Hongjun An¹, Hyeong-Geol Shin¹, Woojin Jung¹, and Jongho Lee^1
1Department of Electrical and computer Engineering, Seoul National University, Seoul, Korea, Republic of

B₀ fluctuation from respiration can induce significant artifacts in MRI images. In this study, a new retrospective correction method that requires no modification in sequences (e.g. no navigator) is proposed. This method utilizes a convolution neural network (CNN), DeepRespi, to extract a respiration pattern from a corrupted image. The respiration pattern is applied back to the corrupted image for phase compensation. When tested, the CNN successfully extracted the respiration pattern (correlation coefficient = 0.94 ± 0.04) and the corrected images showed on average 68.9 ± 13.2% reduction in NRMSE when comparing the corrupted vs. corrected images.

Sunghun Seo¹, Won-Joon Do¹, Huan Minh Luu¹, Ki Hwan Kim¹, Seung Hong Choi², and Sung-Hong Park^1
1Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of, ²Department of Radiology, Seoul National University College of Medicine, Seoul, Korea, Republic of

We propose convolutional neural network (CNN) to accelerate Slice Encoding for Metal Artifact Correction (SEMAC). The concept was tested on metal‑embedded agarose phantoms and patients with metallic neuro plates in the cerebral region. CNN was trained to output images with high SEMAC factor from input images with low SEMAC factor, achieving acceleration factors of 2 or 3. The metal artifacts in low SEMAC factor data were visually and quantitatively suppressed well in the output of CNN (p<0.01), which was comparable to that of the high SEMAC factor. The study shows the feasibility of reducing scan time of SEMAC through CNN.

Chitresh Bhushan¹, Dattesh D. Shanbhag², Andre Maximo³, Uday Patil², Radhika Madhavan¹, Matthew Frick⁴, Kimberly K. Amrami⁴, Desmond Teck Beng yeo¹, and Thomas Foo^1
1GE Research, Niskayuna, NY, United States, ²GE Healthcare, Bengaluru, India, ³GE Healthcare, Rio de Janeiro, Brazil, ⁴Mayo Clinic, Rochester, MN, United States

We evaluate the consistency and clinical applicability of our automated deep-learning based intelligent slice placement (ISP) approach for knee scan planning. We use 146 clinical knee exams that were retrospectively selected to have anatomically consistent scan planning along with manual-marking from in-house radiologist to access the variability across MR technicians. The results indicate that our automated ISP approach has better consistency than the variability seen across MR technicians for coronal and sagittal knee scan planning, indicating promising clinical applicability of our automated ISP approach.

Yiran Li¹, Danfeng Xie¹, Abigail Cember², Ravi Prakash Reddy Nanga², Hanlu Yang¹, Dushyant Kumar², Hari Hariharan², Li Bai¹, John A. Detre³, Ravinder Reddy², and Ze Wang^4
1Department of Electrical and Computer Engineering, Temple University, Philadelphia, PA, United States, ²Department of Radiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, United States, ³Department of Neurology, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, United States, ⁴Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of Medicine, Baltimore, MD, United States

Glutamate Chemical Exchange Saturation Transfer (GluCEST) MRI is a noninvasive technique for mapping parenchymal glutamate in the brain. GluCEST signal is sensitive to magnetic field (B0) inhomogeneity. Corrections for B0 inhomogeneity often require repeated data acquisitions at several saturation offset frequencies, which however dramatically prolongs the total acquisition time and can cause practical issues such as increased sensitive to patient motions. Another technique challenge in GluCEST MRI is the low signal-to-noise-ratio (SNR) as the signal is derived from the small z-spectrum difference. Both issues were addressed in this study with a novel deep learning-based algorithm armed with wide activation neurons.

Fang Liu^1,2 and Li Feng^3
1Gordon Center for Medical Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States, ²Radiology, University of Wisconsin-Madison, Madison, WI, United States, ³Biomedical Engineering and Imaging Institute and Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States

The purpose of this work was to develop a novel deep learning-based reconstruction framework for rapid MR parameter mapping. Building upon our previously proposed Model-Augmented Neural neTwork with Incoherent k-space Sampling (MANTIS) technique combining efficient end-to-end CNN mapping and k-space consistency to enforce joint data and model fidelity, this new method further extends to incorporate the latest adversarial training (MANTIS-GAN), so that more realistic parameter maps can be directly estimated from highly-accelerated k-space data. The performance of MANTIS-GAN was demonstrated for fast T2 mapping. Our study showed that MANTIS-GAN represents a promising approach for efficient and accurate MR parameter mapping. 

Jae-Hun Lee¹, Kanghyun Ryu¹, Sung-Min Gho², Ho-Sung Kim³, Mohammed A. Al-masni¹, and Dong-Hyun Kim^1
1Deparment of Electrical & Electronic Engineering, Yonsei Univ., Seoul, Korea, Republic of, ²MR Collaboration and Development, GE Healthcare, Seoul, Korea, Republic of, ³Department of Radiology, Asan medical center, Seoul, Korea, Republic of

 Synthetic MRI or magnetic resonance imaging compilation (MAGiC) uses multiple-dynamic multiple-echo acquisition(MDME) and acquires 8 contrast images in a single scan. SENSE or GRAPPA method is conventionally used to reconstruct undersampled acquisition for respective contrast images. However, the method enables limited acceleration up to 2~3. In this study, combined reconstruction method (Joint Parallel Imaging with Deep Learning) is explored. The proposed method shows acceptable image quality with RMSE (4.6%) at the higher acceleration factor (up to 8) comparable to conventional GRAPPA with acceleration rate of 2~3.

Chaoyi Zhang¹, JeeHun Kim², Hongyu Li¹, Peizhou Huang³, Ruiying Liu¹, Dong Liang⁴, Xiaoliang Zhang³, Xiaojuan Li², and Leslie Ying^1,3
1Electrical Engineering, University at Buffalo, SUNY, Buffalo, NY, United States, ²Biomedical Engineering, Program of Advanced Musculoskeletal Imaging (PAMI), Cleveland Clinic, OH, United States, ³Biomedical Engineering, University at Buffalo, SUNY, Buffalo, NY, United States, ⁴Paul C Lauterbur Research Center for Biomedical Imaging, Shenzhen institutes of Advanced Technology, Shenzhen, China

Magnetic resonance parameter mapping (e.g. T1, T2, T2* and T1ρ) has shown potential in quantitative assessment while the clinical applications are limited by long acquisition time especially in 3D acquisition. In our previous work, we use single-exponential model to generate off-line single-exponential training data instead of low resolution training data, which reduced the reconstruction time. In clinical use, when motion is introduced in acquisition, single-exponential model is not satisfied and the reconstruction may fail. With this motivation, this abstract alternatively reconstruct the images and correct motion in 3D parameter mapping.

Kewen Liu¹, Xiaojun Li¹, Qinjia Bao², Chaoyang Liu³, Hongxia Xiong⁴, Zhao Li³, Yuan Ma¹, Panpan Fang¹, and Yalei Chen^1
1School of Information Engineering, Wuhan University of Technology, Wuhan, China, ²United Imaging of Scientific Instruments, Shanghai, China, ³State Key Laboratory of Magnetic Resonance and Atomic and Molecular Physics, Wuhan Institute of Physics and Mathmatics, Innovation Academy for Precision Measurement Science and Technology, Wuhan, China, ⁴School of Civil Engineering & Architecture, Wuhan University of Technology, Wuhan, China

We proposed a novel deep learning network architecture (MEBC-RCAN) for water-fat separation based on multi-echo GRE sequence. The network architecture contains three main components: the first part is Multi-Echo Bidirectional Convolutional (MEBC) to explore the correlations of successive images in multi-echo GRE; the second part is Residual Channel Attention (RCA) network to mimic the iterative optimization in traditional water-fat separation method; and the third part is Multi-Layer Feature Fusion (MLFF) to combine separation information learned from every RCA network. The results show that the proposed network could effectively obtain the high-quality water and fat images from clinical multi-echo GRE data.

Matthew C Murphy¹, Joshua D Trzasko¹, Jonathan M Scott¹, Armando Manduca¹, John Huston, III¹, and Richard L Ehman^1
1Mayo Clinic, Rochester, MN, United States

An artificial neural network (ANN) was trained to estimate the partial derivatives of a spatially varying field, and compared against a finite difference approach. For the application of elastography, training data were generated using a wave equation. After the training examples were corrupted by noise and missing data, the network was trained to estimate the analytical solution to the partial derivatives. In simulation, the ANN improved accuracy in noisy data but blurred sharp boundaries relative to a finite difference method. In vivo, using the ANN to compute the curl of the displacement field improved confidence in subsequent property estimates.

Li Chen¹, Wenjin Liu ¹, Gador Canton ¹, Niranjan Balu ¹, Thomas Hatsukami ¹, John C. Waterton ², Jenq-Neng Hwang ¹, and Chun Yuan ^1
1University of Washington, Seattle, WA, United States, ²Centre for Imaging Sciences, Manchester Academic Health Science Centre, The University of Manchester, Manchester, United Kingdom

Atherosclerotic plaque information can be extracted from MR vessel wall images through transforming the images into a high dimensional feature space. However, a huge amount of human supervision has traditionally been required to achieve a meaningful feature space representation. We demonstrated that by using a weakly supervised deep learning workflow including transfer learning, active learning, and metric learning, a meaningful feature space for vessel wall analysis can be generated, which can help us to visualize the high dimensional representations of normal and diseased vessel walls images, and lead to a plaque classification area under the curve of 0.93.

Shahrokh Abbasi-Rad¹, Kieran O'Brien^1,2,3, Samuel Kelly¹, Viktor Vegh^1,3, Anders Rodell², Yasvir Tesiram¹, Jin Jin^2,3,4,5, Markus Barth^1,3,4, and Steffen Bollmann^1,3
1Centre for Advanced Imaging, University of Queensland, Brisbane, Australia, ²Siemens Healthcare Pty Ltd, Brisbane, Australia, ³ARC Centre for Innovation in Biomedical Imaging Technology, University of Queensland, Brisbane, Australia, ⁴School of Information Technology and Electrical Engineering, University of Queensland, Brisbane, Australia, ⁵Mark and Mary Stevens Neuroimaging and Informatics Institute, University of Southern California, Los Angeles, CA, United States

In 7T MRI adiabatic pulses enable robust inversion of spins at the cost of increased SAR and longer scan times. A convolutional neural network was used to estimate the B₁⁺ profile from a localizer scan, Bloch equation simulations were used to calculate the required B₁⁺ for adiabaticity, and adiabatic pulse power was scaled accordingly reducing SAR by up to 38%. We investigated the robustness and efficiency of this approach and showed a substantial SAR reduction is possible without an additional B1 map acquisition. This resulted in an up to 27% faster T2-FLAIR acquisition with full brain coverage. 

Chin-Cheng Chan¹ and Justin P. Haldar^1
1Electrical and Computer Engineering, University of Southern California, Los Angeles, CA, United States

As MR image reconstruction algorithms become increasingly nonlinear, data-driven, and difficult to understand intuitively, it becomes more important that tools are available to assess the confidence that users should have about image reconstruction results. In this work, we suggest that a quantity known as the “local perturbation response” (LPR) provides useful information that can be used for this purpose. The LPR is analogous to a conventional point-spread function, but is well-suited to general image reconstruction methods that may have nonlinear and/or shift-varying characteristics. We illustrate the LPR in the context of several common image reconstruction techniques.

Elizabeth Cole¹, Frank Ong¹, John Pauly¹, and Shreyas Vasanawala^2
1Electrical Engineering, Stanford University, Stanford, CA, United States, ²Radiology, Stanford University, Stanford, CA, United States

Many deep learning-based reconstruction methods require fully-sampled ground truth data for supervised training. However, instances exist where acquiring fully sampled data is either difficult or impossible, such as in dynamic contrast enhancement (DCE), 3D cardiac cine, 4D flow, etc. for training a reconstruction network. We present a deep learning framework for reconstructing MRI without using any fully sampled data. We test the method in two scenarios, and find the method produces higher quality images which reveal vessels and recover more anatomical structure. This method has potential in applications, such as DCE, cardiac cine, low contrast agent imaging, and real-time imaging.

Yue Guan¹, Yudu Li^2,3, Ziyu Meng^3,4, Tianyao Wang⁵, Rong Guo^2,3, Ruihao Liu⁴, Yao Li⁴, Yiping Du⁴, and Zhi-Pei Liang^2,3
1Institute for Medical Imaging Technology, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ⁴Shanghai Jiao Tong University, Shanghai, China, ⁵Department of Radiology, The Fifth People's Hospital of Shanghai, Shanghai, China

MPRAGE imaging has been widely used in clinical applications and various attempts have been made for its acceleration. This paper presents a new method to accelerate MPRAGE imaging using sparse and random sampling of k-space and constrained reconstruction incorporating image priors and subject-specific novel features. In our current implementation, the MPRAGE image priors were obtained using deep learning on data from the Human Connectome Project, and novel localized features were recovered by solving a sparsity-constrained reconstruction. In vivo experimental results demonstrated that the proposed method can produce high-quality whole-brain MPRAGE images in 0.7x0.7x0.7 mm3 nominal resolution from a 1.5-min scan.