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

Abstract

Many multi-modal tumor segmentation methods have been proposed to localize diseased areas from the brain images, facilitating the intelligence of diagnosis. However, existing studies commonly ignore the relationship between multiple categories in brain tumor segmentation, leading to irrational tumor area distribution in the predictive results. To address this issue, this work proposes a Multi-category Region-guided Graph Reasoning Network, which models the dependency between multiple categories using a Multi-category Interaction Module (TMIM), thus enabling more accurate subregion localization of brain tumors. To improve the recognition of tumors’ blurred boundaries, a Region-guided Reasoning Module is also incorporated into the network, which captures semantic relationships between regions and contours via graph reasoning. In addition, we introduce a shared cross-attention encoder in the feature extraction stage to facilitate the comprehensive utilization of multi-modal information. Experimental results on the BraTS2019 and BraTS2020 datasets demonstrate that our method outperforms the current state-of-the-art methods.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1255_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Li_Multicategory_MICCAI2024,
        author = { Li, Dongzhe and Yang, Baoyao and Zhan, Weide and He, Xiaochen},
        title = { { Multi-category Graph Reasoning for Multi-modal Brain Tumor Segmentation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15008},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    Authors propose to model the dependency between multiple tumour categories to enhance the segmentation of tumour subregions. Also, to further refine tumour contours, their model captures semantic relationships between regions and contours via a graph learning branch to emphasize these relationships. The proposed approach was validated on the BRATS datasets of 2019 and 2020, demonstrating superiority w.r.t. other state of the art methods.

  • 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.
    • Authors present a clear motivation for including the projection to the domain of graphs to learn relations and preserve structure.
    • Technical novelty: The provessive integration of contours and region information in several layers of the methodology to exploit its relationships.
    • Strong evaluation: The comparison with other state-of-the-art approaches and the ablation study allow to assess the strengths of the proposed approach in the tumour segmentation 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.
    • Lack of details in the reprojection of graph features: authors did not mention how they projected the graph features back into image space (final segmentation).
  • 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.

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

    The authors should make the code for their proposed method available to facilitate easier replication of their results.

  • 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
    • It would be pertinent to explain why the fusion of contour and tumour regions information contributes to a decrease in the scores for WT, and to generally highlight the advantages of utilizing graphs neural networks for learning relationships over other representations like covariances.
    • Furthermore, the manuscript would benefit from carrying out comparisons with other models that explicitly exploit relations such as GCNs.
    • Page 8, Table 3 contains a typographical error: “Prposed”
    • The pipeline contains a typographical error: “Cross Attetention”
  • 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 progressive fusion of contours and tumour information seems like a promising way to refine the tumour limits. The results show that the proposed approach is superior to other approaches, and they assesed the contribution of each of the proposed components.

  • 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 proposes a multi-modal and multi-category guided Graph Reasoning Network as a step towards robust contour-aware brain tumor segmentation. The authors conduct experiments on BRATs adult tumor dataset and compare their method against 5 different SOTA methods.

  • 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 paper emphasizes a significant clinical challenge of contour delineation of tumor regions. -Ablation studies as well as solid baseline comparison add to the technical rigor of the paper -The paper is well written, with a brief but concise description of each component proposed by the authors.

  • 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.

    Major comments: -The BRATS 2020 dataset likely contains some new cases and updates compared to the BRATS 2019 release, but they cover overlapping but distinct patient cases and imaging data. Therefore, comparing the methods of both datasets might yield similar results. -Reproducibility concern: the paper might be technically challenging to reproduce due to insufficient details.

  • 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?

    It would be great if the authors clarified more information about training (how long models were trained for, and how many epochs?) Is the method proposed in the paper more computationally heavy than a standard UNet? It would be beneficial if authors considered sharing GitHub repository, although that is not a requirement.

  • 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

    Detailed comments:
    -Please improve the clarity of the methodology section; in particular, please define what BCF and CRF in formulas (5) and (6). -Please clarify the abbreviations used in Figure 1 in figure legend. I suggest removing “A” since figure does not contain subfigures -It is worth noting that the average score of our proposed method is the highest among all methods. -> please correct to “average dice score”
    -Please conduct statistical significance testing to back up the following claim in section 3.2 “Although the Dice score in the WT region slightly trails behind Nestedformer and Eoformer, it notably surpasses other methods by a significant margin.”

    Future directions (NOTE: Authors do not need to address this for the paper revision, use this as suggestions for journal extension):

    • It would be interesting to see how the model will behave on small tumor regions. -Are there any correlations or degradation/improvement of Dice/HD95 that authors observe based on MRI image quality? -What about nnUnet comparison? Its an important SOTA (https://www.nature.com/articles/s41592-020-01008-z)
  • 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?

    Overall, the paper presents a promising approach for robust contour-aware brain tumor segmentation. The technical rigor, including the ablation studies and baseline comparisons, is commendable. With the suggested improvements to address the reproducibility concerns and enhance the clarity of the methodology, I believe the paper can be accepted for publication.

  • Reviewer confidence

    Somewhat confident (2)

  • [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

    This work proposes a brain tumor segmentation method by developing a region-guided graph reasoning network. The method generates two segmentation branches: one focusing on tumor regions and another based on their contours, propagating knowledge from region to contour to further improve the overall segmentation. Both representations of contour and region segmentations are then fed into a graph convolution network attached with Transformer attention.

  • 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 paper is well written.
    • There is a technical novelty including region-based graph reasoning network.
    • Comparison against recent works.
  • 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.
    • Some parts of methodology need more explanations.
    • Incomplete discussion on limitations.
  • 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
    1. Authors are encouraged to include their code implementation and add a link in the manuscript.
    2. In Figure 1, in the Multi-category Region-guided Graph Reasoning block, there are two graph branches. Is the top one exclusively predicting the contours, while the bottom one predicts the tumor regions? If so, how do contour predictions assist in improving overall brain tumor segmentation? More explanations are needed. Additionally, what about propagating knowledge from contour to region? Could this improve the overall segmentation performance?
    3. It is mentioned that “This helps reduce the computational cost.” How does the cross-attention encoder reduce computational cost?
    4. More details about G are needed. Try to expand Eq. (1) and describe it mathematically if possible. The part of the Graph Convolution Network (GCN) lacks details. Also, add the definition of the abbreviation "GCN" on page 5.
    5. In Eq. (4), where is G in Figure 1? It would be helpful if symbols used in this equation were added to the figure.
    6. Is the proposed framework trained end-to-end? Please state that clearly.
    7. What is the difference between L and L' in Eq.(7)? The same clarification is needed for L.
    8. How are the contour ground-truth prepared?
    9. In Figure 2, add prediction results of the highest competitors Nestedformer and Eoformer and include the Dice scores for all predictions superimposed over images.
    10. Sigma2 in Eq. (2) is not defined; probably it is for sigmoid.
    11. In the Abstract, state “Transformer” in the Multi-category Interaction Module (TMIM).
    12. In Figure 1, “Flair” should be “FLAIR” in the input data. Also, caption of Figure 1: Remove “A.”
    13. Table 3: There is a typo. “Prposed” should be “Proposed.”
  • 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?
    • Technical novelty
    • Writing skills and paper presentation.
    • Application.
  • Reviewer confidence

    Very confident (4)

  • [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

N/A



Meta-Review

Meta-review not available, early accepted paper.



back to top