List of Papers Browse by Subject Areas Author List
Abstract
Three-dimensional (3D) histopathology of unprocessed tissues has the potential to transform disease management by enabling volumetric characterization of tissue microarchitecture and in-vivo assessment. Back-illumination Interference Tomography (BIT) is a new phase microscopy technology that provides rapid, non-destructive volumetric imaging of unprocessed tissues. However, translating BIT volumes into clinically interpretable H&E images remains challenging, particularly due to shift-variant contrast and the absence of quantitative validation benchmarks. We introduce HistoBIT3D, the first voxel-wise paired BIT and fluorescence-labeled nuclei dataset, enabling quantitative evaluation of structural preservation in unsupervised virtual staining against ground-truth nuclear distributions. Using this dataset, we present a novel virtual staining framework that translates BIT volumes with shift-variant contrast into realistic H&E volumes by leveraging bidirectional multiscale content consistency and cross-domain style reuse to enhance structural fidelity and perceptual realism. Our method achieves state-of-the-art realism metrics while significantly improving 3D nuclei segmentation accuracy and boundary preservation under zero-shot Cellpose evaluation. Together, these contributions establish a quantitatively validated, structurally faithful, and scalable pipeline for 3D virtual H&E staining, advancing the paradigm of slide-free, volumetric computational histopathology. Our data and code are available at:
https://github.com/aasong113/HistoBIT3D_VirtualStaining.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4026_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/aasong113/HistoBIT3D_VirtualStaining
Link to the Dataset(s)
https://github.com/aasong113/HistoBIT3D_VirtualStaining
Working out data repo with JHU IRB for public release right now.
BibTex
@InProceedings{SonAnt_Virtual_MICCAI2026,
author = { Song, Anthony A. AND Zhou, Boyan AND Golhar, Mayank AND Morakis, Marisa AND Baras, Alexander AND Durr, Nicholas J.},
title = { { Virtual 3D H&E Staining from Phase-Contrast Back-Illumination Interference Tomography } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16890},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper studies virtual 3D H&E staining from Back-illumination Interference Tomography (BIT), a label-free volumetric imaging modality with shift-variant contrast. The contributions: first, it introduces HistoBIT3D, described as the first voxel-wise paired 3D BIT-fluorescence nuclei dataset for quantitative evaluation of structural preservation in unsupervised virtual staining; second, it proposes a GAN-based BIT-to-H&E framework built on a ViT-CycleGAN backbone with two key additions: bidirectional multiscale content consistency and cross-domain style fusion using AdaIN-based reuse of H&E style tokens. The evaluation combines FID/KID with downstream 3D structural assessment via zero-shot Cellpose segmentation against fluorescence ground truth, using 3D Dice, HD95, and nuclei volume. The proposed method outperforms some baselines, and the paper claims that code and data will be released after peer review.
- 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 HistoBIT3D benchmark is the most compelling aspect of the paper. A voxel-wise paired BIT-fluorescence dataset is genuinely useful because it provides a way to evaluate structural fidelity quantitatively rather than relying only on FID/KID. This is a meaningful contribution for the virtual staining community. 2, the paper evaluates structural preservation through zero-shot Cellpose segmentation on registered fluorescence volumes. This is a thoughtful and relevant validation strategy. 3, BIT has shift-variant contrast, making the task harder than more standard virtual staining scenarios. The paper motivates why existing unpaired image translation methods may be inadequate.
- 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 paper’s strongest novelty is the dataset and evaluation protocol. The modeling contribution is more incremental from existing methods. 2, Although the paper shows examples from duodenum and kidney tissues, it would help to better quantify generalization across tissues or acquisition conditions. 3, The ablation is still fairly narrow. It would be stronger to separately analyze the contribution of each scale in the consistency loss, the benefit of the ViT bottleneck, the effect of the style prototype update. 4, The work is exciting as a step toward slide-free volumetric pathology, but the current validation remains technical rather than clinical. There is no pathologist study or task-level diagnostic endpoint.
- 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.
- 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?
The paper addresses an important MIC problem and is strengthened by a genuinely valuable dataset/benchmark contribution. The voxel-wise paired 3D BIT-fluorescence dataset and the use of downstream 3D nuclei segmentation against fluorescence ground truth make the evaluation substantially more convincing than standard virtual staining papers. However, the methodological novelty is moderate, since the model is largely an extension of an existing unpaired translation framework with added multiscale consistency and style fusion. The empirical results are solid, but the ablation and robustness analysis could be deeper.
- 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 #2
- Please describe the contribution of the paper
This paper introduces the first voxel-wise paired dataset of BIT and fluorescence-labeled nuclei, enabling quantitative evaluation of structural preservation in unsupervised virtual staining against ground-truth nuclear distributions. The authors also present a novel virtual staining framework that translates BIT volumes with shift-variant contrast into realistic H&E volumes by leveraging bidirectional multi-scale content consistency and cross-domain style reuse, thereby enhancing structural fidelity and perceptual realism. This method demonstrates significant clinical potential for intraoperative real-time analysis and in vivo assessment, offering a promising pathway toward rapid pathology diagnosis without traditional tissue 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.
1.The authors introduce the first voxel-wise paired dataset of BIT and fluorescence-labeled nuclei. 2.The authors propose a Bidirectional Multiscale Content Consistency Loss to tackle the shift-variant contrast issue in BIT imaging, where nuclear intensity varies with focal position. Using multiscale features from a ViT and a U-Net encoder, this loss enforces feature alignment in a bidirectional cycle, thereby preserving global structure during style transfer. 3.A cross-domain style reuse module is proposed to enhance realism. To improve visual fidelity, the authors leverage AdaIN-based injection of learned H&E style features into the BIT-to-H&E generator. This module enables the transfer of realistic staining characteristics from the target H&E domain while preserving the underlying tissue structure.
- 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.
The evaluation in this paper is limited. All metrics are computed only on nuclear segmentation results, with no assessment of other important structures such as cytoplasm or extracellular matrix. Furthermore, while a 3D dataset is introduced, the proposed virtual staining framework operates in 2D and does not take advantage of the available 3D information. No 3D virtual staining results are generated, which limits the impact of the 3D dataset contribution. In Section 2.4, the authors state:” We update a running-mean prototype…consistent cross-dataset staining “. The prototype stably estimates the global H&E staining style distribution. However, could this result in overly smooth staining and the loss of local information?
- 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.
- 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?
1.A data contribution. The paper introduces the first HistoBIT3D dataset—a voxel-wise precisely registered 3D dataset combining BIT imaging with fluorescently labeled nuclei. 2.The paper proposes a novel virtual staining framework that converts BIT volumes into realistic H&E-stained images. 3.A limitation of this paper is that all evaluations are based only on nuclear segmentation results, and the rich 3D information available in the dataset is not used.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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 paper proposes a deep learning framework based on a 3D Vision Transformer (ViT) Generative Adversarial Network (GAN) to translate Backscattering Interferometric Tomography (BIT) images into 3D H&E stained images. Additionally, the authors construct and contribute a novel dataset comprising paired/unpaired BIT and H&E images to facilitate research and establish a benchmark in slide-free histology.
- 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.High Task Novelty and Clinical Relevance: This paper is among the first to explore the cross-modal image translation from BIT to 3D H&E. Given BIT’s tremendous potential for non-destructive, low-cost, and high-speed imaging without photobleaching, developing a highly accurate virtual staining algorithm for it is a significant step toward the clinical translation of slide-free pathology. 2.Valuable Dataset Contribution: The introduction of a new BIT and H&E image dataset is a substantial contribution to the community. High-quality 3D medical microscopic datasets are notoriously difficult to acquire and align. Releasing this dataset will undoubtedly benefit the broader medical image computing community and stimulate further research in this emerging 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.
1.Limited Methodological Innovation: The proposed 3D ViT-GAN architecture relies heavily on existing and well-established 3D Transformer and GAN components commonly used in other medical image translation tasks. While the authors mention the unique physical challenges of BIT, such as shift-variant contrast, the network design does not seem to incorporate specific physical priors or customized modules (e.g., specialized attention masks or physics-informed loss functions) to address these specific optical artifacts. 2.Insufficient Experimental Validation (Baselines): The experiments lack comprehensive comparisons with strong, contemporary baselines. To establish a solid benchmark for this new dataset, the authors should compare their method against widely used 3D image translation models (e.g., 3D CycleGAN, 3D CUT) or recent diffusion-based medical image translation models. 3.Lack of Clinical/Downstream Validation: For a clinically motivated application study, relying solely on standard computer vision metrics (like PSNR, SSIM, or FID) is insufficient. The paper lacks a clinical Turing test (blind evaluation by expert pathologists) or validation on downstream automated tasks (e.g., nuclei segmentation or tissue classification) to prove that the generated H&E images preserve accurate biological morphology. 4.Inadequate Ablation Studies: The paper lacks an in-depth ablation analysis of the core components. For instance, it is unclear how the depth of the 3D receptive field (number of Z-axis slices) impacts the mitigation of BIT artifacts, or what specific quantitative gains the Transformer module provides over a pure 3D CNN baseline.
- 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.
- 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?
I recommend a Weak Accept for this manuscript primarily due to its high task novelty, clinical relevance, and the substantial contribution of introducing a new 3D BIT and H&E dataset. Exploring cross-modal virtual staining from BIT to 3D H&E is an exciting step toward slide-free pathology, and releasing this notoriously difficult-to-acquire dataset will undoubtedly catalyze future research in the medical imaging community. However, the recommendation is marginal because the methodological innovation is limited; the proposed 3D ViT-GAN relies heavily on off-the-shelf components without incorporating specific physical priors to address BIT’s unique optical artifacts. Furthermore, the experimental validation requires significant strengthening, as it currently lacks comparisons with strong contemporary baselines (e.g., 3D CycleGAN, 3D CUT, or diffusion models), clinical Turing tests or downstream task validation, and in-depth ablation studies on the 3D receptive field. Ultimately, while the algorithmic contribution is somewhat incremental and the evaluation has notable gaps, the immense value of the curated dataset and the pioneering nature of the clinical application outweigh these weaknesses, making this work a valuable proof-of-concept for the field.
- 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 thank all reviewers for their thoughtful comments. R1Q7: We are currently expanding HistoBIT3D to include more tissue types and larger-scale 3D samples. We also acknowledge that an isolated ablation of the multiscale consistency loss would strengthen the paper. Finally, we agree that additional clinical validation is important and are currently designing pathology-specific studies on our virtually stained 3D data. R2Q7: Future work will expand the dataset with additional pathological annotations, including cytoplasmic and ECM structures, ideally with pathologist-provided labels. While the current framework performs slice-wise 2D translation, the stacked volumes remain spatially continuous and support meaningful 3D downstream validation. We also explored adjacent-slice consistency losses, but these did not significantly improve performance. We plan to investigate fully 3D virtual staining models in future work. R4Q7: To clarify, the proposed framework is a 2D image-to-image translation model, with 3D volumes generated through slice stacking. Despite this, the resulting volumes are spatially continuous and enable quantitative 3D downstream validation. We explored incorporating adjacent-slice consistency losses, but these added complexity without significant performance gains. Future work will investigate fully 3D and diffusion-based architectures, as well as additional pathologist-driven clinical validation.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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 reviewers view the main strengths as the paper’s valuable HistoBIT3D dataset contribution, high task novelty, and clinical relevance. The voxel-wise paired BIT–fluorescence/H&E data enable quantitative evaluation of structural fidelity, which is especially important for virtual staining. They also appreciate that the proposed method is well motivated for BIT’s shift-variant contrast, with content consistency losses designed to preserve structure and a style reuse module to improve H&E realism. The use of zero-shot Cellpose segmentation on registered fluorescence volumes is also considered a thoughtful and relevant validation strategy.
However, the reviewers also identified several weaknesses. First, the methodological novelty is considered limited, with the main contribution lying more in the dataset and evaluation protocol than in the model itself. Second, the experimental validation is not fully convincing, due to insufficient comparisons with stronger 3D image translation or diffusion-based baselines and limited evidence of generalization across tissues or acquisition conditions. Third, the ablation studies are viewed as too narrow, particularly with respect to the individual consistency loss scales, the ViT bottleneck, the style prototype update, and the 3D receptive field. Reviewers also note that, despite introducing a 3D dataset, the proposed method appears to operate primarily in 2D and does not fully exploit the available 3D information. Finally, the clinical validation is considered insufficient, as the paper lacks a pathologist study, diagnostic endpoint, or broader assessment of biological structures beyond nuclei.
I believe the strongest aspects of this work are the dataset contribution and the novelty of the task. Despite the weaknesses noted above, all three reviewers recommend acceptance, and I agree with their assessment. Therefore, I recommend acceptance.