List of Papers Browse by Subject Areas Author List
Abstract
Polyp segmentation in colonoscopy images is crucial for early
detection and diagnosis of colorectal cancer. However, this task remains a significant challenge due to the substantial variations in polyp shape, size, and color, as well as the high similarity between polyps and surrounding tissues, often compounded by indistinct boundaries. While existing encoder-decoder CNN and transformer-based approaches have shown promising results, they struggle with stable segmentation performance on polyps with weak or blurry boundaries. These methods exhibit limited abilities to distinguish between polyps and non-polyps and capture essential boundary cues. Moreover, their generalizability still falls short of meeting the demands of real-time clinical applications. To address these limitations, we propose SAM-MaGuP, a groundbreaking approach for robust polyp segmentation. By incorporating a boundary distillation module and a 1D-2D Mamba adapter within the Segment Anything Model (SAM), SAM-MaGuP excels at resolving weak boundary challenges and amplifies feature learning through enriched global contextual interactions. Extensive evaluations across five diverse datasets reveal that SAM-MaGuP outperforms state-of-the-art methods, achieving unmatched segmentation accuracy and robustness. Our key innovations—a Mamba-guided boundary prior and a 1D-2D Mamba block—set a new benchmark in the field, pushing the boundaries of polyp segmentation to new heights.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3165_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/deepak1113/SAM-MaGuP
Link to the Dataset(s)
N/A
BibTex
@InProceedings{DutTap_Mamba_MICCAI2025,
author = { Dutta, Tapas K. and Majhi, Snehashis and Nayak, Deepak Ranjan and Jha, Debesh},
title = { { Mamba Guided Boundary Prior Matters: A New Perspective for Generalized Polyp Segmentation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
page = {385 -- 395}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces SAM-MaGuP, a framework specifically designed for polyp segmentation in colonoscopy images. It addresses the original SAM limitations in detecting weak polyp boundaries by introducing two key modules: a boundary distillation module that enhances boundary detection and a 1D/2D Mamba adapter that facilitates feature channel fusion.
- 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 authors present a novel Mamba-guided framework for training a SAM backbone that effectively tackles the challenge of weak boundaries often found in images of polyps.
-
The paper presents a well-executed evaluation with comprehensive results evaluated on both seen and unseen datasets. The comparison includes SOTA methods which enhances the validity of their findings and their highlights the paper’s contribution.
-
A well structured introduction section that outlines the challenge of transferring SAM to polyp detection, particularly focusing on overcoming the weak boundary issue. It clearly establishes the context and the significance of the problem being addressed.
-
An ablation study demonstrates the incremental contribution of the presented modules to the performance.
-
- 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.
Clarity and Reproducibility in the Proposed Method Section:
- While the paper presents intriguing ideas, it seems to lack precise implementation details, such as the number of convolutional filters and training parameters, making it challenging to fully assess reproducibility (the submission guidelines indicate that supplementary materials should be limited to multimedia content, excluding PDFs with additional results or implementation details. Unfortunately, the appendix provided does not adhere to these guidelines, so it was set aside during the review process).
- The paper often uses the terms “adapter” and “Mamba adapter,” which I find confusing as Figure 1(a) shows only one type of “Adapter” module. The MaGuP module is called an “adapter-based” approach, but its connection to the “Adapter” module and the 1D/2D Mamba is unclear. It might be helpful to use different naming conventions. Additionally, the paper lacks detailed explanations or references to the Adapter modules depicted in Figure 1(a).
- The term “Saliency pool” is not clearly defined, and it would be helpful to understand the rationale behind the naming of the saliency and contextual maps in the Mamba Spatial and Channel Interaction.
- 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 provide sufficient information for reproducibility.
- 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/2025/en/REVIEWER-GUIDELINES.html
Minors:
- Please include a reference to support the claim regarding the miss rate of polyp detection (6-27%).
- The E^hat term in the loss in formula (2) is not introduced to the reader, and it appears to have been confused with E^star
- The loss {L_D} is mentioned in section 2.3 without prior introduction. It might be helpful to present it with an enumerated formula before referencing it.
- 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 — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I suggest a weak rejection of this paper because it lacks clear implementation details, making it hard to fully evaluate and reproduce the work. Additionally, the confusing use of terms and unclear definitions make it difficult to fully understand the method. These issues are especially concerning for a paper that claims to be groundbreaking. Addressing them could significantly enhance its impact and contribution.
- 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 author’s response, I now recommend accepting the paper. They have addressed my concerns about reproducibility by agreeing to include important details in the main paper and to share their code publicly if the paper is accepted. They explained the trade-offs in model size and computational cost, showing that the small increase in GFLOPs is justified by better performance. Additionally, they clarified terms like “Adapter” to make the paper clearer and more understandable. With these improvements, along with their claim to achieve state-of-the-art results, I find the paper suitable for publication.
Review #2
- Please describe the contribution of the paper
This paper proposes SAM-MaGuP, a groundbreaking approach for robust polyp segmentation. By incorporating a boundary distillation module and a 1D-2D Mamba adapter within the Segment Anything Model (SAM), SAM-MaGuP excels at resolving weak boundary challenges and amplifies feature learning through enriched global contextual interactions.
- 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.
-
This paper introduces a Mamba-guided boundary prior in SAM, the first of its kind, to effectively tackle the weak-boundary challenge in polyp segmentation.
-
This paper proposes a boundary distillation component that empowers SAM to accurately detect polyps in weak-boundary scenarios, enhanced by a 1D-2D Mamba block that optimizes feature interactions across spatial and channel dimensions.
-
This paper demonstrates the outstanding performance of SAM-MaGuP on five diverse datasets, consistently surpassing existing state-of-the-art polyp segmentation techniques.
-
- 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 segmentation model based on SAM is already quite large. Would introducing a complex structure like Mamba significantly reduce computational efficiency? This paper lacks comparisons and discussions regarding model size.
-
Compared with other methods, the final segmentation results of this paper do not seem to show significant improvement.
-
This paper lacks an introduction to the hardware conditions used for training.
-
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
-
The segmentation model based on SAM is already quite large. Would introducing a complex structure like Mamba significantly reduce computational efficiency? This paper lacks comparisons and discussions regarding model size.
-
Compared with other methods, the final segmentation results of this paper do not seem to show significant improvement.
-
This paper lacks an introduction to the hardware conditions used for training.
-
In images where boundaries are difficult to accurately obtain, will the boundary distillation component have an impact on the SAM segmentation model?
-
It is recommended that the used datasets and code be open-sourced to improve the reproducibility of the proposed method.
-
- 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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed method still lacks detailed elaboration in terms of computational efficiency and hardware conditions.
- Reviewer confidence
Very confident (4)
- [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.
The author’s response has essentially resolved my doubts, and I will maintain my previous rating.
Review #3
- Please describe the contribution of the paper
The paper introduces a novel polyp segmentation framework called SAM-MaGuP, which enhances the performance of the Segment Anything Model (SAM) for medical image segmentation—specifically targeting polyps in colorectal cancer screening. It does this by integrating a Mamba-based adapter and a boundary distillation strategy to overcome key limitations in existing methods such as poor boundary detection
- 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-A new framework that builds upon SAM by injecting domain-specific knowledge using a Mamba-based adapter module and boundary-aware learning mechanisms. 2-A feature fusion technique that integrates channel-wise (1D) and spatial (2D) representations. 3-A boundary-focused distillation unit that learns precise boundary cues. 4-Strong experimental results for polyp segmentation.
- 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 paper contains weak or insufficient explanations of several key components. Important modules such as DWConv and notations like 1-M are introduced without adequate clarification, making it difficult for readers to fully grasp their roles or implementations.
Additionally, the main modules—such as BDC and the 1D-2D Mamba mechanism—are described only briefly, lacking the necessary conceptual depth. A more thorough breakdown of their architectural design, functionality, and novelty would significantly enhance readability and comprehension.
The experimental comparisons are limited to models developed specifically for polyp segmentation, which restricts the scope of evaluation. It would be beneficial to include general-purpose segmentation models like Mask2Former or OneFormer as baselines to provide a broader and more robust performance context.
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper has more strengths than weaknesses, but some concerns still need clarification.
- Reviewer confidence
Very confident (4)
- [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.
stated above
Author Feedback
We thank the reviewers for their insightful feedback and recognition of the novelty of our work. We will address the following clarifications in the revision.
Hyperparameter, hardware, reproducibility (R1, R2, R3): The supplementary material covers hyperparameters and implementation details- Adam optimizer (LR=1e-5), image size (352×352), scaling factors of multi-scale training ({0.75, 1.25}), combined (BCE + Dice) loss, and hardware (NVIDIA A100, 80GB VRAM). As suggested by R2, we will include these details in the main paper. Additionally, we will open-source the code upon acceptance.
Comparison of Model Size and Computational Cost (R1, R2): SAM-MaGuP has higher segmentation accuracy than SAM-Mamba with minimal added cost: 106M vs. 103M trainable parameters, identical inference parameters (103M), and a slight GFLOPs increase (431 vs. 423). This trade-off is justified by improved performance.
Clarifying “Adapter” and “Mamba Adapter” (R2): In Figure 1(a), “Adapter” refers to a general fine-tuning module (from [1]) for integrating domain-specific data into the SAM backbone. The “Mamba adapter,” however, is our novel design using the 1D-2D Mamba block for refining features. To avoid ambiguity, we’ll use clearer terms like “General Adapter” for standard modules and “Mamba Adapter” for ours. The MaGuP module integrates the “Mamba Adapter” to boost SAM’s features, a key driver for the boundary distillation mechanism. [1] Zhe et al., “Vision Transformer Adapter for Dense Predictions,” ICLR 2023.
Clarifying “Saliency pool”, “saliency Map”, and “contextual map” (R2): The “Saliency Pool” is a 2×2 MaxPooling operator that highlights spatial regions with higher semantic importance. The 1D Mamba layer outputs the “Saliency Map”, emphasizing key channel-wise features, while the 2D Mamba layer produces the “Contextual Map”, integrating global spatial relationships.
Do not show Significant Improvement (R1): While quantitative improvement on some metric may seem incremental, our approach delivers sharper boundary precision, driving significant qualitative gains, shown in Figure 2. Qualitative improvements are vital in clinical settings, particularly for detecting complex polyps (like sessile, flat, serrated, and diminutive) that are often missed during colonoscopy. Our method enhances diagnostic efficacy and reduces false positives for such cases.
Impact of Boundary Distillation Component (R1): As shown in Table 3, the BDC module consistently enhances segmentation performance across datasets. Furthermore, Figure 2 (Rows 1, 4) reinforces this claim, illustrating weak-boundary polyp cases where our method surpasses SoTA methods in segmentation precision by producing sharp, clinically meaningful delineations.
Insufficient explanation on DWConv and “1-M” (R3): DWConv (depth-wise convolution) is a lightweight operation that processes input channels independently, reducing computation while preserving spatial details. The “1-M” notation represents the complement of the ground truth mask (M), highlighting non-polyp regions. This is key in the BDC, enabling sharper differentiation between polyp and non-polyp areas for better feature refinement.
Details of BDC and the 1D-2D Mamba (R3): Due to page limitations, the components were described briefly. We will include more detailed diagrams, architecture breakdowns, and their contributions in the revision to enhance clarity.
Comparison with General Purpose Segmentation model like Mask2Former (R3): We compared SAM-MaGuP with 11 domain-specific methods. In future, we will explore Mask2Former and OneFormer for comparative analysis, adhering to reviewer guidelines.
Minor Error (R2): We will: (1) include a reference to [2] for polyp miss rate of (6-27%); (2) replace the E^hat with E^star; (3) introduce the loss {L_D} and its complete formula prior to Section 2.3. [2] Ahn et al., “The miss rate for colorectal adenoma determined by quality-adjusted, back-to-back colonoscopies,” Gut Liver, 2012.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All three reviewers are positive to accept this work after the rebuttal. Following these ratings, I think this work can be published in MICCAI 2025.
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A