Melissa Haskell1,2, Stephen Cauley1,3, and Lawrence Wald1,3,4
1Athinoula A. Martinos Center for Biomedical Imaging, MGH/HST, Charlestown, MA, United States, 2Graduate Program in Biophysics, Harvard University, Cambridge, MA, United States, 3Harvard Medical School, Boston, MA, United States, 4Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, United States
Synopsis
We approach the reconstruction of artifact-free images from
an object undergoing unknown rigid-body transformations using a joint
optimization of the final uncorrupted image and motion parameters. To
characterize motion, the joint optimization must estimate 6 additional
parameters for each shot in the image acquisition. We demonstrate an efficient
method for reconstruction from translation-corrupted kspace data by examining
iterative improvements to only a small, targeted subset of imaging voxels. The
method can be enhanced by providing incomplete or noisy information from motion
sensors or navigator measurements. We discuss generalizing our hybrid greedy
and global step non-linear optimization to full rigid-body motion.Purpose
To
improve the robustness of clinical brain MRI with respect to patient motion.
Methods
Numerous
methods to detect and correct for motion in MRI have been explored, including
the use of motion detection markers or devices1, and MR navigators2–4. The use of motion detection
markers in clinical settings has been limited due to the difficulty of robustly
attaching them to patients, and the use of MR navigator scans presents a
challenge due to low sensitivity and disruption of the optimal timing of the MR
pulse sequence into which they are inserted. There have also been attempts to determine
non-rigid motion in a data driven manner through the use of receive coil
locality and auto-focusing motion metrics5. We present TArgeted Motion Estimation and Reduction (TAMER), which retrospectively estimates and corrects for rigid-body
motion by jointly solving for image and motion parameters. TAMER does not
require prior motion information, but the efficiency of the method can be
enhanced with the inclusion of noisy or incomplete navigator or detector
measurements. TAMER uses a targeted application of the underlying parallel
imaging data consistency model for motion estimation and correction through
model reduction and fast optimization techniques.
To demonstrate the benefits of TAMER, we incorporate translation-based
motion into the SENSE6 model. Although our initial
investigation is limited to translations, we hope to generalize the method to
full rigid-body motion. For the 2D RARE(TSE,FSE) clinical T2 images we analyze
here, each shot acquires “ETL” kspace lines. Ideally we need to determine 6 parameters
for each shot, but here we limit our estimation to two in-plane translation parameters. See
Fig. 1 for an overview of the method. We initialize TAMER with a conventional
reconstruction of the corrupted kspace data ignoring motion, i.e. $$$x_0=\textrm{arg}\min_x\sum_{i=1}^{N}||M_0UFC_ix-k_i||_2$$$
. From this corrupted image, pixels of interest (typically ~5% of image) are
selected by identifying high signal regions or by adding small random
motion to highlight motion susceptible pixels. The joint optimization is
restricted to these key image-space areas in order to remove degeneracy in the
solution space and ensure final solution quality. For each iteration of TAMER, a
greedy search sequentially determines an estimate of the motion parameters at
each shot. The solution of this greedy search is used to jumpstart a global
optimization for all motion parameters. Note that at each evaluation in both
the greedy and global search, we only consider data consistency across the small
targeted subset of pixels. The final motion parameters are used to reconstruct
the corruption-free object
$$$x_{final}=\textrm{arg}\min_x\sum_{i=1}^{N}||M_{final}UFC_ix-k_i||_2$$$.
In
order to highlight the performance of our approach, we consider both simulated
data, where we apply TAMER to a motion-free image that has been retrospectively
corrupted by simulated motion, as well as experimental data from an anthropomorphic head phantom translated during acquisition by a motion actuator. In both cases we can compare
directly to a ground truth (motion-free) image. For the first approach, the motion
corrupted kspace data was simulated by adding translation appropriate phase to kspace
shots of a T2-weighted 2D TSE acquisition from a healthy volunteer on a 3T
Siemens Trio with ETL=8, 224x224mm2 FOV, 1x1x3mm3
resolution, 32-channel, TR=6.1s, TE=98ms, refocus angle=150° and R=1,2.
The motion-free image was corrupted using the measured translation parameters taken
from an Alzheimer's disease patient’s fMRI study. TAMER was also tested on an anthropomorphic
phantom using ETL=11, 220x220mm2 FOV, and resolution 0.9x0.9x3mm3.
To create motion corrupted data the phantom was placed on a translational stage
and moved in the A-P direction intermittently throughout the scan. Due to the
uniformity of the brain phantom, 17% of the pixels were targeted for the data
consistency optimization.
Fig.
2 shows the reduction of motion induced ringing artifacts (compared to ground
truth) when TAMER is applied to the volunteer scan, as well as the close
agreement between the estimated motion parameters (solid) and the ground truth simulated
motion (dotted). Fig. 3 highlights the performance of TAMER applied to the
motion corrupted phantom data, also using no prior motion information. The approximate
1D motion path is validated through close agreement of the motion trajectories for
two imaging slices.
Discussion & Conclusions
We
have demonstrated the effectiveness of TAMER for correcting translational
motion in simulated accelerated data and brain phantom data corrupted by a motion
actuator. The method is able to efficiently and accurately estimate motion trajectories
through the use of targeted parallel imaging and fast greedy search. With the extension
to a full 3D approach and modeling terms for rotational motion, TAMER should
facilitate retrospective motion correction in clinical settings.
Acknowledgements
This project was supported by a training grant from the NIH Blueprint for Neuroscience Research (T90DA022759/R90DA023427). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.
Additional funding sources: U01MH093765, R01EB017337, P41EB015896
The authors would like to thank Andre van der Kouwe for the use of his translator, and David Salat and Jean-Philippe Coutu for the Alzheimer's disease fMRI motion trajectories.
References
[1] M. Zaitsev, C. Dold, G. Sakas, J. Hennig,
and O. Speck, “Magnetic resonance imaging of freely moving objects: prospective
real-time motion correction using an external optical motion tracking system,” Neuroimage,
vol. 31, no. 3, pp. 1038–1050, 2006.
[2] J. G. Pipe, “Motion correction with
PROPELLER MRI: application to head motion and free-breathing cardiac imaging.,”
Magn. Reson. Med., vol. 42, no. 5, pp. 963–9, Nov. 1999.
[3] A. Alhamud, M. D. Tisdall, A. T. Hess, K.
M. Hasan, E. M. Meintjes, and A. J. W. Van Der Kouwe, “Volumetric navigators
for real-time motion correction in diffusion tensor imaging,” Magn. Reson.
Med., vol. 68, pp. 1097–1108, 2012.
[4] N. White, C. Roddey, A. Shankaranarayanan,
E. Han, D. Rettmann, J. Santos, J. Kuperman, and A. Dale, “PROMO: Real-time
prospective motion correction in MRI using image-based tracking.,” Magn.
Reson. Med., vol. 63, no. 1, pp. 91–105, Jan. 2010.
[5] J. Y. Cheng, M. T. Alley, C. H.
Cunningham, S. S. Vasanawala, J. M. Pauly, and M. Lustig, “Nonrigid motion
correction in 3D using autofocusing with localized linear translations,” Magn.
Reson. Med., vol. 68, no. 6, pp. 1785–1797, 2012.
[6] K. P. Pruessmann, M. Weiger, M. B.
Scheidegger, and P. Boesiger, “SENSE: sensitivity encoding for fast MRI.,” Magn.
Reson. Med., vol. 42, no. 5, pp. 952–62, Nov. 1999.