Elisa Marchetto1,2, Kevin Murphy2,3, and Daniel Gallichan1,2
1School of Engineering, Cardiff University, Cardiff, United Kingdom, 2Cardiff University Brain Research Centre (CUBRIC), Cardiff University, Cardiff, United Kingdom, 3School of Physics and Astronomy, Cardiff University, Cardiff, United Kingdom
Synopsis
We investigated the artifacts arising from
different types of head motion during brain structural MR imaging and how well
these artifacts can be compensated for using retrospective correction based on
two different motion-tracking techniques: FatNavs and Tracoline systems.
High image quality could be recovered
in our slow-motion scenarios using both motion estimation techniques. Masking
the non-rigid part of the neck during FatNav volumes registration led to higher
image quality when large pitch-motion was present. The fast continuous motion
scenario led to more severe image artifacts that could not be fully compensated
by the retrospective motion correction techniques used.
Introduction
In
this study, we compared FatNav1-based and Tracoline
(TCL)2-based retrospective motion-correction techniques: the first uses the
natural sparsity of fat images to apply the GRAPPA3 parallel imaging
technique at exceptionally high acceleration factors, enabling the detection
and correction of motion for high resolution imaging in compliant subjects. TCL
is a 3D tracking system that uses a camera pointed at the face to enable motion-correction
of PET and MRI brain images, without using markers, at high frequency rate.
Our
aim is to understand which motion leads to the worst artifacts and how well
image quality can be restored with different motion-tracking estimates.
Moreover, we wanted to understand how to achieve the best image quality in
different motion scenarios, which is clinically relevant to reduce the need for
rescans4. Methods
MPRAGE
images were acquired from a group of 9 healthy subjects on a Prisma scanner
(Siemens Healthcare, Erlangen, Germany) using a 64 channel RF Coil array for
signal reception. Autocalibration lines (ACS) were acquired once at the
beginning of the scan for the FatNavs GRAPPA reconstruction. TCL data were
calibrated at the end of the acquisition via TracSuite software (v3).
Subjects moved their head to generate
continuous motion (4 or 6 cycles/min)5, small/large stepwise motion
and ‘simulated realistic’ motion, based on given instructions. ‘Simulated
realistic’ motion patterns were generated, based on an existing motion trace in
a non-compliant subject during an fMRI experiment acquired without deliberate
motion. From these, we derived: pitch-wise motion and slow continuous motion
along the diagonals. One MPRAGE scan without deliberate motion was always
acquired as reference.
The image reconstruction was performed
using retroMoCoBox ToolBox6, which uses SPM (Statistical Parametric Mapping)7
to perform 6-dof rigid-body alignment between FatNav volumes to generate motion
estimates. If substantial signal is detected in the non-rigid neck region when
FatNavs are acquired, this can affect the quality of the motion estimates,
which is particularly noticeable using the Siemens 64-channel coil. We
therefore tested whether masking the non-rigid part of the head would improve
the motion parameter estimation and image quality in all our motion scenarios,
expecting it to be particularly beneficial in the case of strong pitch-motion.
The
image quality after the motion-correction was assessed visually and using three
different mathematical metrics for comparison: the Feature Similarity Index
(FSIM)8, pixel intensity-based Gradient Entropy (GE)9 and
the Normalized Gradient Squared (NGS)10. GE and NGS decrease and
increase respectively as the image quality improves, while FSIM requires a
reference image for comparison and its value varies between 0-1, with 1
obtained when the two images compared are identical.Results
In our data, NGS and GE showed an unclear
behaviour in almost all experiments, not matching what visually seems like a
good improvement from the uncorrected to the corrected image. Despite the
visible improvements show in Figure 1A, NGS values of the corrected images are
lower than the uncorrected one. Similarly, the uncorrected image GE
value is lower than motion-corrected ones. For this reason, we based our
evaluations on FSIM metric values, which better matches our visual assessment.
Slow up-right diagonal
motion
as well as nodding motion could be well-corrected by FatNavs and TCL as shown
in Figure 1 and 2 respectively.
Both
FatNavs and TCL improved the image quality in all our stepwise motion
scenarios. FatNav motion-correction was further improved by using the neck-mask
in 8/12 experiments. One example of our discrete motion scenario is reported in
Figure 3.
Based on the FSIM score, 3D-FatNavs
outperforms TCL correction in all experiments of continuous motion, although
motion artifacts are still visible as shown in Figure 4. Discussion
GE and NGS exhibited an unclear behaviour
in almost all the experiments performed for this study: this might be due to their sensitivity to different artifacts.
Both TCL and FatNavs can achieve good
image quality in case of slow changes in the head position, although motion
artifacts were still visible in case of large stepwise motion. However, they struggled to restore good image
quality in the continuous motion case: despite the higher sampling rate (0.4 Hz
vs 30 Hz), the FSIM-based image quality metric even decreased after TCL motion-correction
in some cases. This might be caused by extensive violations of the Nyquist
criterion due to the head rotations involved, which could not be compensated for
by the single-step NUFFT-based11 retrospective reconstruction: iterative methods
for applying the motion-correction may help to reduce some of these residual
artifacts.
In Figure 5 is shown how
the neck-mask is applied during the registration process of the FatNav volumes:
this process was shown to be particularly beneficial in case of strong pitch
rotations.Conclusion
In
conclusion, both methods can achieve good image quality in case of slow changes
in the head position. Image quality cannot be fully recovered in this
retrospective implementation when strong violations of the Nyquist criterion
are present. Further improvements were possible masking the non-rigid part of
the neck during the FatNav volumes registration, especially in case of strong
pitch-wise rotations.Acknowledgements
This
work is partly funded by research support from TracInnovations (Ballerup,
Denmark).References
- Gallichan D, Marques J, Gruetter R, Retrospective correction of
involuntary microscopic head movement using highly accelerated fat image
navigators (3D FatNavs) at 7T. Magnetic Resonance in Medicine 2016; 75(3): 1030-1039.
-
Slipsager J, Ellegaard A,
Glimberg S, et al. Markerless motion tracking and correction for PET, MRI, and
simultaneous PET/MRI. PLoS ONE 2019; 14(4):e0215524.
-
Griswold A, Jakob P, Heidemann R, et al. Generalized
Autocalibrating Partially Parallel Acquisitions (GRAPPA). Magnetic Resonance in
Medicine 2002; 47:1202–1210.
-
Slipsager J, Glimberg S,
Søgaard J, et al. Quantifying
the Financial Savings of Motion Correction in Brain MRI: A Model‐Based Estimate
of the Costs Arising From Patient Head Motion and Potential Savings From
Implementation of Motion Correction, J Magn Reson Imaging 2020; 52:731–738.
-
Frost R, Wighton P, Karahanoğlu F, et al. Markerless
high‐frequency prospective motion correction for neuroanatomical MRI. Magnetic
Resonance in Medicine 2019;82:126–144.
-
RetroMoCoBox Toolbox. https://github.com/dgallichan/retroMoCoBox.
-
SPM-Statistical
Parametric Mapping https://www.fil.ion.ucl.ac.uk/spm/doc/manual.pdf.
-
Wang Z, Bovik A, Sheikh
H, et al. Image Quality Assessment: From Error Visibility to Structural
Similarity. IEEE Transactions on Image Processing. 2004;13(4):600-12.
-
McGee KP. Image Metric-Based Correction (Autocorrection) of Motion
Effects: Analysis of Image Metrics. Journal of Magnetic Resonance Imaging.
2000;11:174 –81.
-
Loktyushin A, Nickisch H,
Pohmann R, et al. Blind retrospective motion correction of MR images: Blind
Retrospective Motion Correction. Magnetic Resonance in Medicine 2013;
70(6):1608-18.
- Fessler JA, Sutton BP. Nonuniform fast fourier
transforms using min-max interpolation. IEEE Transactions on Signal Processing.
2003;51(2):560-74.