Shanshan shan1, Mingyan Li1, Haiwei Chen1, Feng Liu1, and Stuart Crozier1
1The University of Queensland, Brisbane, Australia
Synopsis
Radial-encoding sequences can be more
sensitive to gradient trajectory deviations caused by gradient nonlinearity
(GNL) than Cartesian encoding. In this
work, we developed a method to alleviate image distortion caused by GNL
specifically for radial-encoding.
Experimental results acquired from a 9.4 T pre-clinical MRI scanner with
using Ultra-short TE sequence were used to validate the proposed method.
Introduction
Due
to design and construction limitations, gradient nonlinearity (GNL) could
deviate actual k-space trajectories from nominal ones, undermining
orthogonality of Fourier encoding and producing images with blurring, ghosting
artifacts and increased noise 1. These image distortions could be more dramatic
when using radial sequences, which require accurate gradient trajectories for
image reconstruction with regridding 2 or filtered back-projection 3. Various methods have been developed to correct
GNL-induced image distortions. The field camera 4 measures actual field spatiotemporal profile used
for image reconstruction and correction 5. Non-uniform Fourier transform (NUFFT) is
employed to realise integrated GNL correction during image reconstruction 6. Parametric models of gradient waveforms with
multiple functions can be built for trajectory correction 7. In this work, like SENSitivity Encoding
(SENSE) 8, we considered “GNL radial encoding” as a
special encoding superimposed on standard Fourier encoding. The proposed method
was tested on in vivo mice brain
images acquired from a 9.4T pre-clinical MRI scanner with using Ultra-short TE
(UTE) sequence. Primary results demonstrated that the method was capable of
dramatically reducing image distortion caused by GNL. Methods
The forward process of acquiring a
radial k-space $$${b_{radial}}$$$can
be expressed as:$${F_{radial}x=b_{radial}} \tag{1a}$$where$$${F_{radial}}$$$denotes non-distorted radial Fourier encoding
matrix, in which each element is a radial Fourier kernel in the form of$$${e_{k,r}=e^{-2\pi jk_{radial}r}}$$$. $$${k_{radial}}$$$is the radial k-space locations and r is the voxel locations of imaging subject
x.
When the GNL occurs, it resides in the
forward encoding process and the Eq.1a can be modified to:$${E_{GNL}\cdot F_{radial}x=b_{radial}} \tag{1b}$$where$$$ {E_{GNL}}$$$denotes the GNL
encoding in the form of $$${e_{k,r}^{GNL}=e^{-2\pi j{\triangle k_{radial}}r}}$$$. $$${\triangle k_{radial}}$$$ is the radial
trajectory deviation caused by GNL. The image x can be recovered by minimising the following function$${\parallel E_{GNL}\cdot F_{radial}x-b_{radial}\parallel_{2}\tag{2} }$$The experiments were performed on a
9.4T Bruker Biospin MRI scanner (Ettlingen, Germany). A surface coil (Ø: 40 mm)
with 60° open angle and 26 mm length was used to acquire in vivo mouse brain images. The TE and TR of the UTE sequence was
set as 0.4ms and 100ms. Paravision 6
provided radial trajectory locations which were used for the proposed method to
correct image distortion.Results
The k-space trajectories of the UTE
sequence exported from Paravision 6 with and without GNL distortion are shown
in Fig. 1b and Fig. 1d, respectively. As showed in Fig. 1a and c, trajectory
errors occurred at various parts of radial spokes, especially outer region of
the k-space, which potentially result in resolution loss and undesired noise. With
such deviated radial trajectories, images reconstructed from routine algorithms
such as regridding are incapable of removing these artifacts as demonstrated in
Fig. 2a. The blood vessel (arrow 1) which is vague becomes more visible in
Fig.2b after applying the proposed GNL correction method, despite it being
distant to the surface coil. Severe
signal loss (arrow 2) observed in Fig. 2a was well compensated with the proposed
method as indicated by Fig. 2b. The low proton density area marked with red
arrow 3 in Fig. 2a is seriously contaminated with ghost artifacts, which are later
eliminated with the proposed method and the boundary between different tissues is
more distinct. Fig. 3b is the image
reconstructed by the proposed method. Compared to Fig. 3a without GNL
correction, the background noise was substantially reduced and the image
quality was greatly increased. Specifically, the noise (white circular region)
in Fig. 3b was reduced to 0.0011, which is 38% of uncorrected image (0.0029). The
SNR (calculated with NEMA standard 9) of GNL-corrected image has a
2.6-fold increase from 186.7 (Fig. 3a) to 483.9 (Fig. 3b). Details at tissues boundaries are also well preserved. Discussion
The proposed method employs GNL
encoding concept to correct image distortion during image reconstruction
process. Formula 2 was iteratively solved by using Matlab function lsqr. Through
the iterative optimisation process, the proposed method is expected to reduce
more GNL distortions than the regridding method when inaccurate k-space
trajectory information is acquired. The construction of GNL encoding and
Fourier encoding matrix is computationally expensive, which shall be handled
with a more efficient optimization algorithm in the future.Conclusion
An image distortion correction method
based on GNL radial encoding was proposed. This method superimposes the GNL
encoding onto the standard radial encoding, so that the image can be
reconstructed with reduced artifacts through an iterative optimisation process.
In vivo experiments performed on a
9.4 T pre-clinical system showed that the proposed method is capable of effectively
correcting the GNL-induced distortions and thus improve the image quality. In
the future, the proposed algorithm will be enhanced for wider MRI applications
such as spiral imaging.Acknowledgements
No acknowledgement found.References
1. Janke
A, Zhao H, Cowin GJ, Galloway GJ, Doddrell DM. Use of spherical harmonic
deconvolution methods to compensate for nonlinear gradient effects on MRI
images. Magn Reson Med 2004;52:115-122.
2. Block KT, Uecker M, Frahm J.
Undersampled radial MRI with multiple coils. Iterative image reconstruction
using a total variation constraint. Magnetic resonance in medicine
2007;57:1086-1098.
3. Lauterbur PC, Liang Z. Principles of
Magnetic Resonance Imaging: IEEE Press, New York, NY, USA, 2000.
4. Dietrich BE, Brunner DO, Wilm BJ, et
al. A field camera for MR sequence monitoring and system analysis. Magnetic
resonance in medicine 2016;75:1831-1840.
5. Tao S, Trzasko JD, Gunter JL, et al.
Gradient nonlinearity calibration and correction for a compact, asymmetric
magnetic resonance imaging gradient system. Phys Med Biol 2017;62:N18-N31.
6. Tao S, Trzasko JD, Shu Y, et al.
NonCartesian MR image reconstruction with integrated gradient nonlinearity
correction. Medical physics 2015;42:7190-7201.
7. Takizawa M, Hanada H, Oka K,
Takahashi T, Yamamoto E, Fujii M. A robust ultrashort TE (UTE) imaging method
with corrected k-space trajectory by using parametric multiple function model
of gradient waveform. IEEE Trans Med Imaging 2013;32:306-316.
8. Pruessmann KP, Weiger M, Scheidegger
MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magnetic resonance in
medicine 1999;42:952-962.
9. Sano R. NEMA standards: Performance
standards for clinical magnetic resonance systems. RL Dixon 1988:185-189.