pith. sign in

arxiv: 2411.16719 · v4 · submitted 2024-11-23 · 💻 cs.CV · cs.LG

Learn2Synth: Learning Optimal Data Synthesis Using Hypergradients for Brain Image Segmentation

Xiaoling Hu , Xiangrui Zeng , Oula Puonti , Juan Eugenio Iglesias , Bruce Fischl , Yael Balbastre This is my paper

Pith reviewed 2026-05-23 17:34 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords brain image segmentationdata synthesishypergradientsdomain randomizationsynthetic data trainingaugmentation learningMRI segmentation
0
0 comments X

The pith

Tuning synthesis parameters with hypergradients lets a segmentation network trained only on synthetic brain images reach optimal accuracy on real scans.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that synthesis parameters can be learned automatically so a network trained exclusively on generated images performs best when tested on real data. This is done by using a small set of real labeled examples only to guide the tuning process via hypergradients, never to train the network itself. The approach sidesteps manual hyperparameter search and avoids techniques that align synthetic images to real ones at the risk of breaking label consistency. A sympathetic reader would care because it promises networks that generalize across domains while drawing on real labels without inheriting their biases or appearance quirks. The method is shown on both synthetic and real brain MRI scans.

Core claim

Learn2Synth optimizes the parameters of an image synthesis engine so that a segmentation network trained solely on the resulting synthetic images achieves the highest possible accuracy when evaluated on real images. The optimization uses hypergradients computed from a small set of real labeled examples; these real examples never enter the training of the segmentation network. Both parametric and nonparametric enhancement strategies are developed to improve the synthetic images in ways that directly benefit real-data performance.

What carries the argument

Hypergradient optimization of synthesis parameters, which back-propagates through the training of the segmentation network to adjust how synthetic images are generated.

If this is right

  • Parametric and nonparametric strategies can be used to enhance synthetic images in a manner that measurably raises real-data segmentation accuracy.
  • The segmentation network remains unbiased toward the appearance statistics of any particular real training set because those images are never shown during its training.
  • The method works on both fully synthetic test cases and real-world brain MRI scans.
  • Label-image pairs stay aligned because no post-synthesis alignment step is required.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same hypergradient loop could be applied to synthesis engines for other medical imaging modalities or non-medical segmentation tasks where labeled real data are scarce.
  • If the learned synthesis parameters transfer across different network architectures, one could amortize the cost of the hypergradient search over multiple downstream models.
  • The approach might reduce the total number of real labeled examples needed overall, since only a small validation set is used for tuning rather than for network training.

Load-bearing premise

The image synthesis process must be fully differentiable with respect to its parameters.

What would settle it

After running the hypergradient procedure, train the segmentation network on images synthesized with the learned parameters and check whether its Dice score or other accuracy metric on held-out real brain scans is higher than the score obtained from the same network trained on images synthesized with manually chosen or randomly sampled parameters.

Figures

Figures reproduced from arXiv: 2411.16719 by Bruce Fischl, Juan Eugenio Iglesias, Oula Puonti, Xiangrui Zeng, Xiaoling Hu, Yael Balbastre.

Figure 1
Figure 1. Figure 1: SynthSeg [6] (top left) is a domain randomization strategy that has shown great success in neuroimaging applications, where an image with random contrast is synthesized from a label map using a set of stochastic rules. This strategy contrasts with typical supervised training [56] (bottom left), where a model is trained on real labeled images (possibly augmented). Our proposed strategy, Learn2Synth (right),… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of learned synthesized image. (a) Noise-free [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative results on OASIS3 and ABIDE. (a), (b) show the input and segmentation. (c)-(f) show the segmentations of different [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Results obtained when training on MPRAGE scans and [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the unpaired approach for learning seg [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
read the original abstract

Domain randomization through synthesis is a powerful strategy to train networks that are unbiased with respect to the domain of the input images. Randomization allows networks to see a virtually infinite range of intensities and artifacts during training, thereby minimizing overfitting to appearance and maximizing generalization to unseen data. Although powerful, this approach relies on the accurate tuning of a large set of hyperparameters that govern the probabilistic distribution of the synthesized images. Instead of manually tuning these parameters, we introduce Learn2Synth, a novel procedure in which synthesis parameters are learned using a small set of real labeled data. Unlike methods that impose constraints to align synthetic data with real data (e.g., contrastive or adversarial techniques), which risk misaligning the image and its label map, we tune an augmentation engine such that a segmentation network trained on synthetic data has optimal accuracy when applied to real data. This approach allows the training procedure to benefit from real labeled examples, without ever using these real examples to train the segmentation network, which avoids biasing the network towards the properties of the training set. Specifically, we develop parametric and nonparametric strategies to enhance synthetic images in a way that improves the performance of the segmentation network. We demonstrate the effectiveness of this learning strategy on synthetic and real-world brain scans. Code is available at: https://github.com/HuXiaoling/Learn2Synth.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces Learn2Synth, a procedure that optimizes the parameters of an image synthesis/augmentation engine via hypergradients using a small set of real labeled brain scans. A segmentation network is then trained exclusively on the resulting synthetic images and evaluated on real data, with the goal of achieving strong real-data accuracy without ever training the network on real examples.

Significance. If the hypergradient optimization produces synthesis parameters that generalize beyond the small tuning set, the method could reduce the need for large real labeled datasets in domain-randomized training while avoiding label-image misalignment risks of adversarial alignment techniques. The public code release supports reproducibility.

major comments (2)
  1. [Method and Experiments] The central claim requires that synthesis parameters optimized on the small real set generalize to unseen real data. The manuscript does not specify (in the method description or experimental protocol) whether this tuning set is held out from the final real-data test set or from any cross-validation; without this separation the reported gains could reflect indirect supervision rather than domain randomization.
  2. [Experiments] No results, ablations, or error bars are described that test sensitivity to the size of the real tuning set or that compare against a baseline where synthesis parameters are tuned without hypergradients; such controls are load-bearing for the claim that hypergradients are necessary and sufficient for the observed real-data improvement.
minor comments (1)
  1. [Abstract] The abstract mentions both parametric and nonparametric strategies but does not name them; a short enumeration in the abstract or §3 would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments below and will revise the manuscript accordingly to improve clarity and strengthen the experimental validation.

read point-by-point responses

  1. Referee: [Method and Experiments] The central claim requires that synthesis parameters optimized on the small real set generalize to unseen real data. The manuscript does not specify (in the method description or experimental protocol) whether this tuning set is held out from the final real-data test set or from any cross-validation; without this separation the reported gains could reflect indirect supervision rather than domain randomization.

    Authors: We agree that explicit specification of the data separation is necessary. The small real labeled set is used exclusively for hypergradient-based optimization of synthesis parameters and is held out from the final test set (as well as from any cross-validation folds used for evaluation). No real test images participate in the synthesis tuning process. We will revise the method section and experimental protocol to state this separation clearly and to emphasize that the reported real-data performance reflects generalization from the optimized synthesis distribution. revision: yes

  2. Referee: [Experiments] No results, ablations, or error bars are described that test sensitivity to the size of the real tuning set or that compare against a baseline where synthesis parameters are tuned without hypergradients; such controls are load-bearing for the claim that hypergradients are necessary and sufficient for the observed real-data improvement.

    Authors: We acknowledge that these controls would strengthen the paper. We will add (i) an ablation varying the number of real tuning scans and (ii) a direct comparison against a non-hypergradient baseline (e.g., manually chosen or randomly sampled synthesis parameters). Standard error bars across multiple random seeds will be reported for all quantitative results. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method uses explicit outer-loop optimization without reducing claims to fitted inputs by construction

full rationale

The paper describes tuning synthesis parameters via hypergradients on a small real labeled set so that a network trained exclusively on the resulting synthetic images achieves good accuracy on real data. No equations, procedures, or self-citations in the abstract or described method reduce the performance claim to a quantity defined from the same inputs by construction. The separation between synthesis tuning (outer loop) and network training (inner loop on synthetic data only) is maintained, and the approach is presented as a search over augmentation parameters rather than a tautological renaming or fit. This is the most common honest finding for self-contained optimization methods evaluated on external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the method implicitly assumes differentiability of the synthesis pipeline.

axioms (1)
  • domain assumption The image synthesis process is differentiable with respect to its governing parameters.
    Required for hypergradient computation; stated implicitly by the use of hypergradients.

pith-pipeline@v0.9.0 · 5786 in / 1151 out tokens · 36068 ms · 2026-05-23T17:34:12.913101+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

102 extracted references · 102 canonical work pages · 5 internal anchors

  1. [1]

    A learning-based data aug- mentation for network anomaly detection

    Mohammad Al Olaimat, Dongeun Lee, Youngsoo Kim, Jonghyun Kim, and Jinoh Kim. A learning-based data aug- mentation for network anomaly detection. In ICCCN, 2020. 3

    work page 2020

  2. [2]

    The medical segmentation decathlon.Nature communications, 2022

    Michela Antonelli, Annika Reinke, Spyridon Bakas, Key- van Farahani, Annette Kopp-Schneider, Bennett A Landman, Geert Litjens, Bjoern Menze, Olaf Ronneberger, Ronald M Summers, et al. The medical segmentation decathlon.Nature communications, 2022. 1

    work page 2022

  3. [3]

    Deep learning does not replace bayesian modeling: Comparing research use via citation counting.Ap- plied AI Letters, 2022

    Breck Baldwin. Deep learning does not replace bayesian modeling: Comparing research use via citation counting.Ap- plied AI Letters, 2022. 1

    work page 2022

  4. [4]

    Online learning rate adaptation with hypergradient descent

    Atilim Gunes Baydin, Robert Cornish, David Mart ´ınez- Rubio, Mark Schmidt, and Frank Wood. Online learning rate adaptation with hypergradient descent. In ICLR, 2018. 2

    work page 2018

  5. [5]

    Deep learners benefit more from out- of-distribution examples

    Yoshua Bengio, Fr ´ed´eric Bastien, Arnaud Bergeron, Nicolas Boulanger-Lewandowski, Thomas Breuel, Youssouf Chher- awala, Moustapha Cisse, Myriam C ˆot´e, Dumitru Erhan, Jeremy Eustache, et al. Deep learners benefit more from out- of-distribution examples. In AISTATS, 2011. 1

    work page 2011

  6. [6]

    SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

    Benjamin Billot, Douglas N Greve, Oula Puonti, Axel Thielscher, Koen Van Leemput, Bruce Fischl, Adrian V Dalca, Juan Eugenio Iglesias, et al. SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining. MedIA, 2023. 2, 3, 7, 14, 16

    work page 2023

  7. [7]

    Robust machine learning segmentation for large-scale analysis of heterogeneous clinical brain MRI datasets

    Benjamin Billot, Colin Magdamo, You Cheng, Steven E Arnold, Sudeshna Das, and Juan Eugenio Iglesias. Robust machine learning segmentation for large-scale analysis of heterogeneous clinical brain MRI datasets. PNAS, 2023. 3

    work page 2023

  8. [8]

    Assessing radiology research on artificial in- telligence: a brief guide for authors, reviewers, and read- ers—from the radiology editorial board, 2020

    David A Bluemke, Linda Moy, Miriam A Bredella, Bir- git B Ertl-Wagner, Kathryn J Fowler, Vicky J Goh, Elkan F Halpern, Christopher P Hess, Mark L Schiebler, and Clif- ford R Weiss. Assessing radiology research on artificial in- telligence: a brief guide for authors, reviewers, and read- ers—from the radiology editorial board, 2020. 1

    work page 2020

  9. [9]

    Semi- supervised and task-driven data augmentation

    Krishna Chaitanya, Neerav Karani, Christian F Baumgartner, Anton Becker, Olivio Donati, and Ender Konukoglu. Semi- supervised and task-driven data augmentation. In IPMI,

    work page

  10. [10]

    Contrastive learning of global and local features for medical image segmentation with limited annotations

    Krishna Chaitanya, Ertunc Erdil, Neerav Karani, and Ender Konukoglu. Contrastive learning of global and local features for medical image segmentation with limited annotations. In NeurIPS, 2020. 3

    work page 2020

  11. [11]

    Gradient descent: The ultimate optimizer

    Kartik Chandra, Audrey Xie, Jonathan Ragan-Kelley, and Erik Meijer. Gradient descent: The ultimate optimizer. In NeurIPS, 2022. 2

    work page 2022

  12. [12]

    Multimodal MR synthesis via modality-invariant latent representation

    Agisilaos Chartsias, Thomas Joyce, Mario Valerio Giuffrida, and Sotirios A Tsaftaris. Multimodal MR synthesis via modality-invariant latent representation. TMI, 2017. 3

    work page 2017

  13. [13]

    Synergistic image and feature adaptation: Towards cross-modality domain adaptation for medical image seg- mentation

    Cheng Chen, Qi Dou, Hao Chen, Jing Qin, and Pheng-Ann Heng. Synergistic image and feature adaptation: Towards cross-modality domain adaptation for medical image seg- mentation. In AAAI, 2019. 3

    work page 2019

  14. [14]

    Enhancing MR image segmentation with realistic adversarial data augmentation

    Chen Chen, Chen Qin, Cheng Ouyang, Zeju Li, Shuo Wang, Huaqi Qiu, Liang Chen, Giacomo Tarroni, Wenjia Bai, and Daniel Rueckert. Enhancing MR image segmentation with realistic adversarial data augmentation. MedIA, 2022. 3, 7

    work page 2022

  15. [15]

    TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation

    Jieneng Chen, Yongyi Lu, Qihang Yu, Xiangde Luo, Ehsan Adeli, Yan Wang, Le Lu, Alan L Yuille, and Yuyin Zhou. Transunet: Transformers make strong encoders for medi- cal image segmentation. arXiv preprint arXiv:2102.04306,

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Semantic image seg- mentation with deep convolutional nets and fully connected CRFs. arXiv preprint arXiv:1412.7062, 2014. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  17. [17]

    Rethinking Atrous Convolution for Semantic Image Segmentation

    Liang-Chieh Chen, George Papandreou, Florian Schroff, and Hartwig Adam. Rethinking atrous convolution for seman- tic image segmentation. arXiv preprint arXiv:1706.05587 ,

    work page internal anchor Pith review Pith/arXiv arXiv

  18. [18]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected CRFs

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolu- tion, and fully connected CRFs. TPAMI, 2018. 13

    work page 2018

  19. [19]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations. In ICML, 2020. 2

    work page 2020

  20. [20]

    Autoaugment: Learning augmentation strategies from data

    Ekin D Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasude- van, and Quoc V Le. Autoaugment: Learning augmentation strategies from data. In CVPR, 2019. 3

    work page 2019

  21. [21]

    The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism

    Adriana Di Martino, Chao-Gan Yan, Qingyang Li, Erin De- nio, Francisco X Castellanos, Kaat Alaerts, Jeffrey S Ander- son, Michal Assaf, Susan Y Bookheimer, Mirella Dapretto, et al. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism. Molecular psychiatry, 2014. 7, 13

    work page 2014

  22. [22]

    PnP- AdaNet: Plug-and-play adversarial domain adaptation net- work at unpaired cross-modality cardiac segmentation.IEEE Access, 2019

    Qi Dou, Cheng Ouyang, Cheng Chen, Hao Chen, Ben Glocker, Xiahai Zhuang, and Pheng-Ann Heng. PnP- AdaNet: Plug-and-play adversarial domain adaptation net- work at unpaired cross-modality cardiac segmentation.IEEE Access, 2019. 3

    work page 2019

  23. [23]

    Whole brain segmentation: automated labeling of neu- roanatomical structures in the human brain

    Bruce Fischl, David H Salat, Evelina Busa, Marilyn Al- bert, Megan Dieterich, Christian Haselgrove, Andre Van Der Kouwe, Ron Killiany, David Kennedy, Shuna Klaveness, et al. Whole brain segmentation: automated labeling of neu- roanatomical structures in the human brain. Neuron, 2002. 5, 7, 8, 13, 14

    work page 2002

  24. [24]

    Sequence-independent segmentation of mag- netic resonance images

    Bruce Fischl, David H Salat, Andr ´e JW Van Der Kouwe, Nikos Makris, Florent S ´egonne, Brian T Quinn, and An- ders M Dale. Sequence-independent segmentation of mag- netic resonance images. Neuroimage, 2004. 8, 13

    work page 2004

  25. [25]

    Forward and reverse gradient-based hyper- parameter optimization

    Luca Franceschi, Michele Donini, Paolo Frasconi, and Mas- similiano Pontil. Forward and reverse gradient-based hyper- parameter optimization. In ICML, 2017. 2

    work page 2017

  26. [26]

    Synthetic data augmentation using GAN for improved liver lesion classification

    Maayan Frid-Adar, Eyal Klang, Michal Amitai, Jacob Gold- berger, and Hayit Greenspan. Synthetic data augmentation using GAN for improved liver lesion classification. In ISBI,

    work page

  27. [27]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pas- cal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks. JMLR, 2016. 3

    work page 2016

  28. [28]

    Synthesizing programs for im- ages using reinforced adversarial learning

    Yaroslav Ganin, Tejas Kulkarni, Igor Babuschkin, SM Ali Eslami, and Oriol Vinyals. Synthesizing programs for im- ages using reinforced adversarial learning. In ICML, 2018. 2

    work page 2018

  29. [29]

    Faster autoaugment: Learning augmentation strategies using backpropagation

    Ryuichiro Hataya, Jan Zdenek, Kazuki Yoshizoe, and Hideki Nakayama. Faster autoaugment: Learning augmentation strategies using backpropagation. In ECCV, 2020. 3

    work page 2020

  30. [30]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR,

    work page

  31. [31]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In CVPR, 2020. 2

    work page 2020

  32. [32]

    Autoencoder based self-supervised test-time adaptation for medical image analysis

    Yufan He, Aaron Carass, Lianrui Zuo, Blake E Dewey, and Jerry L Prince. Autoencoder based self-supervised test-time adaptation for medical image analysis. MedIA, 2021. 3

    work page 2021

  33. [33]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. In NeurIPS, 2020. 2

    work page 2020

  34. [34]

    CyCADA: Cycle-consistent adversarial domain adaptation

    Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei Efros, and Trevor Darrell. CyCADA: Cycle-consistent adversarial domain adaptation. In ICML, 2018. 3

    work page 2018

  35. [35]

    SynthMorph: learning contrast-invariant registration without acquired images

    Malte Hoffmann, Benjamin Billot, Douglas N Greve, Juan Eugenio Iglesias, Bruce Fischl, and Adrian V Dalca. SynthMorph: learning contrast-invariant registration without acquired images. TMI, 2021. 3

    work page 2021

  36. [36]

    SynthStrip: skull-stripping for any brain image

    Andrew Hoopes, Jocelyn S Mora, Adrian V Dalca, Bruce Fischl, and Malte Hoffmann. SynthStrip: skull-stripping for any brain image. NeuroImage, 2022. 3

    work page 2022

  37. [37]

    Synseg-net: Synthetic segmentation without target modality ground truth

    Yuankai Huo, Zhoubing Xu, Hyeonsoo Moon, Shunxing Bao, Albert Assad, Tamara K Moyo, Michael R Savona, Richard G Abramson, and Bennett A Landman. Synseg-net: Synthetic segmentation without target modality ground truth. TMI, 2018. 3

    work page 2018

  38. [38]

    SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution t1-weighted images for 3D morphometry

    Juan E Iglesias, Benjamin Billot, Ya ¨el Balbastre, Colin Magdamo, Steven E Arnold, Sudeshna Das, Brian L Ed- low, Daniel C Alexander, Polina Golland, and Bruce Fis- chl. SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution t1-weighted images for 3D morphometry. Science advances, 2023. 3

    work page 2023

  39. [39]

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmen- tation

    Fabian Isensee, Paul F Jaeger, Simon AA Kohl, Jens Pe- tersen, and Klaus H Maier-Hein. nnU-Net: a self-configuring method for deep learning-based biomedical image segmen- tation. Nature methods, 2021. 1, 13

    work page 2021

  40. [40]

    Image-to-image translation with conditional adver- sarial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adver- sarial networks. In CVPR, 2017. 3

    work page 2017

  41. [41]

    Noise and the reality gap: The use of simulation in evolutionary robotics

    Nick Jakobi, Phil Husbands, and Inman Harvey. Noise and the reality gap: The use of simulation in evolutionary robotics. In Advances in Artificial Life: Third European Con- ference on Artificial Life Granada, 1995. 2, 3

    work page 1995

  42. [42]

    Pulse sequence resilient fast brain segmentation

    Amod Jog and Bruce Fischl. Pulse sequence resilient fast brain segmentation. In MICCAI, 2018. 1, 3

    work page 2018

  43. [43]

    Unsupervised domain adaptation in brain le- sion segmentation with adversarial networks

    Konstantinos Kamnitsas, Christian Baumgartner, Christian Ledig, Virginia Newcombe, Joanna Simpson, Andrew Kane, David Menon, Aditya Nori, Antonio Criminisi, Daniel Rueckert, et al. Unsupervised domain adaptation in brain le- sion segmentation with adversarial networks. In IPMI, 2017. 3

    work page 2017

  44. [44]

    Test-time adaptable neural networks for robust medical image segmentation

    Neerav Karani, Ertunc Erdil, Krishna Chaitanya, and Ender Konukoglu. Test-time adaptable neural networks for robust medical image segmentation. MedIA, 2021. 3

    work page 2021

  45. [45]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    work page internal anchor Pith review Pith/arXiv arXiv

  46. [46]

    Glow: Generative flow with invertible 1x1 convolutions

    Durk P Kingma and Prafulla Dhariwal. Glow: Generative flow with invertible 1x1 convolutions. In NeurIPS, 2018. 2

    work page 2018

  47. [47]

    Imagenet classification with deep convolutional neural net- works

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural net- works. In NeurIPS, 2012. 13

    work page 2012

  48. [48]

    OASIS-3: longitudinal neuroimaging, clin- ical, and cognitive dataset for normal aging and alzheimer disease

    Pamela J LaMontagne, Tammie LS Benzinger, John C Mor- ris, Sarah Keefe, Russ Hornbeck, Chengjie Xiong, Eliza- beth Grant, Jason Hassenstab, Krista Moulder, Andrei G Vlassenko, et al. OASIS-3: longitudinal neuroimaging, clin- ical, and cognitive dataset for normal aging and alzheimer disease. medrxiv, 2019. 5, 7, 13

    work page 2019

  49. [49]

    Semantic segmentation with generative models: Semi-supervised learning and strong out-of-domain generalization

    Daiqing Li, Junlin Yang, Karsten Kreis, Antonio Torralba, and Sanja Fidler. Semantic segmentation with generative models: Semi-supervised learning and strong out-of-domain generalization. In CVPR, 2021. 3

    work page 2021

  50. [50]

    Fast autoaugment

    Sungbin Lim, Ildoo Kim, Taesup Kim, Chiheon Kim, and Sungwoong Kim. Fast autoaugment. In NeurIPS, 2019. 3

    work page 2019

  51. [51]

    Undoing the damage of label shift for cross- domain semantic segmentation

    Yahao Liu, Jinhong Deng, Jiale Tao, Tong Chu, Lixin Duan, and Wen Li. Undoing the damage of label shift for cross- domain semantic segmentation. In CVPR, 2022. 2

    work page 2022

  52. [52]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In CVPR, 2015. 13

    work page 2015

  53. [53]

    Gradient-based hyperparameter optimization through re- versible learning

    Dougal Maclaurin, David Duvenaud, and Ryan Adams. Gradient-based hyperparameter optimization through re- versible learning. In ICML, 2015. 2

    work page 2015

  54. [54]

    Deep adversarial training for multi-organ nuclei segmentation in histopathology images

    Faisal Mahmood, Daniel Borders, Richard J Chen, Gre- gory N McKay, Kevan J Salimian, Alexander Baras, and Nicholas J Durr. Deep adversarial training for multi-organ nuclei segmentation in histopathology images. TMI, 2019. 3

    work page 2019

  55. [55]

    Generalizability of machine learning models: quantitative evaluation of three methodological pitfalls

    Farhad Maleki, Katie Ovens, Rajiv Gupta, Caroline Rein- hold, Alan Spatz, and Reza Forghani. Generalizability of machine learning models: quantitative evaluation of three methodological pitfalls. Radiology: Artificial Intelligence ,

    work page

  56. [56]

    V-net: Fully convolutional neural networks for volumetric medical image segmentation

    Fausto Milletari, Nassir Navab, and Seyed-Ahmad Ahmadi. V-net: Fully convolutional neural networks for volumetric medical image segmentation. In 3DV, 2016. 2

    work page 2016

  57. [57]

    When does label smoothing help? In NeurIPS, 2019

    Rafael M ¨uller, Simon Kornblith, and Geoffrey E Hinton. When does label smoothing help? In NeurIPS, 2019. 2

    work page 2019

  58. [58]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In ICML, 2021. 2

    work page 2021

  59. [59]

    Synthetic data for deep learning

    Sergey I Nikolenko. Synthetic data for deep learning . Springer, 2021. 2

    work page 2021

  60. [60]

    Learning deconvolution network for semantic segmentation

    Hyeonwoo Noh, Seunghoon Hong, and Bohyung Han. Learning deconvolution network for semantic segmentation. In ICCV, 2015. 13

    work page 2015

  61. [61]

    Masked autoregressive flow for density estimation

    George Papamakarios, Theo Pavlakou, and Iain Murray. Masked autoregressive flow for density estimation. In NeurIPS, 2017. 2

    work page 2017

  62. [62]

    Fast and sequence-adaptive whole-brain segmentation using parametric bayesian modeling

    Oula Puonti, Juan Eugenio Iglesias, and Koen Van Leemput. Fast and sequence-adaptive whole-brain segmentation using parametric bayesian modeling. NeuroImage, 2016. 7

    work page 2016

  63. [63]

    Playing for data: Ground truth from computer games

    Stephan R Richter, Vibhav Vineet, Stefan Roth, and Vladlen Koltun. Playing for data: Ground truth from computer games. In ECCV, 2016. 3

    work page 2016

  64. [64]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015. 7, 13, 14

    work page 2015

  65. [65]

    Data augmentation using generative adversarial networks (cyclegan) to improve generalizability in CT seg- mentation tasks

    Veit Sandfort, Ke Yan, Perry J Pickhardt, and Ronald M Summers. Data augmentation using generative adversarial networks (cyclegan) to improve generalizability in CT seg- mentation tasks. Scientific reports, 2019. 3

    work page 2019

  66. [66]

    Different scaling of linear models and deep learning in ukbiobank brain images versus machine-learning datasets

    Marc-Andre Schulz, BT Thomas Yeo, Joshua T V ogelstein, Janaina Mourao-Miranada, Jakob N Kather, Konrad Kord- ing, Blake Richards, and Danilo Bzdok. Different scaling of linear models and deep learning in ukbiobank brain images versus machine-learning datasets. Nature communications,

    work page

  67. [67]

    A survey on image data augmentation for deep learning

    Connor Shorten and Taghi M Khoshgoftaar. A survey on image data augmentation for deep learning. Journal of big data, 2019. 1

    work page 2019

  68. [68]

    Learning from simulated and unsupervised images through adversarial training

    Ashish Shrivastava, Tomas Pfister, Oncel Tuzel, Joshua Susskind, Wenda Wang, and Russell Webb. Learning from simulated and unsupervised images through adversarial training. In CVPR, 2017. 2

    work page 2017

  69. [69]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  70. [70]

    A ves- sel segmentation-based CycleGAN for unpaired multi-modal retinal image synthesis

    Aline Sindel, Andreas Maier, and Vincent Christlein. A ves- sel segmentation-based CycleGAN for unpaired multi-modal retinal image synthesis. In BVM Workshop, 2023. 3

    work page 2023

  71. [71]

    Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations

    Carole H Sudre, Wenqi Li, Tom Vercauteren, Sebastien Ourselin, and M Jorge Cardoso. Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: Third International Workshop, DLMIA 2017, and 7th International Workshop, ML-C...

    work page 2017

  72. [72]

    One-shot synthesis of images and segmentation masks

    Vadim Sushko, Dan Zhang, Juergen Gall, and Anna Khoreva. One-shot synthesis of images and segmentation masks. In WACV, 2023. 3

    work page 2023

  73. [73]

    Going deeper with convolutions

    Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In CVPR, 2015. 13

    work page 2015

  74. [74]

    Improving deep learning with generic data augmentation

    Luke Taylor and Geoff Nitschke. Improving deep learning with generic data augmentation. In SSCI, 2018. 1

    work page 2018

  75. [75]

    Domain randomization for transferring deep neural networks from simulation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Woj- ciech Zaremba, and Pieter Abbeel. Domain randomization for transferring deep neural networks from simulation to the real world. In IROS, 2017. 1

    work page 2017

  76. [76]

    Training deep networks with synthetic data: Bridging the reality gap by do- main randomization

    Jonathan Tremblay, Aayush Prakash, David Acuna, Mark Brophy, Varun Jampani, Cem Anil, Thang To, Eric Cam- eracci, Shaad Boochoon, and Stan Birchfield. Training deep networks with synthetic data: Bridging the reality gap by do- main randomization. In CVPR workshops, 2018. 1

    work page 2018

  77. [77]

    Repurposing GANs for one-shot se- mantic part segmentation

    Nontawat Tritrong, Pitchaporn Rewatbowornwong, and Su- pasorn Suwajanakorn. Repurposing GANs for one-shot se- mantic part segmentation. In CVPR, 2021. 3

    work page 2021

  78. [78]

    Cross-validation failure: Small sample sizes lead to large error bars

    Ga ¨el Varoquaux. Cross-validation failure: Small sample sizes lead to large error bars. Neuroimage, 2018. 1

    work page 2018

  79. [79]

    Machine learn- ing for medical imaging: methodological failures and rec- ommendations for the future

    Ga ¨el Varoquaux and Veronika Cheplygina. Machine learn- ing for medical imaging: methodological failures and rec- ommendations for the future. NPJ digital medicine , 2022. 1

    work page 2022

  80. [80]

    Deep visual domain adapta- tion: A survey

    Mei Wang and Weihong Deng. Deep visual domain adapta- tion: A survey. Neurocomputing, 2018. 3

    work page 2018

Showing first 80 references.