Filters








3,147 Hits in 5.5 sec

Distribution Matching Losses Can Hallucinate Features in Medical Image Translation [article]

Joseph Paul Cohen, Margaux Luck, Sina Honari
2018 arXiv   pre-print
This paper discusses how distribution matching losses, such as those used in CycleGAN, when used to synthesize medical images can lead to mis-diagnosis of medical conditions.  ...  However, the basis of how these image translation models work is through matching the translation output to the distribution of the target domain.  ...  In order to match the target distribution, image features can be hallucinated and information to reconstruct an image in the other domain can be encoded [13] .  ... 
arXiv:1805.08841v3 fatcat:2ncyhq5pzzfplfynfuxtjp3dpq

Discover, Hallucinate, and Adapt: Open Compound Domain Adaptation for Semantic Segmentation

KwanYong Park, Sanghyun Woo, Inkyu Shin, In So Kweon
2020 Neural Information Processing Systems  
Then, it hallucinates multiple latent target domains in source by using image-translation (hallucinate). This step ensures the latent domains in the source and the target to be paired.  ...  ., robot control, autonomous driving, medical imaging, etc.).  ...  Hallucinate: Latent Target Domains Hallucination in Source We now hallucinate K latent target domains in the source domain. In this work, we formulate it as image-translation [22, 44, 15, 18] .  ... 
dblp:conf/nips/ParkWSK20 fatcat:6ms5scfdpnfphnjdflgjqm2rxy

Unsupervised Domain Adaptation via CycleGAN for White Matter Hyperintensity Segmentation in Multicenter MR Images [article]

Julian Alberto Palladino, Diego Fernandez Slezak, Enzo Ferrante
2020 arXiv   pre-print
Automatic segmentation of white matter hyperintensities in magnetic resonance images is of paramount clinical and research importance.  ...  We aim at learning a mapping function to transform volumetric MR images between domains, which are characterized by different medical centers and MR machines with varying brand, model and configuration  ...  Cycle consistency loss. Adversarial losses alone do not guarantee that images can be converted back and forth from X to Y and vice versa.  ... 
arXiv:2009.04985v1 fatcat:vnk77kisdrf7teo6eo26zeqs2q

Concerns in the use of adversarial learning for image synthesis in cardiovascular intervention

Akinori Higaki, Toru Miyoshi, Osamu Yamaguchi
2021 European Heart Journal - Digital Health  
In this article, the authors depict the various 7 benefits of medical image synthesis, by showing actual OCT images generated from 8 conditional generative adversarial networks (GAN).  ...  Considering the differences in 20 physical properties and image resolution between IVUS and OCT, image translation 21 between them is not an easy task and will require an extremely large amount of data  ...  Distribution matching losses can hallucinate 6 features in medical image translation. in International conference on medical 7 image computing and computer-assisted intervention 529-536 (Springer, 2018  ... 
doi:10.1093/ehjdh/ztab064 fatcat:mci53pddb5cg7lsjbskmfc2hl4

Discover, Hallucinate, and Adapt: Open Compound Domain Adaptation for Semantic Segmentation [article]

KwanYong Park, Sanghyun Woo, Inkyu Shin, In So Kweon
2021 arXiv   pre-print
Then, it hallucinates multiple latent target domains in source by using image-translation (hallucinate). This step ensures the latent domains in the source and the target to be paired.  ...  ., robot control, autonomous driving, medical imaging, etc.).  ...  Hallucinate: Latent Target Domains Hallucination in Source We now hallucinate K latent target domains in the source domain. In this work, we formulate it as image-translation [23, 47, 16, 19] .  ... 
arXiv:2110.04111v1 fatcat:udbcpkspyngvlasaywt5xzczpi

Residual CycleGAN for robust domain transformation of histopathological tissue slides

Thomas de Bel, John-Melle Bokhorst, Jeroen van der Laak, Geert Litjens
2021 Medical Image Analysis  
Variation between stains in histopathology is commonplace across different medical centers. This can have a significant effect on the reliability of machine learning algorithms.  ...  We compare segmentation performances in a colon tissue segmentation and kidney tissue segmentation task, covering data from 6 different centers.  ...  The cycle-consistency loss term allows Cycle-GANs to be trained without paired data, while still allowing for a targeted image translation.  ... 
doi:10.1016/j.media.2021.102004 pmid:33647784 fatcat:cj7oyqnlezbkvntdyi4vd76b3i

PSIGAN: Joint probabilistic segmentation and image distribution matching for unpaired cross-modality adaptation based MRI segmentation [article]

Jue Jiang, Yu Chi Hu, Neelam Tyagi, Andreas Rimner, Nancy Lee, Joseph O. Deasy, Sean Berry, Harini Veeraraghavan
2020 arXiv   pre-print
We developed a new joint probabilistic segmentation and image distribution matching generative adversarial network (PSIGAN) for unsupervised domain adaptation (UDA) and multi-organ segmentation from magnetic  ...  Our UDA approach models the co-dependency between images and their segmentation as a joint probability distribution using a new structure discriminator.  ...  But this loss alone is insufficient to preserve the SOI geometry and appearance in I2I translation of medical images [33] .  ... 
arXiv:2007.09465v1 fatcat:mzm7lgdjjzc63jaesk2jwmesqq

Self domain adapted network [article]

Yufan He, Aaron Carass, Lianrui Zuo, Blake E. Dewey, Jerry L. Prince
2020 arXiv   pre-print
Current UDA methods need both source and target data to train models which perform image translation (harmonization) or learn domain-invariant features.  ...  In this paper, we propose a novel self domain adapted network (SDA-Net) that can rapidly adapt itself to a single test subject at the testing stage, without using extra data or training a UDA model.  ...  Thirdly, deep networks can hallucinate features [4] , which is a severe problem for medical data.  ... 
arXiv:2007.03162v1 fatcat:f2tf2zpx5jajbocvjvebhonkw4

Advancing Medical Imaging Informatics by Deep Learning-Based Domain Adaptation

Anirudh Choudhary, Li Tong, Yuanda Zhu, May D. Wang
2020 IMIA Yearbook of Medical Informatics  
Objective: In this survey, we review the state-of-the-art DL-based DA methods for medical imaging.  ...  Domain adaptation (DA) has been developed to transfer the knowledge from a labeled data domain to a related but unlabeled domain in either image space or feature space.  ...  Acknowledgments The work was supported in part by grants from the National Science Foundation  ... 
doi:10.1055/s-0040-1702009 pmid:32823306 fatcat:gtlhoh6m3fh4hcumfzdlpdohr4

Generative Adversarial Network in Medical Imaging: A Review [article]

Xin Yi, Ekta Walia, Paul Babyn
2019 arXiv   pre-print
This has proven to be useful in many cases, such as domain adaptation, data augmentation, and image-to-image translation.  ...  These properties have attracted researchers in the medical imaging community, and we have seen rapid adoption in many traditional and novel applications, such as image reconstruction, segmentation, detection  ...  But using adversarial loss to match the generated and real data distribution may make the model hallucinate unseen structures.  ... 
arXiv:1809.07294v3 fatcat:5j5i6shlcvbbjm74ceidzg6rc4

Segmentation-Renormalized Deep Feature Modulation for Unpaired Image Harmonization [article]

Mengwei Ren, Neel Dey, James Fishbaugh, Guido Gerig
2021 arXiv   pre-print
to modulate features at every level of translation.  ...  Importantly, these methods are prone to instability, contrast inversion, intractable manipulation of pathology, and steganographic mappings which limit their reliable adoption in real-world medical imaging  ...  Yet, ImageNet-derived features may not yield good representations for comparing the distributions of embedded medical images.  ... 
arXiv:2102.06315v2 fatcat:lozfvb5cpnbq7ixrade7zou36e

Bridging the Gap Between Paired and Unpaired Medical Image Translation [chapter]

Pauliina Paavilainen, Saad Ullah Akram, Juho Kannala
2021 Lecture Notes in Computer Science  
Medical image translation has the potential to reduce the imaging workload, by removing the need to capture some sequences, and to reduce the annotation burden for developing machine learning methods.  ...  The proposed pix2pix variants outperform baseline pix2pix, pix2pixHD and CycleGAN in terms of FID and KID, and generate more realistic looking CT and MR translations.  ...  Since these methods can hallucinate features in images [5] , they require extensive validation of their image quality and fidelity before clinical use.  ... 
doi:10.1007/978-3-030-88210-5_4 fatcat:f5rpjmyumjcqrdr7wvvojidhhi

Unified cross-modality feature disentangler for unsupervised multi-domain MRI abdomen organs segmentation [article]

Jue Jiang, Harini Veeraraghavan
2020 arXiv   pre-print
Finally, we introduce a joint distribution matching discriminator that combines the translated images with task-relevant segmentation probability maps to further constrain and regularize image-to-image  ...  Our contribution is a unified cross-modality feature disentagling approach for multi-domain image translation and multiple organ segmentation.  ...  Ablation tests We evaluated the impact of mode-seeking loss [19] and joint distribution matching losses introduced in this work for T2w MRI segmentation.  ... 
arXiv:2007.09669v1 fatcat:lgrw64rshnfnrjqpsoueqp57ku

A Review of Generative Adversarial Networks in Cancer Imaging: New Applications, New Solutions [article]

Richard Osuala, Kaisar Kushibar, Lidia Garrucho, Akis Linardos, Zuzanna Szafranowska, Stefan Klein, Ben Glocker, Oliver Diaz, Karim Lekadir
2021 arXiv   pre-print
The recent advancements in Generative Adversarial Networks (GANs) in computer vision as well as in medical imaging may provide a basis for enhanced capabilities in cancer detection and analysis.  ...  Despite technological and medical advances, the detection, interpretation, and treatment of cancer based on imaging data continue to pose significant challenges.  ...  Feature Hallucinations in Synthetic Data As displayed in Figure 6 and denoted in Figure 3 (b), conditional GANs can unintentionally 7 hallucinate nonexistent artifacts into a patient image.  ... 
arXiv:2107.09543v1 fatcat:jz76zqklpvh67gmwnsdqzgq5he

Deep learning-based bias transfer for overcoming laboratory differences of microscopic images [article]

Ann-Katrin Thebille and Esther Dietrich and Martin Klaus and Lukas Gernhold and Maximilian Lennartz and Christoph Kuppe and Rafael Kramann and Tobias B. Huber and Guido Sauter and Victor G. Puelles and Marina Zimmermann and Stefan Bonn
2021 arXiv   pre-print
The same source tissue can be represented by vastly different images if the image acquisition or processing protocols vary.  ...  In the scope of our analysis, U-Net cycleGANs trained with an additional identity and an MS-SSIM-based loss and Fixed-Point GANs trained with an additional structure loss led to the best results for the  ...  In Figure 4c an image transformed by cycleGAN with combined losses can be seen.  ... 
arXiv:2105.11765v1 fatcat:u65e4dqswfdcjcolbp3fjg3vle
« Previous Showing results 1 — 15 out of 3,147 results