Xueying Zhao¹, Pierre-Francois Van de Moortele², Ying-Hua Chu³, and He Wang^1,4
1Institute of Science and Technology for Brain-inspired Intelligence, Fudan University, Shanghai, China, ²Center for Magnetic Resonance Research, University of Minnesota, minneapolis, MN, United States, ³MR Collaboration, Siemens Healthineers Ltd., Shanghai, China, Shanghai, China, ⁴Human Phenome Institute, Fudan University, Shanghai, China

Inspired by the success of low-rank and sparsity (L+S) representation in 4D flow MRI acceleration of great vessels (e.g. hearts or aortic), we managed to implement a generalized L+S reconstruction framework for cerebrovascular 4D flow at 7T. The iterative hard-thresholding approach was adopted instead of the popular used soft-thresholding method because we found soft-thresholding underestimated the reconstructed phase value while hard-thresholding well preserved the phase information. The L+S reconstructed magnitude and phase value at an acceleration factor of 10 is comparable to the fully sampled reference and represents less errors compared with other compressed sensing based approaches.  

Ruoyou Wu¹, Chen Hu¹, Cheng Li¹, Wenxin Fan¹, Hairong Zheng¹, and Shanshan Wang^1
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Self-supervised deep learning for MR reconstruction has shown high potential in accelerating MR imaging as it doesn’t need fully sampled dataset for model training.  However, the performances of the current self-supervised methods are limited as they don’t take full utilization of the available under-sampled data. We propose a physics-based data-augmented deep learning method to enable faster and more accurate parallel MR imaging.  Novel augmenting losses are calculated, which can effectively constrain the model optimization with better utilization of the collected dataset. Extensive experiments are conducted, and better reconstruction results are generated by our method compared to the current state-of-the-art methods.

Uten Yarach¹, Suwit Saekho¹, Kittichai Wantanajittikul¹, Salita Angkurawaranon², Chakri Madla², Charuk Hanprasertpong³, and Prapatsorn Sangpin^4
1Department of Radiologic Technology, Faculty of Associated Medical Science, Chiang Mai University, Chiang Mai, Thailand, ²Department of Radiology, Faculty of Medicine, Chiang Mai University, Chiang Mai, Thailand, ³Department of Otolaryngology, Faculty of Medicine, Chiang Mai University, Chiang Mai, Thailand, ⁴Philips (Thailand) Ltd., Bangkok, Thailand

Propeller FSE-DWI has been a method of choice in particular for Cholesteatoma. However, its natures such as prolong scan time, low signal-to-noise ratio (SNR), and phase variations among the blades have been challenging. Parallel imaging technique can be applied to reduce scan time, but phase estimation/correction is often compromised due to g-factor noise penalty. In this work, we develop an iterative reconstruction with locally low rank (LLR) regularization to maximize quality of propeller FSE-DWI images. As a result, LLR enable improving SNR up to SENSE factor of 4 compared to images obtained by vendor’s provided reconstruction engine.

Shihui Chen¹, Chia-Wei Lee², and Hing-Chiu Chang^1,3
1Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong, Hong Kong, ²GE Healthcare, Taiwan, Taiwan, ³Department of Biomedical Engineering, The Chinese University of Hong Kong, Hong Kong, Hong Kong

High-resolution 4-shot multi-echo resting-state fMRI (RS-fMRI) can be achieved by acquiring data in a sliding window manner, and thus the composite full k-space for MUSE reconstruction can be obtained by combining the segments from the consecutive time points. However, the sliding window operation may smooth the time courses and lead to a similar effect as applying a moving average filter. In this study, we aimed to investigate the effect of k-space data sharing from consecutive time points acquired with different TRs, and to determine a feasible TR for the acquisition and reconstruction scheme for high-resolution 4-shot multi-echo RS-fMRI with MUSE.

Yuekai Sun¹, Jun Lv¹, Weibo Chen², He Wang^3,4, and Chengyan Wang^4
1School of Computer and Control Engineering, Yantai University, Yantai, China, ²Philips Healthcare, Shanghai, China, ³Institute of Science and Technology for Brain-inspired Intelligence, Fudan University, Shanghai, China, ⁴Human Phenome Institute, Fudan University, Shanghai, China

To solve the problem that fully-sampled reference data is difficult to acquire, we proposed a self-supervised triple-network (SSTN) for fast MRI reconstruction. Each pipeline of SSTN is composed of multiple parallel ISTA-Net blocks which consists of different scales dilated convolution layers. The results demonstrated that the proposed SSTN can generate better quality reconstructions than competing methods at high acceleration rates.

Xinyu Ye¹, Yuan Lian¹, Hai Luo ², Ziyue Wu ², and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ²Marvel Stone Healthcare Co., Ltd.,, Wuxi, China

Owing to hardware advancements, a resurgence of interest has emerged in low field MRI systems recently. However, the intrinsic low signal to noise ratio (SNR) due to the low magnetic field strength has hindered the acquisition of high-resolution images on low field MRI systems. Here we propose a non-local robust PCA-based method to jointly denoise multi-contrast images acquired on a 0.5 T MRI system. In-vivo brain data are used to test the proposed method. The results show that the proposed method outperforms BM3D and BM4D denoising methods.

Wenjing Xu¹, Sen Jia¹, Qing Zhu², Yikang Li³, Hongying Zhang⁴, Shuai Shen¹, Fuliang Lin¹, Ye Li¹, Dong Liang¹, Xin Liu¹, Hairong Zheng¹, and Na Zhang^1
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²Faculty of Information Technology, Beijing University of Technology, Beijing, China, ³Department of computing，Imperial College London, London, United Kingdom, ⁴Department of Radiology, Northern Jiangsu People's Hospital, Jiangsu, China

To develop a super-resolution method based on the 3D high-resolution MR vessel wall images for generating high-resolution images from low-resolution, a 3D complex-valued super resolution (CVSR) neural network was proposed, which maintained complex algebraic structure of the original acquired images. CVSR was trained on 20 pairs of data sets and tested on 5 pairs. Ground truth with 0.44 mm were compared with Fourier interpolation method, EDSR with two real-valued channels and CVSR. Evaluations were performed using structural similarity (SSIM), peak signal-to-noise ratio (PSNR), and error map quality metrics. The CVSR achieved the best performance when compared with the other methods.

Xiaorui Xu¹, Liyuan Liang¹, Xiaoxi Liu², and Hing-Chiu Chang^3
1Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong, Hong Kong, ²Department of Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States, ³Department of Biomedical Engineering, The Chinese University of Hong Kong, Hong Kong, Hong Kong

3D multi-slab multi-shot diffusion-weighted EPI can enable high-resolution diffusion-tensor imaging (DTI). Furthermore, in order to avoid the slab boundary artefacts, 3D-MUSER was proposed to enable 3D phase correction for effectively reconstructing nearly whole brain 3D DTI data acquired with only a single slab. Compressed sensing (CS) was also combined with 3D-MUSER to further reduce the scan time, but the previously proposed pseudo-random sampling strategy was still suboptimal for CS reconstruction. Therefore, we propose a novel variable-density k-space sampling strategy that is compatible with 3D-MUSER and able to achieve highly-accelerated and high-quality 3D DTI by taking the full advantage of CS.