Jason Peter Stockmann¹, Gigi Galiana², Vicente Parot^3,4, Leo Tam¹, and Robert Todd Constable^1,2
1Biomedical Engineering, Yale University, New Haven, CT, United States, ²Diagnostic Radiology, Yale University, New Haven, CT, United States, ³Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile, ⁴Department of Electrical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile

The variable-order fractional Fourier transform (FrFT) is used to describe signals acquired using both linear and quadratic encoding gradients during readout. The FrFT is a generalization of the Fourier transform which imparts a rotation of angle α in time-frequency space. As quadratic phase evolves during readout, α changes continuously. We reconstruct images on a point-by-point basis using the variable-order FrFT for each α along with a radial k-space density compensation function. FrFT images show markedly reduced sensitivity to off-resonance phase and gradient calibration errors as compared with images reconstructed using an iterative “brute force” solver with the full encoding matrix.

Yu Li¹, and Charles L. Dumoulin^1
1Radiology, Cincinnati Children's Hospital Medical Center, Cincinnati, Ohio, United States

The presented work introduces a new reconstruction framework, "correlation-based reconstruction", to improve parallel imaging by taking advantage of all available data relationships in parallel acquisition. In this framework, correlation functions are used to mathematically describe a generic data relationship, and the reconstruction relies on the estimation of correlation functions from prior knowledge about imaging data. In a high-resolution brain imaging experiment, it is demonstrated that correlation-based reconstruction has the potential to overcome the limit posed by coil array in a conventional parallel imaging technique.

Berkin Bilgic¹, Audrey P. Fan¹, and Elfar Adalsteinsson^1,2
1EECS, MIT, Cambridge, MA, United States, ²Harvard-MIT Division of Health Sciences and Technology, Cambridge, MA, United States

Quantitative Susceptibility Mapping (QSM) aims to quantify tissue magnetic susceptibility χ with applications such as tissue contrast enhancement, venous blood oxygenation, and iron quantification. Estimation of χ from phase is ill posed due to zeros on a conical surface in the Fourier space; hence χ inversion benefits from additional regularization. In our work, we propose enhanced regularization with demonstrated performance benefits by incorporation of magnitude priors. By encoding spatial priors derived from magnitude into l1 regularization scheme via the Focal Underdetermined System Solver (FOCUSS) algorithm, we report high quality QSM both on a numerical phantom and in-vivo data at 7T.

Markus Weiger^1,2, and Klaas Paul Pruessmann^3
1Bruker BioSpin AG, Faellanden, Switzerland, ²Bruker BioSpin MRI GmbH, Ettlingen, Germany, ³Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

ZTE is an MRI technique with 3D radial centre-out encoding and zero echo time, particularly suited for imaging samples with short T2. ZTE data is inherently incomplete in the k-space centre, which can be addressed by algebraic reconstruction. In this work, a noise analysis of ZTE imaging is presented, revealing peculiar and specific mechanisms of noise amplification and correlation, which lead to an artefact-like manifestation of noise in the intermediate 1D images. However, it is also shown that the strong noise averaging in central k-space inherent to 3D radial scanning compensates for this effect, thus making remarkably large acquisition gaps feasible.

Li Feng¹, Ricardo Otazo², Monvadi B Srichai^2,3, Ruth P Lim², Daniel K Sodickson², and Daniel Kim^2
1Sackler Institute of Graduate Biomedical Sciences, New York University School of Medicine, New York, New York, United States, ²Radiology, New York University School of Medicine, New York, New York, United States, ³Medicine, New York University School of Medicine, New York, New York, United States

Real-time cine MRI is a valuable technique for patients with reduced breath-hold capacity and/or arrhythmia. Recently, highly-accelerated real-time cine MRI using compressed sensing and parallel imaging that jointly exploits image sparsity in time series data and coil sensitivity was proposed. However, use of temporal FFT as sparsifying transform yielded temporal blurring of pixels that occupy both myocardium and blood over time. In this work, we performed cardiac motion constrained reconstruction to minimize the aforementioned artifact by generating multiple reconstructions with a different number of cardiac frames and combining the results to minimize temporal blurring.

Michael Schacht Hansen¹, Thomas Sangild Sørensen², and Peter Kellman^1
1National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, United States, ²Department of Computer Science, Aarhus University, Aarhus, Denmark

A method for iterative reconstruction of high spatial and temporal resolution images from free breathing real-time acquisitions is presented. The golden angle radial acquisition scheme is used to enable reconstruction of low temporal resolution images, and respiration deformation maps are obtained from these images using non-rigid registration. The reconstruction includes the deformation maps into the encoding equations and thus compensates for the motion as part of the reconstruction. The method was successfully applied in five healthy volunteers producing high quality, retrospectively gated cardiac cine images during free breathing.

Yijing Wu¹, Steven Kecskemeti¹, Patrick A Turski², and Charles A Mistretta^2
1Medical Physics, University of Wisconsin, Madison, MADISON, WI, United States, ²University of Wisconsin, Madison

HYPR FLOW technique utilizes a separately acquired phase contrast image as a spatial constraint for HYPR processing of undersampled time frames acquired during the first pass of injected contrast and is able to provide high resolution 4D CE MRA. However, the spatially constraining image (complex difference) is susceptible to signal loss due to complex/slow flow, which in turn will degrade the final images. We present here a reconstruction method called IMAC HYPR FLOW, which combines both the magnitude image (MAG) and the complex difference image (CD) as the spatial constraint and utilizes the iterative HYPR (I-HYPR) to generate each individual time frame, such that the signal loss due to the complex/slow flow can be greatly recovered in the time-resolved contrast-enhanced time frames.

Johan Berglund¹, Håkan Ahlström¹, Lars Johansson¹, and Joel Kullberg^1
1Oncology, Radiology and Clinical Immunology, Uppsala University, Uppsala, Sweden

The three-point Dixon method enables high-resolution reconstruction of separate water and lipid images, by sampling the chemical shift dimension at three points. A simple single-peak model of the lipid spectrum is typically used. However, more accurate spectrum modeling has been shown to give better water and lipid estimates. The non-linear problem is typically solved by optimization in each voxel. Here, a closed-form solution is presented. The method was demonstrated in vivo for an abdominal dataset. The water/lipid separation of the 288 × 228 image matrix took 0.25 sec, compared to 141 sec using optimization, and resulted in high-quality fat suppression.

Holger Eggers¹, Peter Boernert¹, and Peter Koken^1
1Philips Research, Hamburg, Germany

Spiral imaging promises high scan efficiency due to long readouts, but suffers from strong susceptibility to off-resonance effects. Water-fat imaging with chemical shift encoding provides exactly the information required for an off-resonance correction. The combination of both thus seems particularly attractive. In the present work, the software of a clinical scanner is modified and extended to provide a fully automated and integrated reconstruction, separation, and correction for spiral water-fat imaging in clinical applications. An initial evaluation in abdominal and cardiac imaging suggests that good image quality and acceptable processing times are achieved.

Diego Hernando¹, Catherine DG Hines¹, Huanzhou Yu², and Scott B Reeder^1,3
1Radiology, University of Wisconsin, Madison, WI, United States, ²Global Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States, ³Medical Physics, University of Wisconsin, Madison, WI, United States

Accurate fat quantification is important for the detection and classification of non-alcoholic fatty liver disease. Chemical shift based methods that rely on the complex signal are sensitive to phase errors, such as those caused by eddy currents, whereas methods that use only the signal magnitude result in poor SNR for certain acquisition parameters. In this work, we propose a mixed fitting method that provides accurate fat quantification with good SNR behavior. The performance of the proposed method is characterized theoretically, and demonstrated using phantom data and in vivo liver acquisitions.