Shanshan Wang1, Xin Liu1, and Hairong Zheng1
1Paul C. Lauterbur Research Center for Biomedical Imaging, SIAT, Chinese Academy of Sciences, Shenzhen, P.R.China, shenzhen, China
Synopsis
Machine
learning, especially deep learning, has shown great potential in accelerating
MR imaging lately. To accelerate MR imaging with deep learning, the sampling
trajectories can be Cartesian or Non-Cartesian subsampling patterns. While the
reconstruction methods can be roughly categorized into end-to-end data-driven
learning reconstruction methods and model based unrolled iterative learning
reconstruction methods. This educational lecture will briefly go through these
methods and provide a starting point for researchers interested in this field.
Introduction
Machine learning
has been very popular. Even in 1994 there were works using neural network for
NMR spectral estimation [1]. Lately, deep learning, as an important branch of
machine learning, has attracted unprecedented public attention, showing great
potential for diverse medical imaging tasks. Specifically, from 2016, deep
learning has been incorporated into the fast MR imaging task, which draws
valuable prior knowledge from big datasets to facilitate accurate MR image
reconstruction from limited measurements [2,3]. In the meantime and thereafter,
there are different works proposed [4-22]. Methods
Let x be the image
to be recovered and y be the corresponding k-space measurements, its
typical reconstruction model can be described as follows
$\arg\min_{x}||y-Ex||_2^2+R(x)$
where E is the
acquisition encoding matrix, which can consist of the sampling trajectory,
Fourier encoding matrix and multi-coil sensitivities for parallel imaging, and
R(x) is the prior knowledge regularization that assists solving the inverse
reconstruction problem. To accelerate MR imaging with deep learning, the
sampling trajectories can be Cartesian or Non-Cartesian subsampling patterns.
The former one includes 1D uniform, 1D random sampling and 2D random sampling
patterns, etc. While for the latter one, there are radial, spiral sampling
schemes and so on so forth. Cartesian sampling is widely used in conventional
MRI and also can be easily incorporated into deep learning based MR imaging
techniques. Non-Cartesian sampling schemes (e.g. radial or spiral sampling)
offer advantages over Cartesian schemes but its reconstruction normally needs
extra regridding layers. For different body parts scanning, it is normally a
balance of flexibility and efficiency of k-space sampling, motion
insensitivity, and the ability to generate images with high spatio-temporal
resolution from limited data.
For deep learning MR
reconstruction methods, based on the framework and its inputs and outputs, reconstruction
methods can be roughly categorized into two types, end-to-end data-driven
learning reconstruction methods and model based unrolled CSMRI learning
reconstruction. The first category learns a mapping between data pairs, the
input of which can be undersampled k-space data or aliased image while the
output is fully-sampled k-space or artifact-free groundtruth images. For the latter
one, these methods unrolled the traditional iterative reconstruction or CS-MRI
methods for solving the above formulation. Both two category methods have their
pros and cons. While data-driven one can assist more accurate MR reconstruction
if the testing data follow the similar distribution as the training data and
the dataset is big, the model driven method has stronger generalization
capability and better theoretical explanations.Summary
Deep learning
methods have shown big potential for the next generation of fast MR image
technique. While most of the methods show encouraging performances and point
out interesting directions, there are also some misconceptions happening from
time to time and ignorance of the data bias and domain shift issues, which care
must be given to, so as to fight for a brighter future. Opportunity always lies
in Challenges. More robust methods with strong theoretical explanations could be
available in the near future if we continue to devote efforts in this
direction. Acknowledgements
Supported by the National Natural Science Foundation of China (61871371, 81830056), Key-Area Research and Development Program of GuangDong Province (2018B010109009), Science and Technology Planning Project of Guangdong Province (2017B020227012), the Basic Research Program of Shenzhen (JCYJ20180507182400762), Youth Innovation Promotion Association Program of Chinese Academy of Sciences (2019351).References
- S.Venkataraman, Z-P
Liang, R.L. Magin, A neural network approach to NMR spectral estimation, ISMRM
1994.
- Wang S, Su Z, Ying L,
Peng X, Zhu S, Liang F, Feng D, Liang D. Accelerating magnetic resonance
imaging via deep learning. IEEE 13th International Symposium on Biomedical
Imaging (ISBI); 2016. p 514-517.
- Yang Y, Sun J, Li H, Xu
Z. Deep ADMM-Net for compressive sensing MRI. Advances in neural information
processing systems; 2016. p 10-18.
- Kinam Kwon Dongchan Kim
HyunWook Park, A parallel MR imaging method using multilayer perceptron,
Medical physics, 2017
- Hammernik K, Klatzer T,
Kobler E, Recht MP, Sodickson DK, Pock T, Knoll F. Learning a variational
network for reconstruction of accelerated MRI data. Magn Reson Med
2018;79(6):3055-3071.
-
Lee D, Yoo J, Tak S, Ye
JC. Deep Residual Learning for Accelerated MRI Using Magnitude and Phase
Networks. IEEE Trans Biomed Eng 2018;65(9):1985-1995.
- Zhu B, Liu JZ, Cauley
SF, Rosen BR, Rosen MS. Image reconstruction by domain-transform manifold
learning. Nature 2018;555(7697):487-492.
-
Mardani M, Gong E, Cheng
JY, Vasanawala SS, Zaharchuk G, Xing L, Pauly JM. Deep Generative Adversarial
Neural Networks for Compressive Sensing MRI. IEEE Trans Med Imaging
2019;38(1):167-179.
- Han Y, Sunwoo L, Ye JC.
k-Space Deep Learning for Accelerated MRI. IEEE Trans Med Imaging 2019.
-
Aggarwal HK, Mani MP,
Jacob M. MoDL: Model-Based Deep Learning Architecture for Inverse Problems.
IEEE Trans Med Imaging 2019;38(2):394-405.
- Zhang J, Ghanem B. ISTA-Net: Interpretable optimization-inspired deep
network for image compressive sensing. Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition; 2018. p 1828-1837.
- Schlemper J, Caballero J, Hajnal JV, Price AN, Rueckert D. A Deep
Cascade of Convolutional Neural Networks for Dynamic MR Image Reconstruction.
IEEE Trans Med Imaging 2018;37(2):491-503.
- Qin C, Schlemper J, Caballero J, Price AN, Hajnal JV, Rueckert D.
Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction.
IEEE Trans Med Imaging 2019;38(1):280-290.
- Akcakaya M, Moeller S, Weingartner S, Ugurbil K. Scan-specific robust
artificial-neural-networks for k-space interpolation (RAKI) reconstruction:
Database-free deep learning for fast imaging. Magn Reson Med
2019;81(1):439-453.
- Mardani M, Sun Q, Donoho D, Papyan V, Monajemi H, Vasanawala S, Pauly J.
Neural proximal gradient descent for compressive imaging. Advances in Neural
Information Processing Systems; 2018. p 9573-9583.
- J. A. Fessler, "Optimization Methods for Magnetic Resonance Image
Reconstruction: Key Models and Optimization Algorithms," in IEEE Signal
Processing Magazine, vol. 37, no. 1, pp. 33-40, Jan. 2020.
16)
- F Liu, A Samsonov, L Chen, R Kijowski, L Feng, SANTIS: Samplingâaugmented neural network
with incoherent structure for MR image reconstruction, Magnetic resonance in
medicine, 2019
- Eo T, Jun Y, Kim T, Jang J, Lee HJ, Hwang D. KIKI-net: Cross-domain
convolutional neural networks for reconstructing undersampled magnetic
resonance images. Magn. Reson. Med. [Internet] 2018. doi: 10.1002/mrm.27201
- Yang G, Yu S, Dong H, et al. DAGAN: Deep De-Aliasing Generative
Adversarial Networks for Fast Compressed Sensing MRI Reconstruction. IEEE
Trans. Med. Imaging [Internet] 2018;37:1310–1321. doi:
10.1109/TMI.2017.2785879.
19)
- TH Kim, JP Haldar, Learning-based computational MRI reconstruction
without big data: from linear interpolation and structured low-rank matrices to
recurrent neural networks, - Wavelets and Sparsity XVIII, 2019
20)
- Shanshan Wang, Huitao Cheng, Leslie Ying, Taohui Xiao, Ziwen Ke, Hairong
Zheng and Dong Liang, DeepcomplexMRI: Exploiting deep residual network for fast
parallel MR imaging with complex convolution, Magnetic resonance imaging, 2020,
DOI: 10.1016/j.mri.2020.02.002
21)
- Shanshan Wang, Ziwen Ke, Huitao Cheng, Sen Jia, Leslie Ying, Hairong
Zheng, Dong Liang. DIMENSION: Dynamic MR Imaging with Both K-space and Spatial
Prior Knowledge Obtained via Multi-Supervised Network Training, NMR in
Biomedicine: 2019 , DOI:10.1002/nbm.4131