List of Papers Browse by Subject Areas Author List
Abstract
Polarization, as a new optical imaging tool, has been explored to assist in the diagnosis of pathology. Moreover, converting the polarimetric Mueller Matrix (MM) to standardized stained images becomes a promising approach to help pathologists interpret the results. However, existing methods for polarization-based virtual staining are still in the early stage, and the diffusion-based model, which has shown great potential in enhancing the fidelity of the generated images, has not been studied yet. In this paper, a Regulated Bridge Diffusion Model (RBDM) for polarization-based virtual staining is proposed. RBDM utilizes the bidirectional bridge diffusion process to learn the mapping from polarization images to other modalities such as H&E and fluorescence. And to demonstrate the effectiveness of our model, we conduct the experiment on our manually collected dataset, which consists of 18,000 paired polarization, fluorescence and H&E images, due to the unavailability of the public dataset. The experiment results show that our model greatly outperforms other benchmark methods. Our data and code are available at https://github.com/xiaoyu-z/RBDM/.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3549_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/xiaoyu-z/RBDM/
Link to the Dataset(s)
MPPD dataset: https://github.com/xiaoyu-z/RBDM/
BibTex
@InProceedings{ZheXia_Diffusionbased_MICCAI2025,
author = { Zheng, Xiaoyu and Wen, Jing and Zhuang, Jiaxin and Du, Yao and Cong, Jing and Guo, Limei and Luo, Lin and He, Chao and Chen, Hao},
title = { { Diffusion-based Virtual Staining from Polarimetric Mueller Matrix Imaging } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15960},
month = {September},
page = {166 -- 176}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces the Regulated Bridge Diffusion Model (RBDM) for translating polarimetric Mueller Matrix (MM) images into standardized Hematoxylin and Eosin (H&E) and fluorescence-stained images. Key contributions include: (1) the first public multi-modality polarization pathology dataset, comprising 18,000 aligned patches across polarization, H&E, and fluorescence modalities; (2) a diffusion-based framework for stable cross-modal translation, incorporating regulated diffusion processes; and (3) state-of-the-art performance on both polarization-to-H&E and polarization-to-fluorescence tasks.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Authors explored the diffusion model with polarization imaging and proposed RBDM with two regulatory losses to translate 16-channel MM data into RGB-stained images.
- The MPPD dataset addresses a critical gap in the field, as no public polarization-stained paired dataset exists. The rigorous multi-stage registration pipeline (SuperPoint + B-spline) ensures pixel-level alignment, which is essential for supervised training. The author’s open source of this dataset can promote more exploration in this field.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- While the study on translating polarization microscopy to conventional stains for pathological assessment represents a novel contribution, the clinical relevance and utility of the generated stains remain insufficiently validated. The manuscript lacks critical evaluation by pathology experts or performance assessment on downstream diagnostic tasks (e.g., tumor detection or pathological grading) that would substantiate the diagnostic value of the generated images.
- The results section requires substantial expansion and deeper analysis. Notably absent are comprehensive comparisons with pathological image conversion research and diffusion-based conversion methods, such as: [1] Accelerating histopathology workflows with generative AI-based virtually multiplexed tumour profiling [2] Breaking the dilemma of medical image-to-image translation
- Critical examination of the Visualization Results reveals that the proposed RBDM approach does not demonstrate visually appreciable improvements over the PyramidPix2Pix baseline, despite reporting substantial enhancements in quantitative metrics (SSIM and FID). This discrepancy between qualitative and quantitative results warrants further investigation and explanation. Furthermore, the ablation experiments show minimal performance variations relative to the baseline, raising questions about whether the reported improvements genuinely stem from the proposed SSR and RR architectural components or potentially from other experimental factors or statistical variations.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
The detailed feedback can be referenced in the weakness section. I suggest the authors consider merging the Introduction and Related Work sections. This would allow for more rigorous and comprehensive experimental results to be added to the Results section.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This represents the first diffusion-based study for polarization-to-stain translation incorporating regulatory mechanisms. The authors’ commitment to making their data publicly available will likely facilitate further exploration in this emerging field. However, the manuscript would benefit from more comprehensive and in-depth experimental validation and analysis to fully substantiate its claims.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Thank you for the detailed rebuttal and clarifications. Your responses helped clarify several concepts and address some of my initial concerns. Incorporating these additional analyses and results you provided will strengthen the results section and improve the overall quality of the manuscript.
Review #2
- Please describe the contribution of the paper
The paper presents a novel approach to polarization-based virtual staining using a diffusion model framework, representing a significant advancement in the application of polarimetric imaging for computational pathology. The key contributions of the work are threefold:
First Diffusion-Based Virtual Staining Model from Polarization Images: The authors propose the Regulated Bridge Diffusion Model (RBDM), which extends the Brownian Bridge Diffusion Model to handle high-dimensional polarization data (specifically, 16-channel Mueller Matrix images). Unlike prior works that rely on GAN-based frameworks for image-to-image translation, RBDM leverages the denoising properties of diffusion models, incorporating a bidirectional stochastic process to enhance the fidelity of morphological and staining features in generated images. To address the unique dimensionality of polarization inputs, the authors design a dedicated autoencoder-based dimensionality reduction module and introduce two regulation mechanisms—Starting State Regulation (SSR) and Route Regulation (RR)—to ensure structural consistency and perceptual realism in the generated outputs.
Creation of the First Public Multi-modality Polarization Pathology Dataset (MPPD): Recognizing the lack of publicly available datasets for this application domain, the authors construct a large-scale, well-aligned dataset containing 18,000 patches from 7 breast cancer patient samples. Each patch includes perfectly registered images across three modalities: Mueller Matrix polarization, Hematoxylin & Eosin (H&E), and fluorescence staining. The dataset is generated through a rigorous pipeline involving full-vector polarization imaging, advanced registration (rigid and non-rigid), and quality control filtering, providing a valuable resource for the research community.
State-of-the-Art Performance Across Multiple Metrics and Tasks: Through extensive experiments, the proposed RBDM significantly outperforms benchmark models (Pix2Pix and Pyramid Pix2Pix) in translating polarization images to both H&E and fluorescence domains. Performance gains are observed across standard metrics including PSNR, SSIM, FID, and LPIPS, with notable improvements in structural clarity and perceptual quality. An ablation study further confirms the contribution of the proposed SSR and RR modules to the overall performance.
In summary, this work is the first to successfully introduce diffusion models into the virtual staining of Mueller Matrix polarization images, overcoming dimensionality and stability challenges through novel architectural and training innovations. By combining algorithmic novelty with dataset contribution and empirical validation, the paper offers a compelling step forward in bridging the gap between physical imaging modalities and pathologist-interpretable histological representations.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
Novel Application of Diffusion Models to Polarization Imaging: This paper is the first to apply a diffusion-based image-to-image translation model to polarimetric Mueller Matrix (MM) imaging. The proposed Regulated Bridge Diffusion Model (RBDM) introduces a novel architecture that extends the Brownian Bridge Diffusion framework to handle high-dimensional (16-channel) MM data. This is particularly novel and important because previous works have primarily used GAN-based methods, which often suffer from instability when applied to such complex input domains.
Innovative Architectural Components: To address the dimensionality mismatch between polarization inputs and RGB outputs, the authors propose a lightweight autoencoder for polarization feature embedding, along with two new training strategies—Starting State Regulation and Route Regulation. These components help stabilize training and preserve structural consistency, contributing to performance gains in both H&E and fluorescence virtual staining.
Strong Empirical Evaluation: The paper includes comprehensive experiments with multiple metrics (PSNR, SSIM, FID, LPIPS), and consistently outperforms existing baselines. The ablation study clearly shows the contribution of each module, lending credibility to the architectural choices.
High-Quality Dataset Contribution: The introduction of the Multi-modality Polarization Pathology Dataset (MPPD) is another strength, enabling reproducibility and further research in this underexplored domain.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Limited Novelty in Application Domain: While the use of diffusion models is novel for polarization data, the overall task—virtual staining from MM images—has been previously explored. For instance, Si et al. (2022, J. Biophotonics) and Fan et al. (2024, J. Innov. Opt. Health Sci.) have demonstrated MM-to-brightfield/H&E translation using deep learning frameworks such as cGAN and CycleGAN.
- Dataset Scope and Generalizability: The dataset is limited to 7 breast cancer cases. While well-registered, this small cohort limits the generalizability of the model across tissue types, staining protocols, and disease contexts.
- Absence of Comparison with Recent Diffusion-Based Virtual Staining Methods: Although GAN-based baselines are included, comparisons with other diffusion models used in histopathology, such as StainDiff (MICCAI 2023), are missing.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I gave a “Weak Accept” because the paper introduces a novel application of diffusion models to the domain of polarization-based virtual staining, which is technically interesting and relevant to the MICCAI community. The proposed Regulated Bridge Diffusion Model (RBDM) is well-motivated, and the inclusion of training regulations (SSR and RR) tailored to the challenges of high-dimensional Mueller Matrix data is a notable contribution. The empirical results are strong, with comprehensive comparisons showing consistent improvements over GAN-based baselines across multiple metrics.
However, my score is moderated by a few concerns. The application of deep learning to polarization-to-stain translation is not entirely new, and prior works have addressed similar goals using GANs. Additionally, the dataset, while carefully constructed, is limited in scope (only 7 cases), and the paper lacks clinical validation or feedback from pathologists to assess the real-world utility of the virtually stained images. Comparisons with more recent diffusion-based virtual staining models are also missing.
Overall, I believe the technical contribution is solid and novel enough to warrant acceptance, provided the authors can address the above concerns in their rebuttal.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The topic—polarization-based virtual staining with diffusion models—is timely and relevant as virtual histology continues to grow in importance, especially for real-time or label-free imaging workflows. This work could have a strong impact on the future of label-free pathology workflows if the model proves generalizable.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Applying a diffusion-based model (specifically a Regulated Bridge Diffusion Model) to polarization-based virtual staining is a novel contribution. Most existing virtual staining work relies on GANs or traditional regression-based methods.
- The idea of using a bidirectional bridge di]usion process to translate polarimetric MM images into histological stains (e.g., H&E, fluorescence) shows technical innovation.
- The field of polarization imaging in pathology is still relatively new, which increases the novelty value.
- Creating a method that enables accurate virtual staining from Mueller Matrix polarization images could reduce cost, processing time, and inter-observer variability in diagnostics.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Clarify what distinguishes RBDM from other diffusion models—what does “regulated bridge” entail?
- Consider adding a sentence on potential clinical or diagnostic implications to elevate impact.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper aligns well with the conference focused on computational pathology, biomedical imaging, and virtual staining.
- Novel application of diffusion models to polarization-based virtual staining.
- Public release of data/code increases value and transparency.
- The mention of a large (18,000-image) paired dataset is a significant strength
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We sincerely thank the reviewers for their valuable feedback and acknowledgment of our key innovations: releasing the first public MPPD dataset, pioneering a diffusion model for polarization-based virtual staining, and achieving SOTA performance. Below, we address the raised concerns:
Q1. Dataset Scope and Generalizability (R3) We have been scanning current WSI data, targeting at releasing over 30 WSIs paired dataset consisting of three modalities. Our dataset has been expanded to 19 WSIs containing ~52,000 paired patches, together with annotations of breast cancer labels from pathology experts. The dataset enabled large-scale validation and downstream task evaluation, and the corresponding experiment results showed that our model still outperformed other baselines.
Q2. Benchmarking Against Recent Methods (R2, R3) To address this concern, we performed comparative experiments with RegiStain[1], RegGAN[2] and Fast-DDPM3. For the MM-to-H&E task, these methods achieved average PSNR/SSIM scores of 19.9/0.516(RegiStain), 19.80/0.391(RegGAN), and 10.49/0.403(Fast-DDPM), all lower than our model. To further validate generalizability, we also tested three best baselines, PyramidPix2pix, RegGAN, and RegiStain, on a dataset with 20,000 patches. Here, they obtained PSNR/SSIM scores of 13.17/0.485, 15.97/0.361, and 15.17/0.541, respectively, while our model achieved 22.20/0.592, demonstrating its consistent superiority. References: [1] Virtual histological staining of unlabeled autopsy tissue, NC 2024 [2] Breaking the dilemma of medical image-to-image translation, NeurIPS 2021 [3] Fast-DDPM: Fast Denoising Diffusion Probabilistic Models for Medical Image-to-Image Generation, JBHI 2025
Q3. Novelty in Application Domain (R3) While prior polarization-based virtual staining methods exist, our work advances the field via two key innovations ● Streamlined Framework: Existing approaches rely on PCA for dimension reduction (MM→PCA→3-channel data→GAN→H&E), which discards critical MM information. Our method, however, directly maps MM to H&E via a diffusion model, preserving structural fidelity and simplifying the workflow. ● Pioneering Application: We are also the first to validate diffusion models for both H&E and fluorescence staining from MM data.
Q4. Downstream Diagnostic Validation (R2) We tested our model and two best baselines on downstream diagnostic classification task. Our method achieved an AUC of 89.6, significantly outperforming RegGAN (78.8) and RegiStain (55.3). Notably, classification using original H&E images yielded an AUC of 89.7, indicating our model’s ability to preserve diagnostically critical features.
Q5. Impact of RR & SSR Modules and Visual Quality (R2) To validate the efficacy of our RR and SSR modules, we evaluated the performance on the dataset with 20,000 patches. Without these modules, PSNR=21.37, SSIM=0.563, FID=16.27, LPIPS=0.27. Integrating RR and SSR elevated metrics to PSNR=22.20(+3.9%), SSIM=0.592(+5.1%), FID=8.63(−47.0%), LPIPS=0.22(−18.5%), underscoring their critical role in enhancing fidelity. Visually, in the Polar→H&E task, PyramidPix2Pix introduced disruptive white artifacts in nuclear regions, whereas our method eliminated these distortions, generating more consistent and diagnostically reliable images.
Q6. Technical Clarification of “Regulated Bridge” and Clinical Impact (R1) Our Regulated Bridge Diffusion Model (RBDM) learns a mapping from 16-Channel MM to 3-Channel images. During training, intermediate states generated by the bridge diffusion process may lack consistency, leading to suboptimal results. To address this, we introduce two “regulation modules” that enforce stability across diffusion timesteps to enhance performance. By bridging polarization data to diagnostically familiar stains, RBDM unlocks a staining-free, low-cost, high-fidelity paradigm for precision pathology.
We will carefully update our manuscript based on the valuable suggestions! Sincerely, Authors
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The authors have sufficiently addressed the major concerns.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All reviewers tend to accept the paper