List of Papers Browse by Subject Areas Author List
Abstract
Modern neuroimaging technologies set the stage for studying structural connectivity (SC) and functional connectivity (FC) \textit{in-vivo}. Due to distinct biological wiring underpinnings in SC and FC, however, it is challenging to understand their coupling mechanism using statistical association approaches. We seek to answer this challenging neuroscience question through the lens of a novel perspective rooted in network topology. Specifically, our assumption is that each FC instance is either locally supported by the direct link of SC or collaboratively sustained by a group of alternative SC pathways which form a topological notion of \textit{detour}. In this regard, we propose a new connectomic representation, coined detour connectivity (DC), to characterize the complex relationship between SC and FC by presenting direct FC with the weighted connectivity strength along in-directed SC routes. Furthermore, we present SC-FC Detour Network (SFDN), a graph neural network that integrates DC embedding through a self-attention mechanism, to optimize detour to the extent that the coupling of SC and FC is closely aligned with the evolution of cognitive states.
We have applied the concept of DC in network community detection while the clinical value of our SFDN is evaluated in cognitive task recognition and early diagnosis of Alzheimer’s disease. After benchmarking on three public datasets under various brain parcellations, our detour-based computational approach shows significant improvement over current state-of-the-art counterpart methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1549_paper.pdf
SharedIt Link: https://rdcu.be/dV1Os
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_35
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/1549_supp.pdf
Link to the Code Repository
https://github.com/Chrisa142857/SC-FC-Detour
Link to the Dataset(s)
https://www.humanconnectome.org/
https://adni.loni.usc.edu/
https://sites.wustl.edu/oasisbrains/home/oasis-3/
BibTex
@InProceedings{Wei_Representing_MICCAI2024,
author = { Wei, Ziquan and Dan, Tingting and Ding, Jiaqi and Laurienti, Paul and Wu, Guorong},
title = { { Representing Functional Connectivity with Structural Detour: A New Perspective to Decipher Structure-Function Coupling Mechanism } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {367 -- 377}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper leverages an existing topological measure, i.e., node detour, to introduce a novel detour connectivity measure, which is then independently fed into an GNN architecture for cognitive state 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 paper is well written and structured.
- A rich experimental design and evaluation scheme.
- Using the detour as a topological measure for graph feature extraction and network coupling with application to connectomics seems novel.
- Evaluation on different brain learning tasks (group comparison and cognitive state classification
- 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.
- Independent DC extraction and GNN training
- Limited comparison against SOTA methods
- Limited methodological novelty
See the details in Section 10.
- 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 provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
For scientific reproducibility, it is important to release the source code including benchmarks.
- 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 major concerns regarding the presented work are detailed below:
- Independent DC extraction and GNN training. The authors simply extracted topological features and fed them into a conventional GNN architecture such as GIN and GCN for classification. Several topological measures have been used in network neuroscience for boosting the learning of GNNs, including betweenness centrality, node strength, clustering coefficient, etc. Hence, on a conceptual level, the proposed approach is not novel. The detour-based connectivity coupling is an interesting contribution—however, any other topological measure could have been used.
Please check the following review paper on GNNs in network neuroscience: Bessadok et al. “Graph neural networks in network neuroscience.” IEEE Transactions on Pattern Analysis and Machine Intelligence 45.5 (2022): 5833-5848.
-
Limited comparison against SOTA methods. Since the main claim of the novelty of this work lies in part (A) for Fig 1, it is important to benchmark DC against other topology-driven connectivities such as eigenvector centrality, page rank, etc. The experiment section is poorly designed in the sense that a few variants of GNN architectures have been used, but the main contribution haven’t been evaluated properly against conventional related methods. Different topological measures should be added and benchmarked against to highlight the importance of the detour measure.
-
Limited methodological novelty & overlooked literature. The paper is limited in novelty for the reasons articulated in points (1,2). Deriving a brain connectivity from a topological measure is not novel (although the choice of the detour seems novel here). Several GNN models have been designed and used for classifying, regressing, and generating brain graphs with different variants of topological/attention enhancements. A related work section could have been dedicated to such models. I believe more advanced topology-inspired GNN models exist and none of those were mentioned or benchmarked against in this 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
Reject — should be rejected, independent of rebuttal (2)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Limited novelty and poor benchmarking.
- 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
Reject — should be rejected, independent of rebuttal (2)
- [Post rebuttal] Please justify your decision
After reading the rebuttal, I still hold the same position. The proposed method has a limited novelty and most importantly comparison against existing popular topological measures. No valid baseline to evaluate the merit of the proposed DC.
Review #2
- Please describe the contribution of the paper
The authors introduce a novel connectomic representation, detour connectivity (DC), to fuse structure and functional connectivity in a way that aligns with neurophysiological principles. Then they propose a structure connectivity-functional connectivity detour network (SFDN), essentially a GNN integrated with DC embedding through a self-attention mechanism. The authors trained and tested their method on three datasets with various cortical parcellation strategies, demonstrating the superior performance of their method and potential in cognitive task recognition and AD 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 idea of using detour connectivity to fuse FC and SC is innovative and generally aligns with neurophysiological principles.
The authors tested the model on three datasets (HCP-A, ADNI and OASIS), covering both healthy and disease conditions, as well as resting-state and task-based fMRI data. They also tested different cortical parcellations. - 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.
One possible weakness of this DC metric is that it does not consider the difference between afferent and efferent fibers. Ideally, the structural connectivity graph should be directed, and the direction between two cortical nodes determines the functional cause-and-effect relationship of the two nodes. Functionally effective detour pathway may only exist when the directions of the nodes along the pathway aligns, such as A->B->C->D…
Despite the fact that the three cognitive tasks tested in this study are all heavily related to motor and visual functions due to the task paradigm, as shown in Figure 3, It is still unclear why the SMmouth, SMhand and Visual networks always rank as the top 3 in DC measurements, instead of purely cognition-related networks. - 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?
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
Consider adding a couple of sentences to discuss the limitation mentioned in the first weakness. Testing another cognitive fMRI task with less emphasis on motor or visual functions would provide a more comprehensive view. It would be expected to see the cognition-related networks with the highest ratios. If these expected results won’t happen, further interpretation would be necessary to understand why the proposed DC measurement emphasizes motor and visual networks.
- 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 novelty of detour connectivity in effectively fusing structural and functional connectivity
- The innovative approach of SFDN, which integrates effective DC information as node features fed into GNN.
- The multi-angle testing of the model on three different datasets with various parcellation strategies.
- 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 have successfully addressed my concerns. The reviewer appreciates the idea of detour. It is not a perfect model, but a reasonable start.
Review #3
- Please describe the contribution of the paper
The authors propose a new connectomes detour connectivity (DC) characterizing the complex relationship between structural connectivity (SC) and functional connectivity (FC), and further propose the integration of the graph neural network of DC through self-attentive mechanisms in order to optimize detour so that the coupling of SC and FC is closely related to the evolution of cognitive states.
- 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) To address the gap between functional and structural correlations by incorporating high-level structural detours associated with each functional connection, thereby capturing individual differences and better representing brain network communities. 2) SC-FC Detour Network efficiently integrates structural connectivity (SC) and functional connectivity (FC) data from multimodal MRI, leveraging the self-attention mechanism to incorporate edge-wise directed connectivity embedding for capturing high-order interactions between brain regions across macroscale networks, resulting in precise predictions of human cognition.
- 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 did not indicate whether the density of fiber bundles or a diffusion metric, such as the fractional anisotropy, was used as a weight in the construction of the SC process; 2) No threshold was given in this work to indicate which fiber bundles were discarded.
- 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 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
The probabilistic tracking method enables the construction of a bundle of white matter fibers between any two brain regions if no threshold is set. Therefore, the author should consider the threshold setting problem.
- 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 authors characterize the complex relationship between SCs and FCs by representing the weighted connection strengths of direct FCs and along directed SC routes, which makes the coupling of SCs and FCs more efficient.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
In the paper, the author should conduct a comparative experiment using both DC characteristics and the conventional FC-SC fused features within the GNN framework, rather than relying solely on t-tests to demonstrate the superiority of DC. This approach allows for a more rigorous and convincing evaluation of the DC method’s effectiveness.
Author Feedback
R#3 scored “reject” and provided conflict comments such as “lack novelty” (weakness #3) and “seems novel” (strongness #3). Concerns by R#3 are summarized below:
- Lack of experiments and related work reviews:
1.1 Limited comparison against SOTA GNNs. We proposed a new SC-FC coupling approach (DC) and how its embedding can benefit a graph model (SFDN) for brain network recognition. We focus on the model interpretability (such as coupling performance indicated by our t-test of the functional community detection) and the clinical value (such as improvement on downstream applications by the embedding of DC). Since our primary interest is an explainable deep model for human connectomes data, SOTA performance on general graph data is not the major focus of this work. GNN is the framework of baseline for our DC embedding to plug in, but the page limitation makes it difficult to add other SOTA GNNs as another focus in the current work. In our experimental design, a comprehensive comparison against some baselines without our DC embedding is delivered to show not only the effectiveness of DC for downstream applications but also the robustness under different preprocessing and datasets. On the other hand, we already have additional results of SOTA brain models solely using FC that further show a superior performance by our SFDN based on the SC-FC detour, but we will put them to future works due to page limitation.
1.2 Overlooked literature of advanced topology-inspired GNN. We have covered most related works of SC-FC coupling in the second paragraph of the Introduction section. Compared to these most relevant works, advanced topology-inspired or high-order GNNs are not relevant to the SC-FC coupling topic. Considering the page limit, we didn’t include them.
- Novelty:
2.1 The node detour idea exists. First of all, R#3 comments that the node detour exists without any reference. But, to our best knowledge, our work first proposes the term of node detour to do the topological measurement, and the searching results of “node detour” on Google Scholar can support us.
2.2 The detour connectivity (DC) uses the same conceptual-level idea of other brain models, and hence it is not novel. No work in the literature mentioned by R#3 focused on SC-FC coupling. As our title highlighted our focus on “decipher structure-function coupling”, a new SC-FC coupling approach named detour connectivity is delivered by our work, and it is novel in the field. This is also agreed upon by other reviewers.
2.3 Any other topological measure could have been used to replace the DC. Our idea of detour connectivity is inspired by the neuroscience discoveries on SC-FC coupling, supported by literature mentioned in the 3rd paragraph of the Introduction section. Furthermore, we validated the idea in our t-test experiments. We think other topological measurements, such as the clustering coefficient mentioned by R#3, are interesting. However, it is arguable to replace DC with classic graph theory measurements in SC-FC coupling since the latter is not designed to model the relationship between SC and FC.
- The detour connectivity extraction and GNN training are independent As a non-parametric approach of SC-FC coupling, it is independent from any deep learning model by design. This makes it easier to reproduce our SC-FC coupling results in our t-test experiments and to plug in any deep learning model as an embedding. In fact, our work focuses on SC-FC coupling instead of GNN training.
R#1 and R#4 are concerned about lacking SC preprocessing details, explanation of some t-test results, and limitations of our approach facing directed SC. Also, code release is a concern by all three reviewers. We will add those details, explanations, and limitations to the revised version. The code is already organized and will be released in the revised version.
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’
The rebuttal does a good job of addressing the concerns by R3. As the other two reviewers are positive about the paper, I would recommend it for acceptance.
- 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).
The rebuttal does a good job of addressing the concerns by R3. As the other two reviewers are positive about the paper, I would recommend it for acceptance.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
After reading the paper, reviews and rebuttal, I agree with some reviewers decision towards weakly rejection. The major issues include:
- Reviewer still has the concern that the proposed method has a limited novelty and most importantly comparison against existing popular topological measures. No valid baseline to evaluate the merit of the proposed DC.
- It is suggested to add a comparative experiment using both DC characteristics and the conventional FC-SC fused features within the GNN framework.
- More explanation are welcomed, e.g., further interpretation would be necessary to understand why the proposed DC measurement emphasizes motor and visual networks.
- 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).
After reading the paper, reviews and rebuttal, I agree with some reviewers decision towards weakly rejection. The major issues include:
- Reviewer still has the concern that the proposed method has a limited novelty and most importantly comparison against existing popular topological measures. No valid baseline to evaluate the merit of the proposed DC.
- It is suggested to add a comparative experiment using both DC characteristics and the conventional FC-SC fused features within the GNN framework.
- More explanation are welcomed, e.g., further interpretation would be necessary to understand why the proposed DC measurement emphasizes motor and visual networks.
Meta-review #3
- 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’
Reviewers were mixed, although all agree on the novelty of the proposed detour measure to integrate SC and FC and the experimental evaluation that was performed (both group analysis and predictive). However, there were strong concerns regarding whether appropriate experimental comparisons are made (ie to other topological measures) and novelty of the overall approach (feature used in GNN). Considering the reviews and rebuttal and reading through the paper, I follow the recommendation of the majority of the reviewers and support acceptance of this paper, as it 1) proposes a new SC-FC measure that would be of interest to the community, and 2) even though more comparisons, especially compared to other methods that combine SC-FC, should be compared, the experiments that were performed in the paper under multiple settings (dataset/tasks, parcellations) seem to show improvement of prediction with the detour feature compared to standard FC for edges.
- 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).
Reviewers were mixed, although all agree on the novelty of the proposed detour measure to integrate SC and FC and the experimental evaluation that was performed (both group analysis and predictive). However, there were strong concerns regarding whether appropriate experimental comparisons are made (ie to other topological measures) and novelty of the overall approach (feature used in GNN). Considering the reviews and rebuttal and reading through the paper, I follow the recommendation of the majority of the reviewers and support acceptance of this paper, as it 1) proposes a new SC-FC measure that would be of interest to the community, and 2) even though more comparisons, especially compared to other methods that combine SC-FC, should be compared, the experiments that were performed in the paper under multiple settings (dataset/tasks, parcellations) seem to show improvement of prediction with the detour feature compared to standard FC for edges.