Ayodeji L Adams1, Itamar Terem2, Allen A Champagne3, Samantha J Holdsworth4, and Jaco Zwanenburg5
1Radiology, University Medical Center Utrecht, Utrecht, Netherlands, 2Radiology, Stanford University, Stanford, CA, United States, 3School of Medicine, Queen’s University, Kingston, ON, Canada, 4Anatomy and Medical Imaging, University of Auckland, Auckland, New Zealand, 5University Medical Center Utrecht, Utrecht, Netherlands
Synopsis
aMRI holds great potential for assessing brain
motion/strain using images acquired from readily-available sequences. However,
registration is needed to extract displacements from the motion-amplified
images, which may limit its accuracy. In this study we separately assessed the
errors induced by registration limitations and by imperfections in the aMRI amplification.
Displacements were extracted from aMRI and DENSE-amplified images using a
common registration algorithm, which were then compared to a ground truth.
Although significant differences were found between DENSE-amplified images and
aMRI, the aMRI-derived displacements were comparable to the ground truth,
strengthening the potential of aMRI for investigating brain motion in disease.
Introduction
Brain tissue exhibits
cardiac-induced pulsatile motion which may be altered in diseases that affect
intracranial pressure1, providing a potential means to better quantify and understand
these diseases. Both DENSE2 and amplified MRI3,4 (aMRI) have the sensitivity to discern the subtle
displacements that are associated with brain tissue motion. aMRI
estimates displacements through a post-processing algorithm applied on
conventional cine-images, which yields a great potential for assessing brain motion
on clinically available scanners with readily available sequences. Furthermore,
brain tissue strain is derivable from tissue displacement fields, which could
provide insights into diseases that affect the (visco)elastic properties of
brain tissue, or the cerebral small vessels which act as a conduit for
cardiac-induced tissue strain. However, a
registration algorithm is needed to extract displacements from the amplified
motion images generated by aMRI, which may limit its accuracy. Moreover,
accurate quantification of strain critically depends on good estimation of the
amplification factor. In this study
we aim to separately assess the errors induced both by registration limitations
and by potential imperfections in the amplification of aMRI. This is done by
comparing both DENSE-amplified images and aMRI-derived brain tissue
displacements to a known ground truth.Methods
Anterior-Posterior
(AP) and Feet-Head (FH) 4D DENSE images were acquired and processed from 8
subjects as previously described5 to extract brain tissue displacements covering the
entire cardiac cycle (0.93x0.93x1 mm3 interpolated resolution). A
high-resolution T1-weighted image (resolution = 0.93x0.93x1 mm3) was
also acquired for each subject. Ground truth displacements (GTD) were created
by selecting a mid-sagittal slice of the T1 image (10-20 mm left/right from the
interhemispheric fissure) and then spatially smoothing the AP and FH DENSE
displacements associated with that mid-sagittal slice with a Gaussian filter
(kernel=21x21 pixels, sigma=3.5). The mid-sagittal T1-weighted slice was then
deformed by each cardiac phase of the GTD to create DENSE-animated images
(Dani-MRI). Phase-based amplification4 was used to 10x amplify the Dani-MRI to create aMRI. As
a reference, DENSE-amplified images (Damp-MRI) were similarly created by
linearly scaling the GTD by a factor of 10 and then using the scaled GTD to
deform the mid-sagittal T1 image. Cardiac-induced brain tissue displacements
were extracted from the Damp-MRI and aMRI through registration with elastix6. Linear regression was then used to investigate the amplification
factor (slope) and agreement (R2) between Damp-MRI derived displacements and the GTD,
and also between aMRI-derived displacements and the GTD. Student’s T-test was
used to assess differences between the Damp-MRI and aMRI regressions. Note that
mismatch between Damp-MRI and GTD selectively reflects errors from the
registration, while aMRI vs. GTD reflects both registration errors and
imperfections of the amplification algorithm used by aMRI.Results
Figure
1 shows example displacement maps and curves for the GTD as well as Damp-MRI
and aMRI-derived displacements. The group mean slopes (at peak displacement within brain
tissue mask) of the Damp-MRI derived displacements versus the GTD were (mean±std)
8.2±0.8 and 8.8±0.2 for AP and FH
displacements, respectively. The group mean slopes for the aMRI-derived
displacements versus the GTD were 5.8±0.3 and 6.2±0.4
for AP and FH displacements, respectively. The aMRI slopes were significantly
lower than the Damp-MRI slopes for both AP (p<0.0005) and FH (p<0.0005)
displacements. The spatial R2 agreement between the GTD, Damp-MRI
and aMRI derived displacements is shown in figure 2, whilst figure 3 shows the
temporal stability over the cardiac cycle of the agreement and amplification
factor. The amplification is noticeably biased, for both Damp-MRI and aMRI. The
AP displacements were least well captured (lowest agreement and amplification),
which may partly be due to the generally lower amplitudes of AP motion compared
to FH motion.Discussion
We
investigated the accuracy of Damp-MRI and aMRI-derived displacements in an
ideal case without image artefacts, using a known amplification factor and known
displacements. Both Damp-MRI and aMRI derived displacements showed generally good
agreement to the GTD. The estimation of the amplification factor (slope) from
Damp-MRI displacements was significantly more accurate than those from aMRI, suggesting
the need for further calibration of the amplification parameters. Poorer
agreement of derived displacements in the later phases of the cardiac cycle was
found for both Damp-MRI and aMRI. As such, these are likely related to
limitations of the registration algorithm and underscores the importance of the
registration algorithm used to extract displacements, and the need for
artefact-free images which could otherwise misguide the registration.
Of note, accurate strain computations would be
compounded by the different amplification factors found in this study for the
AP and FH displacements. Future work is warranted to investigate the accuracy
of brain tissue strain obtained from aMRI derived displacements, and to utilise
these measurements to explore changes to brain tissue with disease.Conclusion
aMRI-derived displacements are comparable to DENSE
in the ideal case, strengthening the potential of aMRI as a means for
investigating brain tissue displacements. However, there was limited success in
extracting the amplification factor from the derived displacements, which is
partly attributable to limitations of the required registration step, and
partly to imperfections of the amplification algorithm itself. Future work is
necessary to investigate to what extent these limitations hinder practical use
of aMRI for studying brain motion in health and disease.Acknowledgements
No acknowledgement found.References
1. Saindane AM,
Qiu D, Oshinski JN, et al. Noninvasive Assessment of Intracranial Pressure
Status in Idiopathic Intracranial Hypertension Using Displacement Encoding with
Stimulated Echoes (DENSE) MRI: A Prospective Patient Study with Contemporaneous
CSF Pressure Correlation. Am
J Neuroradiol.
2018;39(2):311-316. doi:10.3174/ajnr.A5486
2. Adams
AL, Kuijf HJ, Viergever MA, Luijten PR, Zwanenburg JJM. Quantifying cardiac-induced brain tissue expansion using
DENSE. NMR Biomed. 2019;32(2):1-13. doi:10.1002/nbm.4050
3. Holdsworth SJ,
Rahimi MS, Ni WW, Zaharchuk G, Moseley ME. Amplified magnetic resonance imaging
(aMRI). Magn Reson Med. 2016;75(6):2245-2254. doi:10.1002/mrm.26142
4. Terem I, Ni
WW, Goubran M, et al. Revealing sub-voxel motions of brain tissue using
phase-based amplified MRI (aMRI). Magn Reson Med. 2018;80(6):2549-2559. doi:10.1002/mrm.27236
5. Adams
AL, Sloots J-J, Luijten PR, Zwanenburg JJM. SNR analysis of retrospectively gated DENSE at 7T for the
measurement of brain tissue pulsatility. In: Proceedings of the Joint Annual
Meeting ISMRM-ESMRMB. Paris, France; 2018.
6. Metz
CT, Klein S, Schaap M, van Walsum T, Niessen WJ. Nonrigid registration of dynamic medical imaging data using
nD+t B-splines and a groupwise optimization approach. Med Image Anal.
2011;15(2):238-249. doi:10.1016/j.media.2010.10.003