Sen Jia^{1}, Lei Zhang^{1}, Xiaoqing Hu^{1}, Yiu-cho Chung^{1}, Jing Cheng^{1}, Xin Liu^{1}, Hairong Zheng^{1}, and Dong Liang^{1}

^{1}Shenzhen Institutes of Advanced Technology, Shenzhen, People's Republic of China

### Synopsis

Three dimensional whole brain
and neck vessel wall imaging of high resolution facilities the imaging of
intracranial and extracranial arteries in stroke patients. However, its
clinical usage is hindered by long scan time. Using a 32-channel head and neck coil
system specially designed for high SNR, this work investigated the feasibility
of highly accelerated parallel imaging equipped with optimally selected uniform
subsampling pattern based on 3D G-factor calculated from separate
calibration data and joint sparsity based denoising at acceleration factor of 6.

### Purpose

High
resolution three dimensional imaging of arterial wall with whole brain and neck
coverage allows the examination of intracranial and extracranial arteries
simultaneously to detect atherosclerotic plaques in stroke patients ^{1}.
However, simultaneous large spatial coverage and high resolution leads to
impractical long scan time. Reduced sampling such as Parallel Imaging and
Compressed Sensing can shorten the scan time, but typically with the maximum acceleration
factor less than 4 due to the limitation of signal-to-noise ratio (SNR). Based
on a 32-channel head and neck coil specially designed for high SNR ^{2},
this work investigated the feasibility of highly accelerated parallel imaging
with optimally selected 2D uniform subsampling and joint sparsity based
denoising to shorten scan time.### Method

A successful parallel imaging relies on adopted receiver coil,
subsampling pattern and reconstruction algorithm. Recently, a 32-channel
receiver coil aimed at promoting high SNR was proposed specifically for whole
brain and neck vessel wall imaging ^{2}. To maximally utilize the spatial
encoding power of this coil, the optimal subsampling pattern given an
acceleration factor should be determined. Firstly, this work proposed to select
the optimal subsampling pattern from all available GRAPPA and CAIPIRINHA
patterns ^{4 }based on the 3D G-factor maps calculated using separate ACS data ^{3}. 6X acceleration with all subsampling patterns were simulated and
reconstructed by GRAPPA. Noise behavior in these reconstructions were compared
to verify the selection of optimal subsampling pattern. Secondly, the Conjugate
Gradient based L1-SPIRiT and GRAPPA algorithms were employed on 6X
prospectively accelerated data with the selected optimal trajectory to
demonstrate the necessity of joint sparsity prior on suppressing amplified
noise ^{5}. Specifically, the 3D recon was decomposed into multiple 2D
reconstructions along readout direction for parallel computing ^{6}. Thirdly, the L1SPIRiT
reconstructions on that 6X accelerated data was compared with 4X accelerated GRAPPA
reconstructions to exhibit the feasibility of highly accelerated parallel
Imaging of whole brain and neck vessel wall.### Experiment

The
IRB approved study was performed on a 3T Siemens Tim Trio MRI system. T1w-SPACE
with flip down pulse ^{1} was used. Firstly, a separate calibration data with
size of 64x24x24 (RO x PE x PAR) was acquired for calculating 3D G-factor maps with
a fixed GRAPPA kernel size of 5x4x4. The optimal selection was verified on a
fully sampled data covering whole brain and neck acquired from one volunteer at
an isotropic resolution of 0.62 mm in 22 min. Imaging parameters were: TE/TR =
12/800ms, ETL=41, matrix size = 320x320x208, FOV = 200x200x130 mm3. Secondly,
the optimal subsampling pattern for 6X and 4X accelerations were implemented into
the T1w-SPACE sequence. Accelerated data was prospectively acquired from
another volunteer with imaging parameters: TE/TR = 13/1000ms, matrix size =
320x298x240, FOV = 212x189x158 mm3, and isotropic resolution = 0.66
mm.### Results

The selection of optimal subsampling patterns using 3D G-factor maps
from separate ACS data before imaging was illustrated in Figure 1. For the coil
set we used and 6X acceleration, the subsampling patterns that gave the highest
and lowest mean G-factor were SP4 and SP5 as shown in Fig 1(a, b). The G-factor
of commonly used parallel imaging sampling patterns SP10 was also illustrated.
Fig 1(c) demonstrated that the noise behaviors in images reconstructed from
simulated subsampled data were consistent with the pre-calculated G-factor
maps.
Figure 2(a) compared GRAPPA and L1SPIRiT reconstructions of 6X accelerated data using the optimal subsampling pattern. No residual aliasing
artifacts were observed in both reconstructions. Amplified noise in GRAPPA
reconstructions were significantly suppressed by L1SPIRiT, and benefited the
depiction of vessel walls. L1SPIRiT reconstructions of 6X accelerated data were
also compared with GRAPPA reconstructions of 4X accelerated data. We can see comparable image
quality in the extracranial arteries due to high SNR from the neck coil, while signal
degradation in the intracranial arteries due to moderate SNR from the head coil. Moreover, the noise behaviors were similar in 4X GRAPPA
reconstructions and 6X L1SPIRiT reconstructions.### Discussion and Conclusion

This study demonstrated selecting an optimal subsampling pattern for
given coil set before scanning based on 3D G-factor maps calculated from separate
calibration data. In vivo results showed that joint-sparse denoising could
suppress noise amplified by parallel imaging efficiently and benefited the depiction
of vessel walls at a high acceleration factor of 6. Further work will improve
the coil design for better parallel imaging performance. Also, 3D parallel
imaging and 3D sparse priors can be employed to replace current 2D
reconstruction.### Acknowledgements

Grant support: China NSFC 61471350, the
Natural Science Foundation of Guangdong 2015A020214019, the Basic Research
Program of Shenzhen JCYJ20140610151856736.### References

[1] Zhang L, Zhang N, Wu J, Zhang L, Huang Y, Liu X, and Chung Y. High
resolution three dimensional intracranial arterial wall imaging at 3T using T1
weighted SPACE. Magnetic Resonance Imaging, 2015;
33(9):1026-1034.

[2] Hu X, Li Y, Zhang L, Zhang
X, Liu X, and Chung Y. A 32-channel coil system for MR vessel wall imaging of
intracranial and extracranial arteries at 3T. Magnetic Resonance Imaging, 2016;
in press.

[3] Breuer FA, Kannengiesser
SAR, Blaimer M, Seiberlich N, Jakob PM, and Griswold MA. General Formulation
for Quantitative G-factor Calculation in GRAPPA Reconstruction. Magnetic Resonance
in Medicine, 2009; 62:739-746.

[4] Breuer FA, Blaimer M,
Heidemann RM, Mueller MF, Griswold MA, and Jakob PM. Controlled Aliasing in
Volumetric Parallel Imaging (2D CAIPIRINHA). Magnetic Resonance in Medicine, 2006;
55:549-556.

[5] Lustig M and Pauly JM.
SPIRiT: Iterative Self-consistent Parallel Imaging Reconstruction From Arbitrary
k-Space. Magnetic Resonance in Medicine, 2010; 64:457-471.

[6]
Murphy M, Alley M, Demmel J, Keutzer K, Vasanawala S, and Lustig M. Fast
l1-SPIRiT compressed sensing parallel imaging MRI: scalable parallel
implementation and clinically feasible runtime. IEEE Trans. On Medical Imaging,
2012; 31(6):1250-1262.