List of Papers Browse by Subject Areas Author List
Abstract
Advances in digital pathology have enabled large-scale automated grading through deep learning-based analysis of whole slide images (WSIs), which is essential for cancer diagnosis and prognosis. However, existing WSI classification methods typically treat grades as independent categories, ignoring the ordinal and progressive nature of their relationships.
Overlooking this intrinsic relationship leads to unstable boundaries and inconsistent predictions. Although ordinal regression and classification models have been explored to model the label continuity, they primarily enforce output orderings instead of learning continuous feature representations, leading to discrete embeddings that ignore inter-grade similarity. To address these problems, we propose GPO-MIL, an ordinal multiple-instance learning framework that constructs a severity-aligned Gaussian prototype space to model disease progression. A global set of Gaussian prototypes is learned and reordered by their correlation with slide grades, forming an interpretable benign-to-severe axis that guides responsibility-based prototype attention. An ordinal-aware optimization further stabilizes this progression structure and enhances grading consistency. Experiments on our in-house pancreatic cytology dataset and two public histopathology grading datasets, including CAMELYON17 and PANDA, demonstrate consistent gains in grading accuracy and calibration, along with interpretable prototypes that align with pathological progression.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2562_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
PANDAS dataset: https://gleason2019.grand-challenge.org/Home/
CAMELYON17 dataset: https://camelyon17.grand-challenge.org/Data/
BibTex
@InProceedings{DenYin_Learnable_MICCAI2026,
author = { Deng, Yingjiao AND Zhang, Qing AND Tang, Hui AND Zhou, Chunhua AND Gao, Lili AND Wang, Jiansheng AND Sun, Li AND Wang, Yan AND Zou, Duowu AND Xiong, Hongkai AND Li, Qingli},
title = { { Learnable Gaussian Prototype Guided Ordinal Learning for Fine-Grained WSI Grading } },
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 GPO-MIL, a Multiple Instance Learning (MIL) framework for fine-grained histopathological grading that explicitly models the ordinal nature of disease progression. The key contributions are: (1) Severity-Aware Ordered Prototype Space: A Gaussian Mixture Model fitted on training features and reordered by Spearman correlation with slide grades to form a benign-to-severe axis; (2) Responsibility-Guided Instance Aggregation: A two-stage attention mechanism that conditions instance features on the ordered prototype semantics; (3) Ordinal-Aware Optimization: A combination of coarse-to-fine hierarchical grading, Multi-Margin N-Pair loss for monotonic embedding structure, and prototype separation regularization.
- 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 paper correctly identifies that existing MIL methods treat grading as independent categorical classes, ignoring the progressive nature of pathological changes. The ordinal formulation is well-motivated for cytology (pancreatic LBC) and histopathology grading (PANDA, CAMELYON17), where adjacent grades share morphological similarities. This is a meaningful advance over standard MIL.
- 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.
GMM prototypes are similar to prototype-based MIL methods though the ordinal reordering is new. Prototype attention are standard self-attention and cross-attention mechanisms; the prototype-conditioned variant is a natural extension. The GMM is pre-fitted on training features before MIL optimization begins, then frozen for warm-up epochs. This two-stage process may lead to suboptimal prototypes that don’t adapt to the specific grading task. While prototype means are updated after warm-up, the covariances remain fixed—limiting the model’s ability to refine the prototype shape. The visualizations (Figures 3-4) show successful cases, but there is no analysis of whether the model struggles with specific morphological patterns The LBC dataset (in-house pancreatic cytology) cannot be verified by reviewers. While the results are promising, the community cannot reproduce or build upon this evaluation. The paper should clarify if the LBC dataset will be released or if additional public cytology datasets could validate the method. Code needs to be open source.
- 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.
(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?
This paper addresses a clinically important problem, ordinal grading in computational pathology, with a technically sound and empirically validated solution. The integration of GMM prototypes with severity-aware ordering and prototype-conditioned attention is coherent and effective. Results are strong across three datasets, and ablation studies are thorough. However, the methodological novelty is incremental (combining existing techniques), and the evaluation lacks ordinal-specific metrics that would fully validate the ordinal learning claim.
- 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
The paper proposes a prototype-guided ordinal multiple instance learning framework for fine-grained whole-slide image grading. Its main contribution is the construction of an ordered Gaussian-based prototype space that explicitly captures the progression structure of pathological grades, together with an ordinal-aware optimization strategy that preserves label ordering during training.
- 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 proposed method introduces an ordered prototype space, which is a reasonably novel way of injecting grade progression structure into MIL-based pathology models. 2.The combination of prototype-guided representation learning and ordinal-aware optimization is methodologically coherent. 3.The empirical evaluation across multiple datasets provides useful evidence that the proposed design is practically relevant.
- 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 evaluation is not fully aligned with the ordinal nature of the task. Given the paper’s ordinal design, ordinal-aware metrics such as MAE or quadratic weighted kappa would be important to show whether the method reduces the severity of grading errors rather than only improving standard classification scores. 2.The fairness of the empirical comparison is not fully convincing. Using the same epoch and learning-rate settings for all baselines is not necessarily fair, since different methods may require different optimization configurations to achieve their best performance.
- 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 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.
(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 problem and proposes an interesting ordinal-aware framework for WSI grading. The overall idea is meaningful and the multi-dataset results are reasonably promising. My main reservations are that the evaluation could better reflect the ordinal nature of the task, and the fairness of the baseline tuning strategy is not fully convincing. Overall, I view this as a borderline but positive submission, hence a weak accept.
- 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
This manuscript proposed GPO-MIL, an ordinal multiple-instance learning framework that constructs a severity-aligned Gaussian prototype space to model disease progression for a crucial diagnostic task (grading) in computational pathology. The main technical contribution lies in learning a severity-aligned Gaussian mixture prototype space reordered by Spearman correlation with pathological grades, then integrate responsibility-guided prototype attention and ordinal-aware optimization to model the continuous disease progression.
- 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] Meaningful motivation. This work addresses the critical limitation of conventional weakly-supervised learning manners that treats grading as independent classes while ignoring the ordinal and progressive nature of pathological grading, which is common in CPath. [2] Comprehensive experimental validation. Validated on one in-house cytology dataset with two public histology datasets, demonstrating the robustness of the proposed method. [3] Good organization. The illustration and logic of the description is easy-to-follow.
- 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] Related works. Lacks discussion on recent ordinal regression or prototype-based MIL in pathology grading. [2] Hyper-parameters. The weight tuning of three losses lacks explanation. The number of prototypes (K=8/16) is set heuristically, have the authors did sensitivity analysis?
- 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?
This paper presents a novel also well-motivated method for ordinal WSI grading. The experimental results are convincing.
- 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
N/A
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 paper proposes a prototype-guided ordinal multiple instance learning framework that constructs a severity-aligned Gaussian prototype space and uses ordinal-aware optimization to explicitly model disease progression for fine-grained histopathological grading in whole-slide images. Though there are concerns about incremental novelty and insufficient discussion of related works, all reviewers give positive scores and the motivation and comprehensive experimental validation are recognized by the reviewers.
