Pith. sign in

REVIEW 5 major objections 4 minor 67 references

Mediffusion: Joint Diffusion for Self-Explainable Semi-Supervised Classification and Medical Image Generation

T0 review · 5 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Mediffusion claims that a single latent diffusion model with a shared parametrization can classify medical images with very few labels and also explain its own decisions by generating counterfactual images.

desk verdict Useful empirical paper with a real table inconsistency that undercuts its main superiority claim; counterfactual evaluation needs controls before the explainability claims can stand. read the letter →

arxiv 2411.09434 v1 pith:UDEPFB7Q submitted 2024-11-12 cs.CV cs.LG

Joanna Kaleta , Pawe{l} Skierś , Jan Dubiński , Przemys{l}aw Korzeniowski , Kamil Deja This is my paper · ORCID
classification cs.CVcs.LG
keywords semi-supervisedlearninglatentdiffusionmodelsmedicalimageclassificationcounterfactualexplanationsclassifierguidancechestX-raydermoscopyjointgenerative-discriminative
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

Mediffusion claims that a single latent diffusion model, with the classifier built on the same UNet that denoises, can do both semi-supervised classification and self-explanations in medical imaging. The paper argues that adding the diffusion denoising objective to the classifier lets the shared representation learn from labeled and unlabeled data alike, so classification improves when labels are scarce. The same shared parameters then guide the denoiser toward or away from a disease class, producing counterfactual images that show what would need to change to flip the prediction. If correct, one model would cover three clinical needs at once: low-label classification, visual explanations, and synthetic data generation.

What carries the argument

The central object is the joint latent diffusion model, in which a single UNet denoiser also serves as the feature backbone of a classifier: the classifier $g_\omega$ reads the UNet's intermediate features $H_t$ at each diffusion step and outputs class probabilities. The joint objective combines a cross-entropy classification loss on labeled data with the standard diffusion noise-prediction loss on all data. At sampling time, classifier guidance adjusts the noise prediction as $\hat{\epsilon}'(z_t) = \hat{\epsilon}_\theta(z_t) - \sqrt{1-\bar{\alpha}_t}\nabla_{z_t}\log g_\omega(y|e_\nu(z_t))$, which lets the same model generate class-conditioned samples and counterfactuals by guiding toward or against a target class. The counterfactual algorithm adds noise to a latent code up to an intermediate step $t\in(50,300)$ and denoises with guidance toward the opposite prediction, preserving the original anatomy while altering the disease-indicating regions.

What would settle it

Compute, per disease, the mean absolute pixel change inside expert-annotated bounding boxes versus outside them over many counterfactual pairs; if the inside-outside gap vanishes or flips for a disease while the external classifier still reports a large confidence drop, the explanation is driven by non-clinical features. A second check is to apply the same counterfactual manipulation to images from a different scanner or population and see whether the guidance still localizes to the expected anatomy.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a joint latent diffusion model with a shared parametrization can learn representations that simultaneously support discriminative and generative tasks, and that the generative path can explain the discriminative one. The authors train a UNet-based latent diffusion model whose intermediate features feed a classifier, optimize the diffusion objective on all data and the classification objective on labeled data, and then use classifier guidance in the latent space to steer denoising. The resulting model reaches a reported AUC of 71.85 at 2% labels on ChestXRay14 against 64.94 for the same UNet without the diffusion objective, and counterfactuals produced by the model lower an external CheXNet classifier's mean confidence for the target disease from 0.70 to 0.30 for Atelectasis while leaving other classes mostly unchanged. The same procedure can also enforce disease indicators on healthy scans and generate class-conditioned synthetic images.

Load-bearing premise

The load-bearing assumption is that the classifier's gradient in the shared latent space points at clinically meaningful disease features rather than at dataset artifacts, so that guiding the denoiser changes exactly the disease-indicating regions and leaves healthy anatomy alone.

Editorial extensions

If this is right

  • A single model can serve as classifier, counterfactual explainer, and synthetic data generator, so clinical deployment would not need separate explanation or generation systems.
  • The diffusion objective yields classification gains that grow as labels shrink, with the largest relative improvement at 2% labeled data on both ChestXRay14 and ISIC2019.
  • Counterfactual explanations localize changes mostly inside expert-annotated disease bounding boxes, and an independent CheXNet classifier confirms the target disease score drops while other class scores stay stable.
  • Medical doctors in the paper's survey rated Mediffusion counterfactuals as showing more noticeable disease-indicator changes than counterfactuals from an external ACPL-trained classifier.
  • Classifier-guided generation can synthesize disease-present images, and guidance strength controls disease severity, which could support data augmentation and clinician education.

Reading between the lines

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

  • I infer that the same joint parametrization could transfer to other low-label settings where unlabeled images are plentiful, such as pathology slides or retinal scans, provided a good latent autoencoder can be trained on the target distribution.
  • The mechanism suggests a testable diagnostic: if the gradient $\nabla_{z_t}\log g_\omega$ points at clinically meaningful features, then the inside-bounding-box pixel change should correlate with the severity and location of the disease; checking this across many patients would separate true explanation from style artifacts.
  • Counterfactual generation could double as a label-quality audit: images whose counterfactual toward 'healthy' barely changes the external classifier's confidence may point to mislabeled or ambiguous training samples.
  • The authors' requirement to keep the noise step small ($50<t<300$) implicitly assumes a sweet spot where enough semantic content survives for guidance; a systematic sweep of $t$ would clarify how robust the explanations are to this choice.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper proposes Mediffusion, a latent diffusion model whose UNet encoder features are shared with a classifier, trained jointly with a diffusion loss on all data and a classification loss on labeled data. The authors claim that this shared parametrization improves semi-supervised medical image classification and enables self-explainable decisions via classifier-guided counterfactual image edits, in addition to class-conditional synthetic generation. Experiments on ChestXRay14 and ISIC2019 report AUC gains over several SSL baselines, and a counterfactual evaluation uses an externally trained CheXNet and a seven-doctor survey to argue that the edits remove or add disease indicators while preserving other anatomy.

Significance. If the central claims held, the method would be a practically valuable single model for low-label medical imaging: it would combine semi-supervised classification, controllable counterfactual explanations, and guided generation without an external explanation model. The evaluation design has genuine strengths: the counterfactual metrics rely on an independently trained CheXNet classifier and on human expert ratings rather than on the model's own predictions, and the generative assessment uses FID/KID against the target class distribution. However, the current evidence for the explanations' disease specificity is undermined by the absence of control conditions, and the classification comparison contains an internal inconsistency in the main baseline table. The core idea is worth pursuing, but the claims require stronger support.

major comments (5)
  1. [Section 4.1, Table 1 vs Appendix Table 8] The ACPL baseline numbers are internally inconsistent. Table 1 reports ACPL without ImageNet pretraining as 63.22/66.68/71.79/74.93 at 2/5/10/20% labels, but Appendix Table 8 lists ACPL*† (no ImageNet) as 72.51/77.10/79.08/80.47 and ACPL*‡ (ImageNet) as 63.22/66.68/71.79/74.93. The main-table row appears to use the ImageNet-pretrained values while labeling them as non-pretrained, or the two tables use different protocol versions. Because this row is the central evidence for the claim of superiority over ACPL without pretraining, the authors must correct the numbers and re-state the comparison.
  2. [Section 3.3, Eq. (1)] The joint-model formulation is not consistent with the implemented classifier. Eq. (1) factorizes as pν,ψ,ω(z0:T,y)=pν,ω(y|z0)pν,ψ(z0:T), conditioning classification on the clean latent z0, while Section 3.3 states that gω takes UNet features Ht from the noisy latent zt, and Eq. (2) uses the gradient with respect to zt. The factorization should be written for the actual feature distribution, or the mismatch should be justified. In addition, the final loss L(ν,ψ,ω)=Lclass(ν,ω)−Σ_t L_t,diff(ν,ψ) has a minus sign before a positive diffusion loss; if L_t,diff is the usual MSE noise-prediction loss, minimizing this expression would drive the diffusion loss upward. The sign convention must be corrected.
  3. [Section 4.2, Tables 4-5, Fig. 4] The counterfactual evaluation lacks any baseline control, so the central 'self-explainable' claim is not yet supported. The observed drops/increases in CheXNet confidence and the inside-vs-outside bounding-box pixel differences could also be produced by generic perturbations of the latent code, by noise-and-denoise without guidance, or by guidance with a gradient that is unrelated to the disease. The authors should add controls such as (i) unguided noising/diffusion at the same timestep, (ii) guidance toward a random or wrong class, and (iii) guidance from an independent classifier, and report the same metrics (CheXNet confidence change and inside/outside bbox difference) for those controls. Without such controls, Tables 4-5 and Figure 4 do not establish that the edits are disease-specific rather than adversarial-like perturbations.
  4. [Section 4.1, Table 3] The ablation in Table 3 does not isolate the effect of the diffusion loss. Moving from DenseNet to 'UNet w/o diffusion' changes the architecture, the optimizer settings, the training budget (100 epochs vs. 200,000 steps), and the use of unlabeled data, and then moving to Mediffusion adds the diffusion objective and unlabeled data simultaneously. The claim 'thanks to the additional diffusion loss' requires a controlled comparison in which the architecture, labeled-data protocol, and training budget are fixed and only the diffusion/unlabeled objective is toggled.
  5. [Section 4.1 and Appendix Table 8] The paper's positioning as comparable to recent semi-supervised methods is weakened by the omission of PEFAT from the main results. Appendix Table 8 shows PEFAT outperforming Mediffusion on ChestXRay14 at every label percentage (e.g., 75.06 vs. 71.85 at 2%), yet Table 1 and the accompanying text do not mention this baseline. Either PEFAT should be included in the main table and the claims reworded, or the authors should explain why they consider their runs directly comparable.
minor comments (4)
  1. [Section 3.5] The counterfactual generation procedure is described only in prose; provide a pseudocode algorithm or precise step list with the exact t values and guidance scales in the main text, since these are critical to reproducing the reported behavior.
  2. [Section 4.2 and Appendix H] The doctor survey asks about visibility of disease-indicator changes; this measures perceptibility, not correctness, minimality, or clinical validity of the counterfactual. The text should avoid claiming 'practical usability' and 'reliable and precise explanations' solely from this survey.
  3. [Appendix A, Table 11] FID/KID are computed on only 300 generated samples per class with no confidence intervals or seed variance; this is a weak basis for the generative-quality comparison.
  4. [Appendix H] The phrase 'guidance scale 100 larger' is ambiguous; specify whether this means adding 100 to the scale or multiplying it.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation; central claims rest on external benchmarks and independent evaluation, with only a transparent self-citation for the shared-parametrization design.

full rationale

Mediffusion's claimed derivation chain is not circular in any step that can be quoted and reduced to its own inputs. The semi-supervised classification results (Tables 1-3) are compared against external baselines (DenseNet, S2MTS2, ACPL) on held-out test splits, and the diffusion-loss ablation (UNet w/o diffusion) is a same-architecture control; no reported quantity is defined in terms of a value the paper itself fitted. The counterfactual evaluation (Tables 4-7) relies on the externally trained CheXNet classifier and, in Appendix H, a seven-doctor survey, rather than on the model's own confidence, so the observed target-class confidence drop or increase is an independent check, not a self-defined metric; the bounding-box concentration in Figure 4 is measured against physician annotations, not against the model's own gradients. Generative quality (Table 11) is evaluated with FID/KID against real target-class samples. The paper does cite the authors' prior joint-diffusion parametrization [23] when adopting H-features as classifier input (Sections 3.3 and B.1), but this is a transparent methodological lineage rather than a load-bearing proof: the present claims are established by the paper's own experiments, and the self-citation is not used as an external theorem, uniqueness argument, or forbidden alternative. The skeptic's concern that the counterfactual evidence lacks a generic-perturbation control is a correctness or validity risk, not a circularity, because the evidence does not reduce by construction to a fitted input.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim relies on a small set of standard diffusion results and several domain assumptions about the medical latent space. The main free parameters are loss weights, learning rates, guidance scales, and noise timesteps, all tuned per experiment. No invented entities appear.

free parameters (4)
  • classification loss weight w_cls = 0.00005 (CXR 2%), 0.00015 (CXR 5%), 0.001 (CXR 10-100%, ISIC)
    Chosen by the authors per label percentage; controls the balance between the classification objective and the diffusion objective.
  • classifier learning rate = 1e-5 to 5e-5 depending on label percentage
    Adjusted per experiment; affects the classifier's contribution.
  • classifier guidance scale = 500 for disease removal, 200 for enforcement, 300 for generation
    Hand-set to make counterfactual changes visible; larger values produce larger modifications. Not derived from data.
  • counterfactual noise timestep t = 200-300
    Selected to balance semantic preservation and modification strength; a tuning choice.
assumptions (6)
  • standard math DDPM forward/reverse process and the score-based classifier guidance approximation (Eq. 2-6)
    Taken from Ho et al. and Dhariwal and Nichol; used as background.
  • domain assumption The UNet encoder features Ht contain sufficient class-discriminative information for the classifier gω
    Section 3.3: the classifier is trained on Ht; the method's success depends on this property holding across noise levels.
  • domain assumption The autoencoder latent space z = E(x) preserves the disease-relevant semantics needed for classification and counterfactual editing
    Section 3.2 and Appendix C: the KL-autoencoder is trained without labels; the paper assumes no critical disease information is lost in compression.
  • domain assumption ChestXRay14 NLP-extracted labels are reliable enough for evaluation
    Dataset section: labels come from radiological reports via NLP; measurement noise in labels is not modeled.
  • domain assumption The custom ISIC2019 20% test split is representative and matches the training distribution
    Section 4.1 states 'our 20% test set split'; split construction is not described in detail, yet all ISIC results depend on it.
  • domain assumption Guidance of the latent diffusion by the joint classifier produces clinically meaningful counterfactuals rather than artifacts
    Section 3.5: the method assumes the classifier gradient direction in latent space corresponds to disease-indicative features.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mediffusion: Joint Diffusion for Self-Explainable Semi-Supervised Classification and Medical Image Generation." pith.science (2026). https://pith.science/paper/UDEPFB7Q

@misc{pith2026241109434,
  author       = {Pith},
  title        = {Pith review of: Mediffusion: Joint Diffusion for Self-Explainable Semi-Supervised Classification and Medical Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UDEPFB7Q}},
  note         = {Machine review of arXiv:2411.09434}
}
read the original abstract

We introduce Mediffusion -- a new method for semi-supervised learning with explainable classification based on a joint diffusion model. The medical imaging domain faces unique challenges due to scarce data labelling -- insufficient for standard training, and critical nature of the applications that require high performance, confidence, and explainability of the models. In this work, we propose to tackle those challenges with a single model that combines standard classification with a diffusion-based generative task in a single shared parametrisation. By sharing representations, our model effectively learns from both labeled and unlabeled data while at the same time providing accurate explanations through counterfactual examples. In our experiments, we show that our Mediffusion achieves results comparable to recent semi-supervised methods while providing more reliable and precise explanations.

Figures

Figures reproduced from arXiv: 2411.09434 by the authors.

Figure 1
Figure 1. Mediffusion training and capabilities. Our pro￾posed method utilizes both labeled and unlabeled data samples to build joint representations suitable for generative and discrimina￾tive tasks. We evaluate our method in 3 tasks (1) semi-supervised classification (2) inherent visual explainability of classifier deci￾sion (3) synthesising new pseudo-labeled data samples. we can provide accurate explanations for the decis… view at source ↗
Figure 2
Figure 2. Joint training of diffusion model and classifier. Data representation ht extracted in the UNet-based architecture of our joint latent diffusion model is utilized as input to the classifier component. adjust the noise prediction based on the class label y, us￾ing the gradient to guide the generated sample towards this target. Following [11], the noise prediction ϵˆ ′ (zt) at diffusion step t is adjusted as: ϵˆ ′ (zt)… view at source ↗
Figure 4
Figure 4. Mean pixel differences (0-1) inside and outside the ground-truth bounding boxes when generating counterfactual examples. For all diseases, the majority of changes of our counter￾factual examples generation method occur within the ground-truth boxes (green) assigned by the trained physicians. Orig Counterfactual Diff ↑ Other diff↓ Atelectasis 0.48 0.73 0.25 0.07 Cardiomegaly 0.22 0.68 0.46 0.07 Effusion 0.45 0.84 0.3… view at source ↗
Figures from the paper (5 more)
Figure 3
Figure 3. Figure 3: Counterfactual examples. We modify the images to reduce the probability of disease prediction across various classes. From top to bottom: Atelectasis, Cardiomegaly, Mass. The bound￾ing boxes annotated by experts highlight regions that indicate re￾spective diseases. Ori…
Figure 5
Figure 5. Figure 5: Examples of enforcing disease indicators. Visualisa￾tion of counterfactual images generated for healthy data samples that are modified towards cardiomegaly disease characterized by an enlarged heart. dicating the labeled disease. As an evaluation, we check how much the…
Figure 7
Figure 7. Figure 7: Examples of generations with increasing classifier guidance strength. Top: cardiomegaly disease, bottom: effu￾sion. We can observe that higher classifier guidance strength lead to more prominent disease features without significantly altering other parts of the image. …
Figure 8
Figure 8. Figure 8: Mean pixel differences (0-1) inside and outside the ground-truth bounding boxes (20% labeled data). The majority of changes in counterfactual examples occur within the bounding boxes assigned by trained physicians. FID ↓ KID ↓ Unguided 84.81 0.0803 Guided 79.22 0.0610 …
Figure 9
Figure 9. Figure 9: Example question for the expert evaluation survey. score-based conditioning method. This technique is based on a score-conditioning trick adapted from Song et al. [20], which draws on the relation￾ship between diffusion models and score matching [50]. In a general case…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 59 canonical work pages

  1. [1]

    Diffusion visual counterfactual explana- tions

    Maximilian Augustin, Valentyn Boreiko, Francesco Croce, and Matthias Hein. Diffusion visual counterfactual explana- tions. arXiv preprint arXiv:2210.11841, 2022. 1, 13

    arXiv 2022

  2. [2]

    Aviles-Rivero, Nicolas Papadakis, Ruoteng Li, Philip Sellars, Qingnan Fan, Robby T

    Angelica I. Aviles-Rivero, Nicolas Papadakis, Ruoteng Li, Philip Sellars, Qingnan Fan, Robby T. Tan, and Carola- Bibiane Sch¨onlieb. Graphx N ET− chest x-ray classification under extreme minimal supervision, 2020. 11, 12

    work page 2020

  3. [3]

    V oynov, Valentin Khrulkov, and Artem Babenko

    Dmitry Baranchuk, Ivan Rubachev, A. V oynov, Valentin Khrulkov, and Artem Babenko. Label-efficient semantic seg- mentation with diffusion models. International Conference On Learning Representations, 2021. 1, 12

    work page 2021

  4. [4]

    Gori, and Is- abelle Bloch

    Giammarco La Barbera, Haithem Boussaid, Francesco Maso, Sabine Sarnacki, Laurence Rouet, P. Gori, and Is- abelle Bloch. Anatomically constrained ct image translation for heterogeneous blood vessel segmentation. In British Ma- chine Vision Conference, 2022. 12

    work page 2022

  5. [5]

    Demystifying mmd gans

    Mikołaj Bi ´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018. 11

    arXiv 2018

  6. [6]

    Celard, E

    P. Celard, E. L. Iglesias, J. M. Sorribes-Fdez, R. Romero, A. Seara Vieira, and L. Borrajo. A survey on deep learn- ing applied to medical images: from simple artificial neural networks to generative models. Neural Computing and Ap- plications, 35(3):2291–2323, 2023. 1

    work page 2023

  7. [7]

    Towards generalizable tumor synthesis, 2024

    Qi Chen, Xiaoxi Chen, Haorui Song, Zhiwei Xiong, Alan Yuille, Chen Wei, and Zongwei Zhou. Towards generalizable tumor synthesis, 2024. 2, 12

    work page 2024

  8. [8]

    Lightdiff: Surgical endoscopic image low-light enhancement with t-diffusion, 2024

    Tong Chen, Qingcheng Lyu, Long Bai, Erjian Guo, Huxin Gao, Xiaoxiao Yang, Hongliang Ren, and Luping Zhou. Lightdiff: Surgical endoscopic image low-light enhancement with t-diffusion, 2024. 12

    work page 2024

Show all 67 references
  1. [9]

    A manifesto on explainability for artificial in- telligence in medicine

    Carlo Combi, Beatrice Amico, Riccardo Bellazzi, Andreas Holzinger, Jason H Moore, Marinka Zitnik, and John H Holmes. A manifesto on explainability for artificial in- telligence in medicine. Artificial Intelligence in Medicine , 133:102423, 2022. 1

    2022

  2. [10]

    Simon Sherratt

    Tribikram Dhar, Nilanjan Dey, Surekha Borra, and R. Simon Sherratt. Challenges of deep learning in medical image anal- ysis—improving explainability and trust. IEEE Transactions on Technology and Society, 4(1):68–75, 2023. 1

    2023

  3. [11]

    Diffusion models beat GANs on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image synthesis. Advances in Neural Infor- mation Processing Systems, 34, 2021. 1, 3, 15

    2021

  4. [12]

    Conditional diffusion mod- els for semantic 3d brain mri synthesis

    Zolnamar Dorjsembe, Hsing-Kuo Kenneth Pao, Sodtavilan Odonchimed, and Furen Xiao. Conditional diffusion mod- els for semantic 3d brain mri synthesis. IEEE Journal of Biomedical and Health Informatics, 28:4084–4093, 2023. 12

    2023

  5. [13]

    Counterfactual visual explanations

    Yash Goyal, Ziyan Wu, Jan Ernst, Dhruv Batra, Devi Parikh, and Stefan Lee. Counterfactual visual explanations. In In- ternational Conference on Machine Learning , pages 2376–

  6. [14]

    David Gutman, Noel C. F. Codella, Emre Celebi, Brian Helba, Michael Marchetti, Nabin Mishra, and Allan Halpern. Skin lesion analysis toward melanoma detection: A chal- lenge at the international symposium on biomedical imaging (isbi) 2016, hosted by the international skin imag...

    2016

  7. [15]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. Advances in neural information processing systems , 30, 2017. 11

    2017

  8. [16]

    Imagen video: High definition video generation with diffusion mod- els

    Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P Kingma, Ben Poole, Mohammad Norouzi, David J Fleet, et al. Imagen video: High definition video generation with diffusion mod- els. arXiv preprint arXiv:2210.02303, 2022. 1

    arXiv

  9. [17]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. Advances in Neural Information Processing Systems, 33:6840–6851, 2020. 1

    2020

  10. [18]

    Densely connected convolutional net- works

    Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kil- ian Q Weinberger. Densely connected convolutional net- works. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017. 4

    2017

  11. [19]

    Huck Yang, Fangyu Liu, Meng Tian, Yi-Chieh Liu, Ting-Wei Wu, I-Hung Lin, Kang Wang, Hiro- masa Morikawa, Hernghua Chang, Jesper Tegner, and Mar- cel Worring

    Jia-Hong Huang, C.-H. Huck Yang, Fangyu Liu, Meng Tian, Yi-Chieh Liu, Ting-Wei Wu, I-Hung Lin, Kang Wang, Hiro- masa Morikawa, Hernghua Chang, Jesper Tegner, and Mar- cel Worring. Deepopht: Medical report generation for retinal images via deep models and visual explanation. In...

    2021

  12. [20]

    Improving conditional score-based generation with cali- brated classification and joint training

    Paul Kuo-Ming Huang, Si-An Chen, and Hsuan-Tien Lin. Improving conditional score-based generation with cali- brated classification and joint training. In NeurIPS 2022 Workshop on Score-Based Methods, 2022. 15

    2022

  13. [21]

    Robust com- pressed sensing mri with deep generative priors

    Ajil Jalal, Marius Arvinte, Giannis Daras, Eric Price, Alexandros G Dimakis, and Jonathan I Tamir. Robust com- pressed sensing mri with deep generative priors. Advances in Neural Information Processing Systems, 2021. 12

    2021

  14. [22]

    Minimal data requirement for re- alistic endoscopic image generation with stable diffusion

    Joanna Kaleta, Diego Dall’Alba, Szymon Płotka, and Prze- mysław Korzeniowski. Minimal data requirement for re- alistic endoscopic image generation with stable diffusion. International Journal of Computer Assisted Radiology and Surgery, 19(3):531–539, Mar 2024. 1, 12

    2024

  15. [23]

    Tomasz Trzcinski Kamil Deja and Jakub M. Tomczak. Learning data representations with joint diffusion models. arXiv preprint arXiv:2301.13622, 2023. 3, 12

    2023 arXiv

  16. [24]

    Diffusion deformable model for 4d temporal medical image generation

    Boah Kim and Jong Chul Ye. Diffusion deformable model for 4d temporal medical image generation. In International Conference on Medical Image Computing and Computer- Assisted Intervention, pages 539–548. Springer, 2022. 1, 12

    2022

  17. [25]

    Diffusion deformable model for 4d temporal medical image generation

    Boah Kim and Jong Chul Ye. Diffusion deformable model for 4d temporal medical image generation. In Linwei Wang, Qi Dou, P. Thomas Fletcher, Stefanie Speidel, and Shuo Li, editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2022 , pages 539–548, Cham, 2...

    2022

  18. [26]

    Adaptive latent diffusion model for 3d medical image to image translation: Multi- modal magnetic resonance imaging study

    Jonghun Kim and Hyunjin Park. Adaptive latent diffusion model for 3d medical image to image translation: Multi- modal magnetic resonance imaging study. In Proceedings of the IEEE/CVF Winter Conference on Applications of Com- puter Vision (WACV), pages 7604–7613, January 2024. 12 8

    2024

  19. [27]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. Auto-Encoding Vari- ational Bayes. In ICLR, 2014. 13

    2014

  20. [28]

    Diffusion models already have a semantic latent space

    Mingi Kwon, Jaeseok Jeong, and Youngjung Uh. Diffusion models already have a semantic latent space. arXiv preprint arXiv:2210.10960, 2022. 12

    2022 arXiv

  21. [29]

    Acpl: Anti- curriculum pseudo-labelling for semi-supervised medical image classification, 2022

    Fengbei Liu, Yu Tian, Yuanhong Chen, Yuyuan Liu, Vasileios Belagiannis, and Gustavo Carneiro. Acpl: Anti- curriculum pseudo-labelling for semi-supervised medical image classification, 2022. 2, 4, 6, 11, 12, 14, 15

    2022

  22. [30]

    Cordeiro, Vasileios Bela- giannis, Ian Reid, and Gustavo Carneiro

    Fengbei Liu, Yu Tian, Filipe R. Cordeiro, Vasileios Bela- giannis, Ian Reid, and Gustavo Carneiro. Self-supervised mean teacher for semi-supervised chest x-ray classification,

  23. [31]

    Diffsinger: Singing voice synthesis via shallow diffu- sion mechanism

    Jinglin Liu, Chengxi Li, Yi Ren, Feiyang Chen, and Zhou Zhao. Diffsinger: Singing voice synthesis via shallow diffu- sion mechanism. AAAI Conference on Artificial Intelligence,

  24. [32]

    Semi-supervised medical image classi- fication with relation-driven self-ensembling model

    Quande Liu, Lequan Yu, Luyang Luo, Qi Dou, and Pheng Ann Heng. Semi-supervised medical image classi- fication with relation-driven self-ensembling model. IEEE Transactions on Medical Imaging, 2020. 11, 12

    2020

  25. [33]

    Diffusion hyperfeatures: Searching through time and space for semantic correspondence

    Grace Luo, Lisa Dunlap, Dong Huk Park, Aleksander Holyn- ski, and Trevor Darrell. Diffusion hyperfeatures: Searching through time and space for semantic correspondence. In Ad- vances in Neural Information Processing Systems, 2023. 12

    2023

  26. [34]

    Diffusion hyperfeatures: Searching through time and space for semantic correspondence

    Grace Luo, Lisa Dunlap, Dong Huk Park, Aleksander Holyn- ski, and Trevor Darrell. Diffusion hyperfeatures: Searching through time and space for semantic correspondence. Ad- vances in Neural Information Processing Systems, 36, 2024. 1

    2024

  27. [35]

    Ganterfac- tual—counterfactual explanations for medical non-experts using generative adversarial learning

    Silvan Mertes, Tobias Huber, Katharina Weitz, Alexander Heimerl, and Elisabeth Andr ´e. Ganterfac- tual—counterfactual explanations for medical non-experts using generative adversarial learning. Frontiers in Artificial Intelligence, 5, Apr. 2022. 2, 13

    2022

  28. [36]

    Virtual adversarial training: A regularization method for supervised and semi-supervised learning

    Takeru Miyato, Shin-Ichi Maeda, Masanori Koyama, and Shin Ishii. Virtual adversarial training: A regularization method for supervised and semi-supervised learning. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(8):1979–1993, 2019. 12

    1979

  29. [37]

    Codcam: A new ensemble visual explanation for classification of medical thermal im- ages

    Ahmet Ornek and Murat Ceylan. Codcam: A new ensemble visual explanation for classification of medical thermal im- ages. Quantitative InfraRed Thermography Journal , 21:1– 25, 01 2023. 2, 13

    2023

  30. [38]

    Muzaffer Ozbey, Onat Dalmaz, Salman U. H. Dar, Hasan A. Bedel, Saban Ozturk, Alper Gungor, and Tolga Cukur. Un- supervised medical image translation with adversarial dif- fusion models. IEEE Transactions on Medical Imaging , 42(12):3524–3539, 2023. 12

    2023

  31. [39]

    Understanding the latent space of diffusion models through the lens of riemannian geome- try

    Yong-Hyun Park, Mingi Kwon, Jaewoong Choi, Junghyo Jo, and Youngjung Uh. Understanding the latent space of diffusion models through the lens of riemannian geome- try. Advances in Neural Information Processing Systems , 36:24129–24142, 2023. 12

    2023

  32. [40]

    Kevin Zhou, Vishal M

    Cheng Peng, Pengfei Guo, S. Kevin Zhou, Vishal M. Patel, and Rama Chellappa. Towards performant and reliable un- dersampled mr reconstruction via diffusion model sampling. In Medical Image Computing and Computer Assisted In- tervention – MICCAI 2022: 25th International Confere...

    2022

  33. [41]

    Walter H. L. Pinaya, Mark S. Graham, Robert Gray, Pe- dro F. da Costa, Petru-Daniel Tudosiu, Paul Wright, Yee H. Mah, Andrew D. MacKinnon, James T. Teo, Rolf Jager, David Werring, Geraint Rees, Parashkev Nachev, Sebastien Ourselin, and M. Jorge Cardoso. Fast unsupervised brain...

    2022

  34. [42]

    Grad-tts: A diffusion prob- abilistic model for text-to-speech

    Vadim Popov, Ivan V ovk, Vladimir Gogoryan, Tasnima Sadekova, and Mikhail Kudinov. Grad-tts: A diffusion prob- abilistic model for text-to-speech. In International Confer- ence on Machine Learning, pages 8599–8608. PMLR, 2021. 1

    2021

  35. [43]

    Rahman, J

    A. Rahman, J. Valanarasu, I. Hacihaliloglu, and V . M. Pa- tel. Ambiguous medical image segmentation using diffusion models. In 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11536–11546, Los Alamitos, CA, USA, jun 2023. IEEE Computer Society. 12

    2023

  36. [44]

    Lungren, and An- drew Y

    Pranav Rajpurkar, Jeremy Irvin, Kaylie Zhu, Brandon Yang, Hershel Mehta, Tony Duan, Daisy Ding, Aarti Bagul, Curtis Langlotz, Katie Shpanskaya, Matthew P. Lungren, and An- drew Y . Ng. Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning, 2017. 5

    2017

  37. [45]

    High-resolution image syn- thesis with latent diffusion models, 2022

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models, 2022. 3, 13

    2022

  38. [46]

    U-net: Convolutional networks for biomedical image segmentation,

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation,

  39. [47]

    Debiasing cardiac imaging with controlled latent diffusion models, 2024

    Grzegorz Skorupko, Richard Osuala, Zuzanna Szafra- nowska, Kaisar Kushibar, Nay Aung, Steffen E Petersen, Karim Lekadir, and Polyxeni Gkontra. Debiasing cardiac imaging with controlled latent diffusion models, 2024. 12

    2024

  40. [48]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Confer- ence on Machine Learning, pages 2256–2265. PMLR, 2015. 1

    2015

  41. [49]

    Cubuk, Alex Kurakin, Han Zhang, and Colin Raffel

    Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi- supervised learning with consistency and confidence, 2020. 4, 12, 14

    2020

  42. [50]

    Generative modeling by esti- mating gradients of the data distribution.Advances in Neural Information Processing Systems, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution.Advances in Neural Information Processing Systems, 32, 2019. 15

    2019

  43. [51]

    Solv- ing inverse problems in medical imaging with score-based generative models

    Yang Song, Liyue Shen, Lei Xing, and Stefano Ermon. Solv- ing inverse problems in medical imaging with score-based generative models. In International Conference on Learning Representations, 2022. 12 9

    2022

  44. [52]

    Santab´arbara-G´omez, Iv ´an Mac ´ıa Oliver, and Miguel A

    Maialen Stephens Txurio, Karen L ´opez-Linares Rom ´an, Andr´es Marcos-Carri ´on, Pilar Castellote-Huguet, Jos ´e M. Santab´arbara-G´omez, Iv ´an Mac ´ıa Oliver, and Miguel A. Gonz ´alez Ballester. Diffusion models for realistic ct image generation. In Yen-Wei Chen, Satoshi Ta...

    2023

  45. [53]

    Semi-supervised classification of radiology images with noteacher: A teacher that is not mean

    Balagopal Unnikrishnan, Cuong Nguyen, Shafa Balaram, Chao Li, Chuan Sheng Foo, and Pavitra Krishnaswamy. Semi-supervised classification of radiology images with noteacher: A teacher that is not mean. Medical Image Anal- ysis, 73:102148, Oct. 2021. 4, 11

    2021

  46. [54]

    Towards generic semi- supervised framework for volumetric medical image seg- mentation

    Haonan Wang and Xiaomeng Li. Towards generic semi- supervised framework for volumetric medical image seg- mentation. In Thirty-seventh Conference on Neural Infor- mation Processing Systems, 2023. 12

    2023

  47. [55]

    Jorge Cardoso, and Razvan Mari- nescu

    Jueqi Wang, Jacob Levman, Walter Hugo Lopez Pinaya, Petru-Daniel Tudosiu, M. Jorge Cardoso, and Razvan Mari- nescu. Inversesr: 3d brain mri super-resolution using a latent diffusion model, 2023. 2, 12

    2023

  48. [56]

    Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mo- hammadhadi Bagheri, and Ronald M. Summers. Chestx- ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of com- mon thorax diseases. In 2017 IEEE Conference on Computer V...

    2017

  49. [57]

    Implicit image-to-image schrodinger bridge for ct super-resolution and denoising,

    Yuang Wang, Siyeop Yoon, Pengfei Jin, Matthew Tiv- nan, Zhennong Chen, Rui Hu, Li Zhang, Zhiqiang Chen, Quanzheng Li, and Dufan Wu. Implicit image-to-image schrodinger bridge for ct super-resolution and denoising,

  50. [58]

    Diffusion models for medical anomaly detection

    Julia Wolleb, Florentin Bieder, Robin Sandk ¨uhler, and Philippe C Cattin. Diffusion models for medical anomaly detection. In International Conference on Medical image computing and computer-assisted intervention, pages 35–45. Springer, 2022. 2, 12

    2022

  51. [59]

    Ddm2: Self-supervised diffu- sion mri denoising with generative diffusion models

    Tiange Xiang, Mahmut Yurt, Ali B Syed, Kawin Setsom- pop, and Akshay Chaudhari. Ddm2: Self-supervised diffu- sion mri denoising with generative diffusion models. In The Eleventh International Conference on Learning Representa- tions, 2023. 12

    2023

  52. [60]

    Qizhe Xie, Minh-Thang Luong, Eduard Hovy, and Quoc V . Le. Self-training with noisy student improves imagenet clas- sification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , June

  53. [61]

    Measurement-conditioned denoising diffusion probabilistic model for under-sampled medical image reconstruction

    Yutong Xie and Quanzheng Li. Measurement-conditioned denoising diffusion probabilistic model for under-sampled medical image reconstruction. In International Conference on Medical Image Computing and Computer-Assisted Inter- vention, pages 655–664. Springer, 2022. 1

    2022

  54. [62]

    Simultaneous tri-modal medical image fusion and super- resolution using conditional diffusion model, 2024

    Yushen Xu, Xiaosong Li, Yuchan Jie, and Haishu Tan. Simultaneous tri-modal medical image fusion and super- resolution using conditional diffusion model, 2024. 12

    2024

  55. [63]

    Diffmic: Dual-guidance diffusion network for medical image classification

    Yijun Yang, Huazhu Fu, Angelica Aviles-Rivero, Carola- Bibiane Sch ¨onlieb, and Lei Zhu. Diffmic: Dual-guidance diffusion network for medical image classification. arXiv preprint arXiv:2303.10610, 2023. 1, 2, 12

    2023 arXiv

  56. [64]

    Synthetic augmentation with large-scale uncon- ditional pre-training

    Jiarong Ye, Haomiao Ni, Peng Jin, Sharon X Huang, and Yuan Xue. Synthetic augmentation with large-scale uncon- ditional pre-training. In International Conference on Med- ical Image Computing and Computer-Assisted Intervention , pages 754–764. Springer, 2023. 12

    2023

  57. [65]

    Stochastic segmentation with conditional categorical diffusion models

    Lukas Zbinden, Lars Doorenbos, Theodoros Pissas, Adrian Thomas Huber, Raphael Sznitman, and Pablo M´arquez-Neila. Stochastic segmentation with conditional categorical diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 111...

    2023

  58. [66]

    Pefat: Boosting semi-supervised medical image classification via pseudo-loss estimation and feature adversarial training

    Qingjie Zeng, Yutong Xie, Zilin Lu, and Yong Xia. Pefat: Boosting semi-supervised medical image classification via pseudo-loss estimation and feature adversarial training. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 15671–15680...

    2023

  59. [67]

    M3dsynth: A dataset of medical 3d images with ai-generated local manipulations

    Giada Zingarini, Davide Cozzolino, Riccardo Corvi, Gio- vanni Poggi, and Luisa Verdoliva. M3dsynth: A dataset of medical 3d images with ai-generated local manipulations. arXiv preprint arXiv:2309.07973, 2023. 2, 12 10 Supplementary Material This is a supplementary material to ...

    2023 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.