Kuang Gong¹, Paul Kyu Han¹, Thibault Marin¹, Georges El Fakhri¹, Quanzheng Li¹, and Chao Ma^1
1Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States

The subspace-based method, known as SPICE, is an emerging technique that achieves rapid high-resolution MRSI with good SNR. In SPICE, the spectrum at each voxel is represented as a low-dimensional subspace or manifold, where the basis functions or features are learned from training data. The spatial coefficients of the subspace model are estimated by fitting the model to the k-space data for image reconstruction. In this work, we propose to extend the SPICE framework by representing the spatial coefficients of the subspace model using deep image prior for improved image reconstruction.

Zidan Yu^1,2,3, Shota Hodono^1,2,3, Bili Wang¹, Olga Dergahyova¹, Bei Zhang^1,3, Ryan Brown^1,3, Daniel K. Sodickson^1,2,3, Guillaume Madelin^1,2, and Martijn A. Cloos^1,2,3
1Center for Biomedical Imaging, Department of Radiology, New York University School of Medicine, New York, NY, United States, ²Sackler Institute of Graduate Biomedical Sciences, NYU Langone Health, New York, NY, United States, ³Center for Advanced Imaging Innovation and Research (CAI2R), NYU Langone Health, New York, NY, United States

In this work, we present a 3D sequence that can simultaneously capture quantitative ¹H density, T₁, T₂, B₁⁺ maps and a ²³Na image of the whole head in a reasonable scan time (~10 min). The gradient momenta are strategically distributed to simultaneously acquire a full-radial trajectory for proton and a center-out radial trajectory for sodium in one single readout. A sodium SNR comparison, verification of the proton multi-parametric maps, and in-vivo results are shown.

Yudu Li^1,2, Kihwan Kim^3,4, Bryan Clifford^1,2, Rong Guo^1,2, Yuning Gu^3,4, Zhi-Pei Liang^1,2, and Xin Yu^3,4,5,6
1Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, ⁴Case Center for Imaging Research, Case Western Reserve University, Cleveland, OH, United States, ⁵Department of Radiology, Case Western Reserve University, Cleveland, OH, United States, ⁶Department of Physiology and Biophysics, Case Western Reserve University, Cleveland, OH, United States

Dynamic ³¹P-MRS/MRSI is a promising tool for in vivo quantification of mitochondrial oxidative capacity. However, its practical utility is limited by the inherently low SNR of the ³¹P signal. This work is built upon our recent progress in accelerating dynamic ³¹P-MRSI using low-rank tensor models. We extended this method by learning the temporal priors with deep generative models and then incorporating them into the reconstruction via an information theoretical framework. This approach enabled high-resolution dynamic ³¹P-MRSI with 1.5x1.5x2 mm³ nominal spatial resolution and 5.1-sec temporal resolution in capturing the kinetics of metabolite changes in rat hindlimb during a stimulation-recovery protocol.

Siyuan Dong¹, Henk M. De Feyter², Monique A. Thomas², Robin A. de Graaf³, and James S. Duncan^4
1Department of Electrical Engineering, Yale University, New Haven, CT, United States, ²Department of Radiology and Biomedical Imaging, Yale University, School of Medicine, New Haven, CT, United States, ³Department of Radiology and Biomedical Imaging, Department of Biomedical Engineering, Yale University, School of Medicine, New Haven, CT, United States, ⁴Department of Radiology & Biomedical Imaging, Department of Electrical Engineering, Department of Statistics & Data Science, Yale University, New Haven, CT, United States

Deuterium Metabolic Imaging (DMI) is a novel approach providing 3D metabolic data from both animal models and human subjects. DMI relies on ²H MRSI in combination with administration of ²H-labeled substrates. Common to all MRI and MRSI methods, DMI's resolution is ultimately limited by the achievable SNR. This work proposes a data-driven method using a deep convolutional autoencoder to improve the SNR and increase the spatial resolution of DMI. The method was tested with simulated, phantom and in vivo experiments at various SNR levels to demonstrate its capability and precision for metabolic mapping using noisy DMI data.

Fabian J. Kratzer^1,2, Sebastian Schmitter^1,3, Armin M. Nagel^1,4,5, Nicolas G. R. Behl⁶, Benjamin R. Knowles¹, Peter Bachert^1,2, Mark E. Ladd^1,2,7, and Sebastian Flassbeck^1
1Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, ²Faculty of Physics and Astronomy, Ruprecht-Karls University Heidelberg, Heidelberg, Germany, ³Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, ⁴Institute of Radiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany, ⁵Institute of Medical Physics, Friedrich-Alexander-Universität (FAU), Erlangen, Germany, ⁶Siemens Healthcare GmbH, Erlangen, Germany, ⁷Faculty of Medicine, Ruprecht-Karls University Heidelberg, Heidelberg, Germany

Sodium relaxation times have been shown to be altered in several diseases. However, due to short relaxation times and low in-vivo signal, measurement times in sodium relaxometry on the order of 1h were reported for both, longitudinal and transversal relaxation constants. In this work, a novel sodium relaxometry method based on Magnetic Resonance Fingerprinting (MRF) principles is presented, which enables simultaneous quantification of T₁, T_2s^*, T_2l^*, T₂^* and ΔB₀, with automatic distinction between bi- and monoexponential transverse relaxation.

Rong Guo^1,2, Yudu Li^1,2, Yibo Zhao^1,2, Yao Li^3,4, and Zhi-Pei Liang^1,2
1Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ⁴Med-X Research Institute, Shanghai Jiao Tong University, Shanghai, China

SPICE has recently provided a unique capability for simultaneous acquisition of metabolite and water spectroscopic signals. While the water signals are often removed as nuisance components in traditional MRSI experiments, SPICE utilizes the water signals for QSM, MWF mapping, etc. In this work, we further extend SPICE data acquisition to achieve much larger k-space coverage and improve its processing scheme for simultaneous MRSI/QSM/SWI/MWF mapping. In vivo experiments demonstrated that this new scheme improved the accuracy of water/lipid removal, reduced the effects of field inhomogeneity, and achieved higher resolution for QSM, SWI and MWF using the unsuppressed water signals.

Chunwei Ying¹, Yasheng Chen², Michael M. Binkley², Meher R. Juttukonda^3,4, Shaney Flores¹, Tammie L. S. Benzinger^1,5, and Hongyu An^1
1Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, MO, United States, ²Department of Neurology, Washington University School of Medicine, St. Louis, MO, United States, ³Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States, ⁴Department of Radiology, Harvard Medical School, Boston, MA, United States, ⁵Department of Neurological Surgery, Washington University School of Medicine, St. Louis, MO, United States

We proposed a 3D patch based residual U-Net method to estimate pseudo CT images for PET/MR attenuation correction by including quantitative R1 maps as input. The proposed deep learning based T1-enhanced selection of linear attenuation coefficients (DL-TESLA) method outperformed the deep learning methods using UTE-R2* or MPRAGE as inputs with a similar network structure. Moreover, we demonstrated that DL-TESLA had an excellent PET test-retest repeatability that was comparable to PET/CT, supporting its use for PET/MR AC in longitudinal studies of neurodegenerative diseases. 

Ileana Ozana Jelescu¹, Jelle Veraart², and Cristina Cudalbu^1
1Center for Biomedical Imaging, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ²Dept. of Radiology, New York University School of Medicine, New York, NY, United States

MRS is an inherently low signal-to-noise technique resulting in substantial spectral averaging and large voxel volumes. The problem is further amplified for diffusion-weighted MRS. Here we test the performance of denoising using principal component analysis coupled with Marchenko-Pastur’s random matrix theory in the context of DW-MRS. We report 50 – 100% increase in SNR, reduction in Cramer-Rao bounds and a potential eight-fold reduction in scan time. This technique is expected to also bring significant improvements in the context of fMRS, X-nuclei MRS and CSI.

Martijn Froeling¹, Tijl A van der Velden¹, Jeanine J Prompers¹, and Dennis WJ Klomp^1
1Department of Radiology, UMC Utrecht, Utrecht, Netherlands

Chemical shift imaging generally suffers from low SNR and low spatial resolution, especially for x-nuclei. State of the art image processing methods from the MRI domain, e.g. DTI pre-processing, can be applied to CSI data. In this study we show the feasibility of PCA denoising combined with deconvolution to enhance CSI SNR and spatial localization.

Bing Ji¹, Zahra Hosseini², Liya Wang³, Lei Zhou⁴, Xinhua Tu⁵, and Hui Mao^6
1Department of Radiology and Imaging Sciences, Emory University School of Medicine, Emory University, Atlanta, GA, Georgia, ²MR R&D Collaborations, Siemens Healthineers,, Atlanta, Georgia, ³Department of Radiology, The People’s Hospital of Longhua, Shenzhen, China, ⁴Department of Radiology and Imaging Sciences, Emory University School of Medicine, Emory University, Atlanta, Georgia, ⁵School of Communication and Information Engineering, Nanjing University of Posts and Telecommunication, Nanjing, China, ⁶Department of Radiology and Imaging Sciences, Emory University School of Medicine, Emory Univeristy, Atlanta, GA, United States

Low signal-to-noise ratio (SNR) and long acquisition time limit the clinical applications of magnetic resonance spectroscopy (MRS). This work presents a data-driven machine-learning assisted Spectral Wavelet-feature Analysis and Classification Assisted Denoising (SWANCAD) approach to extract the specific spectral wavelets of signals and noises for reducing noise and improving SNR of MRS data.  The effective denoise by SWANCAD enabled resolving prominent metabolic peaks but also identify the smaller concentration metabolites which are merged in the noises. Potential applications of the SWANCAD includes the possibility of improving the signal to noise ratio (SNR) of MRS data collected in sub-minute or sub-cm voxels.