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

Abstract

Structural Magnetic Resonance Imaging (sMRI) is a non-invasive technique to get a snapshot of the brain for diagnosing Alzheimer’s disease. Existing works have used 3D brain images to train deep learning models for automated diagnosis, but these models are prone to exploit shortcut patterns that might not have clinical relevance. We propose an Anatomy-Aware Gating Network (AAGN) which explicitly extracts features from various anatomical regions using an anatomy-aware squeeze-and-excite operation. By conditioning on the anatomy-aware features, AAGN dynamically selects the regions where atrophy is most discriminative. Once trained, we can interpret the regions selected by AAGN as explicit explanations for a given prediction. Our experiments show that AAGN selects regions well-aligned with medical literature and outperforms various convolutional and attention architectures. The code is available at \url{https://github.com/hongcha0/aagn}.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2553_paper.pdf

SharedIt Link: https://rdcu.be/dV168

SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72086-4_9

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2553_supp.zip

Link to the Code Repository

https://github.com/hongcha0/aagn

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Jia_AnatomyAware_MICCAI2024,
        author = { Jiang, Hongchao and Miao, Chunyan},
        title = { { Anatomy-Aware Gating Network for Explainable Alzheimer’s Disease Diagnosis } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15005},
        month = {October},
        page = {90 -- 100}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper first points out a significant challenge in DL-based AD diagnosis: the absence of embedded relevant medical knowledge, which carries a risk of exploiting trivial or shortcut patterns for decision-making. To tackle this challenge, it proposes the anatomy-aware gating network (AAGN) that leverages anatomy information in a differentiable manner to simultaneously select ROIs and extract features in an end-to-end manner. AAGN not only enhances the model’s interpretability but also boosts its decision-making capabilities.

  • 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.

    This paper identifies a significant challenge in DL-based AD diagnosis: the lack of model interpretability. It introduces AAGN as a solution, which automatically selects discriminative ROIs in a differentiable manner for different disease stages. This topic is both interesting and critically important for clinical applications.

  • 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 framework of the proposed AAGN, as shown in Fig. 1, is not well organized, which obscures the specific implementations of its components. Additionally, the rationale for utilizing the Bernoulli distribution for region selection is not explained, which weakens its justification. In the experimental section, details about the number of T1-weighted scans and their pre-processing steps are missing. Moreover, how the authors achieve the results of the comparison methods are unclear. It would be better to implement all comparison methods on the same dataset and under the same experimental conditions such as data-set splitting and 5-fold CV.

  • 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 provide sufficient information for reproducibility.

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

    It would be beneficial to make the codes publicly avaliable

  • 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 topic is both intersting and critical important for clinical application. However, the method description lacks clarity, and the novelty and superiority of the approach in enhancing model interpretability require further improvement. For additional comments, please refer to “the main weaknesses of the paper”

  • 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 Reject — could be rejected, dependent on rebuttal (3)

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

    The details of the proposed AAGN’s implementation are not adequately described, and its motivation and nolvety in enhancing model interpretability are insufficiently explained.

  • 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

    This manuscript presents an anatomy-aware gating network that extracts features from brain ROIs using an anatomy-aware squeeze and excite operation. This method adopts the multiple-instance learning learning scheme to classify the set of ROI features to provide the final disease diagnosis. Further, it uses Bernoulli distribution to independently model the importance of each ROI. The authors only used single modality (sMRI) from ADNI dataset.

  • 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.
    • Simple methodology and easy to follow.
    • The source code is given in the supplementary material with other details.
  • 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.
    • Novelty of the proposed methodology is limited as it is based on the well-known SENet [16] and the ROI features.
    • The results are limited to only 2 tasks and 2 evaluation metrics.
    • No in-depth discussion including limitations and future works.
    • No major improvement in performance.
  • 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 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?

    The authors already provided the source code and dataset information in the supplementary materials.

  • 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 is better to briefly add some details about the data and the performance in the Abstract.
    • The major contributions of the manuscript are not mentioned.
    • In page 2, you cannot start a new sentence with only the reference number “[27]”!
    • It is not clear what the ROI features represent, do they represent the GM mask volume of the ROIs?
    • pMCI is not short of prodromal mild cognitive impairment, it is “progressive”.
    • The performance comparison in the first half of Table 1 is not fair as these methods did not use the same dataset as the author. In addition, some of the compared methods are too old.
    • The results are limited to only 2 tasks and 2 evaluation metrics.
    • Generally, I cannot see a major improvement in the proposed method over the SOTA. So many methods performed much better on AD vs. NC and the pMCI vs. sMCI tasks.
    • The cited number of references is too many for a conference paper, especially many of them are relatively old. Writing
    • There is no need to capitalize every character when defining new abbreviations.
    • There are some writing typos. For instances, “a MLP”, “a ROI”.
    • All equations are not punctuated.
  • 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?

    Please se my detailed comments to authors.

  • 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
    • Very important topic about guiding the model to areas that have clinical based featuewa for AD diagnosis.
    • The work supports the area of explainability in healthcare.
  • 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.
    • Paper is well structured and have good-quality figures.
    • Very good ablation study and comparision with previous work
    • Well-written introduction in terms of mentioning and comparing the well-known preprocessing pipelines like VBM and ROI.
  • 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.
    • You can test your method using more than one dataset usually one dataset is not enough. I understand the effort you have to do for a single dataset, but in the future consider more than one dataset and gove information about the size of the collected dataset and the size of each class.
  • 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

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

    Providing code is always good for reproducibility

  • 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
    • You should try your best to make the figure on the same page where it is referred to.
    • Improve the quality of table 2 and table 3
  • 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 is clear and well written. Extensive ablation study is provided. The contribution is novel and clear.

  • 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 proposed an anatomy-aware gating network for AD diagnosis and prognosis by introducing brain atlas-based prior knowledge and MIL strategies. The experimental results are promising, and the corresponding analysis shows that the proposed work is effective.

  • 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.
    1. Clear organization. The introduction and methodological description are clear.
    2. Novel technologies. The proposed method considers the atlas-based prior clinical knowledge, and the combination with the MIL idea is novel.
    3. Excellent diagnosis and prognosis experimental performance.
    4. Excellent explainable atlas-level results and analysis.
  • 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.

    I mainly appreciate this paper, but the information leverage problem exists for different brain regions, and the description could be clearer for some minor but important points.

  • 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?

    no

  • 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. Incomplete uncoupling region features. Unfortunately, the extracted different regions’ underlying representations of f_{\theta}_{CNN} are not ``absolutely’’ uncoupling due to the overlap between different receptive fields of the convolutional network. On the other hand, the extracted single specific region’s feature also contains information from other regions, or information leverage exists for different regions’ extracted features.

    Although I understand this may be due to limited computational resources that use the altas mask-based feature-level extraction'' rather thanthe altas mask-based image-level feature extraction, ‘’ the authors should acknowledge this limitation in the final paper. Moreover, the recent work [doi: 10.1016/j.media.2023.103032] proposed slice-aware MIL strategies that use a shared 2D extractor’’ toindividually’’ extract different slices’ features in the initial stage to explain what slices are more important; perhaps these individual extraction operations could give the authors insights into this problem.

    1. Some description needs to be clarified. When I first read this paper, I felt confused about how to match the brain atlas size to the feature size (simple downsize?), how the resized altas mask reflects the spatial information (simple flatten?), and the concrete generation of the context vector’’. The authors add moretext-wise’’ descriptions for these minor comments.
  • 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

    Strong Accept — must be accepted due to excellence (6)

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

    The clinical knoweldge-based AI method is novel. The experimental results are promising. The experimental analysis is sufficient and 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



Author Feedback

N/A



Meta-Review

Meta-review not available, early accepted paper.



back to top