Xinyu Ye¹, Pylypenko Dmytro¹, Yuan Lian¹, Yajing Zhang², and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ²MR Clinical Science, Philips Healthcare, Suzhou, China

In clinical scans, the acquired DWI images usually has limited resolution. Super resolution method has the potential to improve the image resolution without adding scan time. Here we propose a deep-learning based multi-contrast super resolution network with gradient-map guidance and a novel FA loss function to reconstruct high-resolution DWI images from low-resolution DWI images and high–resolution anatomical images. In-vivo DWI data are used to test the proposed method. The results show that the image quality can be improved.   

Tianshu Zheng¹, Cong Sun², Guangbin Wang², Weihao Zheng¹, Wen Shi¹, Yi Sun³, Yi Zhang¹, Chuyang Ye⁴, and Dan Wu^1
1Key Laboratory for Biomedical Engineering of Ministry of Education, Department of Biomedical Engineering, College of Biomedical Engineering & Instrument Science, Zhejiang University, Hangzhou, Zhejiang, China,, Zhengjiang University, Hangzhou, China, ²Department of Radiology, Shandong Medical Imaging Research Institute, Cheeloo College of Medicine, Shandong University, 324, Jingwu Road, Jinan, Shandong, 250021, People's Republic of China, Shandong University, Jinan, China, ³Department of Radiology 2MR Collaboration, Siemens Healthcare China, Shanghai, China, Siemens Healthcare China, Shanghai, China, ⁴chool of Information and Electronics, Beijing Institute of Technology, Beijing Institute of Technology, Beijing, China

Intravoxel incoherent motion (IVIM) can be used to assess microcirculation in the brain, however, conventional IVIM requires long acquisition to obtain multiple b-values, which is challenging for fetal brain MRI due to excessive motion. Q-space learning helps to accelerate the acquisition but it is hard to be interpreted. In this study, we proposed a sparsity coding deep neural network (SC-DNN), which is a model-driven network based on sparse representation and unfold the parameter optimization process. Compared to conventional IVIM fitting, SC-DNN took only 50% of the data to reach the comparable accuracy for parameter estimation, which outperformed the multilayer perceptron.

Yujiao Zhao^1,2, Linfang Xiao^1,2, Zhe Zhang³, Yilong Liu^1,2, Hua Guo⁴, and Ed X. Wu^1,2
1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, ²Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China, ³China National Clinical Research Center for Neurological Diseases, Beijing Tiantan Hospital, Capital Medical University, Beijing, China, ⁴Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China

Diffusion MRI intrinsically suffers from low signal-to-noise ratio (SNR), especially when spatial resolution or b-value is high. A typical diffusion MRI scanning session produces image sets with same geometries but different diffusion directions and b-values, thus these diffusion-weighted (DW) images often share strong structural similarities. In this study, we developed a joint denoising method for DW images based on low-rank matrix approximation. This denoising method exploits structural similarities of DW image set. Both simulation and in vivo brain experiments demonstrate significant noise reduction in all DW images, revealing more microstructural details in quantitative diffusion maps.

Hongyu Li¹, Zifei Liang², Chaoyi Zhang¹, Ruiying Liu¹, Jing Li³, Weihong Zhang³, Dong Liang⁴, Bowen Shen⁵, Peizhou Huang⁶, Sunil Kumar Gaire¹, Xiaoliang Zhang⁶, Yulin Ge², Jiangyang Zhang², and Leslie Ying^1,6
1Electrical Engineering, University at Buffalo, State University of New York, Buffalo, NY, United States, ²Center for Biomedical Imaging, Radiology, New York University School of Medicine, New York, NY, United States, ³Radiology, Peking Union Medical College Hospital, Peking Union Medical College and Chinese Academy of Medical Sciences, Beijing, China, ⁴Paul C. Lauterbur Research Center for Biomedical Imaging, Medical AI research center, SIAT, CAS, Shenzhen, China, ⁵Computer Science, Virginia Tech, Blacksburg, VA, United States, ⁶Biomedical Engineering, University at Buffalo, State University of New York, Buffalo, NY, United States

The main factor that prevents diffusion tensor imaging (DTI) from being incorporated in clinical routines is its long acquisition time of a large number of diffusion-weighted images (DWIs) required for reliable tensor estimation. This abstract presents SuperDTI to learn the nonlinear relationship between DWIs (reduced in q-space and k-space) and the corresponding tensor-derived quantitative maps as well as fiber tractography. Experimental results show that the proposed method can generate fractional anisotropy and mean diffusivity maps, as well as fiber tractography, from as few as six undersampled raw DWIs with quality comparable to results from 90 DWIs using conventional tensor fitting.

lei hu¹, jungong Zhao^1,2, Caixia fu³, and Thomas Benkert^4
1Department of Diagnostic and Interventional Radiology, Shanghai Jiao Tong University Affiliated Sixt, 上海, China, ²Department of Diagnostic and Interventional Radiology, Shanghai Jiao Tong University Affiliated Sixt, shanghai, China, ³MR Application Development, Siemens Shenzhen magnetic Resonance Ltd, shanghai, China, ⁴MR Application Predevelopment, Siemens Healthcare, Erlangen, Gernmany, Erlangen, Germany

A deep learning framework based on a generative adversarial network (GAN) to synthesize high-b-value DWI (syn-DWIb1500) with high quality using the acquired standard b-value DWI (a-DWIb800-1000) was developed. Reader ratings for image quality and PCa detection were performed on the a-DWI b1500, syn-DWIb1500, and optimized syn-DWIb1500 sets. Wilcoxon signed-rank tests and MRMC-ROC were used to compare the readers’ scores and diagnostic capabilities of each DWI set, respectively. Optimized syn-DWIb1500 resulted in significantly better image quality (all P≤0.001) and a higher mean AUC than a-DWIb1500 and cal-DWIb1500 (all P≤0.042).

Tianyu Zhang¹, Yishi Wang², Chengxiu Yuan¹, Xiaoyu Wang¹, Jia Zhao¹, and Huaqiang Sheng^1
1The First Affiliated Hospital of Shandong First Medical University, Jinan, China, ²Philips Healthcare, Beijing, China

The aim of the study was to compare intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) for evaluating lung cancer using single-shot gradient- and spin-echo (SS-GRASE), single-shot turbo spin-echo (SS-TSE) and single-shot echo-planar imaging (SS-EPI) on a 3.0T MRI scanner. The signal-to-noise ratio (SNR), contrast signal-to-noise ratio (CNR), image distortion of lesions, ADC and IVIM parameters from the three sequences were compared. We found that GRASE-IVIM had the characteristics of short scan time and small distortion, relatively low SNR but high CNR. Our study showed that GRASE-IVIM has great potential for lung cancer.

Sofie Rahbek¹, Tim Schakel², Faisal Mahmood^3,4, Kristoffer H. Madsen^5,6, Marielle E.P. Philippens², and Lars G. Hanson^1,5
1Department of Health Technology, Technical University of Denmark, Kgs. Lyngby, Denmark, ²Department of Radiotherapy, University Medical Center Utrecht, Utrecht, Netherlands, ³Laboratory of Radiation Physics, Odense University Hospital, Odense, Denmark, ⁴Department of Clinical Research, University of Southern Denmark, Odense, Denmark, ⁵Danish Research Centre for Magnetic Resonance, Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark, ⁶Department of Applied Mathematics and Computer Science, Technical University of Denmark, Kgs. Lyngby, Denmark

The RARE-based diffusion-weighted MRI sequence, SPLICE, provides diffusion-weighted images free of geometric distortions. However, the image quality in terms of spatial resolution and SNR needs to be improved. We suggest a framework based on optimization of the flip-angles in the refocusing pulse train to improve the SNR for a chosen spatial point-spread-function. Simulations based on EPG calculations indicate superiority of the optimized flip-angle scheme and experimental recordings on a 1.5 T scanner demonstrate the improvements in a practical setting.  

John Neri¹, Matthew F Koff¹, Kevin M Koch², and Ek Tan^1
1Radiology and Imaging, Hospital for Special Surgery, New York, NY, United States, ²Medical College of Wisconsin, Wauwatosa, WI, United States

Diffusion-weighted MAVRIC is a novel pulse sequence that can obtain quantitative diffusion values in regions with high susceptibility. The goal of this work is to compare the quantitative accuracy of DWI-MAVRIC with sequences currently available on clinical MR scanners, using a diffusion phantom and images acquired at different orientation and spatial offsets in the scanner. Overall, good linearity with diffusivity was observed, with a large positive ADC bias at low diffusivity in both axial and coronal DWI-MAVRIC noted. Right and left spatial offsets increased ADC errors that can be mitigated by gradient nonlinearity correction.

Volkan Emre Arpinar^1,2, Alexander D Cohen¹, Sampada Bhave³, and Kevin M Koch^1,2
1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States, ²Center for Imaging Research, Medical College of Wisconsin, Milwaukee, WI, United States, ³Canon Medical Research USA, Cleveland, OH, United States

Diffusion weighted imaging(DWI) is a workhorse sequence for many clinical and research applications of MRI.  Beyond the role in neuroimaging, DWI has also been deployed for use other anatomies outside the head. However, in some anatomies with air-tissue susceptibility boundaries or proximity to metallic implants, DWI assessment may be challenging due to substantial susceptibility artifacts. In this study, the diffusion quantification accuracy of several DWI sequences within increasing resilience to susceptibility artifacts were compared using an established DWI phantom. In addition, a controlled test of two sequences with substantial differences in susceptibility artifact reduction was performed on spinal cord DWI.

Fanwen Wang¹, Hui Zhang¹, Fei Dai¹, Weibo Chen², Chengyan Wang³, and He Wang^1,3
1Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China, ²Philips Healthcare, Shanghai, China, ³Human Phenome Institute, Fudan University, Shanghai, China

Compared with the standard imaging method, i.e. single-shot echo planar imaging for diffusion MRI, Multi-Shot EPI (MS-EPI) is featured with high spatial resolution, though suffering from longer scanning time and severer phase variations. This study proposed a novel method to reconstruct four-shot high-resolution DWIs from one-shot data for multiple b-values simultaneously, enabling the physiological feature transformation through different b-values. Trained on healthy volunteers, we succeeded in recovering the aliasing images which violates the Nyquist-Shannon theorem from one-shot DWI both on healthy people and tumor patients, showing great clinical generalization.

Hiroshi Hamano¹, Masami Yoneyama¹, Yasutomo Katsumata², Kazuhiro Katahira³, and Kenji Iinuma^1
1Philips Japan, Tokyo, Japan, ²Philips Healthcare, Tokyo, Japan, ³Department of Radiology, Kumamoto Chuo Hospital, Kumamoto, Japan

DWIBS based on single shot echo-planar imaging (EPI) is a first-choice sequence in routine clinical examinations. However, it sometimes suffers from sever image distortion due to the presence of air within and/or at edge of the FOV, especially at the border of chest and abdomen. On the other hand, image based B0 shimming and blip-up blip-down distortion correction improved for image qualities of DWI.  We demonstrated that whole body DWIBS applied both image based B0 shimming and blip-up blip-down distortion correction to provide higher robustness.

Feihong Liu^1,2, Junwei Yang^2,3, Zhiming Cui^2,4, Xiaowei He^1,5, Jun Feng^1,5, and Dinggang Shen^2,6,7
1School of Information Science and Technology, Northwest University, Xi'an, China, ²School of Biomedical Engineering, ShanghaiTech University, Shanghai, China, ³Department of Computer Science and Technology, University of Cambridge, Cambridge, United Kingdom, ⁴Department of Computer Science, The University of Hong Kong, Hong Kong, China, ⁵State-Province Joint Engineering and Research Center of Advanced Networking and Intelligent Information Services, Xi'an, China, ⁶Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China, ⁷Department of Artificial Intelligence, Korea University, Seoul, Korea, Republic of

We cautiously analyze phase correction procedures, unveiling why they falsely introduce dark holes to diffusion-weighted images, and calibrate them with the goal of unbiased white matter microstructure estimation.   The calibrated procedures properly rotate image contents to be negative, hence, dark hole issue is effectively avoided.

Nastaren Abad¹, Luca Marinelli¹, Radhika Madhavan¹, James Kevin DeMarco², Robert Y Shih^2,3, Vincent B Ho^2,3, Gail Kohls², and Tom K.F Foo^1
1General Electric Global Research, Niskayuna, NY, United States, ²Walter Reed National Military Medical Center, Bethesda, MD, United States, ³Uniformed Services University of the Health Sciences, Bethesda, MD, United States

In order to establish a benchmark for future studies, we utilize a data driven approach towards optimizing diffusion sampling for an ultra-high-performance gradient MRI sub-system as a means to establish minimal discrepancy compared to a fully sampled, defined superset. This study focused on b-value contribution and the impact of decreased sampling on uncertainty of diffusion and kurtosis tensor estimates, and fiber orientation to resolve sub-voxel information.

Shihui Chen¹ and Hing-Chiu Chang^1
1Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong, Hong Kong

High angular resolution diffusion imaging (HARDI) is a useful tool for neuroscience research, but the widespread clinical applications are limited by its long scan time and low spatial resolution. Multiple strategies have been proposed to achieve in-plane acceleration to improve the scan efficiency and geometric fidelity for HARDI. However, the feasible in-plane acceleration factor is still limited by the number of coils and the noise amplification associated with parallel imaging reconstruction. In this study, we proposed a reconstruction method based on SVD for HARDI to achieve superior reconstruction performance, even when acceleration factor is greater than number of coils.

Gabriel Ramos-Llordén¹, Rodrigo A. Lobos², Tae Hyung Kim¹, Qiyuan Tian¹, Slimane Tounetki¹, Thomas Witzel³, Boris Keil⁴, Anatasia Yendiki¹, Berkin Bilgic^1,5, Justin P. Haldar², and Susie Huang^1,5
1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Masachusetts General Hospital, Harvard Medical School, Charlestown, MA, United States, ²Ming Hsieh Department of Electrical and Computer Engineering, University of Southern California, Los Angeles, CA, United States, ³Q Bio Inc, San Carlos, CA, United States, ⁴Institute of Medical Physics and Radiation Protection, Mittelhessen University of Applied Sciences, Giessen, Germany, ⁵Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of Technology, Cambridge, MA, United States

Multi-shot EPI diffusion MRI acquired using high diffusion-encoding gradient strengths suffers from severe ghosting artifacts, which can bias and confound the estimation of diffusion microstructural MRI measures at high b-values. In this work, we show that conventional EPI ghost correction techniques fall short in ghosting reduction when high diffusion-encoding gradient strengths ~250mT/m are used, and that advanced reconstruction algorithms based on structured low-rank matrix modeling  can substantially reduce ghosting without introducing additional artifacts.  

Arun Venkataraman¹, Benjamin Risk², Deqian Qiu^3,4, Jianhui Zhong^1,5, Feng (Vankee) Lin^6,7, and Zhengwu Zhang^8
1Physics and Astronomy, University of Rochester, Rochester, NY, United States, ²Biostatistics and Bioinformatics, Emory University, Atlanta, GA, United States, ³Radiology and Imaging Sciences, Emory University School of Medicine, Atlanta, GA, United States, ⁴Biomedical Engineering, Emory University, Atlanta, GA, United States, ⁵Imaging Sciences, University of Rochester Medical Center, Rochester, NY, United States, ⁶Brain and Cognitive Sciences, University of Rochester, Rochester, NY, United States, ⁷Neuroscience, University of Rochester Medical Center, Rochester, NY, United States, ⁸Biostatistics and Computational Biology, University of Rochester Medical Center, Rochester, NY, United States

We sought to understand the implications of slice and phase acceleration factors on diffusion MRI (dMRI) data quality. As part of an on-going study about the impact of acceleration on dMRI quality, reconstruction, and data analysis, we acquired data from young and old, healthy subjects as well as mild cognitive impairment (MCI) subjects. From our study results, we found that there appears to be a trade-off between SNR loss due to higher acceleration and SNR gain from reducing the impact/magnitude of motion. Future studies should examine how the costs and benefits of acceleration impact diffusion metrics, tractography, and reproducibility.

Zifei Liang¹ and Jiangyang Zhang^1
1Center for Biomedical Imaging, Dept. of Radiology, New York University School of Medicine, NEW YORK, NY, United States

Deep-learning/machine-learning based super-resolution techniques have shown promises in improving the resolution of MRI without additional acquisition. In this study, we examined the capability of deep-learning based super-resolution using a newly developed network at resolutions from 0.2 mm to 0.025 mm. We also investigated whether the networks were able to enhance data acquired with a different contrast. Our results demonstrated that the enhancement of deep learning based super-resolution, although better than cubic interpolation, remained limited. In order to achieve the best performance, the network needs to be trained using data acquired at the target resolution and share similar contrasts. 

Lisha Yuan¹, Qing Li², Guojing Wei³, Hongjian He¹, and Jianhui Zhong^4
1Department of Biomedical Engineering, Center for Brain Imaging Science and Technology, Zhejiang University, Hangzhou, China, ²MR Collaborations, Siemens Healthcare Ltd., Shanghai, China, ³SHS DI MR R&D SZN LP, Siemens Shenzhen Magnetic Resonance Ltd., Shenzhen, China, ⁴Department of Imaging Sciences, University of Rochester, Rochester, NY, United States

Since both imaging and diffusion gradient pulses could lead to significant diffusion-related signal attenuation, one must account for the effect of all imaging and diffusion gradient pulses to achieve an accurate estimation of diffusion metrics and tensors. Based on a generalized definition of b matrix, a general framework for an automated and accurate b-matrix calculation was proposed in this study. Its correctness was verified in SE-diffusion sequence, and the importance of accurate inclusion of all gradient pulses was shown in both SE- and SPEN-diffusion sequence. The proposed method is suitable for accurate evaluation of diffusion effects in various sequences.

Xiao Liang¹, Pan Su², Steve Roys¹, Rao P Gullapalli¹, Jerry L Prince³, and Jiachen Zhuo^1
1Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of Medicine, Baltimore, MD, United States, ²Siemens Medical Solutions USA Inc, Malvern, PA, United States, ³Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, United States

In this study, we propose a more robust PITA-MDD method with automatic brain ROI selection and adaptive thresholding for the motion threshold (termed APITA-MDD). Automatic brain ROI is initially identified by Otsu’s method and then improved by erosion and dilation. Haralick’s Homogeneity Index (HHI) of each slice is converted to a deviation score independent of image SNR and ROI size for motion detection. APITA-MDD was tested on brain dMRI data acquired with head motion and leg crossing motion and correctly detected motion slices missed by PITA-MDD due to insufficient coverage at edge slices and single thresholding.

Jianxing Qiu¹, Jing Liu¹, Chenwen Liu², Jinxia Zhu², and Thomas Benkert^3
1Peking University First Hospital, Beijing, China, ²MR Collaboration, Siemens Healthcare Ltd China, Beijing, China, ³MR Application Development, Siemens Healthcare GmbH, Erlangen, Germany

Diffusion-weighted imaging (DWI) with integrated slice-specific dynamic-shimming (iShim) could improve image quality compared with conventional 3D-Shimming single-shot echo-planar-imaging (SS-EPI) DWI in patients with rectal cancer. iShim DWI can help differentiate rectal cancer from normal walls and predict pathologic characterizations. iShim DWI could be an effective and promising approach and a good alternative to conventional 3D-Shimming SS-EPI DWI for patients with rectal cancer.