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

Abstract

Regular screening and early discovery of uterine fibroid are crucial for preventing potential malignant transformations and ensuring timely, life-saving interventions. To this end, we collect and annotate the first ultrasound video dataset with 100 videos for uterine fibroid segmentation (UFUV). We also present Local-Global Reciprocal Network (LGRNet) to efficiently and effectively propagate the long-term temporal context which is crucial to help distinguish between uninformative noisy surrounding tissues and target lesion regions.
Specifically, the Cyclic Neighborhood Propagation (CNP) is introduced to propagate the inter-frame local temporal context in a cyclic manner. Moreover, to aggregate global temporal context, we first condense each frame into a set of frame bottleneck queries and devise Hilbert Selective Scan (HilbertSS) to both efficiently path connect each frame and preserve the locality bias. A distribute layer is then utilized to disseminate back the global context for reciprocal refinement. Extensive experiments on UFUV and three public Video Polyp Segmentation (VPS) datasets demonstrate consistent improvements compared to state-of-the-art segmentation methods, indicating the effectiveness and versatility of LGRNet. Code, checkpoints, and dataset are available at https://github.com/bio-mlhui/LGRNet

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0813_paper.pdf

SharedIt Link: https://rdcu.be/dY6jH

SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72083-3_62

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0813_supp.pdf

Link to the Code Repository

https://github.com/bio-mlhui/LGRNet

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Xu_LGRNet_MICCAI2024,
        author = { Xu, Huihui and Yang, Yijun and Aviles-Rivero, Angelica I. and Yang, Guang and Qin, Jing and Zhu, Lei},
        title = { { LGRNet: Local-Global Reciprocal Network for Uterine Fibroid Segmentation in Ultrasound Videos } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15004},
        month = {October},
        page = {667 -- 677}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This submission presents an ultrasound video uterine fibroid segmentation (UFUV) dataset, and a uterine fibroid segmentation method, Local-Global Reciprocal Net (LGRNet). The collected UFUV dataset includes 100 videos and each video contains 50 frames, and the annotation was performed by two experienced sonographers. The major components of the proposed LGRNet are Local Cyclic Neighborhood Propagation (CNP) and Global Hilbert Selective Scan (HilbertSS). Quantitative comparisons are provided to validate the proposed approach on the UFUV dataset and three video polyp segmentation datasets.

  • Please list the main strengths of the paper; you should write about 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 collected and annotated UFUV dataset is useful to facilitate the research in ultrasound video uterine fibroid segmentation, which may help the screening and early discovery of uterine fibroid.

    • The proposed LGRNet demonstrates improved performance on the UFUV dataset and three polyp segmentation datasets compared to the existing methods that are included in the comparisons (Tables 1-3).

    • The paper is well written and easy to follow.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    • It is stated that Hilbert curve (dilation factor is 6) better preserves the 2D locality structure compared to zigzag curve (dilation factor is much larger). However, Table 5 shows that the differences between using Hilbert scan and zigzag scan are relatively small. This evidence does not support that the performance of using Hilbert curve is significantly better than that of using zigzag scan.

    • Fully connected inter-frame dependencies are mentioned as an option that may lead to worse performance and increased computation (the end of section 2.1). The increased computation is reasonable, however, leading to worse performance is not clearly discussed or validated.

  • Please rate the clarity and organization of this paper

    Very 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    The paper proposes a new dataset (UFUV) for ultrasound video uterine fibroid segmentation and a new approach for uterine fibroid segmentation. The dataset is useful to promote the research in developing ultrasound video uterine fibroid segmentation algorithms. The proposed approach (LGRNet) is validated on the UFUV dataset and three other datasets, where the proposed LGRNet demonstrates improved performance compared to the existing methods.

    The following issues are expected to be addressed in the rebuttal.

    • Add a discussion or evaluation regarding the advantage of using Hilbert scan compared to zigzag scan.
    • Discuss or provide evidence to support the claim that using fully connected inter-frame dependencies cause worse performance.

    Other comment:

    • Table 2 and 3. Tables 2 and 3 should be used.
  • 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

    Weak Accept — could be accepted, dependent on rebuttal (4)

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

    The current rating is based on the following major factors:

    • The proposed dataset (UFUV) is useful for ultrasound video uterine fibroid research.
    • The proposed approach (LGRNet) demonstrates improved performance compared to existing methods.
    • The advantage of using the proposed Hilbert scan is not fully validated.
    • No evidence to support the statement that fully connected inter-frame dependencies cause worse performance.
  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    The paper presents a dataset of ultrasound videos for uterine fibroids. It also presents a new network architecture for uterine fibroid segmentation. The experimental results demonstrate the effectiveness of the proposed model.

  • Please list the main strengths of the paper; you should write about 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 experimental results seem strong and the ablation studies are comprehensive.

    Several state of the art components are integrated very well to solve a very challenging problem. CNP uses neighborhood attention mechanisms to be efficient and effective; using the Hilbert curve to preserve the 2D locality structures seems novel and effective.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Detailed comparisons and error analyses between the proposed method and MS_TFAL [5] would be insightful.

    Giving intuitions underlying the choices of the proposed system would be valuable.

  • Please rate the clarity and organization of this paper

    Very 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Right after Section 2, it would be better to explain how the common dimension is produced later as the features at each scale have the same number of channels.

    It would be better to explain the variables after equation (4) is defined rather than before they are introduced.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

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

    The experimental results are strong and convincing. The method integrates several state of the models to develop an effective solution for a challenging problem. The dataset is valuable to the community.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #3

  • Please describe the contribution of the paper

    The paper introduces the Local-Global Reciprocal Network (LGRNet) that innovatively combines Local Cyclic Neighborhood Propagation (CNP) with Global Hilbert Selective Scan (HilbertSS). It addresses uterine fibroid segmentation in ultrasound videos with a specific focus on managing the complexities of local and global temporal contexts.

  • Please list the main strengths of the paper; you should write about 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.

    • Solid Experiments and Ablation Study: The paper offers detailed experimental validation and an ablation study that highlight the method’s superior performance over current benchmarks, presenting statistically significant results and significant advancements in medical image segmentation.

    • The methodology, incorporating Local Cyclic Neighborhood Propagation (CNP) and Global Hilbert Selective Scan (HilbertSS), is expertly crafted for segmenting uterine fibroids in ultrasound videos. The authors effectively utilize rigorous experimental and theoretical analysis to validate their approach. This demonstrates profound modeling expertise and adeptly addresses the complexities of the task.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Limited Dataset Size: The primary limitation of this study is the dataset size. With only 100 videos and 5,000 annotations, there are concerns about the robustness and generalizability of the proposed method. The dataset may not adequately represent the variability seen in broader clinical settings. Testing the method on additional, larger datasets would strengthen the claims of efficacy and adaptability.

  • Please rate the clarity and organization of this paper

    Very 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Expanding the dataset or applying the method to a larger dataset would enhance the robustness and generalizability of the findings. It’s a great work.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

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

    The method is well designed

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Author Feedback

Dear Riviewer#1,

  1. Our main intuition is to efficiently aggregate context along temporal axis in a local-global manner. The CNP module is token-level (local) operation, while the HilbertSS module is sequence-level(global) operation. Unlike existing Mamba-based models, HilbertSS scans the highly semantic queries, instead of feature maps, which enables efficient temporal propagation.
  2. A non-biased linear layer with GroupNorm (number of groups=32) is used for each scale to transform the backbone features into the common dimension.
  3. We will put the symbol illustration after Eq.4. Moreover, a detailed S6 block illustration is included in our arxiv version.

Dear Reviewer#2,

  1. We may expand our dataset size in our journal extension.
  2. LGRNet achieves SOTA performance on SUN-SEG, which is the existing largest Polyp Segmentation dataset, which validates the generalizability and versatility of our design. We may also include experiments on other surgical (Endo tool segmentation) settings and other semantic segmentation datasets in journal extension.

Dear Reviewer#3,

  1. Table 5 shows that Hilbert scan achieves 65.8/77.5 IoU/Dice score, which is bigger than Zigzag scan (63.9/76.1). Moreover, its S-Measure is also better, which shows Hilbert scan better preserve the region-aware and object-aware structural similarity.
  2. Since CNP is attention-based, CNP is token-level operation. For a query token at one frame, if we build fully-connected-inter-frame-dependencies, its key tokens will include neighborhood distant-frame-tokens. If the motion is severe, these distance-frame-tokens probably are background tokens, which has weak semantics. Therefore, from the view of connectionism, we think fully-connection is not necessary.



Meta-Review

Meta-review not available, early accepted paper.



back to top