List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Tianji Pang, Dajiang Zhu, Tianming Liu, Junwei Han, Shijie Zhao
Abstract
Task-based functional magnetic resonance imaging (fMRI) has been widely used for functional brain network identification. Recently, deep belief network (DBN) has shown great advantages in modeling the hierarchical and complex task functional brain networks (FBNs). However, due to the unsupervised nature, traditional DBN algorithms may be limited in fully utilizing the prior knowledge from the task design. In addition, the FBNs extracted from different DBN layers do not have correspondences, which makes the hierarchical analysis of FBNs a challenging problem. In this paper, we propose a novel prior knowledge guided DBN (PKG-DBN) to overcome the above limitations when conducting hierarchical task FBNs analysis. Specifically, we enforce part of the time courses learnt from DBN to be task-related (in either positive or negative way) and the rest to be linear combinations of task-related components. By incorporating such constraints in the learning process, our method can simultaneously leverage the advantages of data-driven approaches and the prior knowledge of task design. Our experiment results on HCP task fMRI data showed that the proposed PKG-DBN can not only successfully identify meaningful hierarchical task FBNs with correspondence comparing to traditional DBN models, but also converge significantly faster than traditional DBN models.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16431-6_24
SharedIt: https://rdcu.be/cVD46
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This work proposed a novel method to incorporate prior knowledge to DBN for hierarchical brain network decomposition. Sufficient experiments suggest the proposed method can converge faster than the traditional DBN models. In addition, the proposed method can identify better task and intrinsic functional brain networks compared with the traditional DBN models. This work contributes a novel idea on how to introduce prior knowledge to unsupervised deep models for fMRI data analysis.
- 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 work proposed a novel algorithm to incorporate prior knowledge to DBN for fMRI data analysis. The idea of this work is clearly conveyed. The algorithm is clearly formulated. It is an interesting topic how to incorporate prior knowledge into deep models. This work contributes a novel idea on how to introduce prior knowledge to unsupervised deep models for fMRI data 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.
The writing needs improvement and a few details are not clear. For instance, how the overlap rate is calculated is not clear. For Table I, in addition to the average overlap rate between DBN inferred RSNs and their corresponding RSN templates across subjects, the standard deviation should also be reported.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
They will release the code after the paper is accepted.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
In this paper, authors proposed a novel prior knowledge guided DBN (PKG-DBN) model to identify the hierarchical and complex task functional brain networks (FBNs). Specifically, they enforce part of the time courses learnt from DBN to be task-related and the rest to be linear combinations of task-related components. By incorporating such constraints in the learning process, the PKG-DBN model can simultaneously leverage the advantages of data-driven approaches and the prior knowledge of task design. Extensive experiments on both HCP Gambling and Language task fMRI data have well supported the proposed ideas. In general, this is a novel and interesting work and the paper is well structured. There are only a few writing problems need to be improved. For instance, how the overlap rate is calculated is not clear. For Table I, in addition to the average overlap rate between DBN inferred RSNs and their corresponding RSN templates across subjects, the standard deviation should also be reported. The figure resolution need to be improved.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This work proposed a novel method to incorporate prior knowledge to DBN for hierarchical brain network decomposition. Sufficient experiments suggest the proposed method can converge faster than the traditional DBN models. It is a novel method for fMRI analysis field. However, there are still a few unclear descriptions and typo errors which need to be further improved.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #2
- Please describe the contribution of the paper
The authors proposed a prior knowledge-guided version of deep belief network to infer brain networks with the task labels used as prior information. The inferred networks are more aligned with the ground truth resting-state network
- 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 results are interesting as providing task labels will help form more reliable brain networks.
- 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 aim of the proposed framework is not very clear. In my understanding the previous DBN approaches are used as unsupervised method due to the lack of labelling. However, in this proposed framework, as we are aware of the task labels already, why can’t we switch entirely to the supervised method?
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
It is reproducible with codes and data becoming available
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
In this paper, the authors propose a prior knowledge guided version of DBN for brain network discovery. By imposing task paradigm information, the discovered brain networks are able to match better with the RSN templates. However, there are a couple of comments I would like to make.
-
It is a bit confusing to show fig 2 as the authors mentioned that DBN has lower loss but the authors mentioned that the model can be overfitted. Did this happen to all the subjects? How did the author infer that the model were overfitted? In my understanding, the authors concluded the overfit by observing an early turning point of the mismatch loss in PKG-DBN. However, could the authors plot similar task paradigm mismatch plot for the DBN model?
-
How were the hyperparameters chosen in the experiment?
-
Similar as I wrote above in the weakness. As we already have access to the task labels, why don’t we switch to the supervised method entirely to estimate the brain networks?
-
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The advantage of the framework is not very clear to me.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
3
- Reviewer confidence
Somewhat Confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #3
- Please describe the contribution of the paper
In this paper, the authors propose a novel prior knowledge guided DBN (PKG-DBN) model hierarchical brain network decomposition. By incorporating such constraints in the learning process, the proposed method can simultaneously leverage the advantages of data-driven approaches and the prior knowledge of task design.
- 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 work contributes a novel idea on how to introduce prior knowledge to unsupervised deep models for fMRI data analysis. This is the first supervised hierarchical FBN identification method as far as I know.
- 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 writing needs improvement and more details should be clear.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The paper decribed the method clearly, but did not release the code currently.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
In this paper, authors propose a novel prior knowledge guided DBN (PKG-DBN) model hierarchical brain network decomposition. Specifically, they enforce part of the time courses learnt from DBN to be task-related (in either positive or negative way) and the rest to be linear combinations of task-related components. By incorporating such constraints in the learning process, our method can simultaneously leverage the advantages of data-driven approaches and the prior knowledge of task design. In general, it is novel and interesting. The proposed points are well supported by the results. I only have a few concerns/comments. 1) A few Figure resolution is relatively low, such as Fig.2. 2) The model parameters should be clearer. 3) The language need to be improved with native English speaker. There are a few language errors in current version.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This work propose a novel prior knowledge guided DBN (PKG-DBN) model hierarchical brain network decomposition. This is the first supervised hierarchical FBN identification method as far as I know. It is important to fMRI analysis field.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
7
- [Post rebuttal] Please justify your decision
The author has given me a good response.
Primary Meta-Review
- Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The presented idea of brain network decomposition for task-related fMRIs has novelty, and the detailed formulations will be helpful for reproducing the results. However, there are some critiques regarding experimental results. It is not clear that the results in Table 1 display statistically meaningful improvement since all overlap scores are low. It will be helpful if there is a discussion on the results by referring to reported results in the recent work. Also, the results in Fig.2 should have more clarification, as Reviewer 2 mentioned (losses by different methods). More experimental details, including parameter setting, need to be provided as well. It is expected for the authors to answer the aforementioned questions in the rebuttal and improve the writing of the final version of the paper.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and 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).
1
Author Feedback
We thank meta-reviewer and all reviewers for recognizing the novelties and interesting results in our work to incorporate prior knowledge to DBN for hierarchical brain network decomposition. (“all reviewers agree the work is novel and result is interesting”).
- Novelties (“This work proposed a novel method to…” –R1,” The results are interesting as providing task labels will help form more reliable brain networks.” –R2,” This work contributes a novel idea on how to introduce prior knowledge” –R3, “The presented idea of brain network decomposition for task-related fMRIs has novelty…” –Meta)
- Interesting results (The results are interesting as providing task labels will help form more reliable brain networks…” –R2,” In general, it is novel and interesting.” –R3)
We notice that most of the concerns from reviewers are caused by the writing problems of the paper. We will revise the paper thoroughly as suggested. Detailed responses to the concerns raised by the meta-reviewer and reviewers are as follows.
Major Concerns R2 concerns about the motivation of the proposed framework. We apologize for the unclearness and will significant improve its clearness in the revision if we have the chance. There are two notable limitations in current FBN analysis methods: 1) the prior knowledge of task paradigm is not used during feature learning; 2) the FBNs extracted by different DBN layers do not have specific correspondence, which makes the hierarchical analysis of FBNs a challenge problem. The proposed PKG-DBN framework can infuse the prior knowledge of task diagram into the FBN decomposition procedure and help establish a few correspondence between specific components across different layers. Experiments shows the PKG-DBN model could help reduce training loss and accelerate convergence, as well as establish correspondence between different layers. This is also well recognized by R1 and R3.
Meta also concerns statistically meaningful improvement of results in Table 1. The overlaprate improves a lot although they are still in a relatively low score. The low score due to the reason that the template are usually group average result. Thus, the individual overlap ratio in such a level is already meaningful.
R2 and Meta concern how we infer that the traditional DBN models were overfitted. From Fig. 2, there is an early turning point of the mismatch loss in PKG-DBN. But after that turning point, the loss of PKG-DBN still decrease and at the same time the mismatch loss in PKG-DBN increase. This means after the turning point, though the model can learn better representation for the data, it learns worth representation w.r.t task paradigms. We believe the model get overfit after the turning point and the traditional DBN suffers the same problem. In addition, though traditional DBN can get lower loss, PKG-DBN can get better brain networks. This also suggests that the traditional DBN model suffers the overfitting problem. The similar task paradigm mismatch cannot be plotted for the traditional DBN model. This is because the correspondence between the learnt brain networks and task paradigms are unclear during model training.
R2, R3, and Meta all concern about the parameter setting. The learning rate, weight decay, batch size, and epoch number for PKG-DBN are selected according to [19]. The mismatch threshold, the Lagrange multiplier, and the sparsity regularization parameter are selected by experimentation. We will specify how these parameters will influence the performance of the proposed model in the revised manuscript if we have the chance.
Other concerns: Reviewers and meta-reviewer raised questions about the grammar and typos, mathematical formulation, and figure improvement suggestions. We totally agree with reviewers’ questions and suggestions and will definitely address these issues in the revised manuscript if we have the chance.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The proposed network decomposition method’s novelty was recognized by reviewers. The reviewer critiques included discussing experimental results (comparison with other methods) in terms of statistical significance, experimental setting (e.g., hyperparameters), and clarification in formulation and figures. The rebuttal sufficiently addressed those critiques, and the authors are expected to revise the final manuscript according to their responses to the rebuttal.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
3
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
Authors improved fMRI data analysis by assuming a prior by means of linear combination of fMRI signals from pre-assessed tasks for analysis,. Main strengthes including the new approach for prior encoding for fMRI analysis and experimental results. Clarity was raised as an issue by reviewers but according to authors response this wil be addressed at final submission.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
1
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The work proposed a novel algorithm to incorporate prior knowledge to DBN for fMRI data analysis, which is an interesting topic. The main strength lies in its new way of incorporating prior knowledge into deep learning framework. The main weakness is its clarity and writing. The rebuttal has addressed some of the clarification issues and has promised to improve the writing. Given the proposed interesting method, this could be considered for acceptance.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- 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).
4