Time

Prog #

Practical Considerations for GRAPPA-Accelerated Readout-Segmented EPI in Diffusion-Weighted Imaging

Samantha J. Holdsworth¹, Stefan Skare¹, Rexford D. Newbould¹, Anders Nordell², Roland Bammer¹

¹Stanford University, Palo Alto, California , USA; ²Karolinska University Hospital, Stockholm, Sweden

Readout mosaic segmentation (RS-EPI) has been suggested as an alternative approach to EPI for high resolution diffusion-weighted imaging (DWI) with minimal geometric distortions. In this abstract, peripherally cardiac gated and non-gated RS-EPI-DW images are acquired with the use of parallel imaging. The methods used to phase correct and reconstruct the partial Fourier GRAPPA-accelerated RS-EPI-DW data are described. It is shown that patient handling can be simplified with the use of non-gated acquisitions and minimally-overlapping blinds. The efficient acquisition of high resolution RS-EPI images makes this sampling strategy a useful alternative to other navigated methods used for DW imaging.                

Chunlei Liu¹, Jian Zhang¹, Michael E. Moseley¹

¹Stanford University, Stanford, California , USA

A non-iterative parallel imaging reconstruction algorithm that utilizes k-space sparse matrix (kSPA) was recently introduced for arbitrary sampling patterns. The kSPA algorithm computes a sparse reconstruction matrix in k-space. This algorithm was shown to be particularly useful for a wide range of applications including 3D imaging, functional MRI (fMRI), perfusion-weighted imaging, diffusion tensor imaging (DTI) and massive parallel imaging, where a large number of images have to be reconstructed. The original algorithm requires the acquisition of low-resolution coil sensitivity maps. We present an auto-calibrated kSPA algorithm for arbitrary trajectories that does not require the explicit estimation of coil sensitivities.

Redha Boubertakh¹, ², Philip G. Batchelor¹, Sergio Uribe¹, Thomas S. Sørensen², Michael S. Hansen², Reza S. Razavi¹, Tobias Schaeffter¹

¹King's College London, London, UK; ²University College London, London, UK

We present a new 3D acquisition for whole-heart imaging that combines radial k-space phase encoding in the ky-kz plane and Cartesian readout sampling. Fully sampled data were acquired on a volunteer using a 32-channel cardiac coil. The raw data were undersampled offline with different acceleration factors (R = 8 and 12). The images were reconstructed using gridding and iterative SENSE techniques. When compared to the fully sampled volume, iterative SENSE provides good quality images where artifact levels are strongly reduced compared to gridding. This would lead to a significant decrease in the scan time.

Nicole Seiberlich¹, Philipp Ehses¹, Felix A. Breuer², Martin Blaimer², Peter M. Jakob¹, ², Mark A. Griswold³

¹University of Wuerzburg, Wuerzburg, Germany; ²Research Center Magnetic Resonance Bavaria (MRB), Wuerzburg, Germany; ³University Hospitals of Cleveland, Cleveland, Ohio, USA

It has been shown that the Generalized Sampling Theorem of Papoulis can be exploited to reduce measurement time by acquiring points blipped in the phase encoding direction and applying a conjugate gradient (CG) reconstruction.  Such a blipped trajectory can also be mimicked using a standard trajectory in conjunction with the GRAPPA Operator Gridder (GROG) to shift k-space points; a subsequent CG reconstruction results in an unaliased image even when the Nyquist criterion has not been met in all portions of k-space.  The acceleration of in vivo radial, spiral, and rosette images is demonstrated using GROG to generate random blipped points.

Philip James Beatty¹, Wei Sun², Anja C. S. Brau¹

¹GE Healthcare, Menlo Park, California , USA; ²GE Healthcare, Waukesha, Wisconsin, USA

A method is proposed for improving the computational efficiency of data-driven parallel imaging reconstruction, while maintaining good image quality.  The proposed method forgoes the computationally expensive ‘coil-by-coil’ approach introduced by GRAPPA in favor of directly synthesizing ‘virtual coil’ data.  Results show that the proposed method is able to achieve similar SNR to coil-by-coil approaches and offers similar resiliency to phase cancellation artifacts, while reducing the data synthesis computation by a factor of 20X for 32-channel arrays and over 100X for 128-channel arrays.

Jian Zhang¹, ², Chunlei Liu¹, Michael Moseley¹

¹Stanford University, Stanford, California , USA

A new GRAPPA based iterative Cartesian parallel reconstruction method is proposed which is called Parallel Reconstruction Using Null Operations (PRUNO).  In PRUNO, some local null operators are applied on all k-space locations to formulate the reconstruction problem as linear equations. We also demonstrate that it can be solved efficiently and accurately with a conjugate gradient method. According to our preliminary simulation and in vivo results, PRUNO can be used to improve the accuracy of image reconstruction compared to GRAPPA, especially at high image acceleration rate. Besides, since we usually use merely small local operators in PRUNO, only a small number of ACS lines are required, independent of the exact reduction rate.

Felix A. Breuer¹, Martin Blaimer¹, Nicole Seiberlich², Peter M. Jakob¹, ², Mark A. Griswold³

¹Research Center Magnetic Resonance Bavaria, Würzburg, Germany; ²University of Würzburg, Würzburg, Germany; ³University Hospitals of Cleveland, Cleveland, USA

In this work, equivalent to the g-factor in SENSE reconstructions, a theoretical description for quantitative estimation of the noise enhancement in GRAPPA reconstructions is described. The Grappa g-factor is derived directly from the GRAPPA reconstruction weights. In addition, the procedure presented here allows the calculation of quantitative g-factor maps for both the uncombined and combined accelerated GRAPPA images.

Shaihan J. Malik¹, Jo V. Hajnal¹

¹Hammersmith Hospital, Imperial College London, London, UK

In parallel imaging the 'g-factor' provides a vital prospective measure of noise amplification.  For dynamic undersampled techniques such as k-t SENSE, in addition to typical noise amplification from parallel imaging, errors can arise from temporal filtering due to the prior information (training data).  We define an analogous quantity to g termed g^kt which includes both effects, and investigate its correlation with reconstruction error and its spatiotemporal distribution.  Results from retrospectively undersampled cardiac images and numerical phantoms indicate that g^kt is a useful tool for relative comparison between different undersample strategies given an object and receiver coil setup.

Holger Eggers¹, Peter C. Mazurkewitz¹

¹Philips Research Europe, Hamburg, Germany

Results of previously performed statistical estimations of the noise amplification in non-Cartesian sensitivity encoding imaging with Monte Carlo simulations remain questionable due to their dependence on the number of iterations after which the reconstruction is stopped. In this work, the use of explicit regularization is advocated, and it is demonstrated to stabilize the convergence of the reconstruction and to virtually eliminate this dependence. Calculated maps of the noise amplification are thus rendered more reliable and better comparable between different sampling patterns. Potential advantages of non-Cartesian acquisitions, like more homogeneous and lower maximum noise amplification, are confirmed and substantiated with this approach.

Bo Liu¹, Kevin F. King², Michael C. Steckner³, Lei Ying¹

¹University of Wisconsin-Milwaukee, Milwaukee, Wisconsin, USA; ²GE Healthcare, Waukesha, Wisconsin, USA; ³Toshiba Medical Research Institute USA, Inc., Cleveland, Ohio, USA

The ill-conditioning problem has been addressed by Tikhonov regularization inCartesian SENSE with some success. However, a high-quality regularization image is needed to preserve the details, and otherwise the reconstruction is overlysmooth. In this abstract, we propose a new regularization technique usingBregman iteration. Without any need for regularization images, the methoditeratively refine the total variation (TV) regularization such that theregularized image has more fine details than using TV regularization alone. The proposed method is shown to address the oversmooth problem in Tikhonov regularization and the blocky artifacts in TV regularization.