Fan Yang¹, Jian Zhang², Guobin Li², Jiayu Zhu², Xin Tang¹, and Chenxi Hu^1
1Institute of Medical Imaging Technology, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²United Imaging Healthcare Co., Ltd, Shanghai, China

Three-dimensional variable-flip-angle (VFA) T1 mapping is an accurate T1 quantification method suffering from long scan time. SUPER is a contrast-domain acceleration technique with strength in noise suppression and fast reconstruction. Here, we develop regularized SUPER-CAIPIRINHA to accelerate 3D VFA T1 mapping up to 16-fold with 10 flip angles, and perform fast reconstruction with the proposed proximal-Levenberg Marquardt algorithm. The novel method is validated with both retrospective and prospective experiments, compared to Locally Low Rank and Compressed Sensing. The results show that rSUPER-CAIPIRINHA is an accurate and computationally efficient technique, reducing the 3D scan time from 14:10 minutes to 0:59 minutes.

Mojtaba Shafiekhani¹, Vahid Ghodrati², and Abbas Nasiraei-Moghaddam^1
1biomedical engineering, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran (Islamic Republic of), ²Department of Radiological Sciences, University of California Los Angeles, Los Angeles, CA, United States

The two- and three-dimensional radial trajectory has been widely used due to its lower sensitivity to movement and flow and also shorter acquisition time. Re-gridding, the dominant approach for radial data reconstruction, suffers from severe artifacts at high acceleration rates. This study proposes a mathematical approach based on spherical harmonics to reconstruct images in the spherical coordinates without any interpolation in the frequency domain, unlike the conventional re-gridding algorithm. The feasibility of the proposed algorithm for reconstructing the images on both 3D Shepp-Logan and brain digital phantoms was shown. 

Ruihao Liu¹, Yudu Li^2,3, Ziyu Meng¹, Yue Guan¹, Ziwen Ke¹, Tianyao Wang⁴, Yao Li¹, Yiping P. Du¹, and Zhi-Pei Liang^2,3
1School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ⁴Radiology Department, The Fifth People's Hospital of Shanghai, Shanghai, China

Quantitative T2 maps of brain tissues are useful for diagnosis and characterization of a number of diseases, including neurodegenerative disorders. This work presents a new learning-based method for the reconstruction of T2 maps from highly sparse k-space data acquired in accelerated T2-mapping experiments. The proposed method synergistically integrates generalized series modeling with deep learning, which effectively captures the underlying signal structures of T2-weighted images with variable TEs and a priori information from prior scans (e.g., data from the Human Connectome Project). The proposed method has been validated using experimental data, producing improved brain T2 maps over the state-of-the-art methods.

Chenxi Hu¹, Fan Yang¹, Xin Tang¹, Zhiyong Zhang¹, and Dana Peters^2
1Shanghai Jiao Tong University, Xuhui, China, ²Yale University, New Haven, CT, United States

The Locally Low-Rank (LLR) constraint has been increasingly used for MR acceleration. Here we compare two strategies for LLR-constrained reconstruction, namely the Non-overlapped LLR (NLLR) and the Densely-overlapped LLR (DLLR) to show their differences. The NLLR strategy has been used by a number of LLR algorithms, including the most well-known POCS algorithm. On the other hand, the DLLR strategy has not been well-recognized as a different strategy, and algorithms able to employ the strategy have only been developed recently. In this work, we show that DLLR is different and superior to NLLR by yielding faster convergence and reduced undersampling artifacts.

Qingyong Zhu¹, Jing Cheng², Zhuo-Xu Cui¹, Yuanyuan Liu³, Yanjie Zhu², and Dong Liang^1
1Research Center for Medical AI, SIAT, Chinese Academy of Sciences, Shenzhen, China, ²Paul C. Lauterbur Research Center for Biomedical Imaging, SIAT, Chinese Academy of Sciences, Shenzhen, China, ³National Innovation Center for Advanced Medical Devices, Shenzhen, China, Shenzhen, China

The existing accelerated MRI methods have the limitations such as fine-structure loss in the cases of high reduction factors or noisy measurements. This work presents a novel mutual-structure filtering based half-quadratic splitting algorithm for accurate undersampled brain MRI reconstruction. Experimental results on in-vivo images have shown that the proposed approach has a superior ability to capture meaningful details compared with other state-of-the-art reference guided reconstruction technologies.

Haonan Zhang¹, Qingwei Song¹, Ailian Liu¹, and Geli Hu^2
1Department of Radiology, the First Affiliated Hospital of Dalian Medical University, Dalian, Dalian, China, ²Philips Healthcare, Beijing, China, Beijing, China

Lumbar magnetic resonance imaging has a significant diagnostic value for spine lesions. Reducing scan time is critical to improve patient compliance and comfort. The purpose of this study aims to introduce a combination of compressed-sensing and parallel imaging for high acceleration and motion artifact reduction for the lumbar spine MRI, and to determine an optimal acceleration factor, meanwhile, provide high quality images .

Ziwen Ke¹, Yue Guan¹, Yudu Li², Yunpeng Zhang¹, Tianyao Wang³, Ziyu Meng¹, Ting Zhao¹, Yujie Hu¹, Ruihao Liu¹, Huixiang Zhuang¹, Zhi-Pei Liang², and Yao Li^1
1School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Department Radiology, The Fifth People’s Hospital of Shanghai, Fudan University, Shanghai, China

Fast imaging is essential in neonatal brain MRI. Deep learning-based methods can provide high acceleration rates but their performance is instable when limited training data are available. Subspace model-based approach could reduce the dimensionality of imaging and improve reconstruction stability. This work presents a novel method to integrate neonate-specific subspace model and model-driven deep learning, making stable and ultrafast neonatal MR imaging possible. The feasibility and potential of the proposed method have been demonstrated using in vivo data from four medical centers, producing very encouraging results. With further development, the proposed method may provide an effective tool for neonatal imaging. 

Chen Qian¹, Zi Wang¹, Xinlin Zhang¹, Boxuan Shi¹, Di Guo², Boyu Jiang³, Ran Tao³, and Xiaobo Qu^1
1Department of Electronic Science, Biomedical Intelligent Cloud R&D Center, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, China, ²School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China, ³United Imaging Healthcare, Shanghai, China

Diffusion-weighted imaging (DWI) is widely employed in clinical diagnosis and white matter connections mapping. Multi-shot interleaved echo planer imaging (ms-iEPI) is used for acquiring images with higher spatial resolution and fewer distortion but suffers from motion-induced shot-phase variations. In this work, a joint reconstruction model is proposed to obtain the shot-phase and composite magnitude image simultaneously. Specifically, the smoothness of the shot-phase is exploited to construct low-rank constraints. Experiment results show that the proposed method has better performance than state-of-the-art methods on shot-phase estimation and image reconstruction.

Zhao He¹, Ya-Nan Zhu², Yuchen He², Yu Chen¹, Suhao Qiu¹, Linghan Kong¹, Xiaoqun Zhang², and Yuan Feng^1
1School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, ²School of Mathematical Sciences, MOE-LSC and Institute of Natural Sciences, Shanghai Jiao Tong University, Shanghai, China

Interventional MRI (i-MRI) needs fast data acquisition and image reconstruction. We have shown that a Low-rank and Sparsity decomposition with Framelet transform model with Primal dual fixed point optimization (LSFP) could satisfy the reconstruction of real-time i-MRI. In this study, we unrolled the LSFP into a deep neural network, dubbed LSFP-Net, with multi-coil golden-angle radial sampling. Simulation results showed that LSFP-Net outperformed the state-of-the-art methods, and a phantom experiment demonstrated its potential for real-time i-MRI.

Yue Guan¹, Yudu Li^2,3, Ruihao Liu¹, Ziyu Meng¹, Yao Li¹, Leslie Ying^4,5, Yiping P. Du¹, 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, ⁴Department of Biomedical Engineering, The State University of New York at Buffalo, Buffalo, NY, United States, ⁵Department of Electrical Engineering, The State University of New York at Buffalo, Buffalo, NY, United States

Deep learning-based image reconstruction methods are often sensitive to changes in data acquisition settings (e.g., sampling pattern and number of encodings). This work proposes a novel subspace-assisted deep learning method to effectively address this problem. The proposed method uses a subspace model to capture the global dependence of image features in the training data and a deep network to learn the mapping from a linear vector space to a nonlinear manifold. Significant improvement in reconstruction accuracy and robustness by the proposed method has been demonstrated using the fastMRI dataset.

Yuxuan Liu¹, Yongsheng Pan¹, Mancheng Meng¹, and Haikun Qi^1
1School of Biomedical Engineering, ShanghaiTech University, Shanghai, China

A cascaded hybrid domain generative adversarial network is proposed for accelerated MRI reconstruction. A novel multi-scale feature fusion sampling layer is proposed to replace the pooling layers and upsampling layers in the U-Net k-space generator to better recover the missing samplings. The proposed method is extensively validated with low and high acceleration factors against several state-of-the-art reconstruction methods, and achieves competitive reconstruction performance.

Kamlesh Pawar^1,2, Gary F Egan^1,2,3, and Zhaolin Chen^1,4
1Monash Biomedical Imaging, Monash University, Melbourne, Australia, ²School of Psychological Sciences, Monash University, Melbourne, Australia, ³ARC Centre of Excellence for Integrative Brain Function, Monash University, Melbourne, Australia, ⁴Department of Data Science and AI, Faculty of Information Technology, Monash University, Melbourne, Australia

Deep learning (DL) models for accelerated image reconstruction involves retrospective undersampling of the fully sampled k-space data for training and validation. This strategy is not a true reflection of real-world data and in many instances, the input k-space data is corrupted with artifacts and errors, such as motion artifacts. In this work, we propose to improve existing methods of DL training and validation by incorporating a motion layer during the training process. The incorporation of a motion layer makes the DL model aware of the underlying motion and results in improved image reconstruction in the presence of motion.

Jiahao Hu^1,2,3, Zheyuan Yi^1,2,3, Yujiao Zhao^1,2, Christopher Man^1,2, Linfang Xiao^1,2, Vick Lau^1,2, Shi Su^1,2, Ziming Huang^1,2, Junhao Zhang^1,2, Alex T.L. Leong^1,2, Fei Chen³, and Ed X. Wu^1,2
1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, ²Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China, ³Electrical and Electronic Engineering, Southern University of Science and Technology, Shenzhen, China

Low-rank matrix completion has emerged as a potent reconstruction approach for calibrationless parallel imaging. However, in all existing low-rank reconstruction methods, the rank threshold must be carefully chosen slice by slice in a manual and trial-and-error manner, severely hindering the adoption of low-rank reconstruction in routine clinical applications. To tackle the problem, we proposed a fast and automatic rank determination via deep learning. It directly determines optimal rank from undersampled k-space data by exploiting coil sensitivity and finite image support.  Our proposed method enables fast, automatic and robust rank determination for all existing calibrationless reconstruction using low-rank matrix completion.

Lihong Tang¹, Yibo Zhao^2,3, Yudu Li^2,3, Rong Guo^2,3, Bingyang Cai¹, Yao Li¹, Zhi-Pei Liang^2,3, and Jie Luo^1
1School 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 Sciences and Technology, University of Illinois at Urbana Champaign, Urbana, Urbana, IL, United States

Accurate estimation of coil sensitivity functions is essential for SENSE-based image reconstruction. This paper presents a new method for joint estimation of coil sensitivity functions and image reconstruction from sparsely sampled k-space data without any auto-calibration data. The proposed method uses a probabilistic subspace model to capture statistical spatial distributions of a given coil system from prior scan data. The proposed method has been validated using experimental data (fastMRI dataset), producing high-quality reconstruction results from calibrationless sparse data (acceleration factor R = 8 or R = 16). The proposed method may further enhance the practical utility of parallel imaging.

Jieying Zhang¹, Simin Liu¹, Erpeng Dai², Xinyu Ye¹, Yang Yu^3,4, Yajing Zhang⁵, Jie Lu^3,4, and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ²Department of Radiology, Stanford, CA, United States, ³Department of Radiology and Nuclear Medicine, Xuanwu Hospital, Capital Medical University, Beijing, China, ⁴Beijing Key Laboratory of Magnetic Resonance Imaging and Brain Informatics, Beijing, China, ⁵MR Clinical Science, Philips Healthcare, Suzhou, China

Simultaneous multi-slab imaging (SMSlab) can achieve high-resolution DWI with high SNR efficiency, but image distortions remain a problem. Joint reconstruction of blip-up/down EPI data can correct for distortion and reduce g-factors simultaneously. This study proposed a joint reconstruction framework with intrinsic distortion correction for SMSlab. Three sampling patterns were compared in terms of levels of distortions and g-factors. According to the results, a multi-shot acquisition should be used to reduce the distortion. Combined with the controlled aliasing in parallel imaging (CAIPI) sampling pattern, it can further reduce the g-factors.