From Pixels to Pathology: Restoration Diffusion for Diagnostic-Consistent Virtual IHC

Azar Kazemi; Carolin Mogler; Christian Grashei; Gesa Wilkens; Han Li; Jingsong Liu; Nassir Navab; Peter J. Sch\"uffler; Tanja Groll; Xiaofeng Deng

REVIEW 5 major objections 4 minor 1 cited by

From Pixels to Pathology: Restoration Diffusion for Diagnostic-Consistent Virtual IHC

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

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

Pith's one-line read This paper claims that virtual staining from H&E to IHC is better posed as a restoration problem than a translation problem, and that a dual-path diffusion model can preserve tissue structure while modeling biomarker variability.

desk verdict SFS is a genuinely useful evaluation metric; the generation model's core restoration mechanism is built on a pixel-wise residual that its own dataset description says is misaligned, and the training objective is never stated, so the SOTA claim is not yet supported. read the letter →

arxiv 2508.02528 v1 pith:EGLC3GKN submitted 2025-08-04 eess.IV cs.CV

Jingsong Liu , Xiaofeng Deng , Han Li , Azar Kazemi , Christian Grashei , Gesa Wilkens , Xin You , Tanja Groll

show 3 more authors

Nassir Navab Carolin Mogler Peter J. Schüffler

This is my paper · ORCID

classification eess.IVcs.CV

keywords virtualstainingimmunohistochemistrydiffusionmodelimagerestorationHER2breastcancersemanticfidelityscorehistopathology

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

This paper tries to establish that virtual staining from H&E to IHC should be treated as an image restoration task, not a pure translation task, and that a diffusion model with a deterministic restoration path plus a stochastic noise path can balance tissue-structure preservation and biomarker variability. If true, it gives pathologists a fast way to generate HER2 IHC images from routine H&E slides, and it provides a misalignment-robust metric, SFS, for judging whether synthetic images carry the diagnostic signal. On the BCI breast cancer dataset, the authors report top results across both pixel-level quality metrics and diagnostic metrics, ranking first on the challenge leaderboard.

What carries the argument

The load-bearing mechanism is the dual-path reverse diffusion process. The forward process adds both Gaussian noise and a scaled deterministic restoration signal $\bar\beta_t I_{\mathrm{res}}$, where $I_{\mathrm{res}} = I_{\mathrm{ihc}} - I_{\mathrm{he}}$; the reverse process trains two U-Nets, a restoration predictor $r_\theta$ and a noise predictor $\epsilon_\theta$, and samples via $x_{t-1} = x_t - \gamma_t r_\theta - \eta_t \epsilon_\theta$. This decouples structural guidance from stochastic variability, and the model reduces to a standard DDPM when $\gamma_t = 0$. The companion evaluation metric, Semantic Fidelity Score (SFS), calibrates a ResNet classifier's accuracy on generated images against class-wise recall degradation, making it robust to spatial misalignment and classifier bias.

What would settle it

Run Star-Diff on a dataset whose H&E and IHC images come from the same tissue section (pixel-level registration) and compare against the slide-level version; if the diagnostic or quality metrics do not improve, the restoration path contributed little beyond the noise path.

Watch

Extended reading notes

Core claim

The central claim is that a structure-aware diffusion model that adds an explicit deterministic restoration path to the standard denoising path can generate virtual IHC images that preserve tissue architecture while capturing realistic biomarker variation. The paper defines the restoration target as the pixel-wise residual $I_{\mathrm{res}} = I_{\mathrm{ihc}} - I_{\mathrm{he}}$ and injects it through a restoration schedule alongside Gaussian noise, with two networks predicting the residual and the noise during reverse sampling. The forward process writes $x_t = x_0 + \bar\alpha_t \epsilon + \bar\beta_t I_{\mathrm{res}}$, and the reverse step samples via $x_{t-1} = x_t - \gamma_t r_\theta - \eta_t \epsilon_\theta$. This dual-path design is reported to outperform eight baselines on the BCI dataset in both image-quality metrics (PSNR, SSIM) and diagnostic metrics (accuracy and the proposed Semantic Fidelity Score), exceeding the second-best model by over 5% in diagnostic metrics.

Load-bearing premise

The approach depends on the pixel-wise difference between a matched H&E and IHC patch being a meaningful restoration signal, but the slides are only aligned at the level of whole slides, not individual pixels, so that difference can be contaminated by misalignment.

Editorial extensions

If this is right

Virtual HER2 IHC could be produced from H&E slides in seconds, making intraoperative frozen-section assessment feasible where real IHC staining would take too long.
The SFS metric could replace or supplement SSIM and PSNR for evaluating staining translation, since it stays stable under translation, rotation, and deformation that break pixel-level metrics.
The dual-path design—deterministic restoration guidance plus stochastic noise—could be re-used for other stain pairs and other medical image-restoration tasks with paired but misaligned data.
Releasing the pretrained classifier means other labs can score diagnostic relevance of synthetic IHC without retraining a classifier on annotated patches.

Reading between the lines

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

A testable consequence of the paper's logic is that Star-Diff should degrade gracefully as registration error increases, since the residual path only needs coarse anatomical correspondence; this could be checked with synthetic misalignment applied to pixel-aligned data.
The SFS construction generalizes beyond HER2: any clinical classifier with class-wise recall defines a misalignment-robust fidelity metric, so the same evaluation recipe could be reused for grading Ki-67, ER, PR, or even non-pathology imaging tasks.
The paper reports results on a single public dataset; the claim that the residual path carries structure rather than misregistration artifacts would be considerably strengthened by replication on internally collected pixel-registered sections, which the authors state is in progress.

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.

Desk Editor's Note

read the letter

You should know two things about this paper. First, the Semantic Fidelity Score (SFS) is a solid idea: a classifier-based, diagnostic-grounded metric that is robust to the misalignment that plagues pixel-level comparisons in histopathology. The authors validate it against translation, rotation, and deformation, show it is stable even when the classifier is under- or over-fit, and plan to release the classifier weights. That part deserves a close look. Second, the generative model Star-Diff has a serious soft spot that the paper itself hands you. The restoration path is built on a pixel-wise residual I_res = I_ihc - I_he, but Section 4.1 states the H&E and IHC patches are only co-registered at slide level, not pixel level. So the residual is a difference of misaligned images. That makes the 'deterministic restoration signal' ill-defined and likely injects registration artifacts into every denoising step. The authors even acknowledge misalignment as a core challenge and design SFS to be robust to it, yet the method's central mechanism presupposes alignment. That is a real internal inconsistency, not a manufactured one.

What is genuinely new: reformulating virtual staining as restoration with a dual path is a plausible angle, though closely related to prior residual diffusion works like SRDiff and DiffIR, and the authors' own Residual DDPM. The SFS metric is the more original contribution. The experiments do show consistent gains over several baselines on the BCI challenge, and the reduced variance across sampling runs is a nice touch. But the training losses are never specified (what exactly is r_theta optimized against?), and the claim that the framework reduces to DDPM when gamma_t = 0 is false because the alpha-bar schedule remains in the forward process. The baselines also come from different sources—PST-Diff results are lifted from the original paper without a shared protocol, which makes the comparison unfair.

Is the paper hopeless? No. The SFS metric is worth building on, and the virtual staining problem matters. But the generation model's foundational assumption is shaky, and the missing training details prevent reproduction. I would not cite the Star-Diff results in their current form, but I would consider citing the SFS metric once it is peer-reviewed. This deserves a serious referee: the topic is important, the metric is useful, and the flaws are correctable—with a learned alignment or a residual defined on registered features, or by dropping the restoration claim entirely. But your referee should be told to dig into the residual definition and request the full training setup before any acceptance.

Referee Report

5 major / 4 minor

Summary. The manuscript proposes Star-Diff, a dual-path diffusion model for virtual IHC generation from H&E images. The method combines a deterministic restoration path, based on the pixel-wise residual I_res = I_ihc - I_he, with a stochastic noise path, and claims to balance tissue-structure preservation with biomarker variability. The paper also introduces the Semantic Fidelity Score (SFS), a classification-based evaluation metric designed to be robust to spatial misalignment and classifier bias. Experiments on the BCI dataset report state-of-the-art image-quality and diagnostic metrics, along with saliency analyses, perturbation studies, and an ablation of the two sampling paths. The SFS metric is not used in the training objective of Star-Diff, so the evaluation is not circular in that sense.

Significance. If the generation method were valid, the paper would address a clinically important task and the SFS metric would be a useful step toward misalignment-robust evaluation in virtual staining. The perturbation experiments for SFS are well designed, and the decision to release the pretrained classifier is a practical strength. However, the central generation mechanism is not a well-defined diffusion process as written, and the load-bearing residual path is undermined by the paper's own description of the data as only slide-level co-registered. The current manuscript does not provide enough specification to be reproduced, and the mathematical claims connecting the method to DDPM are incorrect. The limitations section acknowledges single-dataset validation and classifier-annotation dependence, but does not address these more fundamental inconsistencies.

major comments (5)

[Section 3.2, Eq. (3)] The residual I_res = I_ihc - I_he is computed as a pixel-wise difference, but Section 4.1 states that H&E and IHC patches from the BCI dataset are co-registered at slide level rather than pixel level and may have spatial misalignments. A pixel-wise subtraction of two misaligned images produces high-magnitude edge artifacts at tissue boundaries, so I_res is not a faithful structural signal. Since the restoration path in Eqs. (4) and (6) uses this residual as deterministic guidance at every diffusion step, the method may inject misregistration artifacts rather than preserve tissue structure. This internal contradiction is load-bearing: the central structure-preservation claim is unsupported unless the authors register the pairs before computing I_res, define a deformation-invariant residual, or provide quantitative evidence that corresponding patches are aligned to sub-pixel accuracy.
[Section 3.2, Eqs. (4)-(6)] The forward and reverse processes are under-specified. No loss function is given for r_theta or epsilon_theta, no transition kernel q(x_t | x_{t-1}) corresponding to Eq. (4) is defined, and no posterior q(x_{t-1} | x_t, x_0) is derived to justify the sampling update in Eqs. (5)-(6). The schedules bar_alpha_t, bar_beta_t, gamma_t, and eta_t are not defined, and at t=T Eq. (4) does not reduce to a Gaussian noise prior because of the bar_beta_T I_res term. As written, the method cannot be implemented, trained, or evaluated by a reader.
[Section 3.2, 'Connection to DDPM'] The statement that the framework reduces to standard DDPM when gamma_t = 0 is false. Even with gamma_t = 0, Eq. (4) still contains the deterministic restoration term bar_beta_t I_res, and the remaining term x_0 + bar_alpha_t epsilon differs from the DDPM forward process sqrt(bar_alpha_t) x_0 + sqrt(1 - bar_alpha_t) epsilon. The reverse update x_{t-1} = x_t - eta_t epsilon_theta is not the DDPM posterior sampling step. The claimed relationship to DDPM is therefore not correct and should either be removed or re-derived within a properly formulated diffusion framework.
[Section 4.1, Table 1] The baseline comparison does not follow a shared protocol. Pix2Pix and Pix2Pix-Pyramid results are taken from [20], and PST-Diff results from [10], while other methods are trained and evaluated locally; PST-Diff's diagnostic metrics are missing. Without common training, hyperparameter, and evaluation settings across methods, the claims of state-of-the-art performance and ranking first on the leaderboard are not established. In addition, Section 4.2 refers to CUT, which is not listed in Section 4.1 or in Table 1.
[Section 3.3, Eq. (10)] The SFS formula averages Accgen with 1 - AvgDeg, where AvgDeg is the average class-wise recall gap between real and generated images. Under a weak or poorly calibrated classifier, both R_real_c and R_gen_c are low, so AvgDeg is small and SFS remains high even when generated images carry little diagnostic signal. The robustness experiments in Figure 4 show stability but do not demonstrate that SFS remains discriminative across translation methods when classifier quality degrades. A calibration study or a minimal classifier-quality threshold is needed before SFS can be claimed as a clinically meaningful evaluation metric.

minor comments (4)

[Figures 1 and Graphical Abstract] The word 'structure' is misspelled as 'structrure' in the figure and graphical abstract; also, 'PNSR' appears in Table 3 and in parts of the text and should be 'PSNR'.
[Section 4.1] The method list omits Vahadane, although it appears in Table 1; the baseline list should be consistent with the reported comparisons.
[Section 5.2] The limitations section acknowledges single-dataset validation and the dependence on a pretrained classifier with patch-level annotations, but it does not address the residual misalignment issue in Section 3.2, which is more fundamental.
[References] Several references appear with incomplete citation formatting, for example reference [9] ends with 'corr abs/1512.03385 (2015)'; the journal's reference style should be applied consistently.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity: the Star-Diff derivation is not forced by its evaluation metric, and the SFS metric is not used to train the model.

full rationale

The paper's central claim is that Star-Diff balances structure preservation and biomarker variability through a dual-path restoration/noise diffusion process. The restoration path is defined by the pixel-wise residual I_res = I_ihc - I_he (Eq. 3) and the reverse update x_{t-1}=x_t - gamma_t r_theta - eta_t eps_theta (Eq. 6). This is an architectural design adopted from residual-denoising diffusion literature (Refs. 15, 17, 31), including the first author's own CVPR 2024 paper (Ref. 17). That self-citation is not load-bearing in a circular sense: the paper does not derive the SOTA result from the citation; it reports held-out BCI test evaluations against eight baselines and provides an ablation (Table 3) showing that the restoration path contributes to performance. The Semantic Fidelity Score (Eq. 10) is a classifier-based evaluation metric computed at test time from a separately trained ResNet classifier; it is not part of the Star-Diff training objective, so the diagnostic 'prediction' is not a fitted input renamed as a result. The main weaknesses are correctness/under-specification concerns rather than circularity: the pixel-wise residual in Eq. 3 may be ill-defined under slide-level co-registration, and the training losses for r_theta and eps_theta are not stated. These do not make the derivation equivalent to its inputs by construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The model relies on several unspecified schedules (alpha_bar, beta_bar, gamma, eta) and on the assumption that a pixel-wise residual is meaningful despite acknowledged misalignment. No new physical entities are introduced.

free parameters (2)

restoration schedule beta_bar_t = not reported
A predefined schedule that controls the deterministic residual guidance in the forward process (Eq. 4). Its values are not given, so the method cannot be reproduced.
sampling weights gamma_t and eta_t = not reported
Balance structure and noise in the reverse update (Eq. 6). Chosen by hand or heuristic, but no values or schedules are provided.

assumptions (4)

standard math Standard DDPM forward and reverse processes are valid for conditional generation.
Used as the base for the proposed noisy path, Section 3.1, Eq. 1-2.
domain assumption The residual I_res = I_ihc - I_he is a meaningful pixel-wise signal despite spatial misalignment.
Required for the restoration path, Eq. 3, and contradicted by the paper's own description of misalignment in Section 4.1.
domain assumption A classifier trained on real IHC images is a valid proxy for pathologist-based HER2 scoring.
Underlies the SFS metric, Section 3.3 and Section 4.1.
ad hoc to paper The reverse sampling update in Eq. 6 correctly approximates the trained forward process.
No derivation is provided; the 'Connection to DDPM' claim is incorrect.

how reviews work

0 comments

Cite this review

Pith. "Pith review of From Pixels to Pathology: Restoration Diffusion for Diagnostic-Consistent Virtual IHC." pith.science (2026). https://pith.science/paper/EGLC3GKN

@misc{pith2026250802528,
  author       = {Pith},
  title        = {Pith review of: From Pixels to Pathology: Restoration Diffusion for Diagnostic-Consistent Virtual IHC},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EGLC3GKN}},
  note         = {Machine review of arXiv:2508.02528}
}

read the original abstract

Hematoxylin and eosin (H&E) staining is the clinical standard for assessing tissue morphology, but it lacks molecular-level diagnostic information. In contrast, immunohistochemistry (IHC) provides crucial insights into biomarker expression, such as HER2 status for breast cancer grading, but remains costly and time-consuming, limiting its use in time-sensitive clinical workflows. To address this gap, virtual staining from H&E to IHC has emerged as a promising alternative, yet faces two core challenges: (1) Lack of fair evaluation of synthetic images against misaligned IHC ground truths, and (2) preserving structural integrity and biological variability during translation. To this end, we present an end-to-end framework encompassing both generation and evaluation in this work. We introduce Star-Diff, a structure-aware staining restoration diffusion model that reformulates virtual staining as an image restoration task. By combining residual and noise-based generation pathways, Star-Diff maintains tissue structure while modeling realistic biomarker variability. To evaluate the diagnostic consistency of the generated IHC patches, we propose the Semantic Fidelity Score (SFS), a clinical-grading-task-driven metric that quantifies class-wise semantic degradation based on biomarker classification accuracy. Unlike pixel-level metrics such as SSIM and PSNR, SFS remains robust under spatial misalignment and classifier uncertainty. Experiments on the BCI dataset demonstrate that Star-Diff achieves state-of-the-art (SOTA) performance in both visual fidelity and diagnostic relevance. With rapid inference and strong clinical alignment,it presents a practical solution for applications such as intraoperative virtual IHC synthesis.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Towards Cellular-Scale Interpretability in Pathology Foundation Models for Biomarker Assessment
cs.CV 2025-11 reject novelty 3.0 of 10

JWTH achieves modest tissue-classification gains by adding attention pooling and stain augmentation to a DINOv3 backbone, but the biomarker claims in the abstract are unsupported by the experiments.

Reference graph

Works this paper leans on

35 extracted references · 32 canonical work pages · cited by 1 Pith paper

[20]

Bci:Breastcan- cerimmunohistochemicalimagegenerationthroughpyramidpix2pix, in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionand pattern recognition, pp

Liu,S.,Zhu,C.,Xu,F.,Jia,X.,Shi,Z.,Jin,M.,2022. Bci:Breastcan- cerimmunohistochemicalimagegenerationthroughpyramidpix2pix, in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionand pattern recognition, pp. 1815–1824

work page 2022
[10]

Pst-diff: achieving high-consistency stain transfer by diffusion models with pathological and structural constraints

He,Y.,Liu,Z.,Qi,M.,Ding,S.,Zhang,P.,Song,F.,Ma,C.,Wu,H., Cai, R., Feng, Y., et al., 2024. Pst-diff: achieving high-consistency stain transfer by diffusion models with pathological and structural constraints. IEEE Transactions on Medical Imaging

work page 2024
[1]

Predicting ki67,er,pr,andher2statusesfromh&e-stainedbreastcancerimages

Akbarnejad, A., Ray, N., Barnes, P.J., Bigras, G., 2023. Predicting ki67,er,pr,andher2statusesfromh&e-stainedbreastcancerimages. arXiv preprint arXiv:2308.01982

arXiv 2023
[2]

Can pathology diagnostic services for cancer be stratified and serve global health? Cancer 126, 2431–2438

Anglade, F., Milner Jr, D.A., Brock, J.E., 2020. Can pathology diagnostic services for cancer be stratified and serve global health? Cancer 126, 2431–2438

work page 2020
[3]

Bouteldja,N.,Klinkhammer,B.M.,Schlaich,T.,Boor,P.,Merhof,D.,

work page
[4]

Region-guided cy- clegans for stain transfer in whole slide images, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer

Boyd, J., Villa, I., Mathieu, M.C., Deutsch, E., Paragios, N., Vakalopoulou, M., Christodoulidis, S., 2022. Region-guided cy- clegans for stain transfer in whole slide images, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 356–365

work page 2022
[5]

The demonstration of pneumococcal antigen in tissues by the use of fluorescent antibody

Coons, A.H., Creech, H.J., Jones, R.N., Berliner, E., 1942. The demonstration of pneumococcal antigen in tissues by the use of fluorescent antibody. The Journal of Immunology 45, 159–170

work page 1942
[6]

Modern Pathology 35, 44–51

Farahmand, S., Fernandez, A.I., Ahmed, F.S., Rimm, D.L., Chuang, J.H.,Reisenbichler,E.,Zarringhalam,K.,2022.Deeplearningtrained onhematoxylinandeosintumorregionofinterestpredictsher2status and trastuzumab treatment response in her2+ breast cancer. Modern Pathology 35, 44–51

work page 2022

Show all 35 references

[7]

Vesseldiffusion:3dvascular structuregenerationbasedondiffusionmodel

Guo,Z.,Tan,Z.,Feng,J.,Zhou,J.,2025. Vesseldiffusion:3dvascular structuregenerationbasedondiffusionmodel. IEEETransactionson Medical Imaging , 1–1doi:10.1109/TMI.2025.3568602

2025
[8]

Deep learning-based transformation of h&e stained tissues into special stains

de Haan, K., Zhang, Y., Zuckerman, J.E., Liu, T., Sisk, A.E., Diaz, M.F., Jen, K.Y., Nobori, A., Liou, S., Zhang, S., et al., 2021. Deep learning-based transformation of h&e stained tissues into special stains. Nature communications 12, 4884

2021
[9]

Deep residual learning for image recognition

He, K., Zhang, X., Ren, S., Sun, J., 2015. Deep residual learning for image recognition. corr abs/1512.03385 (2015)

2015 arXiv
[11]

Denoising diffusion probabilistic models

Ho, J., Jain, A., Abbeel, P., 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33, 6840–6851

2020
[12]

Image quality metrics: Psnr vs

Hore, A., Ziou, D., 2010. Image quality metrics: Psnr vs. ssim, in: 2010 20th international conference on pattern recognition, IEEE. pp. 2366–2369

2010
[13]

Image-to-image translation with conditional adversarial networks, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

Isola, P., Zhu, J.Y., Zhou, T., Efros, A.A., 2017. Image-to-image translation with conditional adversarial networks, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1125–1134

2017
[14]

Keychallengesfordeliveringclinicalimpactwithartificial intelligence

Kelly, C.J., Karthikesalingam, A., Suleyman, M., Corrado, G., King, D.,2019. Keychallengesfordeliveringclinicalimpactwithartificial intelligence. BMC medicine 17, 195

2019
[15]

Srdiff: Single image super-resolution with diffusion probabilistic models

Li, H., Yang, Y., Chang, M., Chen, S., Feng, H., Xu, Z., Li, Q., Chen, Y., 2022. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing 479, 47–59

2022
[16]

Hasd: Hierarchical adaption for pathologyslide-leveldomain-shift

Liu,J.,Li,H.,Yang,C.,Deutges,M.,Sadafi,A.,You,X.,Breininger, K., Navab, N., Schüffler, P.J., 2025a. Hasd: Hierarchical adaption for pathologyslide-leveldomain-shift. arXivpreprintarXiv:2506.23673

arXiv
[17]

Residual denoisingdiffusionmodels,in:ProceedingsoftheIEEE/CVFConfer- ence on Computer Vision and Pattern Recognition, pp

Liu,J.,Wang,Q.,Fan,H.,Wang,Y.,Tang,Y.,Qu,L.,2024. Residual denoisingdiffusionmodels,in:ProceedingsoftheIEEE/CVFConfer- ence on Computer Vision and Pattern Recognition, pp. 2773–2783

2024
[18]

Treatment-aware diffusion probabilistic model for longitudinal mri generation and diffuse glioma growth prediction

Liu, Q., Fuster-Garcia, E., Hovden, I.T., MacIntosh, B.J., Grødem, E.O., Brandal, P., Lopez-Mateu, C., Sederevičius, D., Skogen, K., Schellhorn, T., et al., 2025b. Treatment-aware diffusion probabilistic model for longitudinal mri generation and diffuse glioma growth predictio...
[19]

Unpairedstaintransferusingpathology-consistentconstrainedgener- ativeadversarialnetworks

Liu, S., Zhang, B., Liu, Y., Han, A., Shi, H., Guan, T., He, Y., 2021. Unpairedstaintransferusingpathology-consistentconstrainedgener- ativeadversarialnetworks. IEEEtransactionsonmedicalimaging40, 1977–1989

2021
[21]

Compar- ison of consecutive and restained sections for image registration in histopathology

Lotz, J., Weiss, N., van der Laak, J., Heldmann, S., 2023. Compar- ison of consecutive and restained sections for image registration in histopathology. Journal of Medical Imaging 10, 067501–067501

2023
[22]

A method for normalizing histology slides for quantitative analysis, in: IEEE ISBI, J

Macenko, M., Niethammer, M., Marron, J.S., Borland, D., Woosley, J.T., Guan, X., Schmitt, C., Thomas, N.E., 2009. A method for normalizing histology slides for quantitative analysis, in: IEEE ISBI, J. Liu et al.:Preprint submitted to Elsevier Page 10 of 11 From Pixels to Patho...

2009
[23]

Anintro- duction to the performance of immunohistochemistry

Magaki,S.,Hojat,S.A.,Wei,B.,So,A.,Yong,W.H.,2018. Anintro- duction to the performance of immunohistochemistry. Biobanking: methods and protocols , 289–298

2018
[24]

Rise: Randomized in- put sampling for explanation of black-box models

Petsiuk, V., Das, A., Saenko, K., 2018. Rise: Randomized in- put sampling for explanation of black-box models. arXiv preprint arXiv:1806.07421

2018 arXiv
[25]

Color transfer between images

Reinhard, E., Adhikhmin, M., Gooch, B., Shirley, P., 2001. Color transfer between images. IEEE Computer Graphics and Applications 21, 34–41

2001
[26]

Virtual histological staining of unlabelled tissue-autofluorescence imagesviadeeplearning

Rivenson, Y., Wang, H., Wei, Z., de Haan, K., Zhang, Y., Wu, Y., Günaydın, H., Zuckerman, J.E., Chong, T., Sisk, A.E., et al., 2019. Virtual histological staining of unlabelled tissue-autofluorescence imagesviadeeplearning. Naturebiomedicalengineering3,466–477

2019
[27]

Asurveyontraining challenges in generative adversarial networks for biomedical image analysis

Saad,M.M.,O’Reilly,R.,Rehmani,M.H.,2024. Asurveyontraining challenges in generative adversarial networks for biomedical image analysis. Artificial Intelligence Review 57, 19

2024
[28]

Palette: Image-to-image diffusion models, in: ACM SIGGRAPH 2022 conference proceedings, pp

Saharia,C.,Chan,W.,Chang,H.,Lee,C.,Ho,J.,Salimans,T.,Fleet, D., Norouzi, M., 2022. Palette: Image-to-image diffusion models, in: ACM SIGGRAPH 2022 conference proceedings, pp. 1–10

2022
[29]

Structure- preserving color normalization and sparse stain separation for histo- logical images

Vahadane,A.,Peng,T.,Sethi,A.,Albarqouni,S.,Wang,J.,Baust,M., Steiger, K., Schlitter, A.M., Esposito, I., Navab, N., 2016. Structure- preserving color normalization and sparse stain separation for histo- logical images. IEEE Transactions on Medical Imaging 35, 1962– 1971

2016
[30]

Image qualityassessment:fromerrorvisibilitytostructuralsimilarity

Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P., 2004. Image qualityassessment:fromerrorvisibilitytostructuralsimilarity. IEEE transactions on image processing 13, 600–612

2004
[31]

13095–13105

Xia, B., Zhang, Y., Wang, S., Wang, Y., Wu, X., Tian, Y., Yang, W., VanGool,L.,2023.Diffir:Efficientdiffusionmodelforimagerestora- tion, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 13095–13105

2023
[32]

Utilizing hybrid mask and upsampling attention gate for multipleimmunohistochemistryimagecellrecognition.IEEEJournal of Biomedical and Health Informatics

Zhou, X., Yang, J., Cheng, K., Liu, Q., Sha, H., Wei, R., Jiang, J., 2025. Utilizing hybrid mask and upsampling attention gate for multipleimmunohistochemistryimagecellrecognition.IEEEJournal of Biomedical and Health Informatics

2025
[33]

Breast cancer immunohistochemical image generation: a benchmark dataset and challenge review

Zhu, C., Liu, S., Yu, Z., Xu, F., Aggarwal, A., Corredor, G., Mad- abhushi, A., Qu, Q., Fan, H., Li, F., et al., 2023. Breast cancer immunohistochemical image generation: a benchmark dataset and challenge review. arXiv preprint arXiv:2305.03546

2023 arXiv
[34]

Unpaired image- to-image translation using cycle-consistent adversarial networks, in: ProceedingsoftheIEEEinternationalconferenceoncomputervision, pp

Zhu, J.Y., Park, T., Isola, P., Efros, A.A., 2017. Unpaired image- to-image translation using cycle-consistent adversarial networks, in: ProceedingsoftheIEEEinternationalconferenceoncomputervision, pp. 2223–2232. J. Liu et al.:Preprint submitted to Elsevier Page 11 of 11

2017
[2022]

Journal of pathology informatics 13, 100107

Improving unsupervised stain-to-stain translation using self- supervision and meta-learning. Journal of pathology informatics 13, 100107

Pith tools

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

[1] [20]

Bci:Breastcan- cerimmunohistochemicalimagegenerationthroughpyramidpix2pix, in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionand pattern recognition, pp

Liu,S.,Zhu,C.,Xu,F.,Jia,X.,Shi,Z.,Jin,M.,2022. Bci:Breastcan- cerimmunohistochemicalimagegenerationthroughpyramidpix2pix, in:ProceedingsoftheIEEE/CVFconferenceoncomputervisionand pattern recognition, pp. 1815–1824

work page 2022

[2] [10]

Pst-diff: achieving high-consistency stain transfer by diffusion models with pathological and structural constraints

He,Y.,Liu,Z.,Qi,M.,Ding,S.,Zhang,P.,Song,F.,Ma,C.,Wu,H., Cai, R., Feng, Y., et al., 2024. Pst-diff: achieving high-consistency stain transfer by diffusion models with pathological and structural constraints. IEEE Transactions on Medical Imaging

work page 2024

[3] [1]

Predicting ki67,er,pr,andher2statusesfromh&e-stainedbreastcancerimages

Akbarnejad, A., Ray, N., Barnes, P.J., Bigras, G., 2023. Predicting ki67,er,pr,andher2statusesfromh&e-stainedbreastcancerimages. arXiv preprint arXiv:2308.01982

arXiv 2023

[4] [2]

Can pathology diagnostic services for cancer be stratified and serve global health? Cancer 126, 2431–2438

Anglade, F., Milner Jr, D.A., Brock, J.E., 2020. Can pathology diagnostic services for cancer be stratified and serve global health? Cancer 126, 2431–2438

work page 2020

[5] [3]

Bouteldja,N.,Klinkhammer,B.M.,Schlaich,T.,Boor,P.,Merhof,D.,

work page

[6] [4]

Region-guided cy- clegans for stain transfer in whole slide images, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer

Boyd, J., Villa, I., Mathieu, M.C., Deutsch, E., Paragios, N., Vakalopoulou, M., Christodoulidis, S., 2022. Region-guided cy- clegans for stain transfer in whole slide images, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 356–365

work page 2022

[7] [5]

The demonstration of pneumococcal antigen in tissues by the use of fluorescent antibody

Coons, A.H., Creech, H.J., Jones, R.N., Berliner, E., 1942. The demonstration of pneumococcal antigen in tissues by the use of fluorescent antibody. The Journal of Immunology 45, 159–170

work page 1942

[8] [6]

Modern Pathology 35, 44–51

Farahmand, S., Fernandez, A.I., Ahmed, F.S., Rimm, D.L., Chuang, J.H.,Reisenbichler,E.,Zarringhalam,K.,2022.Deeplearningtrained onhematoxylinandeosintumorregionofinterestpredictsher2status and trastuzumab treatment response in her2+ breast cancer. Modern Pathology 35, 44–51

work page 2022