Muhammad Usman¹, Christoph kolbitsch¹, Ghislain Vaillant¹, David Atkinson², Tobias Schaeffter¹, Philip G. Batchelor¹, and Claudia Prieto^1,3
1King's College London, London, United Kingdom, ²University College London, ³Escuela de Ingenieria, Pontificia Universidad Catolica de Chile, Santiago, Chile

Motion during MRI acquisition can cause inconsistencies in k-space, resulting in strong artefacts in the reconstructed images. Accelerated imaging using compressed sensing (CS) requires motion correction approaches not just to correct the motion related artefacts, but also to retain the sparsity level in the sparse representation, which is one of the requirements of CS reconstruction. Currently, only translational motion correction methods have been combined with CS. In this work, we propose a novel Motion Correction-Compressed Sensing (MC-CS )technique that can correct for any arbitrary non-rigid motion in CS undersample acquisitions. This approach was tested both in simulations and in-vivo data for 2D CINE MRI, and use to reconstruct respiratory motion-free cardiac cycle from free-breathing acquisitions.

Haifeng Wang¹, Dong Liang², King F Kevin³, Gajanan Nagarsekar¹, and Leslie Ying^1
1Department of Electrical Engineering and Computer Science, University of Wisconsin-Milwaukee, Milwaukee, WI, United States, ²Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, Guangdong, China, ³Global Applied Science Laboratory, GE Healthcare, Waukesha, WI, United States

In compressed sensing with Fourier encoding, the low spatial frequency always has to be fully sampled such that the high frequency is insufficiently sampled at high accelerations, leading to serious loss of resolution. In this paper, we propose a novel 3-D acquisition method using hybrid encoding. The method exploits random encoding with a circulant structure along one direction, while keeping conventional Fourier encoding along the other two directions. Both simulation and experimental results demonstrate that the proposed method preserves better resolution than the conventional 3-D Fourier encoding when the same acceleration factor is used.In compressed sensing with Fourier encoding, the low spatial frequency always has to be fully sampled such that the high frequency is insufficiently sampled at high accelerations, leading to serious loss of resolution. In this paper, we propose a novel 3-D acquisition method using hybrid encoding. The method exploits random encoding with a circulant structure along one direction, while keeping conventional Fourier encoding along the other two directions. Both simulation and experimental results demonstrate that the proposed method preserves better resolution than the conventional 3-D Fourier encoding when the same acceleration factor is used.In compressed sensing with Fourier encoding, the low spatial frequency always has to be fully sampled such that the high frequency is insufficiently sampled at high accelerations, leading to serious loss of resolution. In this paper, we propose a novel 3-D acquisition method using hybrid encoding. The method exploits random encoding with a circulant structure along one direction, while keeping conventional Fourier encoding along the other two directions. Both simulation and experimental results demonstrate that the proposed method preserves better resolution than the conventional 3-D Fourier encoding when the same acceleration factor is used.

Mariya Doneva¹, Tim Nielsen¹, and Peter Börnert^1
1Philips Research Europe, Hamburg, Germany

In this work, we present a CS reconstruction based on statistical non-local self-similarity filtering (STAINLeSS). The method provides improved image quality compared to wavelet based CS reconstruction and does not require any parameter adjustments. All the parameters are automatically determined by the noise estimation in the receive channels obtained from a standard noise measurement.

Berkin Bilgic¹, and Elfar Adalsteinsson^1,2
1EECS, Massachusetts Institute of Technology, Cambridge, MA, United States, ²Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, United States

In clinical MRI, it is routine to acquire images with different contrasts for increased diagnostic power. Yet depending on the imaging sequences, acquiring certain contrasts is relatively faster. Here, a Bayesian compressed sensing (CS) algorithm that uses a fully-sampled image as prior information to help reconstruct images from undersampled k-space is presented. This method substantially improves the reconstruction quality, and allows joint reconstruction of multi-contrast images.

Anthony G. Christodoulou^1,2, S. Derin Babacan², and Zhi-Pei Liang^1,2
1Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ²Beckman Institute of Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States

This work highlights a novel dynamic imaging method which jointly uses partial separability (PS), sparsity, and group sparsity constraints to enable sparse sampling in (k,t)-space. The specific formulation of the group sparsity spatially varies the effective model order of the PS constraint as a form of controlling the balance between the PS and sparsity constraints.

Sajan Goud Lingala¹, and Mathews Jacob^2
1Biomedical Engineering, The University of Iowa, Iowa city, IA, United States, ²Electrical and Computer Engineering, The University of Iowa, IA, United States

In this work, we introduce a novel blind compressive sensing frame work for dynamic MRI reconstruction. This models the temporal profile at each voxel as a sparse linear combination of temporal basis functions chosen from a large dictionary, which are also estimated from the data. We show that the model significantly reduces the number of degrees of freedom than what is seen in schemes based on promoting low rank structure of the data. We demonstrate this concept on myocardial perfusion data sets with significant inter-frame motion. Significant improvement in the reconstruction qualities over low rank schemes are observed (eg: better preservation of subtle spatial details, reduced temporal blur and artifacts).

Huisu Yoon¹, and Jong Chul Ye^1
1Bio and Brain engineering, KAIST, Dae-jeon, Korea, Republic of

Compressed sensing (CS) tells us that the perfect reconstruction is possible if the nonzero support in transform domain is sparse and sampling basis are incoherent. By exploiting that dynamic MRI can be sparsified due to the temporal redundancy, we have demonstrated successful application of CS for cardiac imaging. In particular, more accurate prediction using motion estimation/compensation or data-driven optimal temporal sparsifying transforms have proven to be quite effective. However, despite their successes to some extent, there still remain considerable artifacts in edge area when the acceleration factor increases.We propose a non-local motion compensated k-t FOCUSS which generates more accurate prediction images than the existing motion compensated k-t FOCUSS. Non-local motion compensation retrieves similar blocks in another reference frame, not within the processed dynamic frame itself. Non-local motion compensation is shown MMSE optimal and experimental result shows that the proposed algorithm clearly reconstruct the important cardiac structures and improves over k-t FOCUSS.

Kyunghyun Sung¹, Manoj Saranathan¹, Brady Quist^1,2, Shreyas S Vasanawala¹, Bruce L Daniel¹, and Brian A Hargreaves^1
1Radiology, Stanford University, Stanford, CA, United States, ²Electrical Engineering, Stanford University, Stanford, CA, United States

High spatial and temporal resolution is desirable for many dynamic contrast enhanced (DCE) MRI applications. Many view sharing methods have been developed to improve both spatial and temporal resolution, but those methods inherently increase temporal footprints, resulting in temporal blurring. In this work, we show a temporal footprint of the view sharing method can be improved by reconstructing the individual subsampled k-space using a novel CS reconstruction while maintaining excellent image quality in a total of 12 DCE MRI patients.

Ti-chiun Chang¹, Mariappan S. Nadar¹, Jens Guehring², Michael O. Zenge², Kai T. Block³, Peter Speier², Michael S. Hansen⁴, and Edgar Mueller^2
1Siemens Corporate Research, Princeton, NJ, United States, ²Siemens AG, Erlangen, Germany, ³NYU Langone Medical Center, New York, NY, United States,⁴National Institutes of Health, Bethesda, MD, United States

Compressed sensing (CS) theory emerges as a promising approach that can accurately reconstruct a signal f even when its indirect measurement is severely undersampled. In the basic CS framework, the sparsity is the only prior knowledge exploited. In practice, measurement imperfections and noise are present, so the data reduction factor promised by CS is clearly reduced. To achieve better results, effort has been devoted to incorporating more prior estimates. In this work, the reconstruction results for dynamic 2d radial cardiac MRI are improved by using additional prior obtained from combining the interleaved samples in the dynamic image sequence

Li Feng^1,2, Hersh Chandarana¹, Jian Xu³, Kai Tobias Block¹, Daniel Sodickson¹, and Ricardo Otazo^1
1Radiology, New York University School of Medicine, New York, New York, United States, ²Sackler Institute of Graduate Biomedical Sciences, New York University School of Medicine, New York, New York, United States, ³Siemens Medical Solutions, New York, New York, United States

Radial k-space sampling is a good candidate for compressed sensing due to the inherent incoherent aliasing artifacts. For dynamic applications, the incoherence can be further increased by using the golden-angle approach, which completely avoids acquisition of replicate radial lines. Moreover, the golden-angle approach allows for reconstruction of arbitrary time points with arbitrary temporal resolution. In this study, we propose a joint non-Cartesian image-reconstruction technique that combines compressed sensing and parallel imaging for accelerating volumetric dynamic MRI with stack-of-stars golden-angle radial trajectories. The performance of this technique is demonstrated for highly accelerated 3D free breathing liver perfusion imaging.