Husnain Javid Bhatti1,2, Fariha Aamir1, Ibtisam Aslam1,3, Khan Afsar1,4, and Hammad Omer1
1Department of Electrical and Computer Engineering, COMSATS University Islamabad (CUI), Islamabad, Pakistan, 2School of Electrical Engineering and Computers Science, National University of Science and Technology (NUST), Islamabad, Pakistan, 3Department of Radiology and Medical Informatics, Hospital University of Geneva, GENEVA, Switzerland, 4Department of Electronic science and Technology, Xiamen University, Xiamen, China
Synopsis
To reduce MRI scan
time, under-sampled non-Cartesian trajectories are used which lead to
artifacts. This work proposes a new method ‘GROG with calibration-less pMRI
for CS based p-thresholding’ to reconstruct MR
images from the under-sampled radial k-space
data. The proposed method is validated on the
phantom and 1.5T human head data and provides significant improvement both visually and in terms of
quantifying parameters (AP, RMSE & PSNR) e.g. 77% and 86% improvement in AP, 5% and 32%
improvement in RMSE, 7% and 11% improvement in PSNR at AF=4 for the phantom data than POCS and pseudo
Cartesian GRAPPA, respectively.
Introduction
Magnetic Resonance Imaging (MRI) is a non-invasive
imaging technique offering challenge of long scan time1. Non-Cartesian trajectories help to reduce the scan time but require
an extra step called gridding2. In
the recent past, GROG has been proposed to shift the non-Cartesian data points
to the adjacent Cartesian positions via coil-by-coil weight sets2. However, it leaves some unfilled
spaces in the gridded data which leads to incoherent artifacts. Several algorithms have been proposed to overcome these artifacts3,4,5; CS also being a good candidate. CS requires a small number of
samples in the acquired k-space, incoherent artifacts and an appropriate
non-linear reconstruction algorithm to efficiently reconstruct the artifact
free MR image from the under-sampled k-space
data6. This paper presents a new method ‘GROG with calibration-less pMRI
for CS based p-thresholding’ to
reconstruct the artifact free MR images from the radially encoded under-sampled
k-space data.
Method
Gridding is a way to map the acquired non-Cartesian k-space data points onto Cartesian k-space2. GROG transfers the non-Cartesian samples to the nearby Cartesian locations
utilizing coil-by-coil self-calibrated weight sets leaving some unfilled
positions in the gridded data2,7.
Pseudo-Cartesian GRAPPA has been conventionally used for reconstructing
the GROG gridded data to get the solution images using multiple coil-by-coil
weight estimating patterns4. However, it is difficult to determine the correct choice and right number
of the weight estimating patterns.
In the recent past, Projection onto Convex
Set (POCS) algorithm was proposed in CS to recover the un-aliased MR images iteratively from the 2D variable density Cartesian under-sampled data3. p-thresholding is
an extension of the iterative soft thresholding algorithm (ISTA) which has been
used in literature to minimize the non-convex functions in CS9. The use of p-thresholding
function promotes sparsity in the gridded k-space which is a key factor for CS based
image reconstruction.
This paper proposes
a new technique to reconstruct the fully sampled image from the highly
under-sampled radial k-space data. Firstly,
GROG transfers the acquired radial k-space
points onto Cartesian k-space by
using self-calibrated coil-by-coil weight sets. The receiver coil sensitivity
maps are estimated from the center of the GROG gridded data using Eigenvalue
maps estimation approach8. The receiver coil sensitivity
profiles are incorporated with the CS based p-thresholding using
adaptive coil combination method9 to recover the solution image. Figure-1 shows a block diagram of the proposed method. Mathematically,
the optimization equation of the proposed method is:
mins ||Fu.C.m - yG||22 + λ||Xp(ψm)||1 (Eq. 1)
where,
(Xp(ψm))k = sign(ψmk).max{0, |ψm|k - λ|ψm|kp-1}
In
eq.1, Fu is the under-sampled Fourier transform operator, C depicts the receiver coil sensitivity profiles, yG is the
GROG gridded data (yG = G.YR; G is GROG operator and YR is the
radially acquired k-space data), m
is the reconstructed image, λ is the thresholding parameter,
ψ
is taken as a sparsifying transform (Wavelet) and Xp
is the p-thresholding function.
In this work, the values of p (i.e.
p = -0.05) and λ (i.e. λ = 0.3×10-3) are empirically chosen after performing
experiments for a range of values; and the correlation of the central line
profiles of the reconstructed images between the current and previous
iterations is used as a stopping criterion2,10.Results and Discussion
The proposed
method is tested on the Shepp–Logan phantom and 1.5T human head data having
dimensions 256×256×8. Initially, all the data sets are retrospectively encoded
into radial projections using Fessler toolbox with the help of the following
formula: (π/2)×FOV5.
Figure
2 and 3 show the reconstruction results of the proposed scheme (Row A), projection onto
convex set (POCS) (Row B) and GROG
followed by pseudo Cartesian GRAPPA (Row C) for Shepp–Logan phantom data and 1.5T human head data sets at AF 4, 6, and 9 with 101, 67 and
45 radial projections (Proj), respectively.
Table 1 and 2 show the AP, RMSE and PSNR values at
AF =4, 6 and 9 of the Shepp–Logan phantom and 1.5T human head data reconstructed images .The proposed method gives better
results both visually and in terms of quantifying parameters for Shepp–Logan phantom
data (e.g.
77% and 86% improvement in AP, 5% and 32% improvement in RMSE , 7% and 11%
improvement in PSNR at AF=4) and 1.5T human head data (e.g. 37% and 85%
improvement in AP, 20% and 62% improvement in RMSE, 2% and 11% improvement in
PSNR at AF=4) as compared
to POCS and pseudo Cartesian GRAPPA, respectively.Conclusion
In this paper, GROG
with calibration-less pMRI for CS based p-thresholding is
proposed to reconstruct the un-aliased MR images from the highly under-sampled
radial k-space data. The proposed method efficiently
eliminates the need of separate ACS lines and pseudo Cartesian GRAPPA patterns like conventional methods. The
proposed work provides better results both visually and in terms quantifying
parameters than POCS and pseudo Cartesian
GRAPPA.Acknowledgements
No acknowledgement found.References
-
D. W. McRobbie, E. A. Moore et.al. MRI from picture to proton. 2006.
- N. Seiberlich, F. A. Breuer et.al. “Non-Cartesian data
reconstruction using GRAPPA operator gridding (GROG),” MRM. 2007;58(6):1257–1265.
- J. Shah, I. Qureshi, H. Omer et.al. “A modified POCS-based reconstruction method for compressively sampled
MR imaging,” Int. J. Imaging Syst. Technol. 2014;24(3):203–207.
- N. Seiberlich, F. Breuer et.al. “Reconstruction of undersampled
non-Cartesian data sets using pseudo-Cartesian GRAPPA in conjunction with
GROG,” MRM.2008;59(5):1127–1137
- I. Aslam, F. Najeeb, and H. Omer,
“Accelerating MRI Using GROG Gridding Followed by ESPIRiT for Non-Cartesian
Trajectories,” APMR, 2017.
- M. Lustig, D. Donoho, and J. M. Pauly, “Sparse
MRI: The application of compressed sensing for rapid MR imaging,” MRM. 2007;58(6):1182–1195.
- N. Seiberlich, F. Breuer, M. Blaimer, P.
Jakob, and M. Griswold, “Self-calibrating GRAPPA operator gridding for radial
and spiral trajectories,” MRM. 2008;59(4):930–935.
- M. Uecker et al., “ESPIRiT - An
eigenvalue approach to autocalibrating parallel MRI: Where SENSE meets GRAPPA,” MRM.2014;71(3):990–1001.
- M. A. Griswold et al.,
“Autocalibrated coil sensitivity estimation for parallel imaging,” NMR
Biomed. 2006;19(3):316–324.
10. J. X. Ji, J. B. Son, and S. D. Rane,
“PULSAR: A Matlab toolbox for parallel magnetic resonance imaging using array
coils and multiple channel receivers,” Concepts Magn. Reson. Part B Magn.
Reson. Eng. 2007;31B(1):24–36.