检索结果-内蒙古大学图书馆

Propagating Confidences through CNNs for Sparse Data Regression

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Eldesokey, Abdelrahman Felsberg, Michael Khan, Fahad Shahbaz Computer Vision Laboratory Department of Electrical Engineering Linköping University Linköping Sweden Inception Institute of Artificial Intelligence Abu Dhabi United Arab Emirates

In most computer vision applications, convolutional neural networks (CNNs) operate on dense image data generated by ordinary cameras. Designing CNNs for sparse and irregularly spaced input data is still an open problem with numerous applications in autonomous driving, robotics, and surveillance. To tackle this challenging problem, we introduce an algebraically-constrained convolution layer for CNNs with sparse input and demonstrate its capabilities for the scene depth completion task. We propose novel strategies for determining the confidence from the convolution operation and propagating it to consecutive layers. Furthermore, we propose an objective function that simultaneously minimizes the data error while maximizing the output confidence. Comprehensive experiments are performed on the KITTI depth benchmark and the results clearly demonstrate that the proposed approach achieves superior performance while requiring three times fewer parameters than the state-of-the-art methods. Moreover, our approach produces a continuous pixel-wise confidence map enabling information fusion, state inference, and decision support. Copyright © 2018, The Authors. All rights reserved.

关键词： Convolution

NTIRE 2020 Challenge on Real-World image Super-Resolution: Methods and Results

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Lugmayr, Andreas Danelljan, Martin Timofte, Radu Ahn, Namhyuk Bai, Dongwoon Cai, Jie Cao, Yun Chen, Junyang Cheng, Kaihua Chun, SeYoung Deng, Wei El-Khamy, Mostafa Man Ho, Chiu Ji, Xiaozhong Kheradmand, Amin Kim, Gwantae Ko, Hanseok Lee, Kanghyu Lee, Jungwon Li, Hao Liu, Ziluan Liu, Zhi-Song Liu, Shuai Lu, Yunhua Meng, Zibo Navarrete Michelini, Pablo Micheloni, Christian Prajapati, Kalpesh Ren, Haoyu Hyeok Seo, Yong Siu, Wan-Chi Sohn, Kyung-Ah Tai, Ying Muhammad Umer, Rao Wang, Shuangquan Wang, Huibing Haoning Wu, Timothy Wu, Haoning Yang, Biao Yang, Fuzhi Yoo, Jaejun Zhao, Tongtong Zhou, Yuanbo Zhuo, Haijie Zong, Ziyao Zou, Xueyi Wang, Li-Wen Cani, Marie-Paule Siu, Wan-Chi Yang, Huan Fu, Jianlong Shi, Yukai Chen, Junyang Lee, Kanghyu Park, Jaihyun Lee, Junyeop Min, Jeongki Lee, Bokyeung Ko, Hanseok Yoo, Jaejun Sohn, Kyung-Ah Micheloni, Christian Hu, Fengshuo Wang, Yanhong Lu, Yunhua Peng, Jinjia Wang, Huibing Zhuo, Haijie Lee, Junyeop Min, Jeongki Lee, Bokyeung Park, Jaihyun Ko, Hanseok Tai, Lab Ying Jilin Li, Youtu Lab Liu, Shuai Yang, Biao Liu, Xing Chen, Shuaijun Zhao, Lei Wang, Zhan Lin, Yuxuan Jia, Xu Gao, Qinquan Deng, Wei Kheradmand, Amin El-Khamy, Mostafa Wang, Shuangquan Bai, Dongwoon Lee, Jungwon Patel, Heena Chudasama, Vishal Upla, Kishor Ramachandra, Raghavendra Raja, Kiran Busch, Christoph Meng, Zibo Ho, Chiu Man Computer Vision Lab ETH Zurich Switzerland LIX - Computer science laboratory at the cole polytechnique Palaiseau France Center of Multimedia Signal Processing Hong Kong Polytechnic University Hong Kong Shanghai Jiao Tong University China Microsoft Research Beijing China Guangdong University of Technology China Intelligent Signal Processing Laboratory Korea University Korea Republic of Ajou University Korea Republic of EPFL University of Udine Italy BOE Technology Group Co. Ltd Dalian Maritime University China Guangdong OPPO Mobile Telecommunications Corp. Ltd Department of Video Information Processing Korea University Korea Republic of School of Electrical Engineering Korea University Korea Republic of Tencent Youtu Lab Yun Cao Tencent Youtu Tencent Youtu Lab Chengjie Wang Tencent Tencent Youtu Lab Feiyue Huang Tencent Youtu Lab Peking University China North China University of Technology China Huawei Technologies Co. Ltd China Fuzhou University Tong Tong Fuzhou University Imperial Vision Technology China Fuzhou University Imperial Vision Technology China Imperial Vision Technology SOC R&D Samsung Semiconductor Inc. United States Ulsan national institute of science and technology Korea Republic of Sardar Vallabhbhai National Institute Of Technology Surat India Norwegian University of Science and Technology Gjøvik Norway InnoPeak Technology

This paper reviews the NTIRE 2020 challenge on real world super-resolution. It focuses on the participating methods and final results. The challenge addresses the real world setting, where paired true high and low-resolution images are unavailable. For training, only one set of source input images is therefore provided along with a set of unpaired high-quality target images. In Track 1: image Processing artifacts, the aim is to super-resolve images with synthetically generated image processing artifacts. This allows for quantitative benchmarking of the approaches w.r.t. a ground-truth image. In Track 2: Smartphone images, real low-quality smart phone images have to be super-resolved. In both tracks, the ultimate goal is to achieve the best perceptual quality, evaluated using a human study. This is the second challenge on the subject, following AIM 2019, targeting to advance the state-of-the-art in super-resolution. To measure the performance we use the benchmark protocol from AIM 2019. In total 22 teams competed in the final testing phase, demonstrating new and innovative solutions to the problem. Copyright © 2020, The Authors. All rights reserved.

关键词： Optical resolving power

A Generative Appearance Model for End-to-end Video Object Segmentation

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Johnander, Joakim Danelljan, Martin Brissman, Emil Khan, Fahad Shahbaz Felsberg, Michael Computer Vision Laboratory Department of Electrical Engineering Linköping University Inception Institute of Artificial Intelligence Abu Dhabi United Arab Emirates Saab Sweden Zenuity Sweden

One of the fundamental challenges in video object segmentation is to find an effective representation of the target and background appearance. The best performing approaches resort to extensive fine-tuning of a convolutional neural network for this purpose. Besides being prohibitively expensive, this strategy cannot be truly trained end-to-end since the online fine-tuning procedure is not integrated into the offline training of the *** address these issues, we propose a network architecture that learns a powerful representation of the target and background appearance in a single forward pass. The introduced appearance module learns a probabilistic generative model of target and background feature distributions. Given a new image, it predicts the posterior class probabilities, providing a highly discriminative cue, which is processed in later network modules. Both the learning and prediction stages of our appearance module are fully differentiable, enabling true end-to-end training of the entire segmentation pipeline. Comprehensive experiments demonstrate the effectiveness of the proposed approach on three video object segmentation benchmarks. We close the gap to approaches based on online fine-tuning on DAVIS17, while operating at 15 FPS on a single GPU. Furthermore, our method outperforms all published approaches on the large-scale YouTube-VOS dataset. Copyright © 2018, The Authors. All rights reserved.

关键词： Network architecture

Morphological geodesic active contour based automatic aorta segmentation in thoracic CT images

学校读者我要写书评

暂无评论

Morphological geodesic active contour based automatic aorta ...

International Conference on computer vision and image Processing, CVIP 2016

作者： Dasgupta, Avijit Mukhopadhyay, Sudipta Mehre, Shrikant A. Bhattacharyya, Parthasarathi Computer Vision and Image Processing Laboratory Department of Electronics and Electrical Communication Engineering Indian Institute of Technology Kharagpur West Bengal721302 India Institute of Pulmocare & Research KolkataWest Bengal700156 India

ISBN: (纸本)9789811021060

Automatic aorta segmentation and quantification in thoracic computed tomography (CT) images is important for detection and prevention of aortic diseases. This paper proposes an automatic aorta segmentation algorithm in both contrast and non-contrast CT images of thorax. The proposed algorithm first detects the slice containing the carina region. Circular Hough Transform (CHT) is applied on the detected slice to localize ascending and descending aorta (circles with lowest variances) followed by a morphological geodesic active contour to segment the aorta from CT stack. The dice similarity coefficients (DSC) between the ground truth and the segmented output were found to be 0.8845 ± 0.0584 on LIDC-IDRI dataset. © Springer Science+Business Media Singapore 2017.

关键词： image segmentation

Mineral Detection of Neutrinos and Dark Matter. A Whitepaper

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Baum, Sebastian Stengel, Patrick Abe, Natsue Acevedo, Javier F. Araujo, Gabriela R. Asahara, Yoshihiro Avignone, Frank Balogh, Levente Baudis, Laura Boukhtouchen, Yilda Bramante, Joseph Breur, Pieter Alexander Caccianiga, Lorenzo Capozzi, Francesco Collar, Juan I. Ebadi, Reza Edwards, Thomas Eitel, Klaus Elykov, Alexey Ewing, Rodney C. Freese, Katherine Fung, Audrey Galelli, Claudio Glasmacher, Ulrich A. Gleason, Arianna Hasebe, Noriko Hirose, Shigenobu Horiuchi, Shunsaku Hoshino, Yasushi Huber, Patrick Ido, Yuki Igami, Yohei Ishikawa, Norito Itow, Yoshitaka Kamiyama, Takashi Kato, Takenori Kavanagh, Bradley J. Kawamura, Yoji Kazama, Shingo Kenney, Christopher J. Kilminster, Ben Kouketsu, Yui Kozaka, Yukiko Kurinsky, Noah A. Leybourne, Matthew Lucas, Thalles McDonough, William F. Marshall, Mason C. Mateos, Jose Maria Mathur, Anubhav Michibayashi, Katsuyoshi Mkhonto, Sharlotte Murase, Kohta Naka, Tatsuhiro Oguni, Kenji Rajendran, Surjeet Sakane, Hitoshi Sala, Paola Scholberg, Kate Semenec, Ingrida Shiraishi, Takuya Spitz, Joshua Sun, Kai Suzuki, Katsuhiko Tanin, Erwin H. Vincent, Aaron Vladimirov, Nikita Walsworth, Ronald L. Watanabe, Hiroko Stanford Institute for Theoretical Physics Department of Physics Stanford University StanfordCA94305 United States Istituto Nazionale di Fisica Nucleare Sezione di Ferrara via Giuseppe Saragat 1 FerraraI-44122 Italy Kanagawa Yokohama236-0001 Japan SLAC National Accelerator Laboratory Stanford University 2575 Sand Hill Road Menlo ParkCA94025 United States Department of Physics University of Zurich Winterthurerstrasse 190 ZurichCH-8057 Switzerland Department of Earth and Environmental Sciences Graduate School of Environmental Studies Nagoya University Furo-cho Chikusa-ku Nagoya464-8601 Japan Department of Physics and Astronomy University of South Carolina ColumbiaSC29208 United States Mechanical and Materials Engineering Queen’s University 130 Stuart Street KingstonONK7L 2V9 Canada Department of Physics Engineering Physics and Astronomy Queen’s University 64 Bader Lane KingstonONK7L 2S8 Canada Perimeter Institute for Theoretical Physics WaterlooONN2J 2W9 Canada Istituto Nazionale di Fisica Nucleare Sezione di Milano Via Celoria 16 Milan Italy Dipartimento di Scienze Fisiche e Chimiche Università degli Studi dell’Aquila L’Aquila67100 Italy Department of Physics University of Chicago ChicagoIL60637 United States Department of Physics University of Maryland College ParkMD20742 United States Quantum Technology Center University of Maryland College ParkMD20742 United States The William H. Miller III Department of Physics and Astronomy The Johns Hopkins University BaltimoreMD21218 United States Institute for Astroparticle Physics Karlsruhe Institute of Technology Karlsruhe76021 Germany Earth and Planetary Sciences Stanford University StanfordCA94305-2115 United States Texas Center for Cosmology and Astroparticle Physics Weinberg Institute for Theoretical Physics Department of Physics The University of Texas at Austin AustinTX78712 United States The Oskar Klein Centre Department of Physics Stockholm University AlbaNova

Minerals are solid state nuclear track detectors – nuclear recoils in a mineral leave latent damage to the crystal structure. Depending on the mineral and its temperature, the damage features are retained in the material from minutes (in low-melting point materials such as salts at a few hundred ◦C) to timescales much larger than the 4.5 Gyr-age of the Solar System (in refractory materials at room temperature). The damage features from the O(50) MeV fission fragments left by spontaneous fission of 238U and other heavy unstable isotopes have long been used for fission track dating of geological samples. laboratory studies have demonstrated the readout of defects caused by nuclear recoils with energies as small as O(1) keV. This whitepaper discusses a wide range of possible applications of minerals as detectors for ER O(1) keV nuclear recoils: Using natural minerals, one could use the damage features accumulated over O(10) Myr–O(1) Gyr to measure astrophysical neutrino fluxes (from the Sun, supernovae, or cosmic rays interacting with the atmosphere) as well as search for Dark Matter. Using signals accumulated over months to few-years timescales in laboratory-manufactured minerals, one could measure reactor neutrinos or use them as Dark Matter detectors, potentially with directional sensitivity. Research groups in Europe, Asia, and America have started developing microscopy techniques to read out the O(1)–O(100) nm damage features in crystals left by O(0.1)–O(100) keV nuclear recoils. We report on the status and plans of these programs. The research program towards the realization of such detectors is highly interdisciplinary, combining geoscience, material science, applied and fundamental physics with techniques from quantum information and Artificial Intelligence. Copyright © 2023, The Authors. All rights reserved.

关键词： Minerals

The Brain Tumor Segmentation (BraTS-METS) Challenge 2023: Brain Metastasis Segmentation on Pre-treatment MRI

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Moawad, Ahmed W. Janas, Anastasia Baid, Ujjwal Ramakrishnan, Divya Saluja, Rachit Ashraf, Nader Jekel, Leon Amiruddin, Raisa Adewole, Maruf Albrecht, Jake Anazodo, Udunna Aneja, Sanjay Anwar, Syed Muhammad Bergquist, Timothy Calabrese, Evan Chiang, Veronica Chung, Verena Conte, Gian Marco Marco Dako, Farouk Eddy, James Ezhov, Ivan Familiar, Ariana Farahani, Keyvan Iglesias, Juan Eugenio Jiang, Zhifan Johanson, Elaine Kazerooni, Anahita Fathi Kofler, Florian Krantchev, Kiril LaBella, Dominic Van Leemput, Koen Li, Hongwei Bran Linguraru, Marius George Link, Katherine E. Liu, Xinyang Maleki, Nazanin Meier, Zeke Menze, Bjoern H. Moy, Harrison Osenberg, Klara Piraud, Marie Reitman, Zachary Shinohara, Russel Takeshi Tahon, Nourel hoda Nada, Ayman Velichko, Yuri S. Wang, Chunhao Wiestler, Benedikt Wiggins, Walter Shafique, Umber Willms, Klara Avesta, Arman Bousabarah, Khaled Chakrabarty, Satrajit Gennaro, Nicolo Holler, Wolfgang Kaur, Manpreet LaMontagne, Pamela Lin, MingDe Lost, Jan Marcus, Daniel S. Maresca, Ryan Merkaj, Sarah Nada, Ayaman Pedersen, Gabriel Cassinelli von Reppert, Marc Sotiras, Aristeidis Teytelboym, Oleg Tillmans, Niklas Westerhoff, Malte Youssef, Ayda Godfrey, Devon Floyd, Scott Rauschecker, Andreas Villanueva-Meyer, Javier Pflüger, Irada Cho, Jaeyoung Bendszus, Martin Brugnara, Gianluca Cramer, Justin Perez-Carillo, Gloria Guzman J. Johnson, Derek R. Kam, Anthony Kwan, Benjamin Yin Ming Lai, Lillian Lall, Neil U. Memon, Fatima Patro, Satya Narayana Petrovic, Bojan So, Tiffany Y. Thompson, Gerard Wu, Lei Schrickel, E. Brooke Bansal, Anu Barkhof, Frederik Besada, Cristina Chu, Sammy Druzgal, Jason Dusoi, Alexandru Farage, Luciano Feltrin, Fabricio Fong, Amy Fung, Steve H. Gray, R. Ian Ikuta, Ichiro Iv, Michael Postma, Alida A. Mahajan, Amit Joyner, David Krumpelman, Chase Letourneau-Guillon, Laurent Lincoln, Christie M. Maros, Mate E. Miller, Elka Morón, Fanny Nimchinsky, Esther A. Ozsarlak, Ozkan Patel, Uresh Rohatgi, Saurabh Saha, Atin Sayah, Anousheh Schwartz, Eric D. Shih, Robert Shiroishi, Mark Trinity health Mid Atlantic Hosiptals DarbyPA United States Department of Radiology and Biomedical Imaging Yale School of Medicine New HavenCT United States Division of Computational Pathology Department of Pathology and Laboratory Medicine School of Medicine Indiana University IndianapolisIN United States Department of Electical and Computer Engineering Cornell University Cornell Tech New YorkNY United States Department of Radiology Weill Cornell Medicine New YorkNY United States ImagineQuant Children’s Hospital of Philadelphia PhiladelphiaPA United States College of Medicine Alfaisal University Riyadh Saudi Arabia DKFZ Division of Translational Neurooncology The WTZ German Cancer Consortium DKTK Partner Site University Hospital Essen Essen Germany Medical Artificial Intelligence Lab Crestview Radiology Lagos Nigeria Sage Bionetworks SeattleWA United States Montreal Neurological Institute McGill University Montreal Canada lab Crestview Radiology Lagos Nigeria Department of Therapeutic Radiology Yale School of Medicine New HavenCT United States Sheikh Zayed Institute for Pediatric Surgical Innovation Children’s National Hospital WashingtonDC United States Department of Radiology Mayo Clinic RochesterMN United States Department of Radiology Duke University Medical Center DurhamNC United States Department of Neurosurgery Yale School of Medicine New HavenCT United States Center for Global Health Perelman School of Medicine University of Pennsylvania PA United States Department of Informatics Technical University Munich Germany Children’s Hospital of Philadelphia University of Pennsylvania PhiladelphiaPA United States Cancer Imaging Program National Cancer Institute National Institutes of Health BethesdaMD United States Athinoula A. Martinos Center for Biomedical Imaging Massachusetts General Hospital BostonMA United States Children’s National Hospital WashingtonDC United States PrecisionFDA U.S. Food and Drug Administration S

The translation of AI-generated brain metastases (BM) segmentation into clinical practice relies heavily on diverse, high-quality annotated medical imaging datasets. The BraTS-METS 2023 challenge has gained momentum for testing and benchmarking algorithms using rigorously annotated internationally compiled real-world datasets. This study presents the results of the segmentation challenge and characterizes the challenging cases that impacted the performance of the winning algorithms. Untreated brain metastases on standard anatomic MRI sequences (T1, T2, FLAIR, T1PG) from eight contributed international datasets were annotated in stepwise method: published UNET algorithms, student, neuroradiologist, final approver neuroradiologist. Segmentations were ranked based on lesion-wise Dice and Hausdorff distance (HD95) scores. False positives (FP) and false negatives (FN) were rigorously penalized, receiving a score of 0 for Dice and a fixed penalty of 374 for HD95. The mean scores for the teams were calculated. Eight datasets comprising 1303 studies were annotated, with 402 studies (3076 lesions) released on Synapse as publicly available datasets to challenge competitors. Additionally, 31 studies (139 lesions) were held out for validation, and 59 studies (218 lesions) were used for testing. Segmentation accuracy was measured as rank across subjects, with the winning team achieving a LesionWise mean score of 7.9. The Dice score for the winning team was 0.65 ± 0.25. Common errors among the leading teams included false negatives for small lesions and misregistration of masks in space. The Dice scores and lesion detection rates of all algorithms diminished with decreasing tumor size, particularly for tumors smaller than 100 mm3. In conclusion, algorithms for BM segmentation require further refinement to balance high sensitivity in lesion detection with the minimization of false positives and negatives. The BraTS-METS 2023 challenge successfully curated well-annotated, diverse d

关键词： image segmentation

LaG-DESIQUE: A local-and-global blind image quality evaluator without training on human opinion scores 6th

学校读者我要写书评

暂无评论

LaG-DESIQUE: A local-and-global blind image quality evaluato...

6th CCF Academic Conference on Big Data, CCF Big Data 2018

作者： Liu, Ruyi Zhang, Yi Chandler, Damon M. Miao, Qiguang Liu, Tiange School of Computer Science and Technology Xidian University XianShaanxi710071 China Xian Key Laboratory of Big Data and Intelligent Vision XianShaanxi710071 China School of Electronic and Information Engineering Xian Jiaotong University XianShaanxi710049 China Department of Electrical and Electronic Engineering Shizuoka University HamamatsuShizuoka432-8561 Japan School of Information Science and Engingeering Yanshan University QinhuangdaoHebei066004 China

ISBN: (纸本)9789811329210

This paper extends our previous DESIQUE [1] algorithm to a local-and-global way (LaG-DESIQUE) to blindly measure image quality without training on human opinion scores. The local DESIQUE extracts block-based log-derivative features and evaluates image quality through measuring the multivariate Gaussian distance between selected natural and test image patches. The global DESIQUE extracts image-based log-derivative features and image quality is estimated based on a two-stage framework, which was trained on a set of regenerated distorted images with their quality scores estimated by MAD [2] algorithm. The overall quality is the weighted average of local and global DESIQUE scores. Test on several image databases demonstrates that LaG-DESIQUE performs competitively well in predicting image quality. © Springer Nature Singapore Pte Ltd. 2018.

关键词： image quality

Person Re-identification by Integrating Static Texture and Shape Cues 12th

学校读者我要写书评

暂无评论

Person Re-identification by Integrating Static Texture and S...

12th Chinese Conference on Biometric Recognition, CCBR 2017

作者： Madongo, Canaan Tinotenda Huang, Di Chen, Jiaxin Laboratory of Intelligent Recognition and Image Processing School of Computer Science and Engineering Beihang University Beijing100191 China Department of Electrical and Computer Engineering New York University Abu Dhabi Abu Dhabi United Arab Emirates

ISBN: (纸本)9783319699226

Person Re-Identification (Re-ID) is a challenging task with wide ranging applications in various fields. This paper presents a novel hand-crafted method for this issue, enhancing the state of the art ones in literature in two aspects. In contrast to most current studies that analyze texture features, we propose a discriminative and compact shape feature by applying Procrustes shape analysis. It not only retains shape distinctiveness of an individual sample, but also alleviates cross-view impacts. Furthermore, we combine the shape feature with some current popular texture features, namely LOMO and mid-level filters, so that the advantages of multiple clues can be jointly used. A score level fusion strategy is finally adopted to optimally integrate their credits. Evaluated on two public benchmarks, i.e. VIPeR and CUHK03, the proposed method achieves very competitive results, indicating its effectiveness in person Re-ID. © 2017, Springer International Publishing AG.

关键词： computers

Injecting and removing malignant features in mammography with CycleGAN: Investigation of an automated adversarial attack using neural networks

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Becker, Anton S. Jendele, Lukas Skopek, Ondrej Berger, Nicole Ghafoor, Soleen Marcon, Magda Konukoglu, Ender Institute of Diagnostic and Interventional Radiology University Hospital of Zurich Department of Health Sciences and Technology ETH Zurich Department of Computer Science ETH Zurich Department of Radiology Memorial Sloan Kettering Cancer Center New York City United States Computer Vision Laboratory Department of Information Technology and Electrical Engineering ETH Zurich

Purpose To train a cycle-consistent generative adversarial network (CycleGAN) on mammographic data to inject or remove features of malignancy, and to determine whether these AI-mediated attacks can be detected by radiologists. Material and Methods From the two publicly available datasets, BCDR and INbreast, we selected images from cancer patients and healthy controls. An internal dataset served as test data, withheld during training. We ran two experiments training CycleGAN on low and higher resolution images (256×256 px and 512×408 px). Three radiologists read the images and rated the likelihood of malignancy on a scale from 1-5 and the likelihood of the image being manipulated. The readout was evaluated by ROC analysis (Area under the ROC curve = AUC). Results At the lower resolution, only one radiologist exhibited markedly lower detection of cancer (AUC=0.85 vs 0.63, p=0.06), while the other two were unaffected (0.67 vs. 0.69 and 0.75 vs. 0.77, p=0.55). Only one radiologist could discriminate between original and modified images slightly better than guessing/chance (0.66, p=0.008). At the higher resolution, all radiologists showed significantly lower detection rate of cancer in the modified images (0.77-0.84 vs. 0.59-0.69, p=0.008), however, they were now able to reliably detect modified images due to better visibility of artifacts (0.92, 0.92 and 0.97). Conclusion A CycleGAN can implicitly learn malignant features and inject or remove them so that a substantial proportion of small mammographic images would consequently be misdiagnosed. At higher resolutions, however, the method is currently limited and has a clear trade-off between manipulation of images and introduction of *** Codes 68T45 Copyright © 2018, The Authors. All rights reserved.

关键词： Generative adversarial networks