Pith. sign in

REVIEW 4 major objections 5 minor 41 references

Enabling PSO-Secure Synthetic Data Sharing Using Diversity-Aware Diffusion Models

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Diversity-aware diffusion plus a re-identification filter makes synthetic chest X-rays both GDPR-safe and within one point of real-data performance.

desk verdict Useful empirical recipe undercut by an unsupported privacy guarantee: the paper equates a re-ID filter with PSO-security, a load-bearing conflation. read the letter →

arxiv 2506.17975 v1 pith:YMBUBK4I submitted 2025-06-22 cs.CV

Mischa Dombrowski , Bernhard Kainz This is my paper · ORCID
classification cs.CV
keywords syntheticdataprivacy-preservingsharingdiffusionmodelspredicatesingling-outGDPRmedicalimagingdiversity-awaregenerationre-identification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that synthetic medical images can be both legally shareable under the GDPR and nearly as useful as real images for training classifiers. Its route is to treat diversity, not fidelity, as the privacy mechanism: a diffusion model conditioned on self-supervised visual features reproduces disease-relevant predicates while a re-identification filter rejects any generated image that still matches a training identity. On three chest X-ray datasets the filtered synthetic sets close the downstream AUCROC gap to real data from roughly three to four percentage points down to 0.95 points, and combining synthetic sets across datasets improves cross-institution generalization. A sympathetic reader would take the central claim to be that singling-out protection and downstream utility are the same optimization target, diversity.

What carries the argument

The load-bearing object is the re-identification filter $P$ combined with diversity-aware conditioning. SwAV features serve as pseudo-conditional labels $c_s$ that decouple image quality from sample diversity; the diffusion model is asked to generate images sharing $c_s$ while the filter $P$, trained on subject labels, deletes candidates for which $P(x,x')$ is not zero. A final BCE-based selection step chooses the surviving candidate whose classifier prediction best matches the real image, preserving task-relevant predicates. The paper's equation (1) packages this as $x' = \arg\min_{x'\in D^-_x} \mathrm{BCE}(C_r(x), C_r(x'))$ over the filtered set, and the authors read $P(x,x')=0$ as the operational translation of legal non-singling-out.

What would settle it

Release the generated datasets and run an adversarial predicate search: if any synthetic image that the filter accepts ($P=0$) can be matched to exactly one real patient using a predicate combination absent from the filter's training labels, the PSO-security claim is false.

Watch

Extended reading notes

Core claim

The authors claim that a PSO-secure synthetic dataset, one where no combination of observable predicates lets an attacker single out a real patient, can be generated by conditioning a DiADM-style diffusion model on pseudo-labels extracted by a SwAV feature encoder and then filtering candidates through a re-identification model. They formalize the filter as $P(x,x')=0$, meaning knowing the predicate features of a real image $x$ does not imply knowledge of the identity behind a synthetic image $x'$. The selected synthetic image is the one minimizing the binary cross-entropy between a real-data classifier's prediction on $x$ and on the candidate. Empirically, models trained only on these synthetic sets reach 84.04 AUCROC against 84.99 for real data, and datasets combined across NIH, CXR, and CXP generalize better than models trained on a single real dataset.

Load-bearing premise

The whole privacy claim rests on the re-identification filter catching every image that could single out a real person; the paper itself concedes the results depend heavily on this filter.

Editorial extensions

If this is right

  • If the central claim holds, hospitals can share synthetic versions of chest X-ray datasets across institutions without entering GDPR personal-data territory, since the data are treated as non-personal.
  • Models trained on synthetic data from this pipeline stay within about one AUCROC point of real-data models, making synthetic-only training a realistic option for in-domain classification.
  • The IRS value above one indicates the conditioning scheme produces more diversity than an ideal unconditional sampler, so diversity and fidelity are not in conflict here.
  • Because the method outperforms the EDM-2, EDM-2-AG, and DiADM baselines on FID, IRS, and downstream AUCROC, the same framework is offered as a generalisable template for other personal-data domains beyond chest X-rays.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's privacy guarantee is only as strong as the re-identification filter's coverage; I would not infer GDPR anonymity beyond the specific predicates and subject labels that filter was trained on.
  • A natural stress test the authors did not run is an adversarial predicate search: enumerate rare attribute combinations in the real dataset and check whether any synthetic candidate realizes a unique combination despite passing $P=0$.
  • The same diversity-as-privacy mechanism could be ported to non-image personal data such as tabular or time-series records by replacing pixel predicates with feature vectors, but the legal equivalence would need to be re-validated for each modality.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a framework for generating synthetic chest X-ray datasets using a diffusion model conditioned on SwAV pseudo-conditional features, followed by a re-identification filter that removes samples deemed too similar to real training images. The authors claim that the resulting datasets are PSO-secure under the GDPR and that downstream models trained on them reach within one percentage point of real-data AUCROC (Table 1, gap -0.95). Experiments are reported on MIMIC-CXR, CheXpert, and ChestXray8 with FID, IRS, and AUCROC metrics, and the authors include a data-sharing experiment in Table 2.

Significance. If the privacy claim were substantiated, the paper would be significant: it offers a concrete pipeline for sharing medical synthetic data while claiming a formal legal privacy guarantee, and it releases code and uses public benchmarks. The diversity-oriented generation results are also interesting, as the SwAV conditioning achieves FID 5.0 and IRS 1.58, outperforming the compared baselines. However, the central PSO-security claim is not established: the formal argument equates a learned re-identification classifier with the Cohen-Nissim predicate singling-out property, and the paper's own Limitations section concedes that the results depend on the privacy filter. The headline performance claim is based on a single generated dataset without error bars or repeated sampling. The significance is therefore conditional on a substantially revised privacy analysis.

major comments (4)
  1. [Section 3, Eq. (1)] The only formal link between the method and PSO-security is invalid. Equation (1) defines the accepted synthetic sample as the one minimizing BCE(Cr(x), Cr(x')) among candidates with P(x,x')=0, and the text states that P(x,x')=0 'ensures that knowing predicates cs does not imply knowledge of the identity.' However, P is a pairwise re-identification model trained on subject labels (Section 2, refs. [10,33]); it does not implement or approximate the Cohen-Nissim notion of predicate singling out, which is a dataset-level property about whether a set of predicates uniquely identifies a record. A synthetic image can be pairwise dissimilar from its source image (so P=0) and still be unique with respect to a combination of attributes such as sex, age, disease labels, or support devices. The paper never evaluates the released dataset against a singling-out attack, and the Limitations section explicitly states that the results 'highly depend on the privacy filters' and only claims 'generalization according to a privacy filter.' The central claim that the datasets are PSO-secure and 'unpersonal' under the GDPR is therefore unsupported.
  2. [Section 4, Tab. 1 and Fig. 3] The headline performance claim of 'within one percentage point of real-data models' rests on a single generated dataset per method: the text states that 'each generative model samples one synthetic dataset, D′, which is then split according to the ten-fold cross-validation.' Table 1 reports one FID, one IRS, and one AUCROC gap value with no error bars, confidence intervals, or repeated sampling trials, so the -0.95 gap is not statistically characterized. The ten-fold cross-validation in Fig. 3 splits this one sample, and the paper explicitly notes that subject overlap is ignored, a choice that affects the real-data baseline. Thus the abstract's quantitative claim is not supported as reported.
  3. [Section 4, privacy evaluation] The paper's only privacy-related quantitative result is the re-identification filter's own 'combined test performance of 96% AUCROC on re-identification.' This measures how well P detects whether two images come from the same subject; it is not a measure of predicate singling-out. Because the claimed privacy guarantee is defined as the output of this same filter, the evaluation is circular: a dataset is declared PSO-secure if P does not flag it, and P's accuracy is then presented as evidence of PSO-security. No membership-inference or singling-out attack is applied to the released synthetic dataset, and no adversarial predicate set is considered.
  4. [Section 1 and Table 2] The contribution statement says that 'models trained exclusively on our synthetic data generalize better than those trained on real data alone,' but Table 2 shows the opposite in all nine training/test settings: for training on NIH, Real vs. Ours is 85.41 vs. 83.65 on NIH, 81.78 vs. 79.60 on CXR, and 79.62 vs. 76.16 on CXP, with the same pattern for training on CXR and CXP. The text later concedes that 'the model is still not on par with real data.' This internal contradiction should be resolved.
minor comments (5)
  1. [Throughout] There are several typos: 'largley' in the Abstract, 'illustraction' in the Fig. 2 caption, 'instituations' in Section 3, 'mimicing' in Section 4, and 'Suprisingly' in Section 4; these should be corrected.
  2. [Table 2] The baseline 'Rec. (SDv2)' is not defined in the text, and the header 'PSO-s.' is unexplained; the reader cannot tell what is being compared.
  3. [Section 3, Eq. (1)] The set D^-_x is not formally defined, and it is unclear whether BCE is computed on predicted class probabilities, logits, or binary disease labels; please clarify the notation.
  4. [Section 2 and 4] The re-identification model P is central to the privacy argument, but its architecture, training data, and operating point are not described; at minimum, the false-negative rate on a held-out re-identification task should be reported.
  5. [Section 1] The paper cites Cohen et al. [6] for PSO-security but does not reproduce or state their formal definition; adding the definition would help readers judge what is being claimed.

Circularity Check

1 steps flagged · score 8.0 of 10

The central PSO-security claim is definitional: the paper equates its re-identification filter's output with the absence of predicate singling out, and the Limitations concede that the guarantee depends entirely on that filter.

  1. self definitional [Section 3, Eq. (1); Limitations]
    "x′ = arg min x′∈D− x BCE(Cr(x), Cr(x′)) for D− x := {x′|P(x, x′) = 0} (1) where P(x, x′) = 0 ensures that knowing predicates cs does not imply knowledge of the identity. ... Additionally, the results highly depend on the privacy filters. Nevertheless, we have successfully shown that our filtering approach can generate images that are sufficiently different to achieve generalization according to a privacy filter."

    The only formal link between the concrete computation and PSO-security is the assertion that P(x,x')=0 guarantees that knowing predicates does not imply knowledge of identity. P is a pairwise re-identification classifier trained on subject labels, not an implementation of the Cohen–Nissim predicate-singling-out property, which is a dataset-level condition over arbitrary predicate combinations. The paper's own Limitations concede that 'the results highly depend on the privacy filters.' Thus the GDPR/PSO claim is defined by the filter's own output: a synthetic image is declared PSO-secure because the filter accepts it, rather than because the external PSO property has been formalized or verified.

full rationale

The downstream performance claim is benchmarked against held-out real-data AUCROC, so the reported -0.95 gap is an empirical result rather than a purely definitional one, though the selection criterion in Eq. (1) uses the real-data classifier Cr and may inflate the apparent improvement. The central novel claim, however, is PSO-security under the GDPR. The paper never formalizes the Cohen–Nissim dataset-level property, never evaluates a singling-out attack, and the only formal bridge is the statement that P(x,x')=0 'ensures that knowing predicates cs does not imply knowledge of the identity.' Because P is a learned re-identification model from the authors' prior work rather than a verified instantiation of predicate singling-out security, the privacy guarantee reduces by construction to the filter's own acceptance criterion. The Limitations reinforce this by saying that the results 'highly depend on the privacy filters' and that the approach 'can generate images that are sufficiently different to achieve generalization according to a privacy filter.' This makes the privacy claim definitional: accepted samples are deemed PSO-secure because the filter accepted them. Score 8 reflects that the central result is forced by definition, while the downstream accuracy component retains some independent empirical content.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method's privacy guarantee rests on a re-identification filter being a complete proxy for Cohen-Nissim predicate singling-out; no proof is given. The downstream performance rests on a hand-chosen guidance strength of 1.2, batch size 32, and BCE selection against a real-data classifier, which are fitted choices rather than derived. The axioms listed are the structural assumptions that, if false, would invalidate the paper's central claims.

free parameters (4)
  • Guidance strength for classifier guidance = 1.2
    Selected as the best-performing setting among guidance strategies; the reported downstream gap depends on this choice.
  • Sampling batch size b = 32
    Used for filtering and selection; sampling time scales by factor b, and the filter's behavior depends on this batch.
  • Privacy filter decision threshold = not specified
    The re-identification model P is used as a binary filter, but the operating threshold is not reported, and the results 'highly depend on the privacy filters' as the authors note.
  • Guidance reduction on resampling = 0.1
    When all generated samples are flagged as privacy risks, the guidance strength is reduced by 0.1 and sampling is repeated. This is an ad hoc rule that affects the output distribution.
assumptions (4)
  • domain assumption Cohen and Nissim's PSO-security definition is the correct formalization of the GDPR notion of singling out.
    The paper adopts this legal interpretation from the cited literature without discussion, and uses it as the basis for the privacy claim.
  • ad hoc to paper The re-identification model P trained on subject labels is a valid and complete detector of predicate singling-out.
    This is the load-bearing assumption: P(x,x')=0 is equated with non-personal data, but P only checks image-level similarity between a synthetic and a real sample, not predicate combinations across the dataset. No connection to the Cohen-Nissim definition is established.
  • domain assumption SwAV features are better aligned with the diffusion reconstruction loss than Inception features.
    Stated in Section 3 as a belief ('we believe these visual features better align'), with empirical gains attributed to this choice but no separate validation.
  • domain assumption Selecting synthetic images by BCE with Cr's predictions preserves the information needed for downstream classification of the same disease classes.
    The selection step optimizes agreement with a real-data-trained DenseNet-121 classifier, which may bias the synthetic dataset toward that architecture and away from preserving the full data distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enabling PSO-Secure Synthetic Data Sharing Using Diversity-Aware Diffusion Models." pith.science (2026). https://pith.science/paper/YMBUBK4I

@misc{pith2026250617975,
  author       = {Pith},
  title        = {Pith review of: Enabling PSO-Secure Synthetic Data Sharing Using Diversity-Aware Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YMBUBK4I}},
  note         = {Machine review of arXiv:2506.17975}
}
read the original abstract

Synthetic data has recently reached a level of visual fidelity that makes it nearly indistinguishable from real data, offering great promise for privacy-preserving data sharing in medical imaging. However, fully synthetic datasets still suffer from significant limitations: First and foremost, the legal aspect of sharing synthetic data is often neglected and data regulations, such as the GDPR, are largley ignored. Secondly, synthetic models fall short of matching the performance of real data, even for in-domain downstream applications. Recent methods for image generation have focused on maximising image diversity instead of fidelity solely to improve the mode coverage and therefore the downstream performance of synthetic data. In this work, we shift perspective and highlight how maximizing diversity can also be interpreted as protecting natural persons from being singled out, which leads to predicate singling-out (PSO) secure synthetic datasets. Specifically, we propose a generalisable framework for training diffusion models on personal data which leads to unpersonal synthetic datasets achieving performance within one percentage point of real-data models while significantly outperforming state-of-the-art methods that do not ensure privacy. Our code is available at https://github.com/MischaD/Trichotomy.

Figures

Figures reproduced from arXiv: 2506.17975 by the authors.

Figure 1
Figure 1. Illustration of PSO-secure dataset generation. A diffusion model, guided by pre-computed features, generates synthetic images that preserve specific predicates (e.g., color) while altering identity. A re-identification model ensures that identities are not retained. As a result, the synthetic dataset maintains relevant attributes of the real data without revealing personal information, making it non-personal under G… view at source ↗
Figure 2
Figure 2. Method illustraction: We adapt recently established diversity aware diffusion models (DiADM) [11] to generate multiple images that share the same predicates ac￾cording to the SwAV feature encoder. A privacy filter ensures that the identity is not preserved. Finally, the images are ranked according to how well they preserve the pred￾icates. Formally, we aim to generate a synthetic dataset D′ of the same size and the … view at source ↗
Figure 3
Figure 3. Critical difference diagram of different generation methods [21]. To better understand why this works so well, we visualize the generated sam￾ples in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Two random examples of four generated samples using our proposed genera￾tion method. The left-most image shows the training image. All other images have a boundary indicating the prediction of P(x, x′ ), with red indicating privacy issues. The scores indicate the agree…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 37 canonical work pages

  1. [1]

    org/news_item/when-are-synthetic-health-data-personal-data/

    GDPR Brief: when are synthetic health data personal data?, https://www.ga4gh. org/news_item/when-are-synthetic-health-data-personal-data/

    work page

  2. [2]

    Opinion 4/2007 on the concept of personal data https://www.pdp.ie/docs/1030. pdf

    work page 2007

  3. [3]

    Recital 26 - Not Applicable to Anonymous Data, https://gdpr-info.eu/recitals/ no-26/

    work page

  4. [4]

    Caron, M., Misra, I., Mairal, J., Goyal, P., Bojanowski, P., Joulin, A.: Unsuper- visedLearningofVisual FeaturesbyContrastingClusterAssignments.In: NeurIPS (2020)

    work page 2020

  5. [5]

    Chebykin, A., Bosman, P.A.N., Alderliesten, T.: Hyperparameter-Free Medical Im- age Synthesis for Sharing Data and Improving Site-Specific Segmentation (2024)

    work page 2024

  6. [6]

    Towards Formalizing the GDPR's Notion of Singling Out

    Cohen, A., Nissim, K.: Towards Formalizing the GDPR’s Notion of Singling Out. Proceedings of the National Academy of Sciences 117(15), 8344–8352 (Apr 2020). https://doi.org/10.1073/pnas.1914598117, http://arxiv.org/abs/1904. 06009, arXiv:1904.06009 [cs]

    work page Pith review arXiv 2020

  7. [7]

    Deshpande, T., Prakash, E., Ross, E.G., Langlotz, C., Ng, A., Valanarasu, J.M.J.: Auto-Generating Weak Labels for Real & Synthetic Data to Improve Label-Scarce Medical Image Segmentation (2024)

    work page 2024

  8. [8]

    In: Medical Imaging with Deep Learning (2025)

    Dombrowski, M., Kainz, B.: Can diffusion models generalize? privacy and fairness trade-offs for medical data sharing. In: Medical Imaging with Deep Learning (2025)

    work page 2025

Show all 41 references
  1. [9]

    In: ICCV 2023

    Dombrowski,M.,Reynaud,H.,Baugh,M.,Kainz,B.:Foreground-BackgroundSep- aration through Concept Distillation from Generative Image Foundation Models. In: ICCV 2023

    2023

  2. [10]

    Dombrowski, M., Reynaud, H., Kainz, B.: Uncovering Hidden Subspaces in Video Diffusion Models Using Re-Identification (Nov 2024), arXiv:2411.04956

    2024 arXiv

  3. [11]

    In: CVPR

    Dombrowski, M., Zhang, W., Cechnicka, S., Reynaud, H., Kainz, B.: Image gener- ation diversity issues and how to tame them. In: CVPR. pp. 3029–3039 (2025)

    2025

  4. [12]

    Elbatel, M., Kamnitsas, K., Li, X.: An Organism Starts with a Single Pix-Cell: A Neural Cellular Diffusion for High-Resolution Image Synthesis (Jul 2024), arXiv:2407.03018 [cs]

    2024 arXiv

  5. [13]

    Frisch, Y.e.a.: Synthesising Rare Cataract Surgery Samples with Guided Diffusion Models (Aug 2023), arXiv:2308.02587 [eess]

    2023 arXiv

  6. [14]

    Han, K., Xiong, Y., You, C., Khosravi, P., Sun, S., Yan, X., Duncan, J., Xie, X.: MedGen3D: A Deep Generative Framework for Paired 3D Image and Mask Generation (Jul 2023), arXiv:2304.04106 [eess]

    2023 arXiv

  7. [15]

    NeurIPS33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. NeurIPS33, 6840–6851 (2020)

    2020

  8. [16]

    Ho, J., Salimans, T.: Classifier-Free Diffusion Guidance (2022), arXiv:2207.12598

    2022 arXiv

  9. [17]

    In: MICCAI 2023, vol

    Hou, Z., Yan, R., Wang, Q., Lang, N., Zhou, X.: Diversity-Preserving Chest Ra- diographs Generation from Reports in One Stage. In: MICCAI 2023, vol. 14224, pp. 482–492. Springer Nature Switzerland, Cham (2023)

    2023

  10. [18]

    Huang, G., Liu, Z., Maaten, L.v.d., Weinberger, K.Q.: Densely Connected Convo- lutional Networks (Jan 2018)

    2018

  11. [19]

    Huang, K., Ma, X., Zhang, Y., Su, N., Yuan, S., Liu, Y., Chen, Q., Fu, H.: Memory- efficient High-resolution OCT Volume Synthesis with Cascaded Amortized Latent Diffusion Models (May 2024), arXiv:2405.16516 [eess]

    2024 arXiv

  12. [20]

    Irvin, J.e.a.: CheXpert: A Large Chest Radiograph Dataset with Uncertainty La- bels and Expert Comparison (2019) 10 Dombrowski and Kainz

    2019

  13. [21]

    Data Mining and Knowledge Discovery 33(4), 917–963 (2019)

    Ismail Fawaz, H., Forestier, G., Weber, J., Idoumghar, L., Muller, P.A.: Deep learn- ing for time series classification: a review. Data Mining and Knowledge Discovery 33(4), 917–963 (2019)

    2019

  14. [22]

    Scientific Data (2019)

    Johnson, A.E.W., Pollard, T.J., Berkowitz, S.J., Greenbaum, N.R., Lungren, M.P., Deng, C.y., Mark, R.G., Horng, S.: MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports. Scientific Data (2019)

    2019

  15. [23]

    Karras, T., Aittala, M., Aila, T., Laine, S.: Elucidating the Design Space of Diffusion-Based Generative Models (Oct 2022), arXiv:2206.00364 [cs]

    2022 arXiv

  16. [24]

    Karras,T.,Aittala,M.,Kynkäänniemi,T.,Lehtinen,J.,Aila,T.,Laine,S.:Guiding a Diffusion Model with a Bad Version of Itself (Jun 2024)

    2024

  17. [25]

    Karras, T., Aittala, M., Lehtinen, J., Hellsten, J., Aila, T., Laine, S.: Analyzing and Improving the Training Dynamics of Diffusion Models (Mar 2024)

    2024

  18. [26]

    Kumar, A., Bhunia, A.k., Narayan, S., Cholakkal, H., Anwer, R.M., Laaksonen, J., Khan, F.S.: Cross-modulated Few-shot Image Generation for Colorectal Tissue Classification (Jul 2023)

    2023

  19. [27]

    In:MICCAI2024,vol.15011.Cham(2024)

    Liu, Z., Zhang, T., He, Y., Zhang, G.: Generating Progressive Images from Patho- logicalTransitions ViaDiffusionModel. In:MICCAI2024,vol.15011.Cham(2024)

    2024

  20. [28]

    Na, I., Kim, J., Ko, E.S., Park, H.: RadiomicsFill-Mammo: Synthetic Mammogram Mass Manipulation with Radiomics Features (2024)

    2024

  21. [29]

    Oh, H.J., Jeong, W.K.: Controllable and Efficient Multi-Class Pathology Nuclei Data Augmentation using Text-Conditioned Diffusion Models (2024)

    2024

  22. [30]

    Scientific Reports12(1) (2022)

    Packhäuser, K., Gündel, S., Münster, N., Syben, C., Christlein, V., Maier, A.: Deep learning-based patient re-identification is able to exploit the biometric nature of medical chest X-ray data. Scientific Reports12(1) (2022)

    2022

  23. [31]

    In: MICCAI 2024, vol

    Peng, Q., Lin, W., Hu, Y., Bao, A., Lian, C., Wei, W., Yue, M., Liu, J., Yu, L., Wang, L.: Advancing H&E-to-IHC Virtual Staining with Task-Specific Domain Knowledge for HER2 Scoring. In: MICCAI 2024, vol. 15004. Cham (2024)

    2024

  24. [32]

    Rajpurkar, P., Irvin, J., Zhu, K., Yang, B., Mehta, H., Duan, T., Ding, D., Bagul, A., Langlotz, C., Shpanskaya, K., Lungren, M.P., Ng, A.Y.: CheXNet: Radiologist- Level Pneumonia Detection on Chest X-Rays with Deep Learning (2017)

    2017

  25. [33]

    Reynaud, H., Meng, Q., Dombrowski, M., Ghosh, A., Day, T., Gomez, A., Leeson, P., Kainz, B.: EchoNet-Synthetic: Privacy-preserving Video Generation for Safe Medical Data Sharing (2024)

    2024

  26. [34]

    Reynaud, H., Qiao, M., Dombrowski, M., Day, T., Razavi, R., Gomez, A., Leeson, P., Kainz, B.: Feature-Conditioned Cascaded Video Diffusion Models for Precise Echocardiogram Synthesis. vol. 14229 (2023)

    2023

  27. [35]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-Resolution Image Synthesis with Latent Diffusion Models (Apr 2022), arXiv:2112.10752 [cs]

    2022 arXiv

  28. [36]

    Shen, Z., Cao, M., Wang, S., Zhang, L., Wang, Q.: CellGAN: Conditional Cervical Cell Synthesis for Augmenting Cytopathological Image Classification (Jul 2023)

    2023

  29. [37]

    In: MICCAI 2024

    Wang, F., Ren, Z., Lian, C., Ma, J.: Controllable Counterfactual Generation for Interpretable Medical Image Classification. In: MICCAI 2024. Cham (2024)

    2024

  30. [38]

    Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., Summers, R.M.: ChestX-ray8: Hospital-Scale Chest X-Ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases (2017)

    2017

  31. [39]

    Ye, J., Ni, H., Jin, P., Huang, S.X., Xue, Y.: Synthetic Augmentation with Large- scale Unconditional Pre-training (Aug 2023), arXiv:2308.04020 [cs]

    2023 arXiv

  32. [40]

    In: MICCAI 2024, vol

    Yuan, Z.e.a.: Adapting Pre-trained Generative Model to Medical Image for Data Augmentation. In: MICCAI 2024, vol. 15005. Cham (2024)

    2024

  33. [41]

    In: MICCAI 2023, vol

    Zhao, Z.e.a.: Label-Preserving Data Augmentation in Latent Space for Diabetic Retinopathy Recognition. In: MICCAI 2023, vol. 14222

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.