shear
waves.
Acknowledgements
The authors gratefully acknowledge image acquisitions from the University of Magdeburg.References
[1] Zhang J, Green MA,
Sinkus R, Bilston LE. Viscoelastic properties of human cerebellum
using magnetic resonance elastography. J Biomech
2011;44(10):1909-1913.
[2] Johnson CL,
McGarry MD, Gharibans AA, Weaver JB, Paulsen KD, Wang H, Olivero WC,
Sutton BP, Georgiadis JG. Local mechanical properties of white matter
structures in the human brain. NeuroImage 2013;79:145-152.
[3] Clayton EH, Genin
GM, Bayly PV. Transmission, attenuation and reflection of shear waves
in the human brain. J R Soc Interface 2012;9(76):2899-2910.
[4] Braun J, Guo J,
Lützkendorf R, Stadler J, Papazoglou S, Hirsch S, Sack I, Bernarding
J. High-resolution mechanical imaging of the human brain by
three-dimensional multifrequency magnetic resonance elastography at
7T. NeuroImage 2014;90:308-314.
[5] Murphy MC, Huston
J, 3rd, Jack CR, Jr., Glaser K, Senjem ML, Chen J, Manduca A, Felmlee
J, Ehman RL. Measuring the Characteristic Topography of Brain
Stiffness with Magnetic Resonance Elastography. PLoS One 2013; 8(12):
e81668.
[6] Hamhaber U, Klatt
D, Papazoglou S, Hollmann M, Stadler J, Sack I, Bernarding J, Braun
J. In vivo magnetic resonance elastography of human brain at 7 T and
1.5 T. J Magn Reson Imaging 2010;32(3):577-583.
[7] Barnhill, E.,
Kennedy, P., Johnson, C. L., Mada, M., & Roberts, N. (2015).
Real-time 4D phase unwrapping applied to magnetic resonance
elastography.Magnetic Resonance in Medicine, 73(6), 2321-2331.
[8] Ong, F., Uecker,
M., Tariq, U., Hsiao, A., Alley, M. T., Vasanawala, S. S., &
Lustig, M. (2015). Robust 4D flow denoising using divergence-free
wavelet transform. Magnetic Resonance in Medicine, 73(2), 828-842.
[9] Liu, X., Tanaka,
M., & Okutomi, M. (2013). Single-image noise level estimation for
blind denoising. Image Processing, IEEE Transactions on, 22(12),
5226-5237.
[10] Selesnick, I.
W., Baraniuk, R. G., & Kingsbury, N. G. (2005). The dual-tree
complex wavelet transform. Signal Processing Magazine,
IEEE, 22(6), 123-151.
[11] Chen, P. Y., &
Selesnick, I. W. (2014). Group-sparse signal denoising: non-convex
regularization, convex optimization. Signal Processing, IEEE
Transactions on, 62(13), 3464-3478.