List of Papers Browse by Subject Areas Author List
Abstract
Specular highlights frequently occur in endoscopic imaging due to directed illumination and moist tissue surfaces, producing saturated regions that obscure anatomical details and degrade downstream tasks. Accurate highlight detection is therefore essential for reliable medical image analysis and restoration.
We propose a confidence-aware multi-cue fusion framework for medical highlight detection. A Selective Confidence Fusion Module adaptively estimates the spatial reliability of structural and chromatic cues, including edge responses, color contrast, background suppression, and semantic consistency. Through per-pixel soft weighting, the model suppresses unreliable signals caused by illumination variation while enhancing anatomically consistent highlight regions, especially along fine boundaries and scattered reflections. Experiments on a public dataset and a clinically collected nasal endoscopy dataset demonstrate consistent improvements over recent methods. Cross-domain evaluation on CVC-ColonDB further verifies strong generalization. In addition, restoration results confirm that more accurate highlight localization leads to cleaner reflection removal and better structural preservation. These findings highlight the importance of confidence-aware cue modulation for robust medical highlight modeling.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3264_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)
N/A
BibTex
@InProceedings{HuDao_Selective_MICCAI2026,
author = { Hu, Daosong AND Li, Zefeng AND Liu, Shuai AND Cui, Mingyue AND Song, Rihui AND Xie, Yanchong AND Huang, Kai},
title = { { Selective Confidence Fusion for Specular Highlight Detection in Medical Images } },
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 studies specular highlight detection in medical images and proposes a confidence-aware multi-cue fusion framework. Specifically, it introduces a Selective Confidence Fusion Module (SCFM) into a U-shaped encoder-decoder architecture to adaptively fuse edge, color, background, and prediction-related cues. The authors conduct experiments on the public CVCD dataset and their self-collected Nasal-Endo Dataset (NED), and further evaluate the method in a cross-domain setting on CVC-ColonDB. Overall, the results show that the proposed method performs well in highlight boundary localization, detection of small reflective regions, and downstream restoration quality.
- 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.Rather than simply concatenating multiple cues, the paper is motivated by the practical observation that different cues may have different reliability in different regions. Based on this, it proposes a confidence-aware fusion strategy to selectively integrate edge, chromatic, background, and semantic cues. This motivation is well aligned with the characteristics of medical highlight detection, including specular reflections, local saturation, and boundary ambiguity.
2.The method outperforms the listed baselines on both CVCD and NED. In addition, the ablation study in Table 2 generally supports the contribution of SCFM, the edge cue, and the color cue to the overall performance improvement.
3.Fig. 2 shows that the proposed method produces more stable predictions in boundary regions and on small reflective spots. Figure 3 further demonstrates how improved highlight detection leads to better restoration quality, which means that the contribution of the paper goes beyond detection metrics alone.
4.The authors train on NED and directly test on CVC-ColonDB. Fig. 4 shows that the method still maintains more stable detection results under a clear domain shift, which supports its robustness and generalization potential.
- 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 tensor shape of $\alpha^{(k)}$ in Eq. (4) is not clearly specified, which makes the interpretation of the method ambiguous. If $\alpha^{(k)}$ is only a global scalar, then it can only modulate the contribution of a cue globally, rather than achieving the local adaptive behavior claimed by the authors, namely reducing the confidence of a cue in uncertain prediction regions. In other words, if a cue exhibits strong responses in both low-confidence and high-confidence regions, a global scalar would suppress both simultaneously rather than selectively down-weighting only the unreliable regions. Therefore, the authors should clarify whether$\alpha^{(k)}$ is a per-pixel weight map aligned with the spatial feature map, and explicitly specify the normalization dimension of the softmax operation in the formulation.
2.The evidence for cross-domain generalization is currently mainly qualitative, and the quantitative support is relatively limited.
3.According to Table 2, on NED the Dice score from A1 to A2 increases slightly (0.941→0.943), whereas the IoU decreases (0.922→0.913). Therefore, the phrase “consistently improves” appears somewhat too strong, and the authors are encouraged to replace it with a more precise description.
- 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?
I am inclined to give an Accept. This paper addresses the specific yet important problem of medical highlight detection and presents a well-motivated and overall effective method design. The main experiments, ablation studies, visualizations, and downstream restoration results are overall sufficient to support the effectiveness of SCFM, and the cross-domain evaluation also suggests good robustness and generalization potential. Although the mathematical formulation in Eq. (4) still requires clarification, and the cross-domain experiment lacks quantitative evidence, these issues are more related to presentation and exposition than to the core contribution itself, and they do not substantially undermine the main findings or conclusions of the paper.
- 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 #2
- Please describe the contribution of the paper
The paper addresses specular highlight detection in medical images, particularly endoscopic scenarios where strong reflections frequently obscure anatomical structures and degrade downstream analysis. The main contribution is a confidence-aware multi-cue fusion framework that integrates structural, chromatic, background, and semantic prediction cues within a U-shaped encoder–decoder architecture.
- 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 targets a practically meaningful problem in medical image computing. Specular highlight is a common issue in endoscopic and surgical imaging, where bright reflections can obscure tissue structures and negatively affect segmentation, navigation, or image restoration. (2) The method combines multiple complementary cues, including edge information, chromatic contrast, background suppression, and semantic predictions. This is intuitively well motivated because specular highlight often exhibits distinctive brightness, boundary, and color characteristics.
- 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 employs a private clinical nasal endoscopy dataset (NED with 10,000 frames) that is not publicly available, which prevents result reproduction. Please clarify whether NED can be made publicly accessible. (2) The paper does not specify the dimension alignment strategy for cue-conditioned feature maps. It is unclear how the edge features and chromatic differences are matched with encoder features of different scales. Please clarify the critical implementation details, like the definitions of functions σ(.) and ϕ(.). (3) The paper claims to meet real-time clinical requirements. However, it only reports the FPS achieved on a high-end NVIDIA GeForce RTX 4070 Ti SUPERGPU, without providing parameters, FLOPs, or computational complexity. This potentially limits its deployment on resource-constrained intraoperative hardware. (4) The paper only presents single-run numerical results, which are insufficient to demonstrate the stability and statistical significance of the proposed method. (5) The framed regions in the first row of Fig. 3 lack zoomed views, making it difficult to judge differences in boundaries, fine details, and artifacts. Please add magnified insets for all marked regions. (6) Please add standard restoration metrics such as PSNR and SSIM to objectively validate the restoration performance. (7) The proposed SCFM appears closely related to existing attention/gating-based fusion modules. The manuscript should better clarify the methodological novelty over prior confidence-weighted fusion strategies.
- 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 appears to closely resemble existing confidence fusion techniques, and the novelty of the method could be further clarified. Omitted details result in replication challenges.
- 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 paper presents a confidence-aware framework for specular highlight detection in medical images. The proposed method introduces a Selective Confidence Fusion Module (SCFM) to adaptively fuse edge, chromatic, background, and prediction cues within a U-shaped encoder-decoder architecture.
- 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 a clinically relevant problem, as specular highlights are common in endoscopic imaging and can negatively affect both visualization and downstream analysis. 2.The overall motivation is reasonable: structural and chromatic cues are not equally reliable in highlight regions, and confidence-aware fusion is a sensible design choice. 3.The method is lightweight and practical, with reported real-time inference speed. 4.The paper includes basic ablation studies and a cross-domain evaluation attempt.
- 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 dataset naming is slightly confusing (e.g., CVC-Clinic vs. CVCD). 2.The real-time claim should be better contextualized, especially with respect to the actual input resolution used for inference. 3.Additional metrics such as sensitivity, precision, or boundary-aware measures would strengthen the evaluation for this task. 4.Can the authors provide quantitative cross-domain results on CVC-ColonDB? 5.Can the authors show quantitative benefit on a downstream task, rather than only qualitative restoration examples?
- 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 studies an important problem and proposes a reasonable confidence-aware fusion design. The method is simple, computationally efficient, and empirically better than the included baselines in the reported in-domain experiments.
- 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
Author Feedback
We sincerely thank the reviewers for their careful reading and constructive suggestions. The comments are highly valuable for improving the clarity and presentation of the manuscript. We will revise the paper accordingly and provide additional explanations and experimental analysis in the revised version.
### Response to Method Clarification
We thank the reviewer for pointing out the ambiguity in Eq. (4). In our implementation, the confidence weights in the Selective Confidence Fusion Module are spatially varying confidence maps rather than global scalar coefficients. The confidence estimation branch predicts per-pixel reliability scores for edge, chromatic, background, and prediction cues, while the softmax normalization is performed across cue branches at each spatial location independently. Therefore, SCFM performs locally adaptive fusion instead of global cue suppression.
This design is particularly important for endoscopic highlight detection, since specular reflections often introduce locally unreliable gradients and chromatic responses around reflective regions. The proposed confidence-aware fusion selectively suppresses misleading cues near reflective tissues while preserving their contribution in anatomically consistent regions. We agree that the current manuscript does not describe this process clearly enough. In the revised version, we will explicitly clarify the tensor dimensions, normalization strategy, and spatial adaptive behavior of the confidence-aware fusion process.
### Response to Cross-domain Evaluation and Quantitative Analysis
We thank the reviewer for this valuable suggestion. We agree that additional quantitative cross-domain evaluation would further enhance the completeness of the experimental analysis. However, due to the strict page limitation of the MICCAI format, we were unable to include this additional experiment in the current manuscript. Therefore, we mainly present the cross-domain results through qualitative visualizations on CVC-ColonDB, which provide evidence for the generalization ability of the proposed method. We sincerely appreciate the reviewer’s suggestion and will consider more comprehensive quantitative cross-domain evaluation in our future work.
Regarding the downstream restoration analysis, we agree that quantitative evaluation would be valuable. However, real endoscopic highlight regions do not have paired ground-truth restoration targets, making direct quantitative restoration evaluation difficult. Therefore, the current manuscript mainly analyzes restoration quality through structural consistency, boundary preservation, and visual artifact reduction.
### Response to Presentation and Writing Issues
We appreciate the reviewers’ careful observations regarding presentation details. In the revised manuscript, we will standardize all dataset naming conventions throughout the paper to avoid ambiguity, including the usage of CVC-ClinicDB and CVC-ColonDB.
We will also revise the description of the real-time capability by explicitly reporting the inference resolution, hardware configuration, and testing settings associated with the reported FPS performance.
In addition, we agree that the phrase “consistently improves” is overly strong given the slight IoU variation observed in Table 2.We will replace it with more precise and conservative wording.
We sincerely appreciate all reviewers for their constructive comments. The feedback has helped us identify several areas where the manuscript can be improved, particularly in terms of methodological clarity, quantitative validation, and presentation quality.
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 highlight that the paper addresses a clinically relevant endoscopic highlight-detection problem with a well-motivated confidence-aware fusion strategy that selectively integrates structural, chromatic, background, and semantic cues according to their reliability. They also note that the method is lightweight and practical, achieves strong performance with useful ablation and cross-domain evaluations, and shows potential benefits for downstream restoration and interpretable clinical decision support.
