arxiv: 2605.14590 · v1 · submitted 2026-05-14 · 💻 cs.CV

Recognition: 1 theorem link

· Lean Theorem

FedStain: Modeling Higher-Order Stain Statistics for Federated Domain Generalization in Computational Pathology

Fengyi Zhang , Junya Zhang , Wenzhuo Sun

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:36 UTC · model grok-4.3

classification 💻 cs.CV

keywords Federated learningDomain generalizationComputational pathologyStain heterogeneityHigher-order statisticsWhole-slide imagesPrivacy preservation

0 comments

The pith

FedStain lets sites share skewness and kurtosis of stain colors to train pathology models that generalize across institutions.

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

Current federated methods in pathology assume stain colors follow simple Gaussian distributions and only share low-order statistics. Real staining creates asymmetric distributions that these methods miss. The paper introduces FedStain to exchange compact higher-order descriptors like skewness and kurtosis during training. This allows the model to capture real stain variability while keeping all patient images private at each site. Tests on standard and new benchmarks show accuracy gains over previous federated and domain generalization approaches.

Core claim

FedStain is the first federated domain generalization method to explicitly model higher-order stain statistics by exchanging skewness and kurtosis as compact descriptors. These descriptors preserve privacy and communication efficiency yet enable the global model to account for non-Gaussian stain heterogeneity that low-order statistics ignore. The framework also uses contrastive cross-site aggregation to learn stain-invariant representations.

What carries the argument

Higher-order stain moments, specifically skewness and kurtosis, used as compact statistical descriptors exchanged in federated optimization, paired with contrastive parameter aggregation.

Load-bearing premise

Skewness and kurtosis of color distributions capture the main non-Gaussian stain differences across sites, and contrastive aggregation builds truly invariant features without leaking private information.

What would settle it

An experiment where models using only these higher-order descriptors fail to improve generalization on a new institution whose stain shifts are driven by factors not reflected in skewness or kurtosis.

Figures

Figures reproduced from arXiv: 2605.14590 by Fengyi Zhang, Junya Zhang, Wenzhuo Sun.

**Figure 2.** Figure 2: Overview of FedStain. Each client calculates local stain style statistics and transmits them, along with updated local [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Statistical characterization of staining intensity distributions across five hospital domains in the Camelyon17 dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 2.** Figure 2: Fig.2. In the Camelyon17 dataset, the samples originating from [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗

**Figure 4.** Figure 4: The t-SNE visualization of histopathology embeddings. Results reveal substantial domain gaps across (a) hospitals in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗

**Figure 5.** Figure 5: Grad-CAM visualizations on Camelyon17 and MvMi [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

read the original abstract

Robust whole-slide image (WSI) analysis under strict data-governance remains challenging due to substantial cross-institutional stain heterogeneity. Domain generalization (DG) mitigates these shifts but typically requires centralized data, conflicting with privacy regulations. Federated learning (FedL) provides a decentralized alternative; however, existing FedL and federated DG (FedDG) approaches rely almost exclusively on low-order statistics, assuming Gaussian-like stain distributions. In contrast, real-world staining processes often produce asymmetric, heavy-tailed color distributions due to biochemical diffusion and scanner nonlinearity. Consequently, current methods fail to model the higher-order, non-Gaussian characteristics dominating real-world stain variability. To address this, we propose FedStain, a stain-aware FedDG framework explicitly incorporating higher-order stain moments--skewness and kurtosis--as compact statistical descriptors exchanged during federated optimization. These descriptors require no pixel-level data transmission, preserving strict privacy and communication efficiency, while enabling the global model to capture stain variability missed by low-order statistics. FedStain also employs a contrastive, cross-site parameter aggregation strategy to promote stain-invariant representations without relaxing data constraints. Extensive experiments on Camelyon17 and our new MvMidog-Fed benchmark show FedStain yields consistent improvements, outperforming state-of-the-art FedL, DG, and FedDG baselines by up to +3.9% absolute accuracy. To our knowledge, FedStain is the first FedDG approach to explicitly model higher-order stain statistics, enabling robust cross-institutional deployment in computational pathology.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

FedStain adds skewness and kurtosis to federated stain modeling and reports modest gains, but the experiments do not isolate whether those moments actually drive the improvement over simpler baselines.

read the letter

FedStain is a reasonable attempt to bring higher-order statistics into federated domain generalization for stain normalization, but the reported gains rest on thin evidence that doesn't separate the contribution of skewness and kurtosis from the rest of the method. The paper proposes exchanging skewness and kurtosis as descriptors during federated rounds, along with a contrastive aggregation to encourage invariant features. This is new in the FedDG literature for pathology, where most work sticks to mean and variance under Gaussian assumptions. It does a clean job of keeping communication low and privacy intact while targeting the asymmetric distributions that come from real staining. The experiments show consistent improvements over baselines on Camelyon17 and the new MvMidog-Fed benchmark, up to 3.9%. That's the positive part. The weak point is that we don't see ablations or controls that would confirm the higher moments are responsible. The abstract doesn't report whether removing the skewness/kurtosis drops performance back to baseline levels, or if the contrastive loss alone would suffice. The stress-test concern about whether two moments capture the dominant variability is fair; if the distributions need quantiles or more, the gains might not generalize. Also, without details on statistical significance or exact baseline re-implementations, it's hard to judge how solid the numbers are. This is for researchers in federated learning applied to computational pathology. Someone looking for ideas on handling non-Gaussian shifts without data sharing would find it useful to read, even if they end up modifying the approach. I would recommend sending it for peer review. The idea has potential and the privacy angle is important, so referees can push for the missing controls.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes FedStain, a stain-aware federated domain generalization (FedDG) framework for computational pathology. It explicitly models higher-order stain statistics by exchanging skewness and kurtosis as compact descriptors during federated optimization rounds, combined with a contrastive cross-site parameter aggregation strategy to promote stain-invariant representations. Experiments on Camelyon17 and the introduced MvMidog-Fed benchmark report consistent accuracy improvements of up to +3.9% over state-of-the-art FedL, DG, and FedDG baselines.

Significance. If the central claims hold, the work is significant as the first FedDG method to target non-Gaussian stain variability through higher-order moments in a privacy-preserving manner. This addresses a key limitation in existing approaches that rely on low-order statistics, potentially enabling more robust cross-institutional deployment of pathology models. The introduction of the MvMidog-Fed benchmark is a positive contribution for future research.

major comments (2)

[Experimental evaluation] The reported gains on Camelyon17 and MvMidog-Fed are not accompanied by ablations that isolate the contribution of the skewness and kurtosis descriptors versus the contrastive aggregation alone. Without such controls, it is difficult to confirm that the +3.9% improvement specifically traces to modeling higher-order statistics rather than other design choices.
[Method description] The assumption that skewness and kurtosis sufficiently capture the dominant non-Gaussian stain variability (as opposed to requiring quantiles, higher moments, or full histogram descriptors) is central to the claim but lacks direct validation, such as comparisons showing that these two scalars dominate cross-site distribution differences in the benchmarks.

minor comments (2)

The abstract claims 'to our knowledge' this is the first such approach; a more thorough literature review section would help substantiate this.
Details on the exact implementation of the contrastive loss and how the descriptors are integrated into the optimization should be clarified for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback and positive assessment of the significance of FedStain. We address each major comment point by point below, agreeing where the manuscript can be strengthened through additional experiments.

read point-by-point responses

Referee: The reported gains on Camelyon17 and MvMidog-Fed are not accompanied by ablations that isolate the contribution of the skewness and kurtosis descriptors versus the contrastive aggregation alone. Without such controls, it is difficult to confirm that the +3.9% improvement specifically traces to modeling higher-order statistics rather than other design choices.

Authors: We agree that isolating the individual contributions is important for validating the central claim. The current experiments report only the full FedStain model. In the revised manuscript we will add ablation tables on both Camelyon17 and MvMidog-Fed that separately evaluate (i) contrastive aggregation without skewness/kurtosis, (ii) higher-order moments with standard FedAvg-style aggregation, and (iii) the complete FedStain pipeline. These controls will quantify the incremental benefit attributable to the higher-order descriptors. revision: yes
Referee: The assumption that skewness and kurtosis sufficiently capture the dominant non-Gaussian stain variability (as opposed to requiring quantiles, higher moments, or full histogram descriptors) is central to the claim but lacks direct validation, such as comparisons showing that these two scalars dominate cross-site distribution differences in the benchmarks.

Authors: We acknowledge that direct empirical validation of sufficiency would strengthen the methodological justification. While the choice of skewness and kurtosis is motivated by their compactness and privacy properties, the manuscript does not include explicit comparisons against richer descriptors. In the revision we will add analyses that measure cross-site stain distribution divergence (e.g., via Wasserstein distance on color histograms) when using only skewness+kurtosis versus additional quantiles or higher moments, thereby demonstrating the coverage of these two scalars on the benchmarks. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper introduces skewness and kurtosis as compact, directly computed statistical descriptors of stain color distributions and exchanges them during federated rounds. These moments are standard third- and fourth-order central moments obtained from the data itself, not fitted parameters or quantities derived from the model's predictions. The contrastive cross-site aggregation is an independent methodological choice applied on top of the descriptors. No equation reduces a claimed prediction to the inputs by construction, no uniqueness theorem is imported via self-citation, and no ansatz is smuggled in. Empirical gains on Camelyon17 and MvMidog-Fed are presented as validation rather than tautological consequences of the method's own definitions. The derivation chain therefore remains independent of its outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only abstract available; core assumptions are that stain color distributions are asymmetric and heavy-tailed (not captured by low-order moments) and that exchanging skewness/kurtosis preserves privacy while providing useful domain information.

axioms (2)

domain assumption Real-world staining processes produce asymmetric, heavy-tailed color distributions due to biochemical diffusion and scanner nonlinearity
Stated directly in abstract as the reason low-order statistics fail.
domain assumption Higher-order moments (skewness, kurtosis) can be exchanged without pixel-level data transmission while preserving strict privacy
Assumed to enable federated optimization without violating data governance.

pith-pipeline@v0.9.0 · 5583 in / 1385 out tokens · 73664 ms · 2026-05-15T01:36:07.728944+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

FedStain explicitly incorporates higher-order stain moments—skewness and kurtosis—as compact statistical descriptors exchanged during federated optimization... Sc = E[(xc − µc)³]/(σ²c)^{3/2}, Kc = E[(xc − µc)⁴]/(σ²c)²

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

76 extracted references · 76 canonical work pages

[1]

A Survey on Deep Learning in Medical Image Analysis,

G. Litjens, T. Kooi, et al., “A Survey on Deep Learning in Medical Image Analysis,”Medical Image Analysis, vol. 42, pp. 60–88, 2017

work page 2017
[2]

Machine Learning Methods for Histopathological Image Analysis,

D. Komura and S. Ishikawa, “Machine Learning Methods for Histopathological Image Analysis,”Computational and Structural Biotechnology Journal, vol. 16, pp. 34–42, 2018

work page 2018
[3]

Summary of the HIPAA Privacy Rule,

Department of Health & Human Services (HHS), “Summary of the HIPAA Privacy Rule,” 1996–2002. [Online]. Available: https://www.hhs.gov. [Accessed: Nov. 8, 2025]

work page 1996
[4]

Regulation (EU) 2016/679: General Data Protection Regulation,

European Union, “Regulation (EU) 2016/679: General Data Protection Regulation,”Official Journal of the European Union, L 119, 2016

work page 2016
[5]

Quantifying the Effects of Data Augmentation and Stain Color Normalization in Convolutional Neural Networks for Computational Pathology,

D. Tellez, M. Balkenhol, et al., “Quantifying the Effects of Data Augmentation and Stain Color Normalization in Convolutional Neural Networks for Computational Pathology,”Medical Image Analysis, vol. 58, Art. no. 101544, 2019

work page 2019
[6]

From Detection of Individual Metastases to Classification of Lymph Node Status: The CAMELYON17 Chal- lenge,

P. Bándi, O. Geessink, et al., “From Detection of Individual Metastases to Classification of Lymph Node Status: The CAMELYON17 Chal- lenge,”IEEE Transactions on Medical Imaging, vol. 38, no. 2, pp. 550– 560, 2019

work page 2019
[7]

Domain-Adversarial Training of Neural Networks,

Y . Ganin, E. Ustinova, et al., “Domain-Adversarial Training of Neural Networks,”Journal of Machine Learning Research, vol. 17, no. 59, pp. 1–35, 2016

work page 2016
[8]

Learning to Generalize: Meta-Learning for Domain Generalization,

D. Li, Y . Yang, Y .-Z. Song, and T. M. Hospedales, “Learning to Generalize: Meta-Learning for Domain Generalization,” inProc. AAAI, 2018, pp. 3490–3497

work page 2018
[9]

In Search of Lost Domain Generaliza- tion,

A. Gulrajani and D. Lopez-Paz, “In Search of Lost Domain Generaliza- tion,” inProc. ICLR, 2021

work page 2021
[10]

Communication-Efficient Learning of Deep Networks from Decentralized Data,

H. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-Efficient Learning of Deep Networks from Decentralized Data,” inProc. AISTATS, 2017, pp. 1273–1282

work page 2017
[11]

Federated Optimization in Heterogeneous Networks (FedProx),

T. Li, A. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated Optimization in Heterogeneous Networks (FedProx),” in Proc. MLSys, 2020

work page 2020
[12]

Federated Cross-Client Contrastive Representation Learning (FedCCRL),

L. Zhao, Q. Yuan, and J. Luo, “Federated Cross-Client Contrastive Representation Learning (FedCCRL),”arXiv preprintarXiv:2405.13235, 2024

work page arXiv 2024
[13]

International Evalu- ation of an AI System for Breast Cancer Screening,

S. M. McKinney, M. Sieniek, V . Godbole, et al., “International Evalu- ation of an AI System for Breast Cancer Screening,”Nature, vol. 577, pp. 89–94, 2020

work page 2020
[14]

Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer,

B. E. Ehteshami Bejnordi, et al., “Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer,”JAMA, vol. 318, no. 22, pp. 2199–2210, 2017

work page 2017
[15]

J. K. Bae, H. J. Roh, J. S. You, et al., “Quantitative Screening of Cervical Cancers for Low-Resource Settings: Pilot Study of Smartphone- Based Endoscopic Visual Inspection After Acetic Acid Using Machine Learning Techniques,”JMIR mHealth uHealth, vol. 8, no. 3, e16467, 2020

work page 2020
[16]

The Artificial Intel- ligence Clinician Learns Optimal Treatment Strategies for Sepsis in Intensive Care,

M. Komorowski, L. A. Celi, O. Badawi, et al., “The Artificial Intel- ligence Clinician Learns Optimal Treatment Strategies for Sepsis in Intensive Care,”Nature Medicine, vol. 24, no. 11, pp. 1716–1720, 2018

work page 2018
[17]

Stain Specific Standard- ization of Whole-Slide Histopathological Images,

B. E. Bejnordi, G. Litjens, N. Timofeeva, et al., “Stain Specific Standard- ization of Whole-Slide Histopathological Images,”IEEE Transactions on Medical Imaging, vol. 35, no. 2, pp. 404–415, 2015

work page 2015
[18]

FedBN: Federated Learning on Non-IID Features via Local Batch Normalization,

X. Li, M. Jiang, X. Zhang, M. Kamp, and Q. Dou, “FedBN: Federated Learning on Non-IID Features via Local Batch Normalization,” inProc. ICLR, 2021

work page 2021
[19]

FedDG: Federated domain generalization on medical image segmentation via episodic learning in continuous frequency space,

Q. Liu, C. Chen, J. Qin, Q. Dou, and P.-A. Heng, “FedDG: Federated domain generalization on medical image segmentation via episodic learning in continuous frequency space,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1013– 1023

work page 2021
[20]

FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment,

S. Gupta, V . Sutar, V . Singh, and A. Sethi, “FedAlign: Federated Domain Generalization with Cross-Client Feature Alignment,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2025, pp. 1801–1810

work page 2025
[21]

Learning Domain-Agnostic Visual Representation for Computational Pathology Using Medically-Irrelevant Style Transfer Augmentation,

R. Yamashita, J. Long, S. Banda, J. Shen, and D. L. Rubin, “Learning Domain-Agnostic Visual Representation for Computational Pathology Using Medically-Irrelevant Style Transfer Augmentation,”IEEE Trans- actions on Medical Imaging, vol. 40, no. 12, pp. 3945–3954, 2021

work page 2021
[22]

StainGAN: Stain Style Transfer for Digital Histological Images,

M. T. Shaban, C. Baur, N. Navab, and S. Albarqouni, “StainGAN: Stain Style Transfer for Digital Histological Images,” inProc. IEEE ISBI, 2019, pp. 953–956

work page 2019
[23]

RandStainNA: Learning Stain-Agnostic Features From Histology Slides by Bridging Stain Augmentation and Normaliza- tion,

Y . Lv, X. Yang, et al., “RandStainNA: Learning Stain-Agnostic Features From Histology Slides by Bridging Stain Augmentation and Normaliza- tion,” inProc. MICCAI, LNCS 13436, 2022, pp. 239–249

work page 2022
[24]

Mitosis Domain Generalization in Histopathology Images — The MIDOG Challenge,

M. Aubreville, N. Stathonikos, et al., “Mitosis Domain Generalization in Histopathology Images — The MIDOG Challenge,”Medical Image Analysis, vol. 84, Art. no. 102684, 2023

work page 2023
[25]

Tackling the Objective Inconsistency in Heterogeneous Federated Optimization (FedNova),

J. Wang, Q. Liu, H. Liang, G. Joshi, and H. V . Poor, “Tackling the Objective Inconsistency in Heterogeneous Federated Optimization (FedNova),” inProc. NeurIPS, 2020

work page 2020
[26]

SCAFFOLD: Stochastic Controlled Averaging for Federated Learning,

S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “SCAFFOLD: Stochastic Controlled Averaging for Federated Learning,” inProc. ICML, 2020, pp. 5132–5143

work page 2020
[27]

Domain Generalization with MixStyle,

K. Zhou, Y . Yang, Y . Qiao, and T. Xiang, “Domain Generalization with MixStyle,” inProc. ICLR, 2021

work page 2021
[28]

Learning Transferable Visual Models From Natural Language Supervision,

A. Radford, J. W. Kim, et al., “Learning Transferable Visual Models From Natural Language Supervision,” inProc. ICML, 2021, pp. 8748– 8763

work page 2021
[29]

Domain Gener- alization via Model-Agnostic Learning of Semantic Features,

Q. Dou, D. C. Castro, K. Kamnitsas, and B. Glocker, “Domain Gener- alization via Model-Agnostic Learning of Semantic Features,” inProc. NeurIPS, 2019

work page 2019
[30]

Episodic Training for Domain Generalization,

D. Li, J. Zhang, Y . Yang, C. Liu, and T. M. Hospedales, “Episodic Training for Domain Generalization,” inProc. ICCV, 2019, pp. 1446– 1455

work page 2019
[31]

Color Transfer Between Images,

E. Reinhard, M. Ashikhmin, B. Gooch, and P. Shirley, “Color Transfer Between Images,”IEEE Computer Graphics and Applications, vol. 21, no. 5, pp. 34–41, 2001

work page 2001
[32]

Quantification of Histochemical Staining by Color Deconvolution,

A. C. Ruifrok and D. A. Johnston, “Quantification of Histochemical Staining by Color Deconvolution,”Analytical and Quantitative Cytology and Histology, vol. 23, no. 4, pp. 291–299, 2001

work page 2001
[33]

RandStainNA++: Foreground-Aware and Self-Distilled Stain Normalization & Augmentation,

Y . Lv, X. Yang, et al., “RandStainNA++: Foreground-Aware and Self-Distilled Stain Normalization & Augmentation,”IEEE Journal of Biomedical and Health Informatics, 2024

work page 2024
[34]

AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty,

D. Hendrycks and N. Dietterich, “AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty,” inProc. ICML, 2020, pp. 4863–4874

work page 2020
[35]

Review of Federated Learning and Machine Learning-Based Methods for Medical Image Analysis,

N. Hernandez-Cruz, P. Saha, M. M. K. Sarker, and J. A. Noble, “Review of Federated Learning and Machine Learning-Based Methods for Medical Image Analysis,”Big Data Cogn. Comput., vol. 8, p. 99, 2024

work page 2024
[36]

Preserving Fairness and Diagnostic Accuracy in Private Large-Scale AI Models for Medical Imaging,

S. Tayebi Arasteh, A. Ziller, C. Kuhl, et al., “Preserving Fairness and Diagnostic Accuracy in Private Large-Scale AI Models for Medical Imaging,”Commun Med, vol. 4, p. 46, 2024

work page 2024
[37]

Supervised Contrastive Learning,

P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised Contrastive Learning,” inProc. NeurIPS, 2020, pp. 18661–18673

work page 2020
[38]

Contrastive Learning With Hard Negative Samples,

T. Wang, J. Zhu, Z. Liu, D. Tao, and G. Hua, “Contrastive Learning With Hard Negative Samples,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2020, pp. 9260–9269

work page 2020
[39]

The Jensen–Shannon Divergence,

M. L. Menéndez, J. A. Pardo, L. Pardo, and M. C. Pardo, “The Jensen–Shannon Divergence,”Journal of the Franklin Institute, vol. 334, no. 2, pp. 307–318, 1997

work page 1997
[40]

Multi-Objective Federated Learning for Medical Image Analysis With Heterogeneous Labels,

X. Chen, Y . Zhang, and Q. Dou, “Multi-Objective Federated Learning for Medical Image Analysis With Heterogeneous Labels,”IEEE Trans- actions on Medical Imaging, vol. 43, no. 5, pp. 1890–1902, 2024

work page 1902
[41]

HistoFS: Non-IID Histopatho- logic Whole Slide Image Classification via Federated Style Transfer With RoI-Preserving,

F. H. Raswa, C.-S. Lu, and J.-C. Wang, “HistoFS: Non-IID Histopatho- logic Whole Slide Image Classification via Federated Style Transfer With RoI-Preserving,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2025, pp. 30251–30260

work page 2025
[42]

PathFL: Multi-Alignment Federated Learning for Pathology Image Segmentation,

Y . Zhang, F. Chen, Y . Qi, G. Yang, and H. Fu, “PathFL: Multi-Alignment Federated Learning for Pathology Image Segmentation,”arXiv preprint arXiv:2505.22522, 2025

work page arXiv 2025
[43]

Generative AI for Synthetic PHI: Privacy-Preserving Training Data for Healthcare LLMs,

K. S. Chadha, “Generative AI for Synthetic PHI: Privacy-Preserving Training Data for Healthcare LLMs,”Frontiers in Emerging Artificial Intelligence and Machine Learning, vol. 1, no. 1, pp. 26–43, 2024

work page 2024
[44]

FedMix: Mixed Supervised Federated Learning for Medical Image Segmentation,

J. Wicaksana, Z. Yan, D. Zhang, X. Huang, H. Wu, X. Yang, and K.- T. Cheng, “FedMix: Mixed Supervised Federated Learning for Medical Image Segmentation,”IEEE Transactions on Medical Imaging, vol. 42, no. 7, pp. 1955–1968, 2023

work page 1955
[45]

Encrypted Federated Learning for Secure Decentralized Collaboration in Cancer Image Analysis,

D. Truhn, S. T. Arasteh, O. L. Saldanha, G. Müller-Franzes, F. Khader, P. Quirke, N. P. West, R. Gray, G. G. A. Hutchins, J. A. James, et al., “Encrypted Federated Learning for Secure Decentralized Collaboration in Cancer Image Analysis,”Medical Image Analysis, vol. 92, p. 103059, 2024

work page 2024
[46]

Truly Privacy-Preserving Federated Analytics for Precision Medicine With Multiparty Homomorphic Encryption,

D. Froelicher, J. R. Troncoso-Pastoriza, J. L. Raisaro, et al., “Truly Privacy-Preserving Federated Analytics for Precision Medicine With Multiparty Homomorphic Encryption,”Nature Communications, vol. 12, p. 5910, 2021

work page 2021
[47]

Federated Learning and Differential Privacy for Medical Image Analysis,

M. Adnan, S. Kalra, J. C. Cresswell, et al., “Federated Learning and Differential Privacy for Medical Image Analysis,”Scientific Reports, vol. 12, p. 1953, 2022

work page 1953
[48]

Multimodal Melanoma Detection With Federated Learning,

B. L. Y . Agbley, J. Li, A. U. Haq, E. K. Bankas, S. Ahmad, I. O. Agyemang, D. Kulevome, W. D. Ndiaye, B. Cobbinah, and S. Latipova, “Multimodal Melanoma Detection With Federated Learning,” inProc. ICCWAMTIP, Chengdu, China, Dec. 17–19, 2021, IEEE, New York, NY , USA, pp. 238–244, 2021

work page 2021
[49]

Federated domain general- ization for image recognition via cross-client style transfer,

J. Chen, M. Jiang, Q. Dou, and Q. Chen, “Federated domain general- ization for image recognition via cross-client style transfer,” inProc. WACV, Waikoloa, HI, USA, Jan. 2–7, 2023, pp. 361–370

work page 2023
[50]

Generalizing to unseen domains: A survey on domain generalization,

J. Wang, C. Lan, C. Liu, Y . Ouyang, T. Qin, W. Lu, Y . Chen, W. Zeng, and P.S. Yu, “Generalizing to unseen domains: A survey on domain generalization,”IEEE Trans. Knowl. Data Eng., vol. 35, no. 8, pp. 8052– 8072, 2022

work page 2022
[51]

Artificial intelligence for medical diagnostics— Existing and future AI technology!

M. A. Al-Antari, “Artificial intelligence for medical diagnostics— Existing and future AI technology!”Diagnostics, vol. 13, no. 4, Art. no. 688, 2023

work page 2023
[52]

Medical diagnostic systems using artificial intelligence (AI) algorithms: Principles and perspectives,

S. Kaur, J. Singla, L. Nkenyereye, S. Jha, D. Prashar, G. P. Joshi, S. El- Sappagh, M. S. Islam, and S. M. R. Islam, “Medical diagnostic systems using artificial intelligence (AI) algorithms: Principles and perspectives,” IEEE Access, vol. 8, pp. 228 049–228 069, 2020

work page 2020
[53]

Artificial intelligence in modern medicine— The evolving necessity of the present and role in transforming the future of medical care,

P. B. Bhattad and V . Jain, “Artificial intelligence in modern medicine— The evolving necessity of the present and role in transforming the future of medical care,”Cureus, vol. 12, no. 5, 2020

work page 2020
[54]

Artificial intelligence in clinical diagnosis: Opportunities, challenges, and hype,

P. A. Kulkarni and H. Singh, “Artificial intelligence in clinical diagnosis: Opportunities, challenges, and hype,”JAMA, vol. 330, no. 4, pp. 317– 318, 2023

work page 2023
[55]

Artificial intelligence in the diagnosis of diseases of various origins,

K. Ryzhova, A. V . Yumashev, M. Klimova, R. Osin, E. Gracheva, and A. Dymchishina, “Artificial intelligence in the diagnosis of diseases of various origins,”J. Complement. Med. Res., vol. 14, no. 2, p. 199, 2023

work page 2023
[56]

Domain generalization via invariant feature representation,

K. Muandet, D. Balduzzi, and B. Schölkopf, “Domain generalization via invariant feature representation,” inProc. ICML, 2013, pp. 10–18

work page 2013
[57]

Domain generalization with adversarial feature learning,

H. Li, S. J. Pan, S. Wang, and A. C. Kot, “Domain generalization with adversarial feature learning,” inProc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2018, pp. 5400–5409

work page 2018
[58]

Deep domain generalization via conditional invariant adversarial networks,

Y . Li, X. Tian, M. Gong, Y . Liu, T. Liu, K. Zhang, and D. Tao, “Deep domain generalization via conditional invariant adversarial networks,” in Proc. Eur. Conf. Computer Vision (ECCV), 2018, pp. 624–639

work page 2018
[59]

Multi-adversarial discriminative deep domain generalization for face presentation attack detection,

R. Shao, X. Lan, J. Li, and P. C. Yuen, “Multi-adversarial discriminative deep domain generalization for face presentation attack detection,” inProc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2019, pp. 10 023–10 031

work page 2019
[60]

Robust domain generalisation by enforcing distribution invariance,

S. Erfaniet al., “Robust domain generalisation by enforcing distribution invariance,” inProc. IJCAI, 2016, pp. 1455–1461

work page 2016
[61]

Unified deep supervised domain adaptation and generalization,

S. Motiian, M. Piccirilli, D. A. Adjeroh, and G. Doretto, “Unified deep supervised domain adaptation and generalization,” inProc. IEEE Int. Conf. Computer Vision (ICCV), 2017, pp. 5715–5725

work page 2017
[62]

Generalizable feature learning in the presence of data bias and domain class imbalance with application to skin lesion classification,

C. Yoon, G. Hamarneh, and R. Garbi, “Generalizable feature learning in the presence of data bias and domain class imbalance with application to skin lesion classification,” inProc. MICCAI, 2019, pp. 365–373

work page 2019
[63]

Collaborative semantic aggregation and calibration for federated domain generaliza- tion,

J. Yuan, X. Ma, D. Chen, F. Wu, L. Lin, and K. Kuang, “Collaborative semantic aggregation and calibration for federated domain generaliza- tion,”IEEE Trans. Knowl. Data Eng., vol. 35, no. 12, pp. 12 528–12 541, 2023

work page 2023
[64]

Federated learning for IoT devices with domain generalization,

L. Zhang, X. Lei, Y . Shi, H. Huang, and C. Chen, “Federated learning for IoT devices with domain generalization,”IEEE Internet Things J., vol. 10, no. 18, pp. 16 219–16 233, 2023

work page 2023
[65]

Collaborative optimization and aggregation for decentralized domain generalization and adaptation,

G. Wu and S. Gong, “Collaborative optimization and aggregation for decentralized domain generalization and adaptation,” inProc. IEEE/CVF Int. Conf. Computer Vision (ICCV), 2021, pp. 6484–6493

work page 2021
[66]

Data-driven color augmentation for H&E stained images in computational pathology,

N. Mariniet al., “Data-driven color augmentation for H&E stained images in computational pathology,”J. Pathol. Informat., vol. 14, Art. no. 100183, 2023

work page 2023
[67]

H and E stain augmentation improves generalization of convolutional networks for histopathological mitosis detection,

D. Tellez, M. Balkenhol, N. Karssemeijer, G. Litjens, J. van der Laak, and F. Ciompi, “H and E stain augmentation improves generalization of convolutional networks for histopathological mitosis detection,” inProc. SPIE Med. Imag., vol. 10581, 2018, pp. 264–270

work page 2018
[68]

Bias reduction using combined stain normalization and augmentation for AI-based classification of histological images,

C. Franchetet al., “Bias reduction using combined stain normalization and augmentation for AI-based classification of histological images,” Comput. Biol. Med., vol. 171, Art. no. 108130, 2024

work page 2024
[69]

A method for normalizing histology slides for quantitative analysis,

M. Macenkoet al., “A method for normalizing histology slides for quantitative analysis,” inProc. IEEE Int. Symp. Biomed. Imag. (ISBI), 2009, pp. 1107–1110

work page 2009
[70]

A nonlinear map- ping approach to stain normalization in digital histopathology images using image-specific color deconvolution,

A. M. Khan, N. Rajpoot, D. Treanor, and D. Magee, “A nonlinear map- ping approach to stain normalization in digital histopathology images using image-specific color deconvolution,”IEEE Trans. Biomed. Eng., vol. 61, no. 6, pp. 1729–1738, 2014

work page 2014
[71]

Colour normalisation in digital histopathology im- ages,

D. Mageeet al., “Colour normalisation in digital histopathology im- ages,” inProc. Opt. Tissue Image Anal. Microsc. Histopathol. Endosc. (MICCAI Workshop), vol. 100, 2009, pp. 100–111

work page 2009
[72]

FedSDA: Federated Stain Distribution Alignment for Non-IID Histopathological Image Classifi- cation,

C.-C. Tsai, K.-W. Cheng, and C.-S. Lu, “FedSDA: Federated Stain Distribution Alignment for Non-IID Histopathological Image Classifi- cation,”arXiv preprintarXiv:2511.12044, 2025

work page arXiv 2025
[73]

StainCUT: Stain Normalization with Contrastive Learning,

J. C. Gutiérrez Pérez, “StainCUT: Stain Normalization with Contrastive Learning,”Vision, vol. 8, no. 7, Art. no. 202, 2022

work page 2022
[74]

Stain Mix-Up: Unsupervised Domain Generalization for Histopathology Images,

J.-R. Chang, M.-S. Wu, W.-H. Yu, C.-C. Chen, C.-K. Yang, Y .-Y . Lin, and C.-Y . Yeh, “Stain Mix-Up: Unsupervised Domain Generalization for Histopathology Images,” inProc. MICCAI, 2021, pp. 117–126

work page 2021
[75]

A Federated Learning System for Histopathology Image Analysis With an Orchestral Stain-Normalization GAN,

Y . Shen, A. Sowmya, Y . Luo, X. Liang, D. Shen, and J. Ke, “A Federated Learning System for Histopathology Image Analysis With an Orchestral Stain-Normalization GAN,”IEEE Trans. Med. Imag., 2022

work page 2022
[76]

Hospital-Agnostic Image Representation Learning in Digital Pathology,

M. Sikaroudi, S. Rahnamayan, and H. R. Tizhoosh, “Hospital-Agnostic Image Representation Learning in Digital Pathology,”arXiv preprint arXiv:2204.02404, 2022

work page arXiv 2022