Zhifeng Chen1, Yujia Zhou1, Xinyuan Zhang1, Peiwei Yi1, Zhongbiao Xu2, Jian Gong1, Zhenguo Yuan3, Xia Kong4, Yaohui Wang5, Ling Xia6, Wufan Chen1, Yanqiu Feng1, and Feng Liu7
1School of Biomedical Engineering, Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical University, Guangzhou, China, 2Department of Radiotherapy, Cancer Center, Guangdong Provincial People's Hospital & Guangdong Academy of Medical Science, Guangzhou, China, 3Shandong Medical Imaging Research Institute, Shandong University, Jinan, China, 4School of Computer and Information Science, Hubei Engineering University, Wuhan, China, 5Division of Superconducting Magnet Science and Technology, Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing, China, 6Department of Biomedical Engineering, Zhejiang University, Hangzhou, China, 7School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane, Australia
Synopsis
High spatiotemporal DCE-MRI is
a valuable tool in liver disease diagnoses and treatments. Recently, there is a
growing research trend which focuses on the motion-robustness of liver DCE-MRI.
However, current techniques cannot simultaneously solve the motion problem when
pursuing high spatiotemporal resolution. In this work, we propose to combine an
accurate registration technique with dynamic artificial sparsity for high
spatiotemporal resolution DCE-MRI of liver. The experiments indicated that the
proposed framework results in better image quality than iGRASP due to
de-enhanced image registration. Compared to motion-sorting techniques, the
proposed framework generates better temporal resolution.
Target Audience
Radiologist and scientists who
take an interest in real-time abdominal MRI and high spatiotemporal motion-free
DCE-MRIPurpose
DCE-MRI
is an emerging tool for diagnoses and treatments of liver diseases 1.
During liver DCE-MRI scanning, subject involuntary movements including respiratory
and cardiac motion, stomach and bowel peristalsis are inevitable 2, 3.
Recently, there has been a substantial increase of research activity in the
field of motion-robustness liver DCE-MRI 4–9. Previous
studies used golden-angle radial acquisition to reduce motion artifacts, which
attracts a lot of attentions from academia and clinic 4–6. However,
it’s not enough for motion-robustness liver DCE-MRI. For example, iGRASP and
L+S average the motion in all directions, resulting in image blurring of anatomical
details 4, 5. Motion-sorting techniques 7–9 resolve the
motion problem to some extent by motion state separation at the sacrifice of temporal
resolution, which impairs its ability of accurate DCE-MRI
parameter study. The aim of this work is to propose a robust abdominal DCE-MRI technique
which not only alleviates the motion problem, but also provides high spatiotemporal
resolution. Specifically, a robust registration scheme is incorporated into the
dynamic artificial sparsity technique to achieve this goal. Compared to
iGRASP-style methods, the proposed scheme provides motion-free results. Meanwhile,
the imaging results indicate that the proposed technique is more robust, and
preserves temporal resolution better than motion-sorting type techniques.Methods
The proposed framework consists of three parts:
dynamic artificial sparsity reconstruction scheme 6, correlation-weighted
de-enhanced signal evaluation 10 and a registration operator based
on intensity-invariant residual complexity 11. The complete framework
is displayed as Fig.1.
Initially, a dynamic artificial sparsity scheme (i.e.,
k-t ARTS-GROWL) is used to obtain the image reconstruction result 6. In this
study, a sliding-window radial GROWL is employed to further improve the SNR of
the reconstructed image comparing to original k-t ARTS-GROWL, as shown in
Fig.1(a).
Subsequently, the correlation-weighted sparse signal s
recovery procedure is performed as the following:
$$\tilde{s}=argmin\left \| y-Ds \right \|_2^2+\lambda \left \| Ws \right \|_1$$
Where y is
the ideal signal-intensity time courses averaged in certain liver ROI, D
is an over-complete dictionary which contains all the signal-intensity curves in
the image series, W is the correlation factor which is related to the dictionary.
In the registration step, an intensity-invariant residual
complexity with B-splined-based free-form deformation is employed in this work 11.
To be more specific, a deformation field Φ is estimated
primarily using de-enhanced DCE-MRI result in the previous step by MIRT toolbox
12. Then, the accurate deformation field Φ is applied to the
original artificial sparsity result to obtain the motion-free DCE-MRI.Experiments
An in vivo liver DCE-MRI experiment was
performed on a 3.0 T Prisma MR scanner (Siemens AG Medical Solutions, Erlangen,
Germany) using a 20-channel body/spine coil array. A 3D stack-of-stars FLASH pulse
sequence with free-breathing golden-angle radial sampling scheme was employed
for this acquisition. The relevant parameters are listed as follows: FOV=350 × 350
× 240 mm3, TR/TE = 3.6/1.6 ms, number of slices 12, number of readout
points in each spoke 512, oversampling ratio 2, number of spokes 1144, and slice
thickness 5 mm.
In this work, each time frame was constructed by 34 consecutive
spokes in all cases, all computations (dynamic artificial sparsity, 2d or 3d registrations)
were implemented in Matlab (R2014a; the Mathworks, Natick, MA, USA), for
off-line reconstruction on an HP workstation (12 Core/2.10 GHz, 128GB, Intel
Xeon E5-2620 v2 CPU).Data Analysis
In
this study, we came up with the concepts of adjacent difference map (ADM) and
its corresponding relative root-mean-square error (rRMSE) in post-contrast
phase of DCE-MRI, which were calculated in the following to exam the effect of
motion correction.
$$ADM_{i-j}=\left | I_i^{post-contrast}-I_j^{post-contrast} \right |(i\not\equiv j)$$
$$rRMSE_{i-j}=\left \|I_i^{post-contrast}-I_j^{post-contrast} \right \|_F/\left \| I_i^{post-contrast} \right \|_F(i\not\equiv j)$$
here $$$I_i$$$ and $$$I_j$$$ represent the ith and jth
time frame, respectively.Results & Discussion
The results
of liver DCE-MRI from different approaches were shown in Fig.2. The corresponding
ADM results were displayed in Fig.3. It can be clearly seen that the proposed framework
has smaller difference between adjacent frames than that of iGRASP, especially
in ADM17-33. The relative RMSE values in Table 1 also support this observation.
From difference maps, we can see motion correction performs well in the
proposed scheme. The movements between different frames in iGRASP were dramatically
removed using the proposed scheme without temporal resolution loss (Fig. 2(b), Fig.
3). This is important for DCE-MRI parameter estimation.
Because XD-GRASP
type techniques impair the temporal continuity of DCE-MRI data, we do not compare
our results with those approaches. In addition, the proposed framework is very
flexible, and can be easily extended to all DCE-MRI acquisitions, while the
motion-sorting techniques can only be used for some certain type trajectories.
Moreover, the proposed framework will have better performance than XD-GRASP in dealing
with aperiodic motions.Conclusion
The proposed scheme
is a registration-based dynamic artificial sparsity framework for high
spatiotemporal resolution DCE-MRI of liver. Compared to iGRASP-type techniques,
it offers advantages such as high image quality, effective elimination of
motion effect and improvement of the accuracy of perfusion parameters. Compared with
motion-sorting techniques, it offers better temporal resolution, which is useful
in clinical DCE-MRI application.Acknowledgements
This work was
supported by NSFC grant (61801205) and CPSF grant (2018M633073). We thank Dr.
Li Feng from Icahn School of Medicine at Mount Sinai for his constructive
suggestions.
References
[1] Aronhime
S, Calcagno C, Jajamovich GH, et al. DCE-MRI of the liver: effect of linear and
nonlinear conversions on hepatic perfusion quantification and reproducibility. J
Magn Reson Imaging. 2014; 40(1): 90–98.
[2] Wood
ML, Henkelman RM. MR image artifacts from periodic motion. Med Phys. 1985; 12(2):
143–151.
[3] Lin W,
Guo J, Rosen MA, et al. Respiratory Motion-Compensated Radial Dynamic Contrast-Enhanced
(DCE)-MRI of Chest and Abdominal Lesions. Magn Reson Med. 2008; 60(5): 1135–1146.
[4] Feng
L, Grimm R, Block KT, et al. Golden-angle radial sparse parallel MRI:
combination of compressed sensing, parallel imaging, and golden-angle radial
sampling for fast and flexible dynamic volumetric MRI. Magn Reson Med. 2014; 72(3):
707–717.
[5] Otazo
R, Candès E, Sodickson DK. Low-rank plus sparse matrix decomposition for
accelerated dynamic MRI with separation of background and dynamic components.
Magn Reson Med. 2015; 73(3): 1125–1136.
[6] Chen
Z, Kang L, Xia L, et al. Sequential combination of parallel imaging and dynamic
artificial sparsity framework for rapid free-breathing golden-angle radial
dynamic MRI: K-T ARTS-GROWL. Med Phys. 2018; 45(1): 202–213.
[7] Feng L, Axel L, Chandarana H, et al. XD-GRASP:
Golden-angle radial MRI with reconstruction of extra motion-state dimensions
using compressed sensing. Magn Reson Med. 2016; 75(2): 775–788.
[8] Chen
Z, Kang L, Xu Z, et al. MR ARTS-GROWL: A Non-Iterative Motion-Resistant
Technique for High Spatiotemporal Liver DCE Imaging. ISMRM, Honolulu, USA, 2017.
p. 3208.
[9] Feng L, Huang C,
Shanbhogue K, et al. RACER-GRASP: Respiratory-weighted, aortic contrast
enhancement-guided and coil-unstreaking golden-angle radial sparse MRI. Magn
Reson Med. 2018; 80(1): 77–89.
[10] Zhou
Y, Sun Y, Yang W, et al. Correlation-weighted sparse representation for robust
liver DCE-MRI decomposition registration. IEEE T Med
Imaging. 2019; doi: 10.1109/TMI.2019.2906493.
[11] Myronenko
A, Song X. Intensity-based image registration by minimizing residual complexity.
IEEE T Med Imaging. 2010; 29(11): 1882–1891.
[12] https://sites.google.com/site/myronenko/research/mirt.