Takahiro Matsuyama¹, Yoshiharu Ohno^1,2, Kaori Yamamoto³, Kazuhiro Murayama², Masato Ikedo³, Masao Yui³, Akiyoshi Iwase⁴, Takashi Fukuba⁴, Satomu Hanamatsu¹, Yuki Obama¹, Takahiro Ueda¹, Hirotaka Ikeda¹, and Hiroshi Toyama^1
1Radiology, Fujita Health University School of Medicine, Toyoake, Japan, ²Joint Research Laboratory of Advanced Medical Imaging, Fujita Health University School of Medicine, Toyoake, Japan, ³Canon Medical Systems Corporation, Otawara, Japan, ⁴Radiology, Fujita Health University Hospital, Toyoake, Japan

    We hypothesize that the newly developed FAST 3D can reduce examination time as well as Compressed SPEEDER and obtain MRCP without any degradation of image quality as compared with conventional parallel imaging technique in patients with hepatobiliary and pancreatic diseases.  In addition, deep learning reconstruction (i.e.AiCE) has a potential to improve image quality of MRCP obtained by different protocols in routine clinical practice.  The purpose of this study was to compare the capability of image quality improvement on MRCP with and without AiCE among FAST 3D, Compressed SPEEDER and conventional parallel imaging (SPEEDER) in patients with hepatobiliary and pancreatic diseases.

Tao Lin¹, Ailian Liu¹, Lihua Chen¹, Qingwei Song¹, Renwang Pu¹, Ying Zhao¹, Xue Ren¹, and yan guo^2
1Department of Radiology, the First Affiliated Hospital of Dalian Medical University, Dalian, China, ²GE Healthcare, Beijing, China

Hepatocellular carcinoma (HCC), often occurring in patients with chronic viral hepatitis and cirrhosis, was one of the most common malignant tumors. Microvascular invasion (MVI) is one of the well-known potential predictors related to the prognosis of HCC. MVI with higher positive expression indicates more aggressive behaviour of HCC and poorer survival outcomes. Early accurate prediction for MVI status plays an important guiding role in surgical selection and adjuvant therapy for HCC. We proposed a radiomics model based on enhanced MRI to predict MVI in HCC before surgery.

Mark A Pinnock^1,2, Yipeng Hu^1,2, Alan Bainbridge³, David Atkinson⁴, Rajeshwar P Mookerjee⁵, Stuart A Taylor⁴, Dean C Barratt^1,2, and Manil D Chouhan^4
1Centre for Medical Image Computing, University College London, London, United Kingdom, ²Wellcome/EPSRC Centre for Interventional and Surgical Sciences, University College London, London, United Kingdom, ³Department of Medical Physics and Biomedical Engineering, University College London Hospitals NHS Foundation Trust, London, United Kingdom, ⁴Centre for Medical Imaging, Division of Medicine, University College London, London, United Kingdom, ⁵Institute for Liver and Digestive Health, Division of Medicine, University College London, London, United Kingdom

Traditional approaches to MRI detection of liver disease require specialist hardware, sequences and post-processing. Here we propose a deep learning (DL) based model for the detection of liver disease using standard T2-weighted anatomical sequences, as an early feasibility study for the potential of DL-based classification of liver disease severity.  Our DL model achieved a diagnostic accuracy of 0.92 on unseen data and achieved a test accuracy of 0.75 when trained with relevant anatomical segmentation masks without images, demonstrating potential scanner/sequence independence. Lastly, we used DL interpretability techniques to analyse failure cases.

Xu Luo^1,2, Ailian Liu³, Yu Yao^1,2, Ying Zhao³, Zhebin Chen^1,2, Meng Dou^1,2, and Han Wen^1,2
1Chengdu Institute of Computer Application, Chinese Academy of Sciences, Chengdu, China, ²University of Chinese Academy of Sciences, Beijing, China, ³Department of Radiology, the First Affiliated Hospital of Dalian Medical University, Dalian, China

The deep learning technology was used on the segmentation automatically and reconstruction of the intrahepatic portal vein to get its volume and three-dimensional spatial position information. The results showed that the dice accuracy was 83.32% in the training and 75.89% in the test set among 12 cases. The current study demonstrated the volume of intrahepatic vessels and 3D spatial location information can be obtained by the tow-stage deep learning model and it can be applied in promising clinical effectively.

Ying Zhao¹, Han Wen^2,3, Ailian Liu¹, Yu Yao^2,3, Tao Lin¹, Qingwei Song¹, Xin Li⁴, Yan Guo⁴, and Tingfan Wu^4
1The First Affiliated Hospital of Dalian Medical University, Dalian, China, ²Chengdu Institute of CoChinese Academy of Sciences, Chengdu, China, ³University of Chinese Academy of Sciences, Beijing, China, ⁴GE Healthcare (China), Shanghai, China

In recent years, convolutional neural networks (CNNs) have become one of the most advanced deep learning networks. Deep learning with CNNs has reportedly achieved good performance in the pattern recognition of images. In the present study, 3D-CNN based on contrast-enhanced (CE)-MR images was demonstrated to be capable to evaluate pathologic grade of hepatocellular carcinoma (HCC) treated with surgical resection, which will provide more prognostic information and facilitate clinical management.

Ruofan Sheng¹, Liyun Zheng², Shu Liao³, Yongming Dai², and Mengsu Zeng^1
1Department of Radiology, Zhongshan Hospital, Shanghai, China, ²United Imaging Healthcare, Shanghai, China, ³Shanghai United Imaging Intelligence, Shanghai, China

Liver magnetic resonance imaging (MRI) is limited by several technical challenges, including relatively long acquisition time and respiratory motion artifacts. Recently, deep learning methods have been proposed to reconstruct undersampled k-space data by training deep neural networks. In this study, we raised a U-net convolutional neural network architecture to improve the reconstruction speed and image quality of liver T2-weighted MRI. This technique was able to cover the whole liver during one breath hold and showed promising performance in image quality and lesion detectability.

Satoshi Funayama^1,2, Utaroh Motosugi³, and Hiroshi Onishi^1
1Department of Radiology, University of Yamanashi, Yamanashi, Japan, ²Graduate School of Medicine, University of Yamanashi, Yamanashi, Japan, ³Department of Radiology, Kofu-Kyoritsu Hospital, Yamanashi, Japan

  Short acquisition time is one of the key features of liver MRI to acquire images during breath-holding. A combination of undersampling and deep learning-based reconstruction would be a powerful reconstruction method to achieve sufficient speed and SNR. However, it is challenging due to high memory consumption in network training. The folded image training strategy (FITS) is one of the methods to handle this problem. In this study, we demonstrated that the model-based deep learning reconstruction using FITS had better image quality in liver MRI acquired with multiple coils.

Fengxian Fan¹, Weiting Huang², Yanli Jiang¹, Wanjun Hu¹, Jing Zhang¹, and Jialiang Ren^3
1LanZhou University Second Hospital, LanZhou, China, ²LanZhou University, LanZhou, China, ³GE Healthcare, Shanghai, China

The purpose of this study is to develop radiomics models by using T1WI for staging liver fibrosis. Two hundred twenty-four patients had available pathologic reports of liver fibrosis and liver MRI were divided into training (n=179) and testing (n=45) cohorts to develop and validate the radiomics models. The results showed that the T1WI-based radiomics models had a powerful ability to stage liver fibrosis.

Junyu Guo¹ and Ivan Pedrosa^1
1Radiology, UT southwestern medical center, Dallas, TX, United States

MR kidney image segmentation is an important enabler of radiomics analysis and assessment of kidney size, morphology and renal disease. Deep learning methods are state-of-the-art techniques for segmentation. Training of a robust model with high accuracy requires a large dataset. Manually drawing masks is time-consuming and labor-intensive for a large number of datasets. Furthermore, different masks are required in training for different MR modalities. In this study, we investigated the feasibility of kidney segmentation using deep learning models trained with MR images from only a few subjects. We tested the hypothesis that few-shot deep learning may achieve accurate kidney segmentation.

Kaixuan Zhao^1,2, Joao dos Santos Periquito³, Thomas Gladytz², Kathleen Cantow³, Luis Hummel³, Jason Millward², Sonia Waiczies², Erdmann Seeliger³, Yanqiu Feng¹, and Thoralf Niendorf^2,4
1School of Biomedical Engineering, Southern Medical University, Guangzhou, China, ²Berlin Ultrahigh Field Facility (B.U.F.F.), Max Delbruck Center for Molecular Medicine in the Helmholtz Association, Berlin, Germany, ³Institute of Physiology, Charité - Universitätsmedizin Berlin, Berlin, Germany, ⁴Experimental and Clinical Research Center, a joint cooperation between the Charité Medical Faculty and the Max Delbrück Center for Molecular Medicine in the Helmholtz Association, Berlin, Germany

Fast renal volume changes during sequential gas challenges might indicate the dynamic balance between renal filtration and reabsorption. In the present work, a deep learning based semantic segmentation method is employed to monitor renal size changes.

Jiang-Ning Dong¹ and Bin Shi^1
1The First Affiliated Hospital of USTC, Anhui Provincial Cancer Hospital, Hefei, China

The combination of IVIM-DWI biomarkers and T2WI-based texture features had good predictive performance to evaluate three different histological types of cervical carcinoma by synergizing diffusion, perfusion and heterogeneous features, especially for cervical squamous cell carcinoma and adenocarcinoma. A novel quantitative preoperative imaging model composed of IVIM-DWI and TA is a useful supplement for histopathological diagnosis, and might be used in clinical practice to assist in the formulation of treatment strategy.

xiang liu¹, chao han¹, and xiaoying wang^1
1department of radiology, peking university first hospital, Beijing, China

This retrospective study aims to perform automated pelvic bones segmentation in multiparametric MRI (mpMRI) using 3D convolutional neural network (CNN). 264 pelvic DWI images and corresponding ADC maps obtained from three MRI vendors from 2018 to 2019 were used for the 3D U-Net CNN development. 60 independent mpMRI data from 2020 were used to externally evaluate the segmentation model using quantitative criteria (Dice similarity coefficient) and qualitative assessment (SCORE system). The results demonstrated that the 3D CNN can achieve fully automated pelvic bone segmentation on multi-vendor DWI and ADC images with good quantitative and qualitative performances.

Ali Pirasteh¹, Lloyd Estkowski², Daniel Litwiller³, Ersin Bayram⁴, and Xinzeng Wang^5
1Department of Radiology, UW Madison, Madison, WI, United States, ²Global MR Applications & Workflow, GE Healthcare, Madison, WI, United States, ³Global MR Applications & Workflow, GE Healthcare, Denver, CO, United States, ⁴Global MR Applications & Workflow, GE Healthcare, Houston, TX, United States, ⁵GE Healthcare, Houston, TX, United States

We evaluated the utility of PROPELLER T2 FSE with deep-learning (DL) reconstruction in the setting of prostate and rectal imaging, with the goal of overcoming respiratory and peristaltic motion, improving image sharpness, and achieving high-resolution imaging within a comparable scan time to the traditional T2 FSE techniques. We demonstrated that in absence of DL reconstruction, the PROPELLER T2 FSE images suffer from excessive noise and lower subjective quality. However, with utilization of DL reconstruction, High-resolution and motion robust images were obtained at clinically acceptable scan times.

Takahiro Ueda¹, Yoshiharu Ohno^1,2, Kaori Yamamoto³, Kazuhiro Murayama², Masato Ikedo³, Masao Yui³, Akiyoshi Iwase⁴, Takashi Fukuba⁴, Satomu Hanamatsu¹, Yuki Obama¹, Hirotaka Ikeda¹, and Hiroshi Toyama^1
1Radiology, Fujita Health University School of Medicine, Toyoake, Japan, ²Joint Research Laboratory of Advanced Medical Imaging, Fujita Health University School of Medicine, Toyoake, Japan, ³Canon Medical Systems Corporation, Otawara, Japan, ⁴Radiology, Fujita Health University Hospital, Toyoake, Japan

There have been no reports of major studies to the utility of DLR for DWI with high b values for improving image quality and detection performance for patients with prostatic cancers.  We hypothesized that DLR could improve image quality and diagnostic performance of DWI with high values, and that an appropriate b value might be determined for this setting.  The purpose of this study was thus to determine the utility of the DLR method for DWI and the appropriate b value for detecting prostatic cancer using a 3T MR system in routine clinical practice.

Mohammed Saleh¹, Sanaz Javadi¹, Manoj Mathew², Jong Bum Son³, Jia Sun⁴, Ersin Bayram⁵, Xinzeng Wang⁵, Jingfei Ma³, Janio Szklaruk¹, and Priya Bhosale^1
1Radiology, MD Anderson Cancer Center, Houston, TX, United States, ²Radiology, Stanford University, Stanford, CA, United States, ³Imaging Physics, MD Anderson Cancer Center, Houston, TX, United States, ⁴Biostatistics, MD Anderson Cancer Center, Houston, TX, United States, ⁵Global MR Applications and Workflow, GE Healthcare, Houston, TX, United States

In oncologic MRI, sagittal T2 weighted images are usually acquired to assess gynecologic malignancy. Motion artifacts may render pathology difficult to detect due to patient or bowel motion or the presence of air in the rectum. PROPELLER sequence has shown promising results to reduce motion-related artifacts. Our work shows that DL Recon can be combined with PROPELLER and further help reduce noise and improve the overall image quality for T2-weighted imaging of gynecological malignancies. The combination of PROPELLER (Non-DL) and DL reconstruction could be synergistic in improving image quality.

Sheena Y Chu¹, Scott B Reeder^1,2,3,4,5, and John W Garrett^1,3
1Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ²Department of Medicine, University of Wisconsin-Madison, Madison, WI, United States, ³Department of Radiology, University of Wisconsin-Madison, Madison, WI, United States, ⁴Department of Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States, ⁵Department of Emergency Medicine, University of Wisconsin-Madison, Madison, WI, United States

By quantifying the efficiency and variability of clinic MRI exams, sources of potential improvement can be identified. An increase in efficiency and decrease in variability of MRI exams are essential to plan appropriate exam slot lengths, which can contribute to better patient access and reduced cost. Analytics are also essential to measure the impact on interventions to improve workflow. In this work we demonstrate the utility of analytics to identify inefficiencies in clinical MRI protocol in the context of MR enterography, a common abdominal MRI exam.

Shifeng Tian¹, Ailian Liu¹, Yan Guo², and Yuan Wei^1
1Department of Radiology, The First Affiliated Hospital of Dalian Medical University, Dalian, China, China, ²GE Healthcare, Dalian City, China, China

Squamous cell carcinoma is the most common pathological type in cervical cancer, followed by cervix adenocarcinoma. There are differences in the evaluation of prognosis between the two. Radiomics can quantitatively analyze a large number of image data, and then quantify tumor heterogeneity and noninvasively evaluate tumor biological behavior. The results suggest that radiomics based on mean kurtosis maps of diffusion kurtosis imaging sequences can identify different types of cervical cancer.

Fu Shen¹, Minglu Liu¹, Zhihui Li¹, Xiaolu Ma¹, Jianping Lu¹, and Yuwei Xia^2
1Changhai Hospital, Shanghai, China, ²Huiying Medical Technology Co., Ltd., Beijing, China

The goal of this study was to investigate the value of high resolution T2-weighted–based radiomics in prediction of treatment response to neoadjuvant chemoradiotherapy (nCRT) in patients with rectal mucinous adenocarcinoma (RMAC). The result demonstrated that the MRI based radiomics machine learning model could assess tumoral treatment response to nCRT in patients with RMAC.

Wen Li¹, Rohan Nadkarni¹, David C Newitt¹, Bo La Yun^1,2, Deep Hathi¹, Alex Nguyen¹, Natsuko Onishi¹, Lisa J Wilmes¹, Ella F Jones¹, Jessica Gibbs¹, Teffany Joy Bareng¹, Bonnie N Joe¹, Elissa Price¹, Rita Mukhtar¹, John Kornak¹, Efstathios Gennatos¹, I-SPY 2 Consortium³, Laura J Esserman¹, and Nola M Hylton^1
1University of California, San Francisco, San Francisco, CA, United States, ²Seoul National University Bundang Hospital, Seoul, Korea, Republic of, ³Quantum Leap Healthcare Collaborative, San Francisco, CA, United States

A previous study demonstrated that tumor sphericity, measured from breast DCE-MRI during neoadjuvant therapy, is predictive of pathologic complete response and adds value to a predictive model based on functional tumor volume (FTV) alone. This study further explores the additive value of alternative radiomic shape features by breast cancer subtype. A subset of shape features were selected using visually assessed MRI morphological patterns as guidance. The analysis of treatment response prediction was conducted retrospectively using data from the multi-center neoadjuvant I-SPY 2 TRIAL. Improved predictive performance when adding shape features was observed at both pre-treatment and early treatment time points.

ZHENYU SHU¹, YUYUN XU¹, and YONG ZHANG^2
1Zhejiang Provincial People’s Hospital, Hangzhou, China, ²MR Research, GE healthcare (China), SHANG HAI, China

This preliminary study explored the application of radiomics MRI in overall survival(OS)  of glioblastoma  patients. We found that EPI, age, and radiomic signature are independent predictors of OS for glioblastoma patients. The nomogram was created by integrating the three independent predictors, had the best performance when stratifying glioblastoma patients into long- versus short-term survival, which could help clinicians develop optimal treatment plans.