Bruno Riemenschneider¹, Matthew Muckley², Alireza Radmanesh¹, Sunwoo Kim³, Geunu Jeong³, Jingyu Ko³, Yohan Jun⁴, Hyungseob Shin⁴, Dosik Hwang⁴, Mahmoud Mostapha⁵, Simon Arberet⁵, Dominik Nickel⁶, Zaccharie Ramzi^7,8, Philippe Ciuciu⁷, Jean-Luc Starck⁷, Jonas Teuwen⁹, Dimitrios Karkalousos¹⁰, Chaoping Zhang¹⁰, Anuroop Sriram¹¹, Zhengnan Huang¹, Nafissa Yakubova², Yvonne W. Lui¹, and Florian Knoll^1
1NYU School of Medicine, New York, NY, United States, ²Facebook AI Research, New York, NY, United States, ³AIRS Medical, Seoul, Korea, Republic of, ⁴Yonsei University, Seoul, Korea, Republic of, ⁵Siemens Healthineers, Princeton, NJ, United States, ⁶Siemens Healthcare GmbH, Erlangen, Germany, ⁷CEA (NeuroSpin) & Inria Saclay (Parietal), Université Paris-Saclay, Gif-sur-Yvette, France, ⁸Département d’Astrophysique, CEA-Saclay, Gif-sur-Yvette, France, ⁹Radboud University Medical Center, Nijmegen, Netherlands, ¹⁰Amsterdam UMC, Amsterdam, Netherlands, ¹¹Facebook AI Research, Menlo Park, CA, United States

The next round of the fastMRI reconstruction challenge took place, this time using anatomical brain data. Submissions were ranked by SSIM and resulting finalists again by 6 radiologists. We observed the cases with clear SSIM separation achieving the highest radiologists’ rankings, in particular the winning reconstructions. Most 4x track reconstructions exhibit desirable image quality, with some exceptions that show anatomy-like hallucinations. Radiologist sentiment decreased for the 8x and Transfer tracks, indicating that these may require further investigation.

Daeun Kim¹, Stephen F. Cauley², Krishna S. Nayak¹, Richard M. Leahy¹, and Justin P. Haldar^1
1Electrical and Computer Engineering, University of Southern California, Los Angeles, CA, United States, ²Radiology, A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States

MRI acquisitions often incidentally excite spatial regions that are not interesting for the application. This unnecessary magnetization can lead to artifacts and/or prolonged acquisitions.  We propose a novel virtual-coil approach, called region-optimized virtual (ROVir) coils, that can localize signal from an ROI and/or suppress signal from unwanted spatial regions while also providing coil compression.  This is achieved using optimal beamforming principles (without requiring modification of pulse sequences or imaging hardware), and can be applied directly to k-space data, which enables simplified image reconstruction.  We illustrate ROVir with reduced-FOV imaging, demonstrating capabilities to suppress aliasing artifacts from outside the nominal FOV.   

Shreya Ramachandran¹, Frank Ong², and Michael Lustig^1
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, CA, United States, ²Electrical Engineering, Stanford University, Stanford, CA, United States

Dynamic MRI reconstruction techniques often use static coil sensitivity maps, but physical sensitivities can change substantially with respiratory and other subject motion. However, time-resolved sensitivity maps occupy a very large amount of memory, and hence, directly employing these maps is often impractical, especially on memory-limited GPUs. Here, we introduce a technique that solves for a compact representation of time-resolved sensitivity maps by leveraging a temporal basis for sensitivity kernels. Our proposed Compact Maps are significantly cheaper (by ~1000x) to store in memory than conventional time-resolved maps and result in lower calibration error and reconstruction error than do time-averaged maps. 

Julio A. Oscanoa^1,2, Frank Ong³, Zhitao Li^2,3, Christopher M. Sandino³, Daniel B. Ennis^2,4, Mert Pilanci³, and Shreyas S. Vasanawala^2
1Department of Bioengineering, Stanford University, Stanford, CA, United States, ²Department of Radiology, Stanford University, Stanford, CA, United States, ³Department of Electrical Engineering, Stanford University, Stanford, CA, United States, ⁴Cardiovascular Institute, Stanford, CA, United States

Parallel imaging and compressed sensing reconstruction of large datasets has a high computational cost, especially for 3D non-Cartesian acquisitions. This work is motivated by the success of iterative Hessian sketching methods in machine learning. Herein, we develop Coil Sketching to lower computational burden by effectively reducing the number of coils actively used during iterative reconstruction. Tested with 2D radial and 3D cones acquisitions, our method yields considerably faster reconstructions (around 2x) with virtually no penalty on reconstruction accuracy.

Zheyuan Yi^1,2,3, Yujiao Zhao^1,2, Yilong Liu^1,2, Yang Gao^1,2, Mengye Lyu⁴, Fei Chen³, and Ed X Wu^1,2
1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong SAR, China, ²Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong SAR, China, ³Department of Electrical and Electronic Engineering, Southern University of Science and Technology, Shenzhen, China, ⁴College of Health Science and Environmental Engineering, Shenzhen Technology University, Shenzhen, China

In conventional parallel imaging, coil sensitivity information can be obtained from calibration data for reconstruction that inevitably prolongs MRI scan. In recent years, structured low-rank matrix completion methods implicitly exploit coil sensitivity that enables calibrationless k-space estimation while prohibitively increases the computational burden. This study presents a fast and calibrationless image-space alternative for reconstruction that derives high-quality coil sensitivity and spatial support maps by structured low-rank tensor estimation. The proposed approach was evaluated with multi-channel multi-contrast brain datasets. It achieves a high convergence rate with significantly reduced reconstruction time, making the calibrationless reconstruction approach more efficient in clinical practice.

Yannick Brackenier^1,2, Lucilio Cordero-Grande^1,2,3, Raphael Tomi-Tricot^1,2,4, Tom Wilkinson^1,2, Jan Sedlacik^1,2, Philippa Bridgen^1,2, Sharon Giles^1,2, Shaihan Malik^1,2, Enrico De Vita^1,2, and Joseph V Hajnal^1,2
1Biomedical Engineering Department, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²Centre for the Developing Brain, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ³Biomedical Image Technologies, ETSI Telecomunicación, Universidad Politécnica de Madrid and CIBER-BNN, Madrid, Spain, ⁴MR Research Collaborations, Siemens Healthcare Limited, Frimley, United Kingdom

A fully data-driven retrospective motion correction reconstruction for volumetric brain MRI at 7T that includes modelling of pose-dependent changes in polarising magnetic (B0) fields in the head has been developed. Building on the DISORDER framework, the use of a physics-based B0 model constrains the number of unknowns to be found, enabling motion correction based solely on data-consistency without requiring any additional probe- or navigator-data. The proposed reconstruction was validated on an in-vivo spoiled gradient echo acquisition in which the subject deliberately moved. Substantial removal of motion artefacts was achieved.

Yanis Djebra^1,2, Isabelle Bloch^2,3, Georges El Fakhri¹, and Chao Ma^1
1Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States, ²LTCI, Telecom Paris, Institut Polytechnique de Paris, Paris, France, ³LIP6, Sorbonne University, CNRS, Paris, France

Many unsupervised learning methods have been proposed to discover the structure of manifolds embedded in high-dimensional input spaces. However, image reconstruction requires mapping the learned low-dimension data in the feature space back to the input space, which can be challenging if the mapping function is implicit. This work presents an image reconstruction scheme closely related to machine learning methods learning manifolds via tangent space alignment. Here, the mapping transform is explicit and learned from the data. This model is a nonlinear generalization of the Low-Rank matrix/tensor model, reconstructing undersampled MR data with lower rank than the standard Low-Rank reconstruction.

Gastao Cruz¹, Kerstin Hammernik^2,3, Thomas Kuestner¹, Daniel Rueckert^2,3, René M. Botnar¹, and Claudia Prieto^1
1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²Technical University of Munich, Munich, Germany, ³Department of Computing, Imperial College London, London, United Kingdom

Motion-resolved reconstructions are standard for cardiac CINE imaging by splitting data into multiple cardiac phases across several heartbeats. Consequently, the reconstruction of each phase is highly ill-posed, since only a subset of the data are used to reconstruct it. In this work a novel motion corrected framework is developed for highly accelerated cardiac CINE imaging. Each cardiac phase is reconstructed with a motion corrected reconstruction and jointly regularized with every other motion corrected cardiac phase via a non-rigidly aligned patch-based denoiser. This approach leads to a substantial improvement in image quality, enabling highly accelerated CINE images from a single heartbeat.

Layla Tabea Riemann¹, Christoph Stefan Aigner¹, Ralf Mekle², Sebastian Schmitter^1,3, Bernd Ittermann¹, and Ariane Fillmer^1
1Physikalisch-Technische Bundesanstalt (PTB), Braunschweig und Berlin, Germany, ²Center for Stroke Research Berlin, Charité Universitätsmedizin, Berlin, Germany, ³Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States

In this work, a Fourier-based technique for ¹H MR spectroscopy based on split-slice-GRAPPA is introduced to decompose simultaneously acquired dual-voxel data. In contrast to the existing sensitivity-map-based approach, this technique does not need any additional image acquisitions. The autocalibration lines are derived by additional low SNR spectral data. 2SPECIAL was used to simultaneously acquire data from two voxels at short echo times. The proposed decomposition algorithm was first validated in a multi-compartment phantom, and its application was then demonstrated at 7T in vivo. 

Zepeng Wang^1,2, Yahang Li^1,2, and Fan Lam^1,2
1Department of Bioengineering, University of Illinois Urbana-Champaign, Urbana, IL, United States, ²Beckman Institute for Advanced Science and Technology, Urbana, IL, United States

J-resolved ¹H-MRSI offers improved molecular specificity by encoding the J-coupling evolution of different molecules at multiple TEs. The addition of another encoding dimension poses both challenges and flexibility for optimizations in data acquisition and reconstruction. This work presents further optimized J-resolved MRSI acquisition and reconstruction strategies for high-resolution, 3D metabolite, and neurotransmitter mapping. Specifically, estimation-theoretic TE selection within a union-of-subspaces (UoSS) framework was analyzed for optimized separation of metabolite and neurotransmitter signals. Both simulation and in vivo studies have been conducted. Promising results in terms of simultaneously high-resolution mapping of major metabolites, Glx, and GABA are provided.