Peng Lai¹, Michael Lustig^2,3, Shreyas S Vasanawala⁴, and Anja C.S Brau^1
1Global Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States, ²Electrical Engineering, Stanford University, Stanford, CA, United States, ³Electrical Engineering and Computer Science, University of California, Berkeley, CA, United States, ⁴Radiology, Stanford University, Stanford, CA, United States

Compressed sensing (CS) parallel imaging (PI) methods, such as L1SPIRiT, provide better image quality than CS or PI alone, but requires highly intensive iterative computation. Efficient L1SPIRiT (ESPIRiT) greatly reduces the computation intensity based on eigenvector computations. This work provides a theoretical analysis of similarities between these two approaches and demonstrates that they should converge to the same solution. Based on our analysis, we show the existence of multiple dominant eigenvectors for overlapped FOV acquisition, where original ESPIRiT generates significant artifacts like mSENSE and identify a solution. Our results based on invivo datasets showed that the proposed modified ESPIRiT can provide reconstruction very similar to L1SPIRiT regardless of FOV overlap. The modified ESPIRiT algorithm is a robust and computationally efficient solution to CS-PI reconstruction.

Ricardo Otazo¹, Daniel Kim¹, Leon Axel¹, and Daniel K Sodickson^1
1Department of Radiology, NYU School of Medicine, New York, NY, United States

First-pass cardiac perfusion MRI studies can be highly accelerated using a combination of compressed sensing and parallel imaging. However, this method is sensitive to respiratory motion, which decreases sparsity in the combined spatial and temporal-frequency domain and produces temporal blurring in the reconstructed images. In this work, we present a rigid respiratory motion correction method for the combination of compressed sensing and parallel imaging, to highly accelerate first-pass cardiac perfusion MRI without the need of strict breath-holding.

Muhammad Usman¹, Claudia Prieto¹, Tobias Schaeffter¹, and Philip G. Batchelor^1
1Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

Over the last few years, the combination of Compressed sensing (CS) and parallel imaging have been of great interest to accelerate MRI. For dynamic MRI, K-t sparse SENSE (K-t SS) has been proposed for combining the CS based K-t Sparse method with SENSE. Recently, K-t group sparse method (K-t GS) has been shown to outperform K-t Sparse for single coil reconstruction, by exploiting the sparsity and the structure within the sparse representation (x-f space) of dynamic MRI. In this work, we propose to extend K-t GS to parallel imaging acquisition in order to achieve higher acceleration factors by exploiting the spatial sensitive encoding from multiple coils. This approach has been called K-t group Sparse SENSE (K-t GSS). In contrast with the previous single-coil based K-t GS method for which a performance parameter is manually optimized for every frequency encode; we propose an entropy based scheme for automatic selection of this parameter. Results from retrospectively undersampled cardiac gated data show that K-t GSS outperformed K-t sparse SENSE at high acceleration factors (up to 16 fold).

Peng Lai¹, Tao Zhang², Michael Lustig^2,3, Shreyas S Vasanawala⁴, and Anja C.S Brau^1
1Global Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States, ²Electrical Engineering, Stanford University, Stanford, CA, United States, ³Electrical Engineering and Computer Science, University of California, Berkeley, CA, United States, ⁴Radiology, Stanford University, Stanford, CA, United States

Compressed sensing (CS) parallel imaging (PI) is computationally intensive due to its need for iterative reconstruction. Autocalibrating PI can improve the initial solution and largely reduce the number of iterations needed. However, random sampling needed for CS generates a huge number of synthesis patterns making PI initialization extremely slow. Also, uniform density k-space sampling currently used for CS-PI is not optimal in terms of reconstruction accuracy. The purpose of this work was to develop a new tiled-random k-space sampling strategy with the desirable features of 1. incoherent k-space sampling with a small number of synthesis patterns and 2. variable density k-space sampling providing more accurate center k-space reconstruction. Based on our evaluations on 4 invivo datasets, the proposed sampling scheme can improve image quality and reconstruction accuracy compared to conventional sampling schemes and meanwhile enables fast PI initialization for CS-PI.

Claudia Prieto¹, Muhammad Usman¹, Eike Nagel¹, Philip Batchelor¹, and Tobias Schaeffter^1
1Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

K-t Group Sparse (k-t GS) has been recently introduced to achieve high acceleration factors in dynamic-MRI. Kt-GS exploits not just the sparsity of dynamic-MRI but also the spatial group structure of the x-f space. However, it presents two drawbacks: a) an additional training-scan is required for group assignment, and b) the group assignment is based only on the connectivity of neighbouring pixels using a time-consuming hard thresholding scheme. Here we propose to modify k-t GS by using the intensity order, estimated from the same acquired data, for a more robust group assignment. This approach has been tested in cine and perfusion cardiac images with acceleration factors up to 9.

Kyunghyun Sung¹, Anderson N Nnewihe^1,2, Bruce L Daniel¹, and Brian A Hargreaves^1
1Radiology, Stanford University, Stanford, California, United States, ²Bioengineering, Stanford University, Stanford, California, United States

Compressed sensing (CS) is a technique that allows accurate reconstruction of images from a reduced set of acquired data. Here, we present a new method, which efficiently combines CS and parallel imaging (PI) by separating k-space sampling and reconstruction for high- and low-frequency k-space data. This maximally utilizes the wavelet-domain sparsity and avoids possible CS failure in low frequency region. This work has been demonstrated for high-resolution 3D breast imaging and the reconstructed image successfully recovered low-frequency content and fine structures with a net acceleration of 10.8.

Berkin Bilgic¹, Vivek K Goyal¹, and Elfar Adalsteinsson^1,2
1EECS, MIT, Cambridge, MA, United States, ²Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, United States

Clinical MRI routinely relies on multiple acquisitions of the same region of interest with several different contrasts. We present a reconstruction algorithm based on Bayesian compressed sensing to exploit such multi-contrast acquisitions for accelerated imaging by jointly reconstructing a set of related images from undersampled k-space. Our method offers better performance than when the images are either reconstructed individually with the algorithm by Lustig et al., or jointly with a previously proposed method, M-FOCUSS.

Kyunghyun Sung¹, Bruce L Daniel¹, and Brian A Hargreaves^1
1Radiology, Stanford University, Stanford, California, United States

Iterative thresholding methods have been extensively studied as faster alternatives to convex optimization for large-sized problems in compressed sensing (CS). A common large-sized problem is dynamic contrast enhanced (DCE) MRI, and the dynamic measurements possess data redundancies, which can be used to estimate non-zero signal locations. In this work, we present a novel iterative thresholding method called LCAMP (Location Constrained Approximate Message Passing) by adding the non-zero location assumption and an approximate message passing term. The method can reduce computational complexity and improve reconstruction accuracy.

Tobias Wech^1,2, Daniel Stäb¹, André Fischer¹, Dietbert Hahn¹, and Herbert Köstler^1
1Institute of Radiology, University of Wuerzburg, Wuerzburg, Bavaria, Germany, ²Center for Applied Medical Imaging, Siemens Corporate Research, Baltimore, Maryland, United States

Compressed Sensing reconstructions are characterized by a non-linear and non-stationary nature of the dedicated algorithms. Therefore image quality estimation as used for regular Fourier Imaging is not feasible. The aim of this work was to develop a workaround that provides a linear PSF-approximation as well as a validity-test to control its quality. The workflow was tested on the example of a sparse temporal difference image of the human heart and showed a positive result for the validity-test.