Tom Meyer1, Stephan Rodrigo Marticorena Garcia1, Heiko Tzschätzsch1, Helge Herthum2, Mehrgan Shahryari1, Jürgen Braun2, Prateek Kalra3,4, Arunark Kolipaka3,4, and Ingolf Sack1
1Department of Radiology, Charité - Universitätsmedizin Berlin, Berlin, Germany, 2Institute of Medical Informatics, Charité - Universitätsmedizin Berlin, Berlin, Germany, 3Department of Biomedical Engineering, The Ohio State University, Columbus, OH, United States, 4Department of Radiology, The Ohio State University, Columbus, OH, United States
Synopsis
Magnetic resonance elastography (MRE) depicts the viscoelastic properties of soft tissues for diagnosis of various diseases such as tumors or fibrosis. However, different MRE inversion methods yield different results. This lack of comparability of different MRE inversion results hinders the dissemination of advanced reconstruction methods. Therefore, we here introduce an extensible, open-access web platform that offers multiple inversion techniques for multifrequency, three-dimensional MRE to promote comparison of values. We demonstrate the utility of the platform in phantom data and in vivo free-breathing multifrequency MRE data of the kidneys of healthy volunteers.
Introduction
Magnetic resonance elastography (MRE)1 is
an established technique for staging liver fibrosis2
and is currently being developed for other diagnostic applications, including
brain disease3
and tumors4. However,
different MRE methods measure different values in the same tissue, leading to
inconsistent thresholds for viscoelasticity-based detection of disease.
We hypothesize that comparison of values, assessment of data quality,
consistency of MRE results, and, ultimately, standardization of thresholds for
quantification of disease can be critically advanced by providing an open
platform where all types of MRE data can be processed using various established
inversion methods. Therefore, we here introduce such a platform, implemented as
an open access web application. We demonstrate the utility of the proposed
platform in a phantom and in a set of prospectively acquired multifrequency MRE
data of the healthy human kidney.Methods
Phantom Data
For validation, repository data of an
agar-based Wirogel (Bego Inc., Bremen, Germany) phantom9 with four
cylindrical inclusions was used.
Kidney Data
The study was approved by our ethics
committee. 23 healthy volunteers (10 women, mean age 35±11 years, age range: 21
to 53 years) were studied by multifrequency MRE in a 1.5-Tesla MRI scanner
(Magnetom Sonata, Siemens, Erlangen)5, after written informed
consent was obtained. Multifrequency vibrations of 40, 50, 60, and 70-Hz
frequencies were consecutively induced by pressurized-air driven actuators.
Full wavefields were acquired in eleven contiguous paracoronal slices using a single-shot,
spin-echo planar imaging sequence with a 256×256mm² field-of-view and 2.46×2.46×2.5mm³
voxels.
Server Architecture
The MRE processing platform (https://bioqic-apps.charite.de/)
was implemented as a Java web application. The Java application handles
downloads and uploads of imaging data and runs compiled MATLAB scripts for data
processing. Communication with web server was encrypted using HTTPS and data is
automatically deleted after logging off or time-out. The pipelines can be parameterized by default
or optimized data-specific settings. The server
architecture is presented in figure 1.
Inversion methods: MDEV, k-MDEV and LFE
MDEV inversion is an approach to Helmholtz
(‘direct’) inversion, providing the magnitude of the shear modulus (|G*| in kPa)
and ϕ, the loss angle6,7. k-MDEV
analyzes the phase-gradient of plane waves by 2D-first-order derivatives, waves
are decomposed into plane-waves using directional filters8, from
which shear-wave speed (SWS in m/s, related to stiffness) and wave penetration
rate (PR in m/s, related to inverse wave attenuation) are obtained by
inversion.9 LFE exploits six pairs of lognormal filters and |G*| is
obtained from the ratio of shear wave fields filtered with a pair of lognormal
filters10.
Data were processed
after 2D- and 3D-motion correction16,11. Stiffness was converted to
SWS by sqrt(|G*|/1kg/L) assuming unit mass density. Regions of interest (ROIs) of the kidney were manually drawn for
the medulla, inner- and outer cortex based on MRE magnitude images. Group was
performed using one-way ANOVA for repeated measures with Bonferroni correction
and Tukey's test for post-hoc analysis.Results
Phantom analysis
Figure 2 shows ground-truth and results
for phantom data using LFE, MDEV and k-MDEV.
While all methods agreed with ground-truth for matrix and softer inclusions,
there was larger disparity for stiff inclusions. Only k-MDEV detected that inclusion 1 was stiffer than the matrix, however
MDEV and k-MDEV correctly detected inclusion
4.
In vivo kidney analysis
SWS values of the full parenchyma were
uninfluenced by motion correction, corroborating previous findings11,
however given the clear improvement in image sharpness (figure 3), 2D-motion
correction was added as a preprocessing module to all methods. Representative slices of kidney SWS
obtained by LFE, MDEV and k-MDEV are shown
in figure 4. k-MDEV SWS values were
higher than MDEV and LFE values (full-kidney SWS: 2.71±0.19m/s [k-MDEV], 2.14±0.16m/s [MDEV], 2.12±0.15m/s [LFE], each p<0.001). k-MDEV visualized renal substructures in
greater detail than the other two methods, however consistently, all methods
showed the inner renal cortex to be stiffer than medulla and outer cortex
without differences between right and left kidney (all p>0.05, figure 5).Discussion
We here presented an open-access
platform for multifrequency, three-dimensional MRE data offering comparative processing
using multiple inversion methods. Beyond wave inversion with standardized
pipelines of LFE, MDEV and k-MDEV,
the platform incorporates modular preprocessing blocks such as 2D- and
3D-motion correction16,11 and distortion correction12 as
well as an repository of MRE data for download and online processing to which
the community has contributed13,14.
Beyond
presenting novel MRE processing tools, we here presented the first comparative
analysis of renal stiffness encompassing multiple inversion pipelines with
motion-corrected multifrequency wave data. Our results are consistent with
renal stiffness measured by tomoelastography (SWS=2.34±0.15m/s8, 2.40±0.17m/s15,
based on k-MDEV inversion) and
single-frequency MRE (60Hz, 1.82±0.54m/s based on LFE10). Tomoelastography
consistently showed that the inner cortex was stiffer than medulla and outer
cortex (2.92±0.25m/s, 2.41±0.13m/s, 2.16±0.18m/s)15, previous work
based on LFE did not resolve inner and outer cortex10.Conclusion
We addressed the
well-known lack of comparability of MRE results by introducing an expandable,
easy-to-use, open-access platform that offers inversion techniques for
multifrequency, 3D-MRE including LFE, MDEV and k-MDEV inversion. We demonstrated the utility of the platform in
phantom data and in kidneys of healthy subjects. The platform might foster the
development of quantitative imaging biomarkers related to tissue
viscoelasticity by facilitating analysis of effects of filter thresholds,
spatial resolution, motion and frequency dispersion on MRE results.Acknowledgements
Support of the German
Research Foundation (GRK2260 BIOQIC, SFB1340, project
number 467843609) is gratefully
acknowledged.References
- Muthupillai, R. & Ehman, R. L. Magnetic resonance elastography. Nature Med. 2, 601-603 (1996).
- Kennedy, P. et al. Quantitative Elastography Methods in Liver Disease: Current Evidence and Future Directions. Radiology 286, 738-763, doi:10.1148/radiol.2018170601 (2018).
- Hiscox, L. V. et al. Magnetic resonance elastography (MRE) of the human brain: technique, findings and clinical applications. Phys Med Biol 61, R401-R437 (2016).
- Garteiser, P. et al. MR elastography of liver tumours: value of viscoelastic properties for tumour characterisation. Eur Radiol 22, 2169-2177 (2012).
- Lang, S. T. et al. Multiparametric Quantitative MRI for the Detection of IgA Nephropathy Using Tomoelastography, DWI, and BOLD Imaging. Invest Radiol 54, 669-674, doi:10.1097/RLI.0000000000000585 (2019).
- Papazoglou, S., Hirsch, S., Braun, J. & Sack, I. Multifrequency inversion in magnetic resonance elastography. Phys Med Biol 57, 2329-2346 (2012).
- Streitberger, K.-J. et al. High-resolution mechanical imaging of glioblastoma by multifrequency magnetic resonance elastography. PLoS One 9, e110588 (2014).
- Manduca, A., Lake, D. S., Kruse, S. A. & Ehman, R. L. Spatio-temporal directional filtering for improved inversion of MR elastography images. Med Image Anal 7, 465-473 (2003).
- Tzschatzsch, H. et al. Tomoelastography by multifrequency wave number recovery from time-harmonic propagating shear waves. Medical Image Analysis 30, 1-10, doi:10.1016/j.media.2016.01.001 (2016).
- Gandhi, D. et al. Magnetic resonance elastography-derived stiffness of the kidneys and its correlation with water perfusion. Nmr in Biomedicine 33, doi:10.1002/nbm.4237 (2020).
- Shahryari, M. et al. Reduction of breathing artifacts in multifrequency magnetic resonance elastography of the abdomen. Magn Reson Med, doi:10.1002/mrm.28558 (2020).
- Fehlner, A. et al. Increasing the spatial resolution and sensitivity of magnetic resonance elastography by correcting for subject motion and susceptibility-induced image distortions. Journal of Magnetic Resonance Imaging 46, 134-141, doi:10.1002/jmri.25516 (2017).
- Barnhill, E. et al. Fast Robust Dejitter and Interslice Discontinuity Removal in MRI Phase Acquisitions: Application to Magnetic Resonance Elastography. IEEE Trans Med Imaging 38, 1578-1587, doi:10.1109/TMI.2019.2893369 (2019).
- Ariyurek, C., Tasdelen, B., Ider, Y. Z. & Atalar, E. SNR Weighting for Shear Wave Speed Reconstruction in Tomoelastography. NMR Biomed, e4413, doi:10.1002/nbm.4413 (2020).
- Marticorena Garcia, S. R. et al. Tomoelastography Paired With T2* Magnetic Resonance Imaging Detects Lupus Nephritis With Normal Renal Function. Invest Radiol 54, 89-97, doi:10.1097/RLI.0000000000000511 (2019).
- Reddy, B. S. & Chatterji, B. N. An FFT-based technique for translation, rotation, and scale-invariant image registration. Ieee Transactions on Image Processing 5, 1266-1271, doi:10.1109/83.506761 (1996).