Image Reconstruction

Monday 22 April 2013

Anh Tu Van¹, Murat Aksoy¹, Samantha J. Holdsworth¹, Daniel Kopeinigg¹, Sjoerd B. Vos², and Roland Bammer^1
1Radiology, Stanford University, Palo Alto, CA, United States, ²Image Sciences Institute, University Medical Center Utrecht, Utrecht, Utrecht, Netherlands

Three-dimensional high-resolution diffusion imaging is feasible in terms of scan time when multislab acquisition is used. However, the main challenge of multislab acquisitions is the slab boundary artifacts caused by an imperfect slab-selective profile along the slice dimension. By reconstructing all the slab collectively using a SENSE-like method with the slab excitation profiles as the “sensitivity maps,” the slab boundary artifacts can be mitigated.

Maryam Vareth^1,2, Anita Flynn³, Wei Bian^1,2, Ye Li¹, Daniel B. Vigneron^1,2, Sarah J. Nelson^1,2, and Xiaoliang Zhang^1,2
1Radiology and Biomedical Imaging, UC San Francisco, San Francisco, CA, United States, ²UC Berkeley/UCSF Joint Graduate Group in Bioengineering, San Francisco, CA, United States, ³EECS, UC Berkeley, Berkeley, CA, United States

Parallel imaging and compressed sensing for traveling wave MR is achievable with a very simple orthogonal microstrip-resonator antenna geometry. We present experimental results with three known undersampling reconstruction algorithms (GRAPPA, SPIRiT andL1-SPIRiT) at an acceleration factor of 1.8.

Daniel B. Rowe^1,2, Andrew S. Nencka², Andrzej Jesmanowicz², and James S. Hyde^2
1Department of Mathematics, Statistics, and Computer Science, Marquette University, Milwaukee, WI, United States, ²Department of Biophysics, Medical College of Wisconsin, Milwaukee, WI, United States

Two MR image slices are simultaneously encoded and a single complex-valued aliased image is measured. From the single complex-valued image, the previously published magnitude-only and the current complex-valued image separation methods are applied. It is shown that the magnitude-only method can have challenges when the difference in phase of the reference images is close to zero. The complex-valued method works extremely well. Reconstructing complex-valued images are important for implementing complex-valued fMRI activation methods. The reconstruction of simultaneously acquired slices alleviates the necessity for slice timing correction and voxels in different slices are temporally aligned to produce proper connectivity maps.

Hans Weber¹, Gerrit Schultz¹, Daniel Gallichan², Jürgen Hennig¹, and Maxim Zaitsev^1
1Department of Radiology, Medical Physics, University Medical Center Freiburg, Freiburg, Germany, ²LIFMET, Ecole Polytéchnique Fédérale de Lausanne, Lausanne, Switzerland

The application of nonlinear spatial encoding magnetic fields for both excitation and geometrically matched encoding allows the acquisition of curved slices with adjustable shape and thus increases the flexibility of MRI. However, both spatially varying slice thickness and in-plane resolution resulting from the nonlinearity of the fields is an unwanted side effect for most applications. The purpose of this study is to reduce the spatial variation of the voxel sizes. This is obtained by applying a concept for alias-free undersampling previously developed for PatLoc imaging.

Kangrong Zhu¹, Adam Kerr¹, and John M. Pauly^1
1Stanford University, Stanford, California, United States

The Blipped-CAIPI technique performs simultaneous multi-slice EPI acquisition with reduced g-factor penalty, but usually requires external calibration scans for image reconstruction. In this work, a data acquisition scheme which does not need any external calibration scans is designed for simultaneous multi-slice EPI and is demonstrated in in vivo brain imaging. The internal auto-calibration in the proposed method minimizes the image artifacts introduced by the mismatch between the calibration data and the accelerated data. The reconstructed images of the proposed method can have higher SNR than the Blipped-CAIPI method if the internal auto-calibration signal is included in the final images.

Paul T. Weavers¹, Eric A. Borisch¹, and Stephen J. Riederer^1
1MR Research Laboratory, Mayo Clinic, Rochester, Minnesota, United States

2D-CAIPIRINHA has been shown to reduce noise amplification when compared to traditional 2D-SENSE. However at high acceleration (R≥8) it is not clear which kernel will best accomplish this. An automated method to select the optimal kernel in a receiver coil and patient-specific manner in less than ten seconds has been developed. It has been validated with a retrospective study of nine sets of 3D foot exams, as well as in a prospective 3D contrast-enhanced MRA study.

Martin Uecker¹, Patrick Virtue¹, Shreyas S. Vasanawala², and Michael Lustig^1
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, California, United States, ²Department of Radiology, Stanford University, Stanford, California, United States

Recently, a new technique to estimate sensitivity maps from the calibration data has been proposed (ESPIRiT). In the ideal case, this technique yields a single set of sensitivity maps, which can be used with SENSE. With data corruption, multiple sets of maps naturally appear when using this method. They correspond to additional signal components which are implicitly taken into account in SPIRiT (and GRAPPA), but do not fit the SENSE model using a single set of maps. Here, we propose “soft SENSE”, which uses multiple weighted sets of maps and has similar properties as SPIRiT.

Peng Lai¹, Shreyas S. Vasanawala², Atsushi Nozaki³, Maggie Fung⁴, and Anja C.S Brau^5
1MR Applications & Workflow, GE Healthcare, Menlo Park, CA, United States, ²Radiology, Stanford University, Stanford, CA, United States, ³MR Applications & Workflow, GE Healthcare, Asahigaoka, Hino, Japan, ⁴MR Applications & Workflow, GE Healthcare, Jersey City, NJ, United States, ⁵MR Applications & Workflow, GE Healthcare, Garching, Munchen, Germany

High acceleration needed for 3D cine MRI potentially results in residual artifacts. High density coils and k-t acceleration methods demand long reconstruction time. In this work, we developed a method that automatically estimates and eliminates static tissue signals from k-t accelerated 3D cine datasets. This method enables k-t reconstruction on dynamic tissue signals with reduced aliasing and furthermore enables selective reconstruction on locations and coil channels around the heart only. Based on our evaluations, the proposed method can improve 3D cardiac cine MRI in both image quality and computation efficiency.

Tobias Wech¹, Johannes Tran-Gia¹, Dietbert Hahn¹, and Herbert Köstler^1
1Institute of Radiology, University of Würzburg, Würzburg, Germany

Eddy currents as well as gradient delays lead to deviations in measured trajectories. Especially in non-Cartesian imaging, these errors can cause severe image artifacts. In this work, we propose a method which allows for a correction of trajectory errors for radial projection imaging with no need of any separately acquired calibration data. The method iteratively translates the measured projections using the GRAPPA operator in order to maximize the self-consistency of data in oversampled k-space regions. This results in shifted coordinates representing the truly measured trajectory.

Samir D. Sharma^1,2, Joshua D. Trzasko³, and Armando Manduca^3
1Radiology, University of Wisconsin - Madison, Madison, WI, United States, ²Electrical Engineering, University of Southern California, Los Angeles, CA, United States, ³Mayo Clinic, Rochester, Minnesota, United States

Conventional image-domain-based methods for chemical shift encoding are limited both by their long scan time, which can restrict spatial resolution and/or volume coverage, and their sensitivity to intraecho off-resonance, which can cause geometric distortion. Previous works have proposed either accelerating image-domain-based methods or using a k-space-based formulation to mitigate the effects of intraecho off-resonance. In this work, we develop and demonstrate a framework for accelerated k-space-based chemical shift encoding. We employ a time-segmented approximation of the multispecies MR signal equation and we exploit prior information on intercoil structure and image sparsity to achieve acceleration. We demonstrate accelerated water-fat separation with reduced geometric distortions as compared to a conventional image-domain-based method.