Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

The advent of Large Vision Models (LVMs) offers new opportunities for few-shot medical image segmentation. However, existing training-free methods based on LVMs fail to effectively utilize negative prompts, leading to poor performance on low-contrast medical images. To address this issue, we propose SynPo, a training-free few-shot method based on LVMs (e.g., SAM), with the core insight: improving the quality of negative prompts. To select point prompts in a more reliable confidence map, we design a novel Confidence Map Synergy Module by combining the strengths DINOv2 and SAM. Based on the confidence map, we select the top-k pixels as positive points set and choose negative points set using a Gaussian distribution, followed by independent K-means clustering for both sets. Then, these selected points are leveraged as high-quality prompts for SAM to get the segmentation results. Extensive experiments demonstrate that SynPo achieves performance comparable to state-of-the-art training-based few-shot methods. Project page: https://liu-yufei.github.io/synpo-project-page/.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1376_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{LiuYuf_SynPo_MICCAI2025,
        author = { Liu, Yufei and Xiao, Haoke and Chai, Jiaxing and Zhang, Yongcun and Wang, Rong and Meng, Zijie and Luo, Zhiming},
        title = { { SynPo: Boosting Training-Free Few-Shot Medical Segmentation via High-Quality Negative Prompts } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15964},
        month = {September},
        page = {596 -- 605}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes SynPo, a training-free few-shot medical segmentation method that leverages high-quality negative prompts to address limitations in existing approaches. The core contributions include:

    1. Confidence Map Synergy Module (CMSM): Combines features from DINOv2 (semantic-rich features) and SAM (spatially aware features) to generate a synergy map, improving anatomical localization and reducing errors in regions with low contrast or ambiguous boundaries.
    2. Point Selection Module (PSM): Introduces a novel strategy for selecting negative prompts based on Gaussian-distributed confidence thresholds, ensuring they are positioned within anatomical regions rather than background areas. This avoids redundancy and enhances segmentation accuracy.
    3. Noise-aware Refine Module (NRM): Refines coarse masks using morphology operations and SAM to suppress noise and improve boundary precision. Experiments on CT and MRI datasets demonstrate that SynPo achieves performance comparable to state-of-the-art (SOTA) training-based methods, particularly in spleen and kidney segmentation.
  • 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 PSM’s approach to selecting negative prompts within anatomical regions (rather than background areas) is a novel contribution. This addresses a critical limitation in prior work (e.g., PerSAM and ProtoSAM) where negative prompts were often clustered in easily distinguishable backgrounds, leading to over-segmentation.
    2. The CMSM effectively integrates the complementary strengths of DINOv2 (semantic features) and SAM (spatial encoding). This hybrid approach mitigates the weaknesses of using either model alone, such as DINOv2’s poor spatial localization or SAM’s weaker semantic discrimination. The fusion mechanism is well-motivated and empirically validated.
    3. The ablation studies clearly demonstrate the incremental gains from each module, highlighting the effectiveness of the synergy map and PSM. The authors also compare with both training-free (e.g., ProtoSAM, PerSAM) and training-based methods (e.g., GMRD), establishing competitive 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.
    1. Limited Innovation in Core Techniques: The CMSM’s feature fusion (combining DINOv2 and SAM features) is not entirely novel. Prior work already explored hybrid feature extraction to improve segmentation. The authors do not sufficiently distinguish their approach from these methods, weakening the novelty claim.
    2. While SynPo outperforms training-free methods like ProtoSAM (73.45% → 79.91% Dice on Synapse-CT), the improvement is modest compared to training-based methods (e.g., GMRD achieves 82.90% vs. SynPo’s 81.15%). Moreover, GMRD itself is a Few-Shot segmentation method.
    3. The ablation study lacks statistical rigor. The Dice gains are not contextualized by p-values or confidence intervals, making it unclear if improvements are statistically significant.
    4. The experimental focus on abdominal organs (e.g., liver) limits the generalizability of SynPo. These organs have well-defined boundaries and are less ambiguous than, say, tumors or soft tissue structures. The paper does not evaluate on more challenging tasks (e.g., brain tumors or fractures), undermining claims of clinical utility.
  • 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.

    (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?

    While SynPo introduces a well-motivated approach to improving negative prompts in training-free few-shot segmentation, its incremental contribution and lack of significant innovation compared to existing methods hinder its potential impact. The experimental results, while promising, do not exceed few-shot baselines. Therefore, I recommend rejecting this 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

    • This paper proposed a novel point prompts selection method for few-shot segmentation

    • This paper proposed a CMSM, PSM, and NRM modules.

    • In CMSM module, it generates reliable confidence map by exploiting both SAM and DINO features.

    • Then, in PSM, it selects the reliable positive and negative points based on the given confidence maps.

    • By using the selected positive and negative points, in inferred SAM model and refine the initial segment results with NRM.

  • 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. Extracting not only positive prompt but also high quality negative point prompts is crucial and the authors successfully extract reliable point prompts.

    2. The proposed CMS and PSM looks novel that exploiting the advantage of each model.

    3. The organization of paper is good.

  • 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. More ablation study is necessary. For example, in Table 2, only DINO results should be included to verify the effectiveness of the proposed approach. Also, ablation study on the number of points and N-shot results should be included.

    2. (Minor) Why do we need clustering scaling factors? How about if we naively select the points based on the order of confidence score?

    3. (Minor) Typo: Right after the equation 7.

    4. (Minor) In terms of writing, in the abstract, the authors address that existing trainig-free methods based on LVMs fail to effectively utilize negative prompts, leading to poor performance on low-contrast medical image. But, the authors demonstrate the proposed method on CT and MRI data that have relatively high contrast images. So, it would be great to demonstrate the effectiveness of the proposed method on low-contrast images like ultrasound images or revise/polish the statement.

  • 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 proposed appoach looks novel. Especially, addressing the importance of both positive and negative point prompts looks interesting.

  • 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 #3

  • Please describe the contribution of the paper

    This paper introduces a novel prompt strategy for few-shot segmentation based on the Segment Anything Model (SAM). The proposed approach combines the strengths of DINOv2 and SAM through an ensemble framework, integrating semantic understanding from DINOv2 with SAM’s precise localization capabilities to generate more reliable confidence maps. Additionally, background voxel value distribution is modeled to sample negative prompts within an optimized confidence interval, ensuring higher specificity for the target segmentation region. Extensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method.

  • 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.

    • Combining semantic understanding from DINOv2 with absolute localization capabilities from SAM is technically sound.

    • The emphasis on high-quality negative prompt sampling holds significance for medical image segmentation, where precise background suppression is critical for reducing false positives in clinical applications.

  • 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.
    • Hyperparameters:
      1. \beta = \alpha - 1.5 needs justification in addition to searching \alpha.
      2. Values of \delta_{S-D}, \delta_{D}, \delta_{S} are absent.
  • 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

    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?

    Soundness of the proposed method and validated effectiveness on different datasets.

  • 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

N/A



Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • 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”.

    All reviewers agree on the practical relevance and technical soundness of the paper. Reviewer #1 highlighted the importance of extracting both positive and high-quality negative prompts, and found the proposed CMSM and PSM modules novel and effective. Reviewer #2 emphasized the technically sound integration of DINOv2 and SAM, and the clinical importance of precise negative prompt sampling to suppress false positives. Reviewer #4 noted that the anatomical-aware negative prompt selection in PSM addresses key limitations of prior work (e.g., PerSAM), and praised the synergy between DINOv2 and SAM in CMSM. The ablation studies and comparisons with both training-free and training-based baselines further support the method’s competitiveness.

    Areas for improvement include expanding ablation studies (e.g., point counts, N-shot settings), clarifying hyperparameters, better distinguishing the fusion strategy from prior work, and including statistical measures.

    Overall, the paper presents a novel and well-validated contribution. I recommend early acceptance and encourage the authors to incorporate reviewer suggestions in the final version.



back to top