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

Abstract

Pituitary tumors often cause deformation or encapsulation of adjacent vital structures. Anatomical structure segmentation can provide surgeons with early warnings of regions that pose surgical risks, thereby enhancing the safety of pituitary surgery. However, pixel-level annotated video stream datasets for pituitary surgeries are extremely rare. To address this challenge, we introduce a new dataset for Pituitary Anatomy Segmentation (PAS). PAS comprises 7,845 time-coherent images extracted from 120 videos. To mitigate class imbalance, we apply data augmentation techniques that simulate the presence of surgical instruments in the training data. One major challenge in pituitary anatomy segmentation is the inconsistency in feature representation due to occlusions, camera motion, and surgical bleeding. By incorporating a \underline{F}eature \underline{F}usion module, F2PASeg is proposed to refine anatomical structure segmentation by leveraging both high-resolution image features and deep semantic embeddings, enhancing robustness against intraoperative variations. Experimental results demonstrate that F2PASeg consistently segments critical anatomical structures in real time, providing a reliable solution for intraoperative pituitary surgery planning. Code: https://github.com/paulili08/F2PASeg.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1527_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{CheLum_F2PASeg_MICCAI2025,
        author = { Chen, Lumin AND Wu, Zhiying AND Lei, Tianye AND Bai, Xuexue AND Feng, Ming AND Wang, Yuxi AND Meng, Gaofeng AND Lei, Zhen AND Liu, Hongbin},
        title = { { F2PASeg: Feature Fusion for Pituitary Anatomy Segmentation in Endoscopic Surgery } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15968},
        month = {September},
        page = {246 -- 255}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper presents the PAS dataset for pituitary anatomy segmentation, introduces the F2PASeg architecture with a feature fusion module, and proposes a data augmentation method to address class imbalance.

  • 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 PAS dataset is introduced, with 7,845 annotated images showing varied pituitary anatomies in endoscopic surgery. 2) A model called F2PASeg is proposed, using a feature fusion module to improve feature integration. 3) A data augmentation method is used by adding surgical tools to images, helping segment rare anatomical structures more accurately.

  • 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 comparative experiments could be strengthened by including comparisons with state-of-the-art (SOTA) methods in video segmentation, as well as SOTA image segmentation algorithms. Additionally, benchmarking against SAM-related models (e.g., SAM, MedSAM) would provide a more comprehensive evaluation of the proposed method. 2)The role of prompts during the model inference phase is not clearly described. Could the authors clarify the prompt strategy used during inference (e.g., types of prompts, frequency of application across frames) 3)The data augmentation approach for simulating surgical instrument placement may not fully capture the dynamic movement of instruments in real surgical scenarios. Additionally, the rationale for applying augmentation selectively to ICA and OCR (but not OP) is unclear. The authors might elaborate on the reasoning behind the selective application of augmentation to address potential data imbalance. 4)The design of the feature fusion module may lack sufficient novelty, and the integration of LoRA with convolutional layers appears unconventional given its typical use in large-model fine-tuning. 5) The description of the fine-tuning process for SAM is incomplete, particularly regarding which modules are frozen or trainable.

  • 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.

  • 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.

    (2) Reject — should be rejected, independent of rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    I greatly appreciate the authors for introducing a new dataset, but I believe the overall design of the algorithm lacks innovation and contains some design flaws. In addition, the authors should consider expanding the comparative analysis to include SOTA video segmentation methods, recent advances in image segmentation, and SAM-related models to better contextualize the novelty and performance of their approach.

  • 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

    1. A Pituitary Anatomy Segmentation dataset.

    2. F2PASeg: a feature fusion model that adapts SAM2’s mask decoder.

    3. Data Augmentation technique to deal with class imbalance.

  • 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.

    Novel Dataset Contribution: The authors introduce a new dataset containing extensive pixel-level anatomical annotations specific to the sellar phase of pituitary surgery. This dataset fills a critical gap in the field by offering detailed, domain-specific segmentation labels for a complex surgical phase. Moreover, the commitment to publicly release the dataset significantly enhances its impact, promising to benefit both methodological development and clinical translation in surgical scene understanding.

    Additionally, the authors propose two technical contributions:

    1. An adaptation to the SAM2’s mask decoder by incorporating residual blocks.

    2. A data augmentation technique that synthetically inserts annotated surgical instruments from one video onto frames of other videos that contain underrepresented anatomical structures but lack instruments.

    Both of these methods demonstrate increases in performance as demonstrated in the ablation studies.

  • 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 manuscript’s discussion of related datasets is lacking in completeness, particularly regarding prior work on pituitary surgery anatomy. Notably, the dataset used in Sarwin et al. (2023) includes both the nasal and sellar phases and comprises a significantly larger number of labeled images, surgical cases, and anatomical classes. This omission undermines the manuscript’s claims about the novelty and comprehensiveness of their dataset, especially in terms of anatomical variability and dataset scale.

    (Sarwin, G. et al. (2023). Live Image-Based Neurosurgical Guidance and Roadmap Generation Using Unsupervised Embedding. In: Frangi, A., de Bruijne, M., Wassermann, D., Navab, N. (eds) Information Processing in Medical Imaging. IPMI 2023. Lecture Notes in Computer Science, vol 13939. Springer, Cham. https://doi.org/10.1007/978-3-031-34048-2_9)

    • Section 2.2: The method description lacks clarity in distinguishing between components inherited from SAM2 and the novel contributions. The intertwined presentation of both architectures reduces overall clarity and makes it difficult to isolate the actual contributions.

    • Section 3.2: The implementation details are insufficiently described and would benefit from further elaboration.

    Suggestions to improve completeness, accuracy and clarity:

    • Incorporate the missing dataset reference and revise claims related to dataset novelty and anatomical diversity.

    • Clarify Section 2.2 by clearly separating inherited components from proposed contributions.

    • Expand implementation details in Section 3.2 to enhance transparency and reproducibility.

  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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

    Introduction, paragraph 3: The beginning of the sentence “U-Net [16] is weakly ….” is missing. Further down, “It has largely achieved the end-to-end efficient segmentation required for intraoperative endoscopy, the current segmentation methods have not fully investigated feature fusion.” also needs reformulation.

  • 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 effort of creating this dataset and its release is highly valuable and is expected to benefit the community. However, the paper would benefit from revisions to improve clarity, and completeness.

  • 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.

    I would like to thank the authors for their responses to the reviewers’ concerns. With the anticipated improvements in clarity and completeness, the overall work—especially the dataset—will make a valuable contribution to the community.



Review #3

  • Please describe the contribution of the paper

    This work proposes a dataset for pituitary anatomy segmentation, as well as a method for real-time segmentation using a feature fusion module. This module enhances the mask decoder of SAM2 by introducing residual blocks when fusing features at different strides as opposed to directly adding the high-dimensional features to the image embeddings.

  • 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.
    • Both quantitative and qualitative results are convincing
    • Comprehensive ablations, showing effectiveness of proposed components
    • Real-time performance in comparison to SAM-Med 2D
  • 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.
    • Incremental algorithmic contribution
  • 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.

  • 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.

    (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?

    The quantitative and qualitative results are convincing, showing that the feature fusion more accurately segments the proper anatomical regions which is important for small and complex structures in pituitary surgery. Considering the critical anatomy present, the improved precision is essential for effective surgical planning and navigation, even if the algorithmic modifications are modest.

    The ablation studies are appreciated to show the efficacy of both the feature fusion module and data augmentation for this segmentation task. The ablation of data augmentation is further important given the natural class imbalance present in the dataset. The presented dataset is more extensive than other publicly available datasets of the same application in both images and cases.

    While the technical novelty is incremental, the performance improvement for all structures over baselines and the expected release of the dataset (as stated) justify the value of 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.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    I appreciate the clarifications provided in the rebuttal. While the technical novelty remains limited, the empirical results appear sound, and the dataset is well-curated and valuable to the community. I continue to view this as a worthwhile contribution, particularly given the dataset’s scope and the practical relevance of segmentation especially in real-time.



Author Feedback

We thank the reviewers for their detailed feedback and positive comments regarding novel dataset contribution (R1, R2&R3), methods (R1&R2) and convincing results (R3). We address the major concerns point-by-point. Q: Novel contributions based on SAM2 (R1&R2&R3). A: F2PASeg stands out in pituitary anatomy segmentation to address both speed and high performance, crucial for clinical use. (1) Model structure: F2PASeg integrates a feature fusion module and LoRA with convolutional layers for efficient fine-tuning. (2) Training Method: We propose an augmentation method that inserts annotated instruments into anatomy-rich, instrument-sparse frames to simulate clinical scenarios and balance class distribution. (3) Dataset: We introduce the first pixel-level annotated dataset for the anatomically complex, clinically important sellar phase, publicly released for reproducibility. Validated on extensive clinical data, F2PASeg shows substantial improvements over existing methods, underscoring its potential for clinical applications. Q: Algorithm comparison (R1). A: Thanks. We have added additional experiments comparing our method with video segmentation algorithm LiVOS (CVPR2025), and SAM, MedSAM. The results are as follows: LiVOS/SAM/MedSAM/Ours: mIoU=0.6226/0.6415/0.6829/0.7796, mDice=0.7523/0.7707/0.8032/0.8635. Our method achieves superior mIoU and mDice. The comparsion results will be added in the revised vision. Q: Prompt strategy (R1). A: Sparse prompts are derived from bounding boxes of structure masks computed every 10 frames. Final scores are averaged across frames. Q: Data augmentation details (R1). A: We agree that ICA, OCR, and OP are underrepresented. In the previous vision, we augmented the dataset with samples containing all three classes, which will be clarified in the revised vision. Given the critical role of ICA in intraoperative decision-making and surgical safety, we prioritized targeted augmentation for ICA. The frames containing both ICA and OCR occur more frequently than those containing ICA and OP, making them suitable for balancing class distribution while preserving realism. Q: Feature Fusion (R1). A: Compared to SAM-based models, SAM2 adds memory components, increasing the encoder-decoder gap and requiring richer ViT features. In our F2PASeg, the design of the feature fusion module is simple and effective to solve this problem. Q: LoRA module function (R1). A: The integration of LoRA with convolutional layers enables efficient fine-tuning with fewer parameters. Without LoRA/With LoRA: Parameters= 39.0M/34.8M, FPS=22.82/28.57. With LoRA, the model better satisfies intraoperative real-time segmentation demands, achieving higher FPS and reduced parameters. Q: The frozen and trainable modules (R1). A: The mask decoder is frozen (due to the use of LoRA), while all other components remain trainable. Total/trainable parameters: 39.0M/34.8M. Q: Dataset comparison (R2). A: The dataset in Sarwin et al. (2023) includes 166 videos with bounding-box annotations across 16 classes, targeting object detection. In contrast, our dataset provides pixel-level annotations for semantic segmentation, which are significantly more labor-intensive. For pituitary anatomy datasets with pixel-level annotations, Nasal phase [17]/Sellar phase [8]/Ours: videos= 23/64/121; classes=3/10/6; Images=549/635/7845. Our dataset offers a substantial advantage in terms of scale and annotation quality. Second, publicly available anatomical datasets for pituitary surgeries remain extremely limited. Sarwin et al.’s dataset and code are unavailable, while ours will be fully open-sourced. The reference will be included in Tab. 1 for comparison. Q: The implementation details (R2). A: Optimization uses the AdamW optimizer (beta1=0.9, beta2=0.999). Q: Writing (R2). A: We will carefully revise the Introduction for clarity and readability. At last, we sincerely thank R3 for the positive feedback on our results, ablations, and real-time performance.



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’

    This paper introduces PAS, a large‐scale pixel‐level pituitary anatomy segmentation dataset, and F2PASeg, a real‐time feature‐fusion model built on SAM2 for intraoperative endoscopy. Although the technical novelty is merited, all of the reviewers highlight the value of the well‐curated dataset and empirical results, and substantive critiques have been addressed. Therefore, I would like to accept the paper.



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



back to top