Abstract

Due to blurred boundaries between the background and the foreground, along with the overlapping of different tumor lesions, accurate segmentation of brain tumors presents significant challenges. To tackle these issues, we propose a causal intervention model designed for brain tumor segmentation. This model effectively eliminates the influence of irrelevant content on tumor region feature extraction, thereby enhancing segmentation precision. Notably, we adopt a front-door adjustment strategy to mitigate the confounding effects of MRI images on our segmentation outcomes. Our approach specifically targets the removal of background effects and interference in overlapping areas across tumor categories. Comprehensive experiments on the BraTS2020 and BraTS2021 datasets confirm the superior performance of our proposed method, demonstrating its effectiveness in improving accuracy in challenging segmentation scenarios.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2795_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{Liu_Causal_MICCAI2024,
        author = { Liu, Hengxin and Li, Qiang and Nie, Weizhi and Xu, Zibo and Liu, Anan},
        title = { { Causal Intervention for Brain tumor Segmentation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15009},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposed a calsal intervention method to adjust and boost the effect of MRI semantic segmentation.

  • 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.

    A disentangle method is proposed which process segmentation and classification independently.

  • 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.

    The theory is clear and the design is reasonable, but the implementation detail is not clear. For example, there are four modalities but all of them are set as “X” or “C”. Other cases would also influence the performance. The main issue would be the few amount of data samples. In this case, the result would be sensitive to the network architecture and parameters. The architecture of encoder and decoder is missing, which would greatly influence the performance. It is better to put it in the supplementary document.

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

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

    It is able to rebuild.

  • 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 hyperlink is missing. The training epochs can be more. Do you use a end-to-end training or a multi-step training?

  • 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 design is novel and effective. The idea is easy to understand.

  • 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



Review #2

  • Please describe the contribution of the paper

    The paper poses brain tumor segmentation as a causal problem where it introduces an intermediate variable through front-door adjustment to mitigate the confounding effects of MRI images on the segmentation results. The causal method is implemented by the region causality module and the category causality module to remove the effect of background and healthy tissue and interference in overlapping areas of different categories of tumors, respectively. Ablation and comparative studies demonstrate the effectiveness of the proposed modules and method. The method outperforms multiple leading methods on the BraTS benchmark.

  • 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 approach of addressing brain tumor segmentation as a causal problem using two causality modules is novel.

  • 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.

    The experiments section needs to be improved.

  • 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
    • Please provide some background for front-door adjustment strategy.
    • “different tumor categories” is confusing. Maybe “different tumor subcompartments” or “different tissue classes present within the tumor”?
    • Glioma is mentioned only once in the paper in the first sentence and never again. The paper title also says “brain tumor segmentation” and not “glioma segmentation”. If the paper is trying to address the interference of different tumor subclasses with each other, then this is very suitable for gliomas (also evident from the experiment and results furnished in the paper). Is this also relevant for other brain tumors? If not, maybe specify it in the paper that this method is targeted towards gliomas.
    • Why use both BraTS 2020 and 2021 datasets when 2021 is a superset of the 2020 data? Usage of BraTS 2020 dataset seems to be redundant.
    • On what basis did the authors select the different methods for comparison? The BraTS 2021 leaderboard methods were:
      1. https://doi.org/10.1007/978-3-031-09002-8_16
      2. https://doi.org/10.1007/978-3-031-09002-8_4
      3. https://doi.org/10.1007/978-3-031-09002-8_2
      4. https://doi.org/10.1007/978-3-031-09002-8_15 Check the following link for the leaderboard: https://www.rsna.org/rsnai/ai-image-challenge/brain-tumor-ai-challenge-2021
    • Please mention the Dice scores in Figure 3.
    • Minor: Please change the font in Figure 2.
  • 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 method is novel and addresses an important problem.

  • 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 a novel approach using causal inference to enhance brain tumor segmentation. The proposed method incorporates a cascade causal model that employs region and category causality modules to improve segmentation accuracy by addressing background and category interference.

  • 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 generally well-organized and clearly written. Figures and tables are effectively used to illustrate the methodology and results. The method is clearly structured and represents a significant conceptual innovation. However, the explanation of how the causal relationships are explicitly defined and quantified could be improved. For instance, the paper would benefit from a more detailed mathematical justification for the choice of causal paths and a discussion on the assumptions underlying the causal model.

  • 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.

    The paper could benefit from a deeper exploration into the robustness of the model across different settings or MRI machines, which is often a challenge in real-world applications. The paper presents comprehensive experiments on the BraTS2020 and BraTS2021 datasets. It would be advantageous to see additional analysis on the computational efficiency of the method, considering its applicability in clinical settings where real-time processing is often crucial.

  • 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

    The methodological framework is well-constructed with a detailed explanation of how causal inference principles are adapted for medical imaging. It would be interesting to see the algorithms performances against several other complex problems in MRI including Knee MRI. Domain generalization problem in MRI segmentation can also be a good choice to explore to see how changes in MRI sequences and apparatus effect the tumor’s region outcome.

  • 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?

    Paper mathematical formulations are convinving. The methodology itself seems novel and workable for the presented problems in medical image segmentation. Results are limited but still convincing.

  • 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



Review #4

  • Please describe the contribution of the paper

    This paper proposes a causal intervention method for brain tumor segmentation. This method consists of a region causality module that removes the effect of background and healthy tissues on the tumor segmentation masks. It also consists of a category causality module that eliminates the interference of overlapping areas of different tumor categories. The proposed causal intervention method outperforms the SOTA methods on two large scale public brain tumor 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 paper is well written and easy to follow.
    • The introduction of the region causality module is novel as it produces a rough segmentation map and when multiplied by the encoded features, the effect of background and healthy tissues could be removed.
    • The category causality module is also a novel contribution as it solves the hard problem of eliminating the interference between different tumor categories.
    • The results of using the proposed model on two brain tumor datasets is promising.
  • 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.
    • In Equation 2, a more elaborate explanation of the physical significance of “c” is needed.
    • Table 1 and 2 shows that the HD95 for the proposed method is not outperforming the other SOTA methods for ET and WT tumor categories. An explanation is needed for this observation. Does this mean that the proposed category causality module cannot completely eliminate the category interference?
  • 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?

    The authors must publicly release the code after the review period.

  • 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
    • More explanation on the confounder set C and its physical significance should be provided for a better understanding of its relevance for brain tumor segmentation task.
  • 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 strengths of this paper far outweigh its weaknesses. The authors of this paper are solving an important and difficult problem. The proposed method could have good clinical usage in terms of brain tumor segmentation. Hence it is a good contribution to the research 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




Author Feedback

Thank you for the supports and the constructive comments. We will carefully consider them and make improvements in our camera-ready version.




Meta-Review

Meta-review not available, early accepted paper.



back to top