Increasing the Spatial Resolution and Sensitivity of High-Resolution Magnetic Resonance Elastography by Correcting for Subject Motion and Susceptibility-Induced Image Distortions

Andreas Fehlner^{1}, Sebastian Hirsch^{1}, Mykola Kadobianskyi^{2}, Patric Birr^{1}, Eric Barnhill^{1,3}, Martin Weygandt^{2,4}, Johannes Bernarding^{5}, Jürgen Braun^{6}, Ingolf Sack^{1}, and Stefan Hetzer^{2,4}

Recent development towards high-resolution multifrequency magnetic resonance elastography (MMRE) enabled the investigation of small areas within in the brain [1]. Echo-planar imaging (EPI) is used to acquire the underlying wave field images enabling retrospective correction of subject motion. However, EPI is susceptible to image distortions caused by local B0 inhomogeneities scaling with image resolution and field strength. Correcting for EPI distortions improves the anatomical localization, thus significantly increasing the statistical power of multi-subject studies [2]. Existing methods for correcting subject motion and image distortions are extracted from and applied to the image magnitude. Therefore the aforementioned correction methods need to be adapted. In this study we developed a novel processing pipeline for MMRE incorporating these correction steps. Furthermore, the increase of spatial specificity of MMRE was analyzed.

Two groups of healthy volunteers were investigated at 7T (N=18) and 3T (N=14). A spin-echo EPI sequence was used to acquire MMRE images at 3 mechanical frequencies with 8 acquisitions over one wave cycle and 3 orthogonal motion encoding gradients resulting in a series of 72 volumes (acquisition time of 10min). For further imaging parameters see Fig.4. For distortion correction two EPI reference volumes (RV) with opposite phase encoding directions (R≫L and L≫R) were acquired with exactly the same parameters as the corresponding MMRE EPI series but without vibration and motion-encoding gradients. Additionally, an MPRAGE scan with 1mm isotropic resolution was acquired for anatomical reference.

Subject motion during the MMRE series was corrected by realigning all magnitude images to the corresponding reference scan (RV). For distortion correction a fieldmap was estimated from the two reference scans (RV), and all MRE volumes were undistorted using FSL5.0-TOPUP [3]. The models behind both correction steps are not applicable to the image phase directly, therefore the complex MRE data was split into real and imaginary parts, which were then corrected separately and re-combined into complex volumes. Finally, the phase of the corrected complex images were processed by an MMRE processing pipeline [4], yielding frequency-averaged maps of the complex shear modulus |*G**| in kPa.

The full width at half maximum (FWHM) of the point-spread function (PSF) of the MMRE process was estimated with FSL-smoothest [5]. The position variability $$PV=2\sqrt{\left(\dfrac{1}{3}\sum\limits_{i}SD(t_{i})\right)^2+\left(\dfrac{1}{3}\sum\limits_{i}SD(r_{i})\right)^2}$$ was calculated from the standard deviations (SD) of the rigid-body realignment parameters $$$(t_{x,y,z},r_{\alpha,\beta,\gamma})$$$ after converting rotational displacements $$$r_{i}$$$ from degrees to millimetres [6], see Fig.1b.

Finally, all images were normalized to the MNI152 template space and to the individual MPRAGE scan, respectively. The corresponding tissue probability maps (TPM) and deformation fields (the non-linear part of the spatial normalisation) were extracted.

Fig.1a shows one subject measured at 7T with above-average motion (PV=1.7mm) before and after motion correction. For all subjects, correction of subject motion significantly reduced the FWHM of the point-spread function by 0.78±0.51mm for the 7T data, and by 0.52±0.63mm for the 3T data. The extent of motion, as quantified by the position variability (PV), averaged over all subjects was 0.85±0.31mm (7T) and 0.77±0.23mm (3T), respectively. A significant linear correlation between the individual PV-values and the FWHM-reduction of the PSF was found for both field strengths, with (r=0.53, p=0.025)@7T and (r=0.69, p=0.006)@3T, see Fig.1c.

Fig.2a demonstrates the effect of distortion correction in areas of strong B0 inhomogeneities in one exemplary subject. In deep slices with strong B0 inhomogeneities, we observed a 6% increase in correlation between the respective tissue masks, and the corresponding standard TPM in MNI space (Fig.3).

Our analysis confirmed that the correction of subject motion significantly sharpened the |*G**| maps, which was demonstrated by a decrease of the width of the PSF. Additionally, we report for the first time, the use of distortion correction for MMRE data and demonstrated that distortion correction enhanced the accuracy of normalization in the MNI space as proven by an increase of the correlation between individual and the standard MNI-TPM.

We interpret the reason for the increased accuracy of the normalization of the distortion-corrected images into MNI space as follows: EPI distortions occur exclusively along the phase-encode direction, whereas the non-rigid registration algorithm implemented by widely-used SPM allows for deformations along all directions. In the presence of large EPI distortions, the normalization algorithm finds a transform with significant deformations along the axes orthogonal to the phase-encoding direction, which ultimately leads to a distortion of the true anatomy. This effect is illustrated in Fig.2c.

The correction methods for MMRE introduced in this work will help to increase the sensitivity of multi-subject studies analysing |*G**| e.g. in small subcortical areas.

[1] 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, 90(1):308-314, 2014.

[2] Cusack R, Brett M, Osswald K. An evaluation of the use of magnetic field maps to undistort echo-planar images. Neuroimage, 18(1):127-142, 2003.

[3] Andersson JLR, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage, 20(2):870-888, 2003.

[4] Hirsch S, Guo J, Reiter R, Papazoglou S, Kroencke T, Braun J, Sack I. MR elastography of the liver and the spleen using a piezoelectric driver, single-shot wave-field acquisition, and multifrequency dual parameter reconstruction. Magn Reson Med, 71(1):267-277, 2014.

[5] Worsley KJ. An unbiased estimator for the roughness of a multivariate Gaussian random field. Technical Report, Department of Mathematics and Statistics, McGill University, 1996. http://www.math.mcgill.ca/keith/smoothness/techrept.pdf

[6] Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. Neuroimage, 59(3):2142-2154, 2012.

Fig.1 a) The high spatial resolution of 1mm
isotropic voxels reveals the significantly sharper point-spread function after
motion correction (FWHM reduced by 1.3mm in the lower images). b) Average
position variability (circle) superposed to the motion trajectory of the 7T
MMRE series. c) Correlation between PV and the FWHM decrease.

Fig.2 a) Example of the significantly increased
spatial accuracy after correcting for distortions in areas with strong B0
inhomogeneities (b). c) Deformation
fields (Dx in phase encoding direction) for spatial normalisation of the MMRE
magnitude image to the individual MPRAGE. After distortion correction the
deformation decrease to noise level.

Fig.3: a) One exemplary slice (#30) with the tissue masks in MNI space
(averaged over all subjects), the corresponding standard TPM and the averaged
B0. The increase of correlation (b) follows significantly the trend of the
averaged B0 (c) with r=0.87, p<0.001.

Fig.4: Imaging
parameters.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)

1945