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

Abstract

Brain structure-function interaction is crucial for cognition and brain disorder analysis, and it is inherently more complex than a simple region-to-region coupling. It exhibits homogeneity at the modular level, with regions of interest (ROIs) within the same module showing more similar neural mechanisms than those across modules. Leveraging modular-level guidance to capture complex structure-function interactions is essential, but such studies are still scarce. Therefore, we propose an interpretable modularity-guided graph convolution network (IMG-GCN) to extract the structure-function interactions across ROIs and highlight the most discriminative interactions relevant to fluid cognition and Parkinson’s disease (PD). Specifically, we design a modularity-guided interactive network that defines modularity-specific convolution operation to learn interactions between structural and functional ROIs according to modular homogeneity. Then, an MLP-based attention model is introduced to identify the most contributed interactions. The interactions are inserted as edges linking structural and functional ROIs to construct a unified combined graph, and GCN is applied for final tasks. Experiments on HCP and PPMI datasets indicate that our proposed method outperforms state-of-the-art multi-model methods in fluid cognition prediction and PD classification. The attention maps reveal that the frontoparietal and default mode structures interacting with visual function are discriminative for fluid cognition, while the subcortical structures interacting with widespread functional modules are associated with PD.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2884_paper.pdf

SharedIt Link: https://rdcu.be/dV545

SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72117-5_44

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2884_supp.pdf

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Xia_IMGGCN_MICCAI2024,
        author = { Xia, Jing and Chan, Yi Hao and Girish, Deepank and Rajapakse, Jagath C.},
        title = { { IMG-GCN: Interpretable Modularity-Guided Structure-Function Interactions Learning for Brain Cognition and Disorder Analysis } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15010},
        month = {October},
        page = {470 -- 480}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The contributions of this manuscript are commendable.

    1. The authors meticulously engineer an analytical synergy between structural and functional imaging modalities within the human cerebral context.

    2. The authors architect and pioneer an innovative and interpretable modularity-guided graph convolutional network (IMG-GCN), which demonstrates marked efficacy. The peer validation of the IMG-GCN is exemplary.

    3. The authors substantiate the robustness of IMG-GCN by benchmarking it against six peer classifiers, showcasing its superior performance.

  • 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. It represents a significant innovation to integrate analytics derived from both structural and functional imaging modalities for the exploration of neurological disorders.

    2. The efficacy of the proposed IMG-GCN model is demonstrably substantial.

    3. The empirical experimentation and subsequent validation of the model are outstanding, reflecting rigorous scientific inquiry and robust results.

  • 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.

    1. The authors are encouraged to elucidate further upon the rationale for their choice of activation functions within the depicted framework in Figure 1, where both ReLU and Sigmoid functions are implemented. Could the authors elaborate on the efficacy of these functions specifically in the context of graph-based learning? Additionally, there is a strong recommendation for the authors to explore the development of novel activation functions to enhance the architectural innovation of their model.

    2. Given the complexity inherent in the IMG-GCN model, it is advisable for the authors to consider the integration of more sophisticated optimizers, such as STORM and STORM+. It is anticipated that the authors will provide metrics on reconstruction accuracy to substantiate the performance and validate the capabilities of the proposed IMG-GCN.

    3. Concerning the methodology of cross-validation, the authors mention the utilization of a 5-fold cross-validation approach yet fail to furnish comprehensive details on the strategies employed to mitigate overfitting. It is imperative that such details be thoroughly articulated to affirm the robustness of the validation process.

  • 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?

    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.Elaborate with greater depth and specificity regarding the selection criteria and underlying rationale for the choice of activation functions utilized within the model. A detailed explanation of the effectiveness of these functions in the specific context of graph-based learning is requested.

    1. Furnish additional empirical studies, such as convergence curves and reconstruction accuracy metrics, to substantiate the effectiveness of the Adam optimizer within the context of your model. This data is essential to validate the optimizer’s performance in achieving reliable and efficient learning outcomes.

    2. Expand the presentation of results from the 5-fold cross-validation process to comprehensively demonstrate the absence of overfitting within the model. Detailed statistical analysis and comparative results should be included to robustly support the model’s generalizability and validity.

  • 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 work lacks detailed exposition on the implementation of the 5-fold cross-validation technique and the specific measures undertaken to prevent overfitting within the model.

  • 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 introduces an interpretable modularity-guided graph convolution network (IMG-GCN) to extract the structure-function interactions across ROIs and highlight the most discriminative interactions relevant to fluid cognition and Parkinson’s disease. The authors also integrated an MLP-based attention model to identify the most contributed interactions. Presented experiments on publicly available HCP and PPMI datasets indicate that proposed method outperforms state-of-the-art multi-model methods in fluid cognition prediction and Parkinson’s disease classification.

  • 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 authors propose their own model to solve tasks of fluid cognition prediction and Parkinson’s disease classification.
    • The mentioned model outperforms SOTA methods in both tasks.
    • The authors provide a comparison to other SOTA methods.
    • Provided research seems to be reproducible, mainly due to publicly available datasets and a detailed description of the proposed method.
    • They also spent some time discussing and interpreting the results.
    • I appreciate the use of the chord diagram (Fig. 4).
  • 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 did not encounter any serious shortcomings. As the authors mention as future work, the work could be extended with more datasets. I am missing some note (at least) on the generalization capability (could be demonstrated by experiments on more different datasets).

  • 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?

    The authors provided references to both publicly available datasets. Presented model is described in detail, so the experiments should be reproducible. The authors do not mention open access to source code.

  • 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

    In section 3 Results, the first part “Competing Methods and Hyperparameter Setting” corresponds more to some other part (Experiments) than the results. I might rename this section to “Experiments and Results”, because even the “Ablation Study” paragraph corresponds more to experiments.

  • 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?

    I find the presented research very clear - the data, method and experiments are well described. The novelty of the presented research is supported by comparison with SOTA methods. I would appreciate the publication of the code to simplify the reproducibility of the research and proof of the generalization capability of the proposed method (at least it could be discussed).

  • 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 proposes novel method to understanding the interaction between brain structure and function. Conventional methods fall short in various ways, such as employing simple coupling to regions to neglecting the interaction between the two when learning from features of both. The paper leverages both the interaction between structure and function as well as the modular system of brain organization. The paper also demonstrates the model is interpretable and biological meaningful.

  • 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.

    I believe the primary strength of the paper lies in strong evaluation. The experiments were thorough with statistical significance for the comparisons of performances of different models as well as thorough ablation study that demonstrates the contribution of each component of their model. The baseline models used to compare the proposed method against were properly described and tuned to fit the applications, with sufficient details of the parameter tuning. Figure 4 also showed evidence that the results were biologically meaningful. I also appreciated that the authors cite specific references regarding some of the design choices/parameters they chose (e.g. 10% cutoff for highest edges in adjacency matrix), that could have been arbitrary without such references.

  • 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 don’t find any particular weakness with the paper.

  • Please rate the clarity and organization of this paper

    Excellent

  • 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?

    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

    Great work and I honestly cannot find much to improve upon. I do find the methods section, particularly relating to the graph representation and the following sections on how the network is constructed to combine the structural and functional connectivity somewhat hard to follow. I realize it likely would not have been possible with page limits on MICCAI submission, but if you are planning similar/future works on longer submissions, I think I would have benefitted a lot from graphical abstract or figure to facilitate the reader to better graph the concepts.

  • 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 experiments and results were solid, demonstrating statistical significance in the performance of the model as well as providing ablation study that showcase the effectiveness of both interactive network and attention model. Coupled with interpretable results that are biologically reasonable, I believe the authors have demonstrated thoroughly the effectiveness of their proposed model.

  • 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

To Reviewer #1: a) We used ReLU and sigmoid as activation functions of the neurons in the model. The ReLU function designed in the modularity-guided interactive network facilitates faster learning and helps avoid the vanishing gradient problem. We also attempted to use the leaky ReLU function with a leak rate of 0.33 as a replacement for ReLU. However, this led to a significant decrease in performance. Our choice of the sigmoid function in the MLP-based attention module is intended to softly map the attention weights to the range (0, 1), with higher contributions above 0.5 and lower contributions below 0.5.

We are able to furnish additional empirical studies, such as convergence curves and reconstruction accuracy metrics, to substantiate the effectiveness of the Adam optimizer within the context of your model. As the reviewer’s suggestied, we will provide the training and testing curves in the supplementary materials.

Due to the page limit, we omitted the details of cross-validation. We conducted 5-fold cross-validation to validate model performance. We ran 10 runs of cross-validation and reported average and standard deviation of cross-validation accuracies. To avoid overfitting, we used dropout layers after the readout layer and between two fully connected layers in the output layer. Moreover, we adopted L2 regularization with a weight setting of 0.001 and utilized early stopping to prevent overfitting. We will add this information in the final version.

To Reviewer #2: Our model’s generalization capability was demonstrated on both HCP and PPMI datasets. As the reviewer suggested, we plan to apply the model to ABIDE 2 and COBRE datasets for autism disorder and schizophrenia classifications in future work.

a) Rename the results section We will rename it as ‘Experiments and Results’ in the final version according to the suggestion.



Meta-Review

Meta-review not available, early accepted paper.



back to top