List of Papers Browse by Subject Areas Author List
Abstract
Many barriers remain before the clinical translation and deployment of prognostic and predictive models utilizing deep learning in digital pathology. In particular, models need to be generalizable to widespread variations in image characteristics resulting from differences in slide preparation protocols and inter-scanner variability. Yet, most existing stain deconvolution methods that correct for the variability in image appearances were developed and validated on specific datasets and perform poorly on unseen data. We developed Physics-Guided Deep Image Prior network for Stain deconvolution (PGDIPS), a self-supervised method guided by a novel optical physics model to perform zero-shot stain deconvolution and normalization. PGDIPS outperformed state-of-the-art approaches for the deconvolution of conventional stain combinations, enabled analysis of previously unsupported special stains, and provided superior interpretability by explicitly encoding representations for stain properties and the light transmittance/absorbance process. PGDIPS is publicly available as an end-to-end off-the-shelf tool at https://github.com/GJiananChen/PGDIPS.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4825_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/4825_supp.zip
Link to the Code Repository
https://github.com/GJiananChen/PGDIPS
Link to the Dataset(s)
MIDOG dataset: https://imi.thi.de/midog/the-dataset/
DeepLIIF dataset: https://deepliif.org
BibTex
@InProceedings{CheJia_Physicsguided_MICCAI2025,
author = { Chen, Jianan and Liu, Lydia Y. and Han, Wenchao and Cheung, Alison and Tsui, Hubert and Martel, Anne L.},
title = { { Physics-guided deep image prior network for general zero-shot stain deconvolution } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15966},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes PGDIPS, a zero-shot stain deconvolution method that combines deep image prior networks with a physics-guided optical model. It addresses generalizability issues by integrating spectral and background correction components without requiring training data. The method outperforms classical approaches in both qualitative and quantitative evaluations across various stains. PGDIPS is designed to be an off-the-shelf tool for robust stain normalization in digital pathology.
- 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.
1） The most innovative aspect is the integration of an adapted Beer-Lambert optical model into a self-supervised deep image prior (DIP) framework. 2） PGDIPS is presented as an end-to-end and readily deployable tool that can operate without expert tuning. 3） The paper provides both qualitative and quantitative evaluations on several types of staining
- 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.
1） It would be interesting to see whether PGDIPS can improve performance on downstream pathology tasks such as classification or segmentation. Since stain deconvolution is often a preprocessing step, evaluating its impact on these tasks could further support the practical value of the method. 2)While the design integrates physically meaningful components, a brief ablation study on elements such as spectral correction or background illumination could strengthen the justification for their inclusion and help isolate their impact on performance.
- 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- 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?
The novelty of proposed Methodology
- Reviewer confidence
Somewhat confident (2)
- [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 #2
- Please describe the contribution of the paper
The paper presents a novel zero-shot stain deconvolution method for histopathology images to address limited generalizability of the existing approaches. The proposed method uses Deep Image Prior modules to learn image priors and introduces spectral correction to beer-lambert law for stains that does not strictly follow that law, e.g. DAB. The method is evaluated on multiple datasets and compared against two existing methods.
- 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.
Good novelty with introducing DIPs for stain-deconvolution. Strong evaluation on multiple datasets and measures.
- 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.
NA
- 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 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
page 4: Please explain: B ∈ R1×3 We used the same set of loss weights α = 1 and β = 0.01 for all experiments. -> the values were chosen based on..? …spectral correction factors and background illumination factor from the reference image -> what is reference image here? Please discuss exclusion loss wrt stain mixing as detailed in Vahadane [27]
- 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?
Good novelty with introducing DIPs for stain-deconvolution. Strong evaluation on multiple datasets and measures.
- 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 authors address key obstacles hindering the clinical deployment of deep learning models in digital pathology, particularly the lack of generalizability across variations in slide preparation and scanner types. To overcome these challenges, they propose PGDIPS (Physics-Guided Deep Image Prior network for Stain deconvolution), a self-supervised method based on an optical physics model. PGDIPS performs zero-shot stain deconvolution and normalization, surpasses existing methods on standard stain combinations, supports analysis of previously incompatible special stains, and offers enhanced interpretability by modeling stain properties and light transmittance. The tool is publicly available as an end-to-end, ready-to-use solution.
- 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.
The authors identify a persistent need for a stain deconvolution (SD) method that merges the advantages of conventional and deep learning-based approaches—specifically, zero-shot capability, accuracy, and generalizability. To meet this need, they introduce PGDIPS (Physics-Guided Deep Image Prior network for Stain Deconvolution), an end-to-end, zero-shot algorithm designed to robustly and accurately deconvolute a wide range of stain types and combinations.
- 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.
Checkerboard artifacts resulting from intra-slide stain variability remain a limitation. Efficient post-processing using overlapping patches could have mitigated this issue. Additionally, benchmarking PGDIPS against non-zero-shot generative approaches would be valuable to enhance its usability and foster its adoption in clinical practice.
- 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
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 authors introduce PGDIPS, a zero-shot algorithm for general stain deconvolution that demonstrates superior performance across diverse stain types compared to existing methods. Designed for broad applicability, PGDIPS emphasizes generalizability and ease of use, requiring no expert input, data collection, or hyperparameter tuning. As an off-the-shelf tool, it can be readily integrated into standard digital pathology workflows. Overall, PGDIPS marks a shift toward physics-guided, general-purpose stain deconvolution and normalization, opening new avenues for research and clinical applications in digital pathology.
- 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 and meta-reviewer for their thoughtful feedback and for recognizing the strengths and relevance of our work. We appreciate the constructive comments that have improved the clarity of our paper. Specifically, we have included an of the exclusion loss in greater detail in the context of stain mixing, and clarified our choice of loss weights.
We have released the source code and are preparing to release the dataset to support further research. In the extended version of this work, we plan to broaden our evaluations to include downstream pathology tasks such as classification and segmentation. We also aim to address the checkerboard effect for whole slide image processing, which we expect to resolve efficiently using the physical parameters deconvoluted by PGDIPS.
We are really excited for PGDIPS’ potential to define a new paradiam of physics-guided methods for stain deconvolution and normalization in computational pathology.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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”.
The paper presents a DIP-based (deep image prior) stain deconvolution method for histopathology images (for both HE and DAB stains). It uses the Beer-Lambert Law to model the relationship between light absorption and stain concentration and includes a spectral correction term for DAB. It also models the effect of background illumination. The model is then evaluated against two other methods on several datasets. All reviewers appreciate the novelty of the method and the solid experimental evaluation, although several reviewers also suggest potential further evaluation on downstream tasks. The paper is suggested to be accepted.