Data acquisition is an important step in magnetic resonance imaging (MRI). Acquisition time and image quality heavily depend on how we sample k-space and how we reconstruct images from the acquired data. In other words, data acquisition and image reconstruction are inseparable. Usually, new imaging theory raises the request on novel sampling scheme. The first part of this talk will briefly review traditional data acquisition connected with traditional reconstruction methods. The second part of the talk will discuss non-traditional acquisition motivated by advanced reconstruction ideas such as compressed sensing (CS), low-rank and deep learning based methods.
Highlights
Introduction
Data acquisition is an important step in magnetic resonance imaging (MRI). K-space data contain valuable information on the objective to be scanned and on the system (off-resonance etc.). They are the source where the diagnosis images come from. Acquisition time and image quality heavily depend on how we sampled k-space and how we reconstruct images from the acquired data. In other words, data acquisition and image reconstruction are inseparable. Sometimes, data acquisition motivates the invention of new reconstruction techniques. At other times, new imaging theory requires novel sampling strategy. The first part of this talk will briefly review traditional data acquisition combined with traditional reconstruction methods, such as parallel imaging, partial Fourier. The second part of the talk will discuss non-traditional acquisition motivated by advanced reconstruction ideas such as compressed sensing (CS), low-rank and deep learning based methods.Traditional acquisition
Parallel imaging [1,2]and partial Fourier[3] are two commonly used reconstruction methods in clinics and have been embedded in commercial scanners. Parallel imaging utilizes the coil sensitivity to reduce the acquisition time, and thus many undersampling trajectories can be used. The trajectories are the same for all channels and can be Cartesian or non- Cartesian [4, 5], uniformly undersampled or variable-density undersampled. While partial Fourier utilizes the characteristic of Hermitian symmetry of Fourier transform and thus only half data need to be acquired with additional central lines for estimating the phase information.Non-traditional acquisition
The further need for fast, high-resolution imaging motivates the emergence of advanced imaging method. The new imaging theory requires new strategy on data acquisition. Compressed sensing [6,7] is one successful representative and has attracted much attention in past ten years, which requires incoherent measurements to generate noise-like artifacts. Obviously, traditional uniformly undersampling acquisition cannot meet this requirement. Thus, variable-density random sampling trajectories were proposed and achieve promising performance in compressed sensing MRI [8]. Surprisingly, non-Cartesian undersampling trajectories exhibit good performance in the framework of CS-MRI since they can produce incoherent artifacts even with uniform undersampling scheme [9]. Low-rank MRI [10-13] also shares the same underampling trajectories with CS-MRI, with differing on organizing the k-space data to form a matrix that pose the low rank property. Since the prior used by different reconstruction method are usually complementary, it has to be carefully considered when combining different methods. For example, theoretically combining CS and parallel imaging is straightforward; however, the sampling trajectory has to meet the requirements from both methods. Some works like CS-SENSE [14] and Possion-disk [15]were proposed for addressing this issue. There are also other attempts for combining CS, parallel imaging and partial Fourier. [16] Employing deep learning for accelerating MR scan (DeepLearnMRI) has been a very appealing new research direction since past year[17-21]. It aims to train a network off-line by using large-scaled dataset for identifying the end-to-end nonlinear mapping between input and labeled output, and then performed as a predicator to reconstruct MR images online from undersampled data. The input is usually a linear reconstruction from undersampled k-space data with aliasing artifacts and the output can be the reconstruction from full samples or just the artifacts themselves. No matter removing the artifacts or extract the artifacts, the type of artifacts which heavily depend on the sampling trajectory play a very important role. Therefore, incoherently sampling, which is required for compressed sensing, may not be the optimal undersampling trajectory for DeepLearnMRI. The optimal strategy for data acquisition in the framework of DeepLearnMRI is still an open question.[1] Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med. 1999; 42(5):952-962.
[2] Griswold M, Jakob P, Heidemann R, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med. 2002; 47:1202–1210.
[3] Margosian P, Schmitt F, Purdy D. Faster MR imaging: imaging with half the data. Health Care Instrum. 1986; 1(6): 195.
[4] Pruessmann KP, Weiger M, Bornert P, Boesiger P. Advances in sensitivity encoding with arbitrary k-space trajectories. Magn Reson Med. 2001; 46(4): 638-651
[5] Seiberlich N, Breuer FA, Blaimer M, Barkauskas KJ, Jakob PM, Griswold MA. Non-Cartesian data reconstruction using GRAPPA operator gridding (GROG). Magn Reson Med. 2007; 58(6): 1257-1265. [6] Donoho D. Compressed sensing. IEEE Trans Inf Theory. 2006; 52(4):1289–1306.
[7] Candès EJ, Romberg J, Tao T. Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information. IEEE Trans Inf Theory. 2006; 52(2):489-509.
[8] Lustig M, Donoho DL, Pauly JM. Sparse MRI: The Application of Compressed Sensing for Rapid MR Imaging. Magn Reson Med. 2007; 58(6):1182-1195.
[9] Block KT, Uecker M, Frahm J. Undersampled radial MRI with multiple coils. Iterative image reconstruction using a total variation constraint. Magn Reson Med. 2007; 57(6):1086-1098.
[10] Brinegar C, Schmitter SS, Mistry NN, Johnson GA, Liang Z-P. Improving temporal resolution of pulmonary perfusion imaging in rats using the partially separable functions model. Magn Reson Med. 2010; 64:1162–1170.
[11] Lingala SG, Hu Y, Dibella EV, Jacob M. Accelerated Dynamic MRI Exploiting Sparsity and Low-Rank Structure: k-t SLR. IEEE Trans Med Imag. 2011; 30(5):1042-1054.
[12] Lee D, Jin KH, Kim EY, Park S, Ye JC. Acceleration of MR parameter mapping using annihilating filter-based low rank hankel matrix (ALOHA). Magn Reson Med. 2016; 76(6): 1848-1864.
[13] Haldar JP. Low-Rank Modeling of Local k-Space Neighborhoods (LORAKS) for Constrained MRI. IEEE Trans Med Imag.2014; 33(3): 668-681.
[14] Liang D, Liu B, Wang J, Ying L. Accelerating SENSE using compressed sensing. Magn Reson Med. 2009; 62(6):1574-1584.
[15] Lustig M, Pauly JM. SPIRiT: Iterative self‐consistent parallel imaging reconstruction from arbitrary k‐space. Magn Reson Med. 2010; 64(2):457-471.
[16] Liu F, Duan Y, Peterson BS, Kangarlu A. Compressed sensing MRI combined with SENSE in partial k-space. Physics in Medicine and Biology, 2012; 57(21).
[17] 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; 514-517.
[18] Wang S, Ying L, Peng X, Liang D. Exploiting deep convolution neural network for fast magnetic resonance imaging. Proceedings of the 24th Annual Meeting of ISMRM; 2016; p. 1778.
[19] Wang S, Huang N, Zhao T, Yang Y, Ying L, Liang D. 1D Partial Fourier Parallel MR imaging with deep convolutional neural network. Proceedings of the 25th Annual Meeting of ISMRM. 2017. (Oral Presentation)
[20] Wang S, Xiao T, Tan S, Liu Y, Ying L, Liang D. Undersampling trajectory design for fast MRI with super-resolution convolutional neural network. Proceedings of the 25th Annual Meeting of ISMRM. 2017.
[21] Wang S, Zhao T, Huang N, Tan S, Liu Y, Ying L, Liang D. Feasibility of Multi-contrast MR imaging via deep learning. Proceedings of the 25th Annual Meeting of ISMRM. 2017.