List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yanqing Kang, Ruoyang Wang, Enze Shi, Jinru Wu, Sigang Yu, Shu Zhang
Abstract
Understanding the relationship between brain functional connectivity and structural connectivity is important in the field of brain imaging, and it can help us better comprehend the working mechanisms of the brain. Much effort has been made on this issue, but it is still far from satisfactory. The brain transmits information through a network architecture, which means that the regions and connections of the brain are significant. The main difficulties with this issue are currently at least two aspects. On the one hand, the importance of different brain regions in structural and functional integration has not been fully addressed; on the other hand, the connectome skeleton of the brain, plays the role of common and key connections in the brain network, has not been clearly studied. To alleviate the above problems, this paper proposes a transformer-based self-supervised graph reconstruction framework (TSGR). The framework uses the graph neural net-work (GNN) to fuse functional and structural information of the brain, and uses the self-supervised model to identify contribution scores of regions for the reconstruction task. Regions with high scores are considered as key connectome regions which play an essential role in the communication connectivity of the net-work in the brain. Based on key brain regions, the connectome skeleton can be obtained. Experimental results demonstrate the effectiveness of the proposed method, which obtains key regions and connectome skeleton in the brain net-work. This provides a new angle of view to explore the relationship between brain function and structure.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43993-3_30
SharedIt: https://rdcu.be/dnwNv
Link to the code repository
https://github.com/kang105/TSGR
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The authors deligently combined structural and functional MRI using a deep-learning framework to identify key regions and connectome skeleton in the brain 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 authors combined multimodal MRI data to identify the relationship between structure and function, and to identify a connectome skeleton that expresses the most key connections of the brain in functional and structural networks. More importantly they have demonstrated their claims with extensive sets of experiments, which is a major strength of the paper.
- 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.
Not many weaknesses, but certain terms and measures they used are not defined / not clear. I have detailed them below.
- 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
The method seems reproducible.
- 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/2023/en/REVIEWER-GUIDELINES.html
- ‘contribution scores’ - please difine this.
- “the functional and structural information is characterized as node and edge features of the graph, respectively” - I think it is the other way.
- State the use of dMRI in introduction (you stated as if you used only fMRI)
- Please define IoU
- Please define the different centrality measures you used and mention how it is calculated.
- 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?
Good demonstartion of intersting findings about brain structure and function.
- Reviewer confidence
Very confident
- [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
I still gave ‘strong accept’. I might have changed my recommendation to ‘weak accept’ after reading the concerns raised by the other reviewers, but I think the authors have reasonably addressed/ answered all the concerns (novelty, defining key regions, biological rationale, and reproducibility- explained and agreed to publish the code) in their rebuttal. I will leave it to them to comment, but I think they will agree.
Review #2
- Please describe the contribution of the paper
The authors proposed a transformer-based graph self-supervised graph reconstruction framework for studying brain function-structure relationship. This method is used for obtain key connectome regions and backbones for combining functional and structural connectivity.
- 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 contain a graph convolution network based generation model for estimating important nodes and edges in brain structural and functional couplling.
- 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.
Applying graph convolution network model for caputuring the relatinship between structural connectome and functional connectome is not novel. Lots of efforts have been made in this topic. https://link.springer.com/chapter/10.1007/978-3-031-16431-6_22 https://proceedings.mlr.press/v143/dsouza21a.html https://link.springer.com/chapter/10.1007/978-3-030-87586-2_14
Almost each framework could give a new measurements of nodal and edge significance. The authors soluld further show clearly how their approach is better than previous methods in addressing either basic brain organization or individual differences in behaviors and disease states. In the current work, some results are hard to understand. For instance, the common hubs in both the resting functional network and structural network located in bilateral PCUN and PCC. But in Fig.3, it seems that they identified visual and motor related brain regions as high contributors.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 HCP brain images used here is public dataset. No codes is given.
- 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/2023/en/REVIEWER-GUIDELINES.html
I suggest an evaluation should be given for comparing the current model with previous methods in accurately capturing brain-behavior associations.
- 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 authors are failed to address the innovativeness of this work.
- Reviewer confidence
Very confident
- [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 a transformer-based self-supervised graph reconstruction (TSGR) framework to fuse structural and functional connectivity of the human brain, and further proposes a self-supervised framework to identify key important contributing nodes/regions of the brain. The authors obtain scores for each region of interest (ROI) and test whether the ROIs with high scores are key connectome ROIs. They further propose a connectome skeleton.
- 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) A novel approach to fusing structural and functional connectome to identify key ROIs via a fused approach. b) Seeming rigorous validation of the important regions across the task-based functional connectivity studies from the HCP project.
- 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) No clear biological rationale for the regions identified as the key region. b) The method is described in a way that is difficult to reproduce.
- Please rate the clarity and organization of this paper
Satisfactory
- 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
Not applicable
- 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/2023/en/REVIEWER-GUIDELINES.html
a) There were seven task fMRI and 1 resting-state functional connectivity. There was 1 structural connectivity for all tasks. It is unclear from the authors’ description whether all task + resting-state data were used to obtain the scores of each ROI or whether the process was repeated for all task + resting-state data. I understand the scores of each ROI were eventually averaged to obtain a group-level ranking of the ROIs but whether this was done as a two-step procedure, i.e. first average the score for each ROI across all tasks for each participant or average the score for each ROI across all participant for each task/resting-state fMRI. Can the authors clarify? b) How was the threshold determined for the structural connectivity adjacency matrix? c) What was the definition of scale-1 and scale-2, or how were these determined? d) In Fig.4A, there are a lot of zeroes in the functional centrality measure across ROIs which is driving the overall correlation. Can the authors redo their correlation analysis or fit a different model respecting the data? e) On average, the authors could also conclude that functional-structure coupling is about 50%, similar to https://www.sciencedirect.com/science/article/pii/S1053811920310946. Can the authors compare their study with the study by Sarwar et al. to point out the key differences? f) Other than using the anatomical atlas, I think the study will have more power when a functional atlas such as Brainnetome (https://pubmed.ncbi.nlm.nih.gov/27230218/) is used. What are the author’s thoughts on this? g) Both the structural and functional connectivity suffers from various analysis choices such as parcellation atlas, resolution, data quality, etc. How confident are the authors that the methods proposed in this paper are reproducible? h) Since the authors use HCP data, can the authors test their skeleton on a few of the HCP dataset to prove that the results obtained from their discovery data is not overfitting and is representative of a healthy population? i) Finally, can the authors shed some light on the regions identified in the connectome skeleton? How does this inform pathology or development or evolution?
- 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?
Difficult to follow
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
3
- [Post rebuttal] Please justify your decision
Not satisfactory; the code is not open access and I doubt it is easy to reproduce.
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.
This paper proposed to use multimodal MRI to study the connectome skeleton, and extensive sets of experiments were performed. Reviewers have some concerns on the clarity of the paper, novelty with existing studies, and reproducible method details.
Author Feedback
Dear Reviewers, We appreciate all the reviewers for the constructive comments and suggestions on our manuscript. Due to space limits, we address the key concerns in detail as follows. Reviewer#1: We appreciate your positive comments.
Reviewer#2: Appreciate for the comments.
Question #1: The novelty of this work and the differences with other work using GCN are not fully elaborated. Response #Q1: We have carefully reviewed the three works given. They explore the relationship between structural and functional connectomes based on predictive tasks such as age. As a comparison, the main innovation of our work is self-supervised model without label. Specifically, we integrate brain structure and function information to explore key regions through a self-supervised model, and then obtain the structure and function connectome skeleton. For the motivation, much work has explored important regions of brain and the relationship between structure and function, but they are based either on graph theoretic centrality or on supervised learning, which rely too much on prior knowledge and may omit the characteristics of the brain network itself. In contrast, self-supervised learning can learn useful representations of the data itself. So, we believe that self-supervised learning can reveal intrinsic properties of brain networks, such as key regions and connections. In addition, self-supervised deep learning has been proven to be able to detect meaningful features from brain data. Question #2: Some key regions are difficult to understand. For example, PCUN and PCC are hub regions, but visual and motor related brain regions are identified as high contributors in Fig. 3. Response #Q2: We explain the process of defining key regions. First, we represent the brain as a graph. Then, the self-supervised model is used to reconstruct the graph and extract contribution scores of nodes, where scores represent the importance of reconstruction both on the functional and structural perspectives. Finally, the regions with high scores are considered as key regions. On one hand, key regions in this work are different from the traditional hubs. They are based on brain structure and functional connectome features derived from the reconstruction task, so they represent some connectome features, such as participate in more functional networks, long-distance and dense connections of structures. Many studies show that visual and motor related regions participate in multiple functional networks and are closely connected locally, thus they appeared in the results. On the other hand, PCUN has shown in some experiments, due to the angle of view, it is hard to observe. We will optimize visualization results.Reviewer#3: Appreciate for the comments. Question #1: No clear biological rationale for the regions identified as the key region. Response #Q1: Due to space limits, we choose a few regions to explain. These are the frontal gyrus, superior parietal gyrus, precentral, and middle occipital gyrus, which are identified in most experiments. In details, Cole et al. confirmed that the lateral prefrontal cortex and the right premotor cortex are the core of cognitive control (10.1523/JNEUROSCI.0536-12.2012). Heuvel et al. confirmed that the parietal and prefrontal cortex contain multiple hubs in almost all species (10.1016/j.tics.2013.09.012). Millet et al. confirmed the lateral occipital region is essential for higher-order visual function (10.1016/j.cortex.2013.11.004). We also analyze the connectome skeleton, e.g. frontal lobe has the most total connections. This also reflects the key role of the frontal lobe. We will add more discussion of the biological rationale for key regions in the revised version. Question #2: The method is difficult to reproduce. Response #Q2: The method is straightforward. The input is multimodal brain graphs. Then a self-supervised model is proposed to identify key regions, which is also stable in results. The code will be open-sourced.
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 authors proposed to combine structural and functional MRI to identify key regions and connectome skeleton in the brain network. There are mixed reviews after the rebuttal. Considering the rebuttal mentioned the reproducibility will be addressed, I tend to accept this paper.
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.
The authors have reasonably addressed/ answered all the concerns, and therefore, I recommend accept.
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.
Although one reviewer strongly recommended to accept the paper, two reviewers recommended reject. I looked at the paper and rebuttal. I agreed with R2 and R3 on the issues of the paper, such as the novelty, result interpretations, practical uses, and method description. So I recommend to reject the paper.