Sagar Mandava¹, Xinzeng Wang², Ty Cashen³, Tetsuya Wakayama⁴, and Ersin Bayram^2
1GE Healthcare, Atlanta, GA, United States, ²GE Healthcare, Houston, TX, United States, ³GE Healthcare, Madison, WI, United States, ⁴GE Healthcare, Hino, Japan

Radial imaging is becoming increasingly popular due to its ability to support highly accelerated imaging. However, it is plagued by streak artifacts that often arise from undersampling which can lead to poor image quality. The problem is particularly acute in time resolved imaging where the need for high spatio-temporal sampling usually leads to large amount of streaks. In this work, we propose a method for separate spatial and temporal deep learning for streak artifact reduction. The utility of the method is demonstrated on free breathing time resolved volumetric DCE MRI acquired using the stack-of-stars trajectory.

Junshen Xu¹, Esra Abaci Turk², Borjan Gagoski², Polina Golland³, P. Ellen Grant^2,4, and Elfar Adalsteinsson^5
1Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, ²Fetal-Neonatal Neuroimaging and Developmental Science Center, Boston Children’s Hospital, Boston, MA, United States, ³Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA, United States, ⁴Harvard Medical School, Boston, MA, United States, ⁵Massachusetts Institute of Technology, Cambridge, MA, United States

Fetal motion is an important measure for monitoring fetal health and neurological function. However, current clinical MRI and ultrasound assessments of fetal motion are qualitative and cannot reflect detailed 3D motion of each body part. In this work, we propose a method for fetal motion analysis in MRI using a deep pose estimator. We train a neural network to estimate fetal pose from MR volumes, and extract quantitative metrics of motion from the time series of fetal pose. In the experiments, we study how different conditions affect fetal motion, such as gestational age and maternal position during scan.

Yasuhisa Kurata¹, Mizuho Nishio¹, Yusaku Moribata², Aki Kido¹, Yuki Himoto¹, Koji Fujimoto³, Masahiro Yakami², Sachiko Minamiguchi⁴, Masaki Mandai⁵, and Yuji Nakamoto^1
1Diagnostic Imaging and Nuclear Medicine, Kyoto university hospital, Kyoto, Japan, ²Preemptive Medicine and Lifestyle-Related Disease Research Center, Kyoto university hospital, Kyoto, Japan, ³Real World Data Research and Development, Graduate School of Medicine Kyoto University, Kyoto, Japan, ⁴Diagnostic Pathology, Kyoto university hospital, Kyoto, Japan, ⁵Gynecology and Obstetrics, Kyoto university hospital, Kyoto, Japan

Endometrial cancer is the most common gynecological malignant tumor in developed countries, and accurate preoperative risk stratification is essential for personalized medicine. For realizing tumor feature extraction by radiomics approach, the segmentation of the tumor is usually required. The model developed in this study has achieved high-accuracy automatic segmentation of endometrial cancer on MRI using a convolutional neural network for the first time. Using multi-sequence MR images were important for high accuracy segmentation. Our model will lead to efficient medical image analysis of a large number of cases using the radiomics approach and/or deep learning methods.

Thomas Lilieholm¹, Ruiming Chen¹, Ruvini Navaratna¹, Daniel Seiter¹, Walter F Block^1,2,3, and Oliver Wieben^1,2,3
1Medical Physics, University of Wisconsin at Madison, Madison, WI, United States, ²Biomedical Engineering, University of Wisconsin at Madison, Madison, WI, United States, ³Radiology, University of Wisconsin at Madison, Madison, WI, United States

Quantitative investigation of placental volumes can be used for characterization of Zika virus (ZIKV) infection, which causes several complications for developing fetuses. To provide more rapidly available image segmentation for analysis, efforts are being made to produce Convolutional Neural Networks (CNN) for autonomous segmentation of placental volume images. We investigated a number of data augmentation techniques for training machine learning models to determine which methods may be most suited for further development of ZIKV-quantifying placental segmentation models. We found rotational and reflective data augmentation to produce the greatest improvement in machine-segmentated Dice Coefficient comparisons.

Ruiqi Geng^1,2, Mahalakshmi Sundaresan³, Jitka Starekova¹, Collin J Buelo^1,2, Nikolaos Panagiotopoulos¹, Marcin Ignaciuk¹, Thekla Helene Oechtering¹, Scott B Reeder^1,2,4,5,6, and Diego Hernando^1,2
1Radiology, University of Wisconsin-Madison, Madison, WI, United States, ²Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ³Electrical and Computer Engineering, University of Wisconsin-Madison, Madison, WI, United States, ⁴Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁵Emergency Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁶Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States

To enable fully free-breathing, single button-push liver exams, an automated AI-based method for image prescription of liver MRI was developed and evaluated. A total of seven classes of rectangular bounding boxes covering the liver, torso, and arms for each localizer orientation were manually and automatically labeled to enable 3D prescription in any orientation. The intersection over union (IoU) between manual and automated 2D liver detection had a median > 0.88 and interquartile range < 0.11 for all classes. The shift in the resultant 3D axial prescription was less than 9 mm in S/I dimension for 91% of the test dataset.

Fasil Gadjimuradov^1,2, Thomas Benkert², Marcel Dominik Nickel², and Andreas Maier^1
1Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, ²Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany

Signal-dropouts due to pulsation are one of the most prominent artifacts in diffusion-weighted imaging (DWI) of the liver. It can affect a significant portion of the repetitions acquired for a given slice. Instead of performing uniform averaging which might result in locally attenuated liver signal, this work proposes to train a convolutional neural network (CNN) to estimate smooth weight maps for individual repetitions. This allows to locally suppress signal-dropouts, resulting in more homogeneous liver signal while maintaining signal-to-noise ratio (SNR) in artifact-free image regions.

FengLing Gan¹, Shuohui Yang², Feng Xing³, Zheng Qu¹, Gaiying Li¹, Chenyao Yang², Rongfang Guo², Jiling Huang², Fang Lu², Caixia Fu⁴, Xu Yan⁴, Kelly Gillen⁵, Yi Wang⁵, Chenghai Liu³, Songhua Zhan², and Jianqi Li^1
1Shanghai Key Laboratory of Magnetic Resonance, School of Physics and Electronic Science, East China Normal University, shanghai, China, ²Department of Radiology, Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, shanghai, China, ³Department of Liver Diseases, Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, shanghai, China, ⁴MR Collaboration NE Asia, Siemens Healthcare, shanghai, China, ⁵Department of Radiology, Weill Medical College of Cornell University, New York, NY, United States

Hepatic fibrosis is characterized by excessive accumulation of extracellular matrix (ECM) proteins including collagen, which would contribute strong diamagnetic susceptibility in the liver tissue due to enhanced density of orbiting electrons. This study measured the texture features on susceptibility maps in patients with chronic liver diseases. The results showed that some of the second-order texture parameters were significantly different between cohorts of significant hepatic fibrosis (Ishak-F ≥ 3) and non-significant hepatic fibrosis (Ishak-F < 3). The texture analysis on susceptibility maps may have the potential to stage hepatic fibrosis.

Wu Zhou¹, Shangxuan Li¹, Wanwei Jian¹, Guangyi Wang², Lijuan Zhang³, and Honglai Zhang^1
1School of Medical Information Engineering, Guangzhou University of Chinese Medicine, Guangzhou, China, ²Department of Radiology, Guangdong General Hospital, Guangzhou, China, ³Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

The combination of context information from multi-modalities is remarkably significant for lesion characterization. However, there are still two remaining challenges for multi-modalities based lesion characterization including features overlapping between different tumor grades and large differences in modal contributions. In this work, we proposed a discriminative feature learning and adaptive fusion method in the framework of deep learning architecture for improving the performance of multimodal fusion based lesion characterization. Experimental results of grading of clinical hepatocellular carcinoma (HCC) demonstrate that the proposed method outperforms the previously reported fusion methods, including concatenation, correlated and individual feature learning, and deeply supervised net.

Rencheng Zheng¹, Luna Wang², Chengyan Wang³, Xuchen Yu¹, Weibo Chen⁴, Yan Li⁵, Weixia Li⁵, Fuhua Yan⁵, He Wang^1,3, and Ruokun Li^5
1Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China, ²Department of Radiology, Shanghai Chest Hospital, Shanghai, China, ³Human Phenome Institute, Fudan University, Shanghai, China, ⁴Market Solutions Center, Philips Healthcare, Shanghai, China, ⁵Department of Radiology, Ruijin Hospital, Shanghai, China

This study presented an algorithm for small hepatocellular carcinoma (sHCC) detection and segmentation in cirrhotic liver based on diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) images. The model included two-steps: screening of suspicious lesions in DWI using pattern matching algorithm; identification and segmentation of true lesions in DCE based on deep learning. The proposed model exhibited superior performance in sHCC (≤2 cm) detection and segmentation, which significantly outperformed the Liver Imaging Reporting and Data System (LI-RADS) based diagnosis. 

Huan-Huan Chong^1,2, Yu-Da Gong³, Lei Chen⁴, Xian-Pan Pan⁴, Ai-E Liu⁴, Chun Yang², and Meng-Su Zeng^1,2,5
1Shanghai Institute of Medical Imaging, Shanghai, China, ²Department of Radiology, Zhongshan Hospital, Fudan University, Shanghai, China, ³Department of General Surgery, Zhongshan Hospital, Fudan University, Shanghai, China, ⁴Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China, ⁵Department of Medical Imaging, Shanghai Medical College, Fudan University, Shanghai, China

Preoperatively identifying predisposing predictors of early relapse is crucial for stratifying patient risk, performing prompt intervention and improving long-term outcome. Whether peritumoral dilation radiomics can predict early recurrence in hepatocellular carcinoma(HCC) patients without macrovascular invasion remains unclear. Hence, we developed a bi-regional (the entire tumor and peritumoral zone within 1 cm) radiomics of gadoxetate disodium-enhanced (Gd-EOB-DTPA) MRI, which derived a satisfying discrimination in 2-year recurrence by the 5-fold cross-validation method.