Huiwen Luo¹, Wissam AlGhuraibawi², Kevin Godines², Daniel Gochberg³, Moriel Vandsburger², and William A Grissom^1
1Biomedical Engineering, Vanderbilt University, Nashville, TN, United States, ²Department of Bioengineering, University of California Berkeley, Berkeley, CA, United States, ³Radiology and Radiological Sciences, Vanderbilt University, Nashville, TN, United States

A tailored spectral-spatial saturation pulse was developed to produce a flat flip angle profile across the heart and achieve more uniform CEST saturation despite B₁ inhomogeneity at 3 Tesla. The tailored saturation pulse train was simulated for a two-pool system to evaluate the z-spectrum at each spatial location in the heart, based on an in vivo 3 Tesla B₁ map. Whereas CEST saturation generated with a conventional Gaussian pulse yielded CEST contrast of 2.60±1.59% across the ventricle, the tailored pulse produced more uniform saturation across the heart which resulted in both greater and more uniform CEST contrast of 4.64±0.34%.

Kai Herz¹, Sebastian Mueller¹, Or Perlman², Ruediger Stirnberg³, Tony Stoecker^3,4, Klaus Scheffler^1,5, Christian Farrar², and Moritz Zaiss^1,6
1Magnetic Resonance Center, Max Planck Institute for Biological Cybernetics, Tuebingen, Germany, ²Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Charlestown, MA, United States, ³German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, ⁴Department of Physics and Astronomy, University of Bonn, Bonn, Germany, ⁵Department of Biomedical Magnetic Resonance, Eberhard Karls University Tuebingen, Tuebingen, Germany, ⁶Department of Neuroradiology, University Hospital Erlangen, Erlangen, Germany

Quantitative CEST imaging is still not applied in clinical routine, as both quantification and whole brain coverage require usually long scan times. In this work, we present a CEST-MRF protocol using spin-lock saturation pulses and a fast 3D-EPI readout with whole brain coverage. This enables a fast generation of quantitative amide proton concentration maps of the entire brain at a clinical scanner.

Felix Glang¹, Anagha Deshmane¹, Sergey Prokudin², Florian Martin¹, Kai Herz¹, Tobias Lindig³, Benjamin Bender³, Klaus Scheffler^1,4, and Moritz Zaiss^1,5
1Magnetic Resonance Center, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, ²Department of Perceiving Systems, Max Planck Institute for Intelligent Systems, Tübingen, Germany, ³Department of Diagnostic and Interventional Neuroradiology, Eberhard Karls University Tübingen, Tübingen, Germany, ⁴Department of Biomedical Magnetic Resonance, Eberhard Karls University Tübingen, Tübingen, Germany, ⁵Department of Neuroradiology, University Clinic Erlangen, Erlangen, Germany

Analysis of CEST data often requires complex mathematical modeling before contrast generation, which can be error prone and time-consuming. Here, a probabilistic deep learning approach is introduced to shortcut conventional Lorentzian fitting analysis of 3T in-vivo CEST data by learning from previously evaluated data. It is demonstrated that the trained networks generalize to data of a healthy subject and a brain tumor patient, providing CEST contrasts in a fraction of the conventional evaluation time. Additionally, the probabilistic network architecture enables uncertainty quantification, indicating if predictions are trustworthy, which is assessed by perturbation analysis.

Beomgu Kang¹, Byungjai Kim^1,2, Michael Schar², Hyunwook Park¹, and Hye Young Heo^2,3
1Department of Electrical Engineeering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of, ²Russell H Morgan Department of Radiology and Radiological Science, Johns Hopkin University, Baltimore, MD, United States, ³F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States

Most currently used MTC/CEST imaging protocols depend on the acquisition of qualitative weighted images, limiting the detection sensitivity to quantitative parameters, their exchange rate and concentration. Here, we propose a fast, quantitative 3D MTC/CEST imaging framework based on a combined 1) time-interleaved parallel RF transmission, 2) compressed sensing, 3) MR fingerprinting, and 4) deep-learning techniques. Typically, supervised deep learning requires a massive amount of labeled images for training, which is limited particularly in MTC/CEST MRI field. However, the proposed unsupervised learning architecture requires only small amounts of unlabeled MTC/CEST data.    

Daniel J. West¹, Gastao Cruz¹, Olivier Jaubert¹, Rui P. A. G. Teixeira^1,2, Torben Schneider³, Jacques-Donald Tournier^1,2, Jo Hajnal^1,2, Claudia Prieto¹, and Shaihan J. Malik^1,2
1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²Centre for The Developing Brain, King's College London, London, United Kingdom, ³Philips Healthcare, Guildford, United Kingdom

Inhomogeneous magnetisation transfer (ihMT) is a contrast mechanism that has shown high specificity towards myelinated tissue. Contrast is typically generated using sequences comprising a preparation phase with several RF saturation pulses, followed by multiple readout periods for measurement. Here, we present a transient acquisition scheme that alternates between periods of multi-band and single-band RF pulses, to efficiently generate ihMT contrast during a single data acquisition. Since signal is transiently varying throughout, we use a dictionary-based low-rank inversion reconstruction method originally proposed for magnetic resonance fingerprinting. Simulation, phantom and human in-vivo experiments are included.

Alexey Samsonov¹, Aaron Field¹, Vasily Yarnykh², and Julia Velikina^3
1Radiology, University of Wisconsin, Madison, WI, United States, ²Radiology, University of Washington, Seattle, WA, United States, ³Medical Physics, University of Wisconsin, Madison, WI, United States

Macromolecular proton fraction (MPF), the key two-pool MT model parameter, was established as a robust myelin-sensitive index, with clinical relevance in demyelinating diseases. However, as MPF assesses macromolecules relative to tissue water, its specificity to myelin is  limited. i.e., MPF changes may occur independent of myelin, e.g., in the setting of inflammation and edema. Further, relating MPF to macromolecules may be ambiguous due to unequal concentrations of protons in macromolecular and water compartments. We demonstrate implications of these effects for MPF interpretation using phantom and ex-vivo experiments and propose a new macromolecular measure that explicitly accounts for tissue water effects. 

Giulia MC Rossi^1,2, Tom Hilbert^1,2,3, Adèle LC Mackowiak^1,2, Katarzyna Pierzchała^4,5, Tobias Kober^1,2,3, and Jessica AM Bastiaansen^1
1Department of Diagnostic and Interventional Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ²Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, ³LTS5, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ⁴Laboratory for Functional and Metabolic Imaging, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ⁵Center for Biomedical Imaging (CIBM), Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

A novel quantitative framework for detection of different tissue compartments based on bSSFP signal profile asymmetries (SPARCQ) is reported. SPARCQ uses a dictionary-based weight optimization algorithm to estimate voxel-wise off-resonance frequency and relaxation time ratio spectra from acquired bSSFP signal profiles. From the obtained spectra, quantitative parameters (i.e. fractions of the components of interest, thermal equilibrium magnetization) can be extracted. Validation and proof-of-concept are provided for voxel-wise water-fat separation and fat fraction mapping. Accuracy and repeatability of SPARCQ are demonstrated with phantom and in vivo experiments.

Ante Zhu^1,2, Yuxin Zhang^2,3, Alan McMillan², Fang Liu², Timothy J Colgan², Scott B. Reeder^1,2,3,4,5, and Diego Hernando^1,2,3,6
1Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States, ²Radiology, University of Wisconsin-Madison, Madison, WI, United States, ³Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ⁴Emergency Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁵Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁶Electrical and Computer Engineering, University of Wisconsin-Madison, Madison, WI, United States

Multi-echo chemical shift-encoded (CSE)-MRI techniques enable liver PDFF and R₂^* quantification, which enable staging and treatment monitoring of liver fat and iron content, respectively.  However, the common requirement of breath-holding in CSE-MRI acquisitions is challenging for many patients. Furthermore, the required specialized multi-echo acquisition and reconstruction are not available in all scanners. In this work, we assessed the accuracy of deep learning (DL)-based PDFF and R₂^* quantification using reduced numbers of echoes. Preliminary results demonstrate the potential of this approach and suggest that at least four echoes are needed for quantifying PDFF and R₂^* at 1.5T and 3.0T. 

Beata Bachrata^1,2,3, Bernhard Strasser^1,2,4, Wolfgang Bogner^1,2, Albrecht Ingo Schmid^1,5, Siegfried Trattnig^1,2,3, and Simon Daniel Robinson^1,2,6,7
1High Field MR Centre, Medical University of Vienna, Vienna, Austria, ²Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria, ³Christian Doppler Laboratory for Clinical Molecular MR Imaging, Vienna, Austria, ⁴Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States, ⁵Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria, ⁶Centre for Advanced Imaging, The University of Queensland, Brisbane, Australia, ⁷Department of Neurology, Medical University of Graz, Graz, Austria

Imaging of body regions containing both water-based and fat-based structures is affected by artefacts arising from the chemical shift difference between water and fat. Recently, a single-echo water-fat separation technique was proposed which used multi-band principles to generate separate water and fat images as well as chemical shift-corrected, recombined water-fat images. We demonstrate the performance of gradient-echo and turbo spin-echo variants of this approach in the knee, breasts and abdomen. The separation of water and fat was similar to or better than with current state-of-the-art techniques and chemical shift effects were fully eliminated in recombined water-fat images.

Dominique Franson¹, Yuchi Liu^1,2, Rajiv Ramasawmy³, Adrienne Campbell-Washburn³, and Nicole Seiberlich^1,2
1Case Western Reserve University, Cleveland, OH, United States, ²University of Michigan, Ann Arbor, MI, United States, ³National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, United States

Fat/water separation at low field strengths can be difficult due to the small difference between resonance frequencies.  Rosette trajectories have previously been shown to be effective for spectral separation and fat suppression, and the approach is not dependent on a large frequency difference. Here, a rosette trajectory is used to significantly suppress fat signal in water images, and to produce separate fat images at 0.55T.  B0 maps are calculated from two of the rosette echoes, and are used to improve the fat/water separation. Initial examples are shown in an oil/water phantom, and in the heart and abdomen.

Christoph C. Zöllner¹, Sophia Kronthaler¹, Stefan A. Ruschke¹, Holger Eggers², Jürgen Rahmer², Peter Börnert², Rickmer F. Braren¹, Daniela Franz¹, and Dimitrios C. Karampinos^1
1Department of Diagnostic and Interventional Radiology, School of Medicine, Technical University of Munich, Munich, Germany, ²Philips Research Laboratory, Hamburg, Germany

Multi-echo Stack-of-stars-type radial k-space trajectories employing golden-angle ordering have been becoming popular for abdominal fat quantification. Gradient chain imperfections including eddy currents and gradient delays are known to affect the image quality of radial imaging. Most methods for compensating radial k-space trajectory errors are based either on the acquisition of calibration lines with opposite polarity or on the processing of approximately anti‐parallel spokes from the actual radial acquisition. This work shows that a trajectory correction based on a gradient system impulse response function improves fat quantification in gated golden-angle radial Dixon imaging.

Liangdong Zhou¹, Qihao Zhang^1,2, Pascal Spincemaille¹, Thanh D Nguyen¹, John Morgan¹, Weiying Dai¹, Ajay Gupta¹, Martin R Prince¹, and Yi Wang^1,2
1Weill Medical College of Cornell University, New York, NY, United States, ²Cornell University, Ithaca, NY, United States

Perfusion quantification is important for the diagnosis of many diseases. Validation of perfusion quantification methods remains challenging due to the various assumptions and lack of the ground truth. We built a numerical phantom of microvascular network in the kidney. In the phantom, the ground truth blood velocity and flow were computed from Navier-Stokes equation. Tracer concentration was simulated based on the mass transport equation. Comparison between Kety's method and our recently proposed AIF-free QTM method was performed using the numerical phantom. It turns out that QTM method reduces the flow error by more than 3 folds compare with Kety's method.

Srikant Kamesh Iyer¹, Brianna F Moon², Nicholas J Josselyn¹, Eileen Hwuang², Jeffrey B Ware¹, David Roalf³, Jae W Song¹, S. Ali Nabavizadeh¹, and Walter R Witschey^1
1Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States, ²Bioengineering, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States, ³Psychiatry, University of Pennsylvania, Philadelphia, PA, United States

This abstract presents a novel non-local filtering based reconstruction approach for high quality quantitative susceptibility mapping (QSM). Popular QSM techniques that use fixed sparsity priors such as total variation or total generalized variation often suffer from blurring of fine features (e.g. edges). Since QSM images have non-local spatial redundancies in the form of self-similarity, we develop an approach that uses non-local grouping by 4D cube-matching and collaborative filtering in a plug-and-play (PnP) alternating direction method of multiplier (ADMM) framework. We show that the proposed non-local filtering based reconstruction approach achieves sharper edges and better preservation of fine features.

Rasim Boyacioglu¹ and Mark Griswold^1
1Radiology, Case Western Reserve University, Cleveland, OH, United States

Magnetic Resonance Fingerprinting with Quadratic RF Phase (MRFqRF) can simultaneously map T1, T2, T2* and off-resonance. It has been shown that local field inhomogeneities due to susceptibility is encoded in MRFqRF off-resonance maps. Here publicly available standard QSM processing tools were used to analyze two high resolution 3D MRFqRF datasets from 3T. Susceptibility contrast is revealed after phase unwrapping, background removal and B1 correction. QSM preprocessed data was further analyzed with two dipole kernel inversion algorithms. Susceptibility encoding in MRF framework is novel and brings immediate additional value to MRI exam.

Kuo-Wei Lai^1,2, Jeremias Sulam¹, Manisha Aggarwal³, Peter van Zijl^2,3, and Xu Li^2,3
1Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, United States, ²F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States, ³Department of Radiology and Radiological Sciences, Johns Hopkins University, Baltimore, MD, United States

We designed a method called Orientation-Grasp Deep Neural Network (OG-DNN) for Quantitative Susceptibility Mapping (QSM). OG-DNN has dynamically adaptive convolutional filters that adjust themselves according to the input B₀ orientation in the subject frame of reference. Our experimental results demonstrate that OG-DNN can reconstruct high-quality and consistent susceptibility maps from MR phase data acquired at different head orientations with respect to B₀ within a consistent subject frame of reference. OG-DNN is expected to provide improved flexibility in practice and may potentially facilitate the development of deep learning-based Susceptibility Tensor Imaging (STI) reconstructions.