Haifeng Wang1, Yuchou Chang2, Xin Liu1, Hairong Zheng1, and Dong Liang1
1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2University of Houston-Downtown, Houston, TX, United States
Synopsis
Smoothed
Random-like Trajectory
(SRT) is a promising MR imaging method, but the data acquisitions have
some hardware limitations of the gradient amplitude and slew rate. In
order to realize the SRT
in k-space, we proposed a new multiple-leaf SRT in k-space to
reduce hardware requirements for applying the Compressed Sensing (CS) theory. To
guarantee the constrains of the gradient amplitude and slew rate and reduce readout, the proposed multiple-leaf gradient waveforms were optimized by the time-optimal
method for arbitrary k-space trajectories. The simulations have showed that the
proposed method could greatly improve the reconstruction image quality, comparing to spiral trajectories.
Introduction
Recently, many research results
1-11 have found that it could improve reconstructions to add random
perturbed factors into the conventional sampling trajectories.
Their results have showed that using randomly perturbed k-space trajectories
enables more sparsely sampled image reconstruction with higher quality and
fewer artifacts compared to using non-randomly sampled trajectories in CS MRI,
because of satisfying the RIP condition 12-13. The sampling
trajectories, such as, Radial and Spiral, can directly be reconstructed by the
CS framework 14-15. However, random trajectories sampling in
frequency domain is impractical for MRI hardware. Previously,
single-leaf random-like trajectory (SRT) has been proposed to better apply the
CS framework to reconstruct images than before 8. To obey the hardware limitations and reduce readout for acquisition
feasibility, we propose a scheme to generate a multiple-leaf SRT to satisfy the
RIP condition 12-13. Firstly, the random-like trajectory per
leaf is created based on the high order Chirp (HOC) sequences 8,16.
Secondly, the Traveling Salesman Problem (TSP) solver 17-18 is applied
to choose a “short” trajectory per leaf. Thirdly, we use a fast algorithm to
design a time optimal gradient waveforms to obey the gradient amplitude and
slew rate limitations per leaf 19. Fourthly, non-uniform fast Fourier
transformation (NUFFT) 20 and nonlinear conjugate gradient (CG) 14-15
are used together by all leaves to reconstruct under the CS framework.
The simulation results show the proposed multiple-leaf SRT can
reduce artifacts than the conventional Spiral under the CS framework.Theory
The CS theory allows sparse or compressible
signals to be sampled at a rate that is close to their intrinsic information
rate and well below their Nyquist rate, and still allows the signal to be recovered
exactly from randomly under-sampled frequency measurements by a non-linear
procedure12-15. And
the HOC sequences are that a family of discrete sequences have the special
“random-like” uniformly decaying auto-correlation properties 16. The
HOC sequences can satisfy RIP property 12-13 under some conditions 16. A sequence of numbers from the HOC sequences has been defined
as,
$$ u(t)=\frac{t}{\beta}\cdot e^{i\cdot2\alpha\pi t^{3}} , (1) $$
where, β is the decay coefficient;α is a random coefficient; t increases from 0 to β. Based on the sequence numbers from Eq.(1), one leaf of multiple-leaf
SRT is generated by,
$$ s(t)=|u(t)|\cdot e^{i\cdot \theta(u(t))\cdot{2\pi} /L} , (2) $$
where, |.| is the magnitude values
of complex numbers; θ(.) is the angle function of complex numbers; L is the leaf number. Based
on Eq.(1), the multiple-leaf SRT can be generated by several leaves from Eq.(2). But now, this trajectory cannot satisfy the maximum gradient and slew
rate limitations, to be implemented by the pulse sequences on MRI scanners. The
proposed multiple-leaf SRT is illustrated as seen as Figure 1. Here, the TSP
solver with the Simulated Annealing (SA) algorithm 17,18 has been applied
to traverse the sampling points. And the algorithm based on
optimal control theory 8,19 is applied to compute the time-optimal
gradient waveform of the paths after the TSP traverse. The sampling points of the proposed multiple-leaf SRT are one type of non-Cartesian sampling, which can be reconstructed
by NUFFT 20
and nonlinear CG 14-15 algorithms
under the CS framework.Methods
The two in vivo human brain data were acquired on a 3T uMR 790 system (United Imaging Healthcare,
Shanghai, China) by the fast spin
echo sequences (FOV: 230×200mm; TE/TR: 93.4/3800s; Resolution: 0.9mm; Thickness:
5mm) with a commercial 36-channel head coil. Both of the data sizes were 256×256. The 4736 and 4735 points were sampled on the
Spiral and proposed trajectories, as seen as Figure 2. The total 4736 and 4735
points of Spiral (Figure 2a) and the proposed
multiple-leaf SRT (Figure 2b) have been sampled
respectively. The proposed multiple-leaf SRT has 5
leaves; the one-leaf gradient magnitudes have the maximum gradients of 0.55 Gauss/cm;
the one-leaf slew-rate magnitudes have the maximum value of 15 Gauss/cm/ms.
All reconstruction calculations were running on the
MATLAB 2016a (MathWorks Inc., Natick, Massachusetts, USA). Here, the
workstation platform has CPU of AMD Athlom(tm) 7750 Dual-Core (2.7G Hz); RAM of
8 GB; OS of Microsoft Windows 7 Professional 64-bit. Results
Figure 3 and 4 are comparing
the reference and images reconstructed by the CS framework with L-1
minimization, total variance (TV) terms and sparse transformation
14-15 between the Spiral and
proposed trajectories from Figure 2. The
results illustrated that the proposed method could improve the image quality, comparing
to the Spiral method. Especially, the proposed method could greatly remove the wrap-around
and other types of artifacts, in comparison to existing
Spiral sampling strategies. The aliasing
artifacts of the proposed method were close to some random noise. The
proposed multiple-leaf strategy could effectively
reduce readout length for actual acquisition feasibility.Conclusion
In sum,
we propose a novel trajectory scheme for reconstructions under the CS
framework. The current results show the proposed trajectory outperforms the
conventional Spiral trajectory in reducing aliasing artifacts. Based on the current results, the proposed
method is able to effectively enhance
reconstruction performance in the application context of CS recon framework and
reduce readout for feasibility. In the future, more actual experiments on the
scanners will be done to validate the proposed multiple-leaf SRT method.Acknowledgements
This work was partially supported by the
National Natural Science Foundation of China (61871373, 81729003, and 61471350),
the Natural Science Foundation of Guangdong Province (2018A0303130132), the
Strategic Priority Research Program of Chinese Academy of Sciences
(XDB25000000), the Shenzhen Peacock Plan Team Program (KQTD20180413181834876), the
Shenzhen Key Laboratory of Ultrasound Imaging and Therapy
(ZDSYS20180206180631473) and the Sanming Project of Medicine in Shenzhen (SZSM201812005).References
-
Lustig M, et al. Faster imaging with randomly
perturbed undersampled spirals and L1 reconstruction. Proc of the ISMRM. Miami,
FL USA, May 2005.
- Chang T, et al. MR image reconstruction from sparse radial
samples using Bregman iteration. Proc of the ISMRM, Seattle, WA, USA, May 2006.
- Bilgin A, et al. Randomly perturbed radial
trajectories for compressed sensing MRI. Proc of the ISMRM, Seattle, Toronto,
Canada, May 2008.
- El-Metwally S, et al. Rapid cardiac MRI
using random Radial trajectories. Proc of the NRSC, Tanta, Egypt, Mar. 2008.
- Curtis A, et al. Random
volumetric MRI trajectories via genetic algorithms. Int. J. of Biomed. Imaging,
2008(2008): 297089, 2008.
- Seeger M, et al. Optimization of
k-space trajectories for compressed sensing by Bayesian experimental design.
Magn Reson Med. 63(1): 116-126, 2010.
- Willett R. Sampling trajectories for sparse image recovery. Proc. of
IEEE ISBI, Chicago, IL, USA, Jun. 2011.
- Wang H, et al. Smoothed random-like
trajectory for compressed sensing MRI. Proc. of IEEE Eng Med Biol Soc. 2012: 404-407,
2012.
- Chauffert N, et al. Variable density sampling with
continuous trajectories. SIAM J Imaging Sci. 7(4): 1962–1992, 2014.
- Lazarus C, et al. SPARKLING: variable‐density k‐space filling curves for
accelerated T2*-weighted MRI. Magn Reson Med. 81(6): 3643-3661, 2019.
- Weiss T, et al. PILOT: Physics-Informed Learned Optimal
Trajectories for Accelerated MRI. arXiv, 1909: 05773, 2019.
- Candès EJ, et al. Robust
uncertainty principles: Exact signal reconstruction from highly incomplete
frequency information. IEEE Trans Inform Theory. 52: 489-509, 2006.
- Donoho D. Compressed sensing. IEEE
Trans Inform Theory. 52: 1289-1306, 2006.
- Lustig M, et al. Sparse
MRI: The application of compressed sensing for rapid MR imaging. Magn Reson
Med. 58: 1182-1195, Dec. 2007.
- Lustig M, et al. Rapid
MR imaging with compressed sensing and randomly undersampled 3DFT trajectories.
Proc of the ISMRM, Seattle, WA, USA, May 2006.
- Saligrama V. Deterministic designs
with deterministic guarantees: Toeplitz compressed sensing matrices, sequence
design and system. arXiv, 0806: 4958, 2008.
- Kirkpatrick S, et al.
Optimization by Simulated Annealing. Science, 220(4598): 671-680, 1983.
- Černý V. Thermodynamical approach
to the traveling salesman problem: An efficient simulation algorithm. J Optim Theory Appl. 45: 41-51, 1985.
- Lustig M, et al. A fast
method for designing time-optimal gradient waveforms for arbitary k-space
trajectories. IEEE Trans Med Image. 27(6): 866-873, 2008.
- Fessler J. On NUFFT-basesd
gridding for non-Cartesian MRI. J Magn Reson, 188: 191-195, 2007.