List of Papers Browse by Subject Areas Author List
Abstract
Dynamic functional connections (dFCs) have been widely used for the diagnosis of brain diseases. However, current dynamic brain network analysis methods ignore the fuzzy information of the brain network and the uncertainty arising from the inconsistent data quality of different windows, providing unreliable integration for multiple windows. In this paper, we propose a dynamic brain network analysis method based on quality-aware fuzzy min-max neural networks (QFMMNet). The individual window of dFCs is treated as a view, and we define three convolution filters to extract features from the brain network under the multi-view learning framework, thereby obtaining multi-view evidence for dFCs. We design multi-view fuzzy min-max neural networks (MFMM) based on fuzzy sets to deal with the fuzzy information of the brain network, which takes evidence as input patterns to generate hyperboxes and serves as the classification layer of each view. A quality-aware ensemble module is introduced to deal with uncertainty, which employs D-S theory to directly model the uncertainty and evaluate the dynamic quality-aware weighting of each view. Experiments on two real schizophrenia datasets demonstrate the effectiveness and advantages of our proposed method. Our codes are available at https://github.com/scurrytao/QFMMNet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2558_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2558_supp.pdf
Link to the Code Repository
https://github.com/scurrytao/QFMMNet
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Hou_QualityAware_MICCAI2024,
author = { Hou, Tao and Huang, Jiashuang and Jiang, Shu and Ding, Weiping},
title = { { Quality-Aware Fuzzy Min-Max Neural Networks for Dynamic Brain Network Analysis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposed a multi-view fuzzy min-max neural networks based on fuzzy sets for dynamic functional connections. Specifically, a quality-aware ensemble module was conducted to solve with uncertainty with D-S theory. Then, a weighted fusion scheme was used to multi-view fusion in model classification. The experiments tested the performance of the proposed method.
- 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 whole layout describes well.
- The procedure of the proposed method is mostly clear.
- The experimental results seem to be good.
- 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 motivations are confusing and insufficient, for example, what is fuzzy information in brain network, and the uncertainty in inconsistent data quality in the introduction.
- Can the order of the three filters be changed, and can experiments be carried out to verify the effect of different orders?
- What is the meaning of (c_1^v,c_2^v,…,c_K^v) and whether is can directly be used to classification task.
- What is the connection between (c_1^v,c_2^v,…,c_K^v) and (e_1^v,e_2^v,…,e_K^v), and how to ensure the credibility of (e_1^v,e_2^v,…,e_K^v)?
- It seems that the proposed method is used for multi-class classification, but the related experiments cannot be found.
- Which combinations of views, and how many views and how to divide the sliding window are conducive to the classification task, that is missing in the experiments.
- The ablation study is missing.
- 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 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
Please refer to the weaknesses.
- 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?
To sum up, the motivations and the experiments are insufficient to support the proposed method. Besides, although the process of the proposed method introduces are not bad, some missing priorities lead to irrationality.
- 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 combination of fuzzy min-max neural networks tailored for handling multi-view data along with ensembling approach to aggregate class predictions accounting for their uncertainty. Resulting method is applied to the dynamic brain network classification on two schizophrenia 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 proposed method shows improved classification results over a comprehensive list of alternative methods.
- 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.
- a related work section is missing, no clear positioning of proposed MFMM w.r.t. competitor methods
- while combined MFMM and QA-EM improves the classification results over competitor methods, MFMM performance is worse than of best competitor methods,
- therefore, it is unclear whether QA-EM would improve results of other methods, if combined with them, and whether the results would be better than of MFMM+QA-EM
- 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 details of optimization are not provided: optimization method, its hyperparameters, a number of trainings of each model. Would authors provide the code of a model? While throughly described a reproducibility is limited due to the niche subject of fuzzy min-max networks-based multi-view learning.
- 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
- could authors provide clear positioning their variant of fuzzy min-max neural networks w.r.t. competitor methods?
- could authors provide results for plain and simple baselines to aggregating predictions over views?
- could authors provide an ablation study for quality-aware ensemble module, combining it with competitor methods, if applicable?
- could authors provide the results on threshold-independent metrics such as ROC AUC, PR AUC or average precision?
- 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 see the possibility to increase rating to 5) Accept, if the comments to authors would be addressed.
- 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
Accept — should be accepted, independent of rebuttal (5)
- [Post rebuttal] Please justify your decision
The authors addressed my key question about simple baselines for information aggregation. The numbers they provide show the benefit of the quality-aware agrregation (QAE) module, which I see as the most important paper contribution.
Review #3
- Please describe the contribution of the paper
This paper employs a multi-view and quality evaluation analysis method for dynamic functional connectivity (DFC), validated using schizophrenia datasets. By integrating multi-view and quality evaluation approaches, the method enables the exploration of discriminative connections, thereby facilitating the detection of schizophrenia.
- 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 concept of multi-view analysis can provide additional information to enhance the diagnostic model. The quality control method helps the model to select the most discriminative information within the multi-view space. Despite the initial skepticism, this approach proves to be effective in selecting the most discriminative features across different windows, thereby improving the accuracy and reliability of the diagnostic 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 title and abstract should clearly reflect the focus and application of the proposed methods. Including information about “schizophrenia detection” would make the paper’s objective more explicit and help readers understand the context and significance of the research.
- 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?
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 claim the application of the proposed method in the abstract ot title.
- increase the font sizes of figures for better readability.
- 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 paper presents a reasonable and well-defined approach. However, to enhance clarity and improve readability, it would be beneficial for the authors to explicitly state in the title and abstract that the proposed method is used for schizophrenia detection.
- 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
We thank all reviewers for their comments, and appreciate the overall enthusiasm on our work. We respond to major comments here and will address all comments in final paper. Code: We will make our well-documented code and pre-trained model publicly available. More experimental results: We will provide explanations for the additional experiments suggested by the reviewers and will include these enhancements in the journal version. R#1 Q1: Related work A1: The purpose of MFMM is to deal with the fuzzy information in brain networks. We did not provide detailed descriptions of MFMM and competitor methods due to space limit. We will clarify this in the revision. Q2: MFMM+QAE A2: NoQAE removes the QAE module, making it unable to assess view quality. Therefore, the performance of NoQAE is inferior to that of TMC and MMD, which both take view quality into account. However, we believe that the results of MFMM+QAE are better than QAE combined with other methods because most other methods use softmax or directly fuse the outputs of networks. During the experiments, we have validated that MFMM performs better than softmax, which is why we designed MFMM. Q3: Positioning A3: We improved FMM into a multi-view method, which is unprecedented, and we have fully considered the quality of each view. The experimental results also fully demonstrate the effectiveness of the model. Q4: simple baselines to aggregating predictions over views A4: We validated simple baselines like concatenation (Acc: 81.99) or addition (Acc: 81.35) for combining predictions from multiple views. Additionally, we tested the performance using single views (Acc: 76.52). Unfortunately, we did not present these results due to space limit. Q5: ablation study for QAE A5: This is a constructive suggestion. QAE can be directly applied to other competitor methods. And we encourage further exploration using our code, specifically by replacing the softmax with QAE and using the loss function in section 2.4. R#4 Q1: Objective A1: Our model is applied to dynamic brain network analysis, particularly in the diagnosis of brain disorders. We validated the performance using two schizophrenia datasets. We will explicitly state our objectives in the revision. Q2: font sizes of figures A2: We will increase the font size of the figures in the revision. R#5 Q1: Motivations A1: We explain how fuzzy information and uncertainty arise [1, 2, 9] in the introduction. Fuzzy information refers to ambiguous data from unreliable sources. Uncertainty arises from the different confidence level in multiple views. We will further clarify this in the revision. Q2: The order of three filters A2: According to equations 1-3, it can be inferred that E2E, E2N, and N2G are performed sequentially, enabling hierarchical extraction of brain network features, thereby reducing the dimensionality of feature mapping. Q3: c_k^v and e_k^v A3: c_k^v can be understood as the probability for each class, which can be directly used for classification in each view. Figure 2 illustrates this result. e_k^v is the input to MFMM, while c_k^v is the output. We ensure the credibility of e_k^v through the loss function elaborated in section 2.4. Q4: Multi-class classification A4: Our model is suitable for multi-class classification tasks. Due to space limit and considerations for the overall experimental setup, we were unable to present the results for multi-class classification. We have made the code publicly available for validation of applicability to multi-class tasks. Q5: views and windows A5: We follow literature guidelines for sliding window division. Exploring the impact of different combinations and quantities of views, as well as ensuring performance in the absence of views, is our main focus in the future. Q6: ablation study A6: We completed extensive ablation study, such as sequentially removing three filters (Acc: 78.57, 76.19, 71.43). We present the most important results (NoQAE and NoMFMM) in Table 1.
Meta-Review
Meta-review #1
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Post rebuttal, two of three reviewers are leaning towards accepting the paper. Rev 5 had recommended a weak reject and did not participate post rebuttal. Upon reading the reviews, response and the paper, it appears that the concerns of the reviewers have largely been addressed by the authors. Therefore, I would be open to accepting this work for MICCAI.
As a follow-up to the authors, please acknowledge the BrainNet CNN paper [1] which introduced the original formulation for E-E, E-N and N-G convolutions. Additionally, their quality aware combination (Eqn 6) is very analogous to Uncertainty based late fusion [2] from the multimodal fusion literature, please make note of this in the paper. Finally, please fix inline references in the text and make sure the supplementary does not violate MICCAI policies against inclusion of text content.
[1] Kawahara, Jeremy, et al. “BrainNetCNN: Convolutional neural networks for brain networks; towards predicting neurodevelopment.” NeuroImage 146 (2017): 1038-1049.
[2] Wang, Hongzhi, Vaishnavi Subramanian, and Tanveer Syeda-Mahmood. “Modeling uncertainty in multi-modal fusion for lung cancer survival analysis.” 2021 IEEE 18th international symposium on biomedical imaging (ISBI). IEEE, 2021.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
Post rebuttal, two of three reviewers are leaning towards accepting the paper. Rev 5 had recommended a weak reject and did not participate post rebuttal. Upon reading the reviews, response and the paper, it appears that the concerns of the reviewers have largely been addressed by the authors. Therefore, I would be open to accepting this work for MICCAI.
As a follow-up to the authors, please acknowledge the BrainNet CNN paper [1] which introduced the original formulation for E-E, E-N and N-G convolutions. Additionally, their quality aware combination (Eqn 6) is very analogous to Uncertainty based late fusion [2] from the multimodal fusion literature, please make note of this in the paper. Finally, please fix inline references in the text and make sure the supplementary does not violate MICCAI policies against inclusion of text content.
[1] Kawahara, Jeremy, et al. “BrainNetCNN: Convolutional neural networks for brain networks; towards predicting neurodevelopment.” NeuroImage 146 (2017): 1038-1049.
[2] Wang, Hongzhi, Vaishnavi Subramanian, and Tanveer Syeda-Mahmood. “Modeling uncertainty in multi-modal fusion for lung cancer survival analysis.” 2021 IEEE 18th international symposium on biomedical imaging (ISBI). IEEE, 2021.