List of Papers Browse by Subject Areas Author List
Abstract
In colonoscopy video polyp segmentation, degradations such as motion blur, specular reflections, and occlusions often render single-frame features incomplete and unreliable. However, most existing methods rely heavily on current-frame features to construct queries for cross-frame retrieval and alignment, leading to performance drops under such degradations. To address this, we propose a query-enhanced cascaded memory framework integrating Quality Awareness (QA) and Prototype Alignment (PA). Specifically, (1) to construct stable query representations for robust retrieval, we propose a quality-aware sensory memory mechanism that adaptively gates memory updates based on observation reliability and state consistency; (2) to mitigate appearance drift, we propose a prototype alignment mechanism that maintains appearance prototypes as long-term anchors. By leveraging stable queries to retrieve relevant prototypes, our framework enables reliable cross-frame association. On the most challenging Unseen-Hard subset of SUN-SEG, our method improves the best baseline by 2.9% in Dice and 4.2% in Sensitivity (Sen), demonstrating superior robustness under degraded conditions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2213_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)
SUN-SEG dataset: https://github.com/GewelsJI/VPS
BibTex
@InProceedings{XioSij_Robust_MICCAI2026,
author = { Xiong, Sijing AND Yin, Yunfei AND Dey, Argho AND Yuan, Zheng AND Timbi, Zumnan AND Ray, Swachha AND Liu, Hongyu},
title = { { Robust Colonoscopy Video Polyp Segmentation via Quality-Aware Memory and Prototype Alignment } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16892},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a query-enhanced cascaded memory framework for video polyp segmentation under degraded imaging conditions. Its key idea is to construct more reliable temporal queries using a Quality Awareness (QA) module with gated sensory memory, and then use these stable queries to retrieve long-term appearance anchors through a Prototype Alignment (PA) module. The method is designed to reduce reliance on noisy current-frame features and to mitigate appearance drift across video frames.
- 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 is well motivated by realistic failure modes in colonoscopy videos, such as motion blur, specular reflections, occlusions, and large inter-frame displacement. The overall method is coherent: QA stabilizes short-term memory updates using reliability and consistency signals, while PA introduces long-term prototype anchors to counter appearance drift. This decomposition is intuitive and well matched to the task. Another strength is that the empirical gains are concentrated on the hardest evaluation setting: the method improves the best baseline by 2.9% Dice and 4.2% Sensitivity on SUN-SEG Unseen-Hard, and the ablation study shows that QA and PA provide complementary gains.
- 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 main weakness is limited evaluation breadth. All experiments are conducted on SUN-SEG, so it is difficult to judge how well the method generalizes beyond this benchmark. A second weakness is that the technical novelty, while solid, is moderate: the paper combines ideas related to memory gating, dynamic memory, and prototype/feature-bank style retrieval that are related to prior works such as quality-aware dynamic memory, adaptive feature bank methods, space-time memory networks, and SALI. The contribution lies more in the task-specific integration than in a fundamentally new temporal modeling formulation. Finally, some implementation details that matter for reproducibility and robustness are not fully analyzed, such as initialization of the prototype bank and sensitivity to design choices like prototype-bank size, top-K retrieval, and update interval.
- 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.
(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?
My overall recommendation is Accept. The paper addresses an important and realistic problem in video polyp segmentation, the method is well aligned with the failure mode it targets, and the experimental improvements on the hardest split are meaningful. The work is not highly groundbreaking, and the evaluation is limited to a single dataset, but overall the paper is technically sound, clearly motivated, and empirically convincing enough to merit acceptance.
- 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 authors propose a query-enhanced cascaded memory framework for robust colonoscopy video polyp segmentation under degraded conditions. The framework includes a Quality Awareness (QA) module that filters unreliable short-term observations through gated sensory memory, and a Prototype Alignment (PA) module that uses stabilized queries to retrieve long-term appearance prototypes to alleviate appearance drift.
- 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 addresses the degradation of single-frame queries and the subsequent noise propagation in colonoscopy video analysis. (2) The Quality-Guided Network presents a well-structured design. (3) The proposed method demonstrates quantitative improvements.
- 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) Real-time performance is crucial for clinical deployment. However, the manuscript does not report inference speed (FPS), computational cost (FLOPs), or model parameters. Given the cascaded architecture with multiple attention modules, the practical feasibility remains unclear. (2) In Section 2.2, Eq. (4) extracts candidate prototypes using the current predicted mask. Unlike the QA module, this step does not apply a quality gate. As a result, high-confidence false positives may be added to the prototype bank. Under the LFU replacement strategy, frequently queried errors could persist in memory and potentially worsen the appearance drift the module aims to address. (3) The paper does not provide a code repository or dataset availability statement, which makes it difficult to verify implementation details or computational cost.
- 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?
This paper presents a structured framework for robust video polyp segmentation, and the QA module offers a reasonable strategy to filter degraded observations. The method also shows clear experimental improvements.
- 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 #3
- Please describe the contribution of the paper
1.Proposed a query-enhanced cascaded memory framework to address single-frame feature degradation in colonoscopy video segmentation.Designed a Quality Awareness (QA) module with a gating mechanism to filter noise and generate stable queries. Introduced a Prototype Alignment (PA) module that uses stable queries to retrieve long-term anchors, mitigating appearance drift. 2.Achieved state-of-the-art results on the SUN-SEG dataset, demonstrating particular robustness on the challenging Unseen-Hard subset.
- 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.Effectively targets prevalent colonoscopy degradations (motion blur, occlusions), ensuring high practical value. 2.Logically separates short-term noise filtering (QA) from long-term drift tracking (PA) using a mathematically sound gating mechanism.
- 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 Endoscopic-Specific Novelty: The QA module resembles generic CV techniques. Authors must clarify specific algorithmic differences from existing works or unique adaptations made for endoscopic scenes. 2.The “reference frame alignment” in the PA stage bypasses the QA gating, directly fusing unfiltered adjacent frames. This contradicts the introduction’s critique of traditional alignment methods and weakens the architecture’s overall consistency. Ablations or justifications are needed. 3.Dataset Limitation: Validation is restricted solely to the SUN-SEG dataset (proposed in 2022). Testing on broader, newer datasets (or cross-modalities like ultrasound) may be necessary to prove true generalizability.
- 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?
This paper presents a logically decoupled framework addressing critical clinical challenges in colonoscopy. Its strict quality-gating mechanism and strong performance on difficult datasets prove its robustness. There are some revisions may be needed regarding endoscopic-specific novelty, architectural self-consistency (specifically the reference frame alignment), and broader dataset validation.
- 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
Author Feedback
We thank the reviewers and AC for their positive evaluation and constructive suggestions. We appreciate the reviewers’ recognition of the clinical relevance and empirical performance of our work. We would like to clarify that prototype-bank related hyperparameters, including bank size, top-K retrieval, and update interval, were considered during method development, and the reported results use the selected configuration. Due to the space limit, detailed sensitivity analyses for all these factors were not included in the submitted manuscript. Regarding the concern on novelty, we note that the contribution of our method lies in the degradation-aware coupling between query stabilization and long-term prototype retrieval, rather than in using a memory or prototype bank in isolation. In colonoscopy videos, degraded frames caused by blur, reflection, occlusion, and large camera motion can corrupt the query features, which then propagate errors through temporal memory. This cascaded dependency makes QA and PA mutually connected: QA improves the quality of the query, while PA uses the improved query to recover stable long-term appearance information. We believe this degradation-aware formulation is different from generic memory gating or feature-bank retrieval methods, and we will revise the related-work and method sections to better clarify these algorithmic differences and their motivation in endoscopic scenes. We also agree that validation on additional datasets and clinical scenarios would further strengthen the evidence of generalization. Our current study follows the standard SUN-SEG protocol adopted by recent representative video polyp segmentation works, such as SALI, including the seen/unseen and easy/hard splits. This benchmark is well aligned with the degradation and robustness issues targeted in our work, enabling meaningful comparison within the established VPS setting. Broader cross-dataset and cross-scenario validation remains an important direction for future work. We thank the reviewers again for their time and constructive feedback.
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.
This paper proposes a query-enhanced cascaded memory framework for robust video polyp segmentation under degraded conditions. All the three reviewers are positive and appreciate the paper’s contribution and good performance. However, there are also some concerns such as the limited experiments and writing issues. Overall this is a good paper and the recommendation is Provisional Accept.
