List of Papers Browse by Subject Areas Author List
Abstract
Label-free virtual staining offers a compelling, non-destructive alternative to standard histopathology; however, its clinical adoption is hindered by the computational bottlenecks inherent to processing gigapixel Whole Slide Images (WSIs). Current deep learning approaches require patch-based inference to avoid memory constraints, which disrupts global tissue continuity and introduces tiling artifacts—displaying visible seams and color shifts. To address this, we introduce the Consistency Memory Bank (COMB), a novel label-free virtual staining framework that enforces spatial and channel consistency across tiles without memory bottlenecks. COMB decouples context storage from computation, utilizing a dynamic retrieval mechanism to fetch feature representations from adjacent tiles. This enables a retrieval-based context integration strategy that adopts local padding to resolve spatial discontinuities and neighbor-aware channel attention to stabilize statistical drift. Further optimized with a sliding window schedule to ensure minimal memory overhead, our method demonstrates superior performance over state-of-the-art baselines, achieving significant improvements in both perceptual fidelity and tiling consistency, while suggesting its downstream utility in tumor segmentation. Code is available at https://github.com/dou0000/COMB.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4124_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/dou0000/COMB
Link to the Dataset(s)
N/A
BibTex
@InProceedings{KwaDou_Seamless_MICCAI2026,
author = { Kwark, Dou Hoon AND Falahkheirkhah, Kianoush AND Oh, Ji-Hun AND Luo, Shirui AND Kindratenko, Volodymyr AND Bhargava, Rohit},
title = { { Seamless Whole Slide Label-Free Virtual Staining } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16891},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes the Consistent Memory Repository (COMB), which is capable of performing unlabeled virtual staining on billion-pixel Whole Slide images. Use a sliding window buffer to manage adjacent tile features, thereby decoupling context storage from computation to address the challenges of memory limitations and global continuity.
- 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.This work attempts to address a practical bottleneck in pathological image processing, where patching reasoning may disrupt the original structure and confuse pathological examination or downstream automated analysis. 2.The method shows significant improvements in metrics specifically designed for seamlessness, such as Focused Sobel Gradient (FSG) and TexTile, closely approaching Ground Truth levels compared to standard baselines like CC-WSI-Net.
- 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.This framework is largely built on VSGD-Net. “Consistent memory” was proposed as a new framework, but it seems to be an incremental combination of existing technologies. Seamless synthesis of local filling is a known concept. 2.Although the dataset uses two modes (IR and SRS), they both come from individual organs and individual clinics. Due to the lack of multi-site data or different tissue types, it is difficult to evaluate the robustness of the “ neighbor-aware “ statistics when facing the high variability of clinical histopathology. 3.The experimental comparison is limited by the small number of baseline models and the omission of recent literature, providing insufficient proof of the method’s superiority. 4.Ablation studies have mentioned that removing local fillings or neighbor perception CBAM can lead to color inconsistency or artifacts. But acknowledged that “in the future work section in the whole network evenly applied local filling is suboptimal, and is a memory intensive. This indicates that the current formula has not been fully optimized for the “standard consumer hardware” it claims to support.
- Please rate the clarity and organization of this paper
Poor
- 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 provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This work is largely a synthesis of established techniques—specifically local padding and prefetching schedules applied to an existing baseline (VSGD-Net). The contribution of this work feels more like a refined project rather than a novel contribution to deep learning. In addition, clinical feasibility requires multi-center datasets and more robust evaluations to prove.
- Reviewer confidence
Somewhat confident (2)
- [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.
N/A
Review #2
- Please describe the contribution of the paper
This paper presents a novel label-free whole-slide virtual staining framework, COMB, designed to mitigate the color drift and structural discontinuities introduced by independent patch-wise processing, which may impair pathological assessment. To address spatial inconsistency, the method leverages a neighboring-feature memory bank to preserve structural continuity across patch boundaries. To reduce channel-wise inconsistency, it further introduces a neighborhood-aware attention mechanism that stabilizes color variation across adjacent regions. The experimental study is relatively comprehensive, and the paper forms a coherent loop from motivation and problem setting to method development and empirical evaluation.
- 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 paper studies an important and practical problem in whole-slide virtual staining, which is more realistic and clinically relevant than the commonly considered patch-level setting.
2.The proposed COMB framework is intuitive and well motivated, as it incorporates neighboring context in both spatial and channel dimensions to address structural discontinuity and color drift. The additional efficiency optimization for WSI inference further enhances its practical relevance.
3.The experiments are fairly comprehensive. In addition to direct artifact-related metrics such as FSG and TexTile, the paper also includes downstream segmentation results, which provide further evidence for the effectiveness of the proposed method
- 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.The use of CBAM may appear somewhat dated. Since CBAM was introduced in 2018, it is a relatively standard attention module rather than a recent architectural contribution. As a result, the novelty of this component seems limited, and it would be helpful for the authors to better clarify what is fundamentally new beyond adapting an existing attention mechanism.
2.The ablation results in Fig. 4 are somewhat difficult to interpret. From the qualitative ablation study, removing the neighbor-aware CBAM appears to cause a more noticeable degradation, whereas removing the local padding does not seem to lead to an equally significant visual difference. This raises the question of whether the two proposed components contribute equally to the claimed improvements, and whether the role of local padding has been sufficiently demonstrated or explained.
3.The paper lacks sufficiently convincing WSI-level evaluation. Although the paper is framed around whole-slide seamless generation, the experimental section does not provide enough analysis at the true WSI level. More comprehensive whole-slide evaluations or large-scale WSI-level visualizations would strengthen the claim that the proposed method effectively addresses seamless generation in realistic whole-slide settings.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 addresses an important and clinically meaningful problem, namely improving patch-wise spatial consistency in WSI-level virtual staining, where discontinuities across independently processed patches can significantly affect the readability and usability of the generated whole-slide images. The proposed COMB framework is intuitive and practically motivated, as it incorporates neighboring information to improve both spatial continuity and channel consistency across patches.
Another major factor behind my score is that the method demonstrates strong empirical performance. The paper reports state-of-the-art results on two experimental datasets, and the improvements are further supported by downstream segmentation performance, where the proposed method also achieves the best overall results. Overall, the experimental design forms a coherent logical loop with the problem being addressed, and the results reasonably support the authors’ claim that the method improves the readability of WSI-level virtual staining.
- 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.
After reading the rebuttal, I maintain my recommendation for acceptance. The paper addresses a well-motivated and clinically relevant problem in WSI-level virtual staining. The proposed COMB framework is intuitive, and the experiments reasonably validate its effectiveness. Although larger-scale WSI-level validation would further strengthen the work, the current evaluation is fairly comprehensive. Overall, I believe the paper makes a meaningful contribution and should be accepted.
Review #3
- Please describe the contribution of the paper
The paper proposes Consistency Memory Bank (COMB), a framework that enforces spatial and channel consistency for label-free virtual staining.
- 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 proposed method preserves local features well while reducing stitching artifacts and color inconsistencies.
- The validation of the method is thorough, including downstream segmentation and ablation studies.
- Compare to the baseline method, the processing time and memory required are not impractical.
- 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.
- Implementation of COMB is unclear. Figure 2 describes the sptial consistency and channel-wise consistency via COMB but how COMB is actually implemented is not clear.
- Figure 1 mentions context g, p, but there’s no reference on what they are, or any details on how they’re computed other than it’s from the encoder decoder.
- Figure 4 provides no ground truth.
- 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 does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
- Novelty of the paper seems to be an incremental improvement over VSGD-Net
- The description of COMB, even with the figures, are not sufficient to replicate the method
- However the results described are promising and the validation is fairly thorough.
- 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.
Although the description of COMB is still insufficient for me to fully grasp the method, the authors did promise to release the code for reproducibility.
Author Feedback
We thank reviewers for recognizing COMB’s practical clinical contribution and its SOTA seamlessness results. We address shared concerns first, then individual items below.
Novelty. The main novelty of COMB lies in its new formulation of seamless label-free Virtual Staining (VS). Prior work follows two paradigms: (1) IN-based (e.g. KIN, DN), which modify normalization statistics, or (2) matching-based (e.g. CC-WSI-Net), which use histogram matching and pixel-adjacency losses. Both rely on consistency mechanisms that are insufficiently robust for spectral-based label-free VS (see intro)—spectral inputs are often heterogeneous, and lack the direct visual correspondence with H&E that these RGB-to-RGB methods assume. In contrast, COMB introduces dynamic feature-level retrieval through a sliding-window memory bank that caches and serves boundary features and channel attention maps, without reliance on those paradigms, representing a distinct strategy tailored to label-free WSI-level VS. Our novelty lies at the integration level rather than component level—not a routine combination of existing techniques, but a deliberate reformulation tailored to the unique bottlenecks of seamless label-free WSI VS. While some components of COMB build on existing techniques, the integration itself requires domain-specific understanding of WSI-level VS bottlenecks, including discontinuities arising not only from tiling artifacts but also color drift, while simultaneously satisfying strict memory constraints. Accordingly, each intervention was deliberately designed as a complementary mechanism within a unified framework; the strong performance gains acknowledged by all reviewers support this. Finally, although implemented on VSGD-Net to align with current SOTA practices (e.g. CC-WSI-Net uses the same backbone), COMB itself is backbone-agnostic: local padding can be integrated into conv block, while CBAM is broadly compatible with diverse architectures. Thus, COMB is not simply an extension of VSGD-Net, but a standalone strategy potentially applicable to other backbones.
Implementation and Reproducibility. We will revise the manuscript for clarity and release the code upon acceptance to support reproducibility.
Individual concerns: R1 (i) Dataset. We appreciate the concern. Unfortunately, no public label-free VS benchmarks exist, and we do not have in-house multi-site data; broader evaluation is thus left to future work, supported by code release. Still, we believe our evaluation demonstrates COMB’s potential with two label-free modalities, multiple task-driven metrics, and downstream segmentation. (ii) Baselines. Our baselines span directly related label-free VS landscape: Pix2Pix (foundational GAN), VSGD-Net (Pix2Pix-variant; strongest patch-based label-free VS), and CC-WSI-Net (SOTA for seamless WSI VS). Other works (e.g. in H&E->IHC) share only partial relevance to label-free VS. (iii) Hardware. Our implementation meets the consumer-hardware bar (Tab 3: 4.38 GB). The Future Work passage refers to further optimization, not an unmet baseline.
R2 (iv) Ablation. We agree Fig 4 could more clearly demonstrate each component’s role. Local padding addresses spatial seams (boundary-localized) and neighbor-aware CBAM addresses channel drift (region-wide), so their removal produces different visual effects rather than equally prominent ones. We will improve Fig 4’s annotation and clarify the discussion in revision. (v) WSI-level evaluation. We appreciate your insight. Expanded WSI-level analyses are planned as future work to demonstrate seamless VS at scale.
R3 (vi) Notation. We appreciate your careful reading. p—local padding width shown in Figs—needs an explicit textual definition, which we will add in revision. The family of g’s is defined in Eq 1–2.(vii) Ground truth. Fig 4 originally emphasized component-wise differences across ablation conditions rather than direct comparison with GT. We will revise Fig 4 to include GT for clearer interpretation.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper addresses a clinically important WSI virtual-staining issue, stitching artifacts and color drift from patch-wise inference. For rebuttal, the key is to defend novelty beyond VSGD-Net/local padding and explain why COMB is more than a routine combination; strengthen true WSI-level evidence ; broaden or justify baselines; and fix clarity/reproducibility gaps by precisely defining terms and detailing COMB implementation. Please do not feel limited by this meta-review and provide a careful, comprehensive response that addresses all reviewer concerns with clear explanations.
- 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.
The reviewers unanimously provided positive recommendations, and I agree with their decision.
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.
The authors have satisfactorily addressed most of the reviewers’ concerns in the revision. The responses clarify the main issues raised during the initial review, and the revised manuscript has improved accordingly. All reviewers now agree that the paper is acceptable. Based on the revision and the reviewers’ consensus, I recommend acceptance.
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.
The committee recommends acceptance for this clinically relevant work on mitigating artifacts in whole-slide virtual staining. The authors’ rebuttal effectively clarified the method’s technical novelty as a deliberate, memory-efficient integration strategy, resolving initial concerns regarding ablations and baselines. Coupled with strong empirical results and a commitment to releasing the code to ensure reproducibility, this paper offers a valuable and practical contribution to the field.