List of Papers Browse by Subject Areas Author List
Abstract
The human cerebral cortex is highly convoluted into convex gyri and concave sulci. It has been demonstrated that gyri and sulci are significantly different in their anatomy, connectivity, and function: besides exhibiting opposite shape patterns, long-distance axonal fibers connected to gyri are much denser than those connected to sulci, and neural signals on gyri are more complex in low-frequency while sulci are more complex in high-frequency. Although accumulating evidence shows significant differences between gyri and sulci, their primary roles in brain function have not been elucidated yet. To solve this fundamental problem, we design a novel Twin-Transformer framework to unveil the unique functional roles of gyri and sulci as well as their relationship in the whole brain function. Our Twin-Transformer framework adopts two structure-identical (twin) Transformers to disentangle spatial-temporal patterns of functional brain networks: one focuses on the spatial patterns and the other is on temporal patterns. The spatial transformer takes the spatially divided patches and generates spatial patterns, while the temporal transformer takes the temporally split patches and produces temporal patterns. We validated our Twin-Transformer on the HCP task-fMRI dataset, for the first time, to elucidate the different roles of gyri and sulci in brain function. Our results suggest that gyri and sulci could work together in a core-periphery network manner, that is, gyri could serve as core networks for information gathering and distributing, while sulci could serve as periphery networks for specific local information processing. These findings have shed new light on our fundamental understanding of the brain’s basic structural and functional mechanisms.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3934_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Yu_Gyri_MICCAI2024,
author = { Yu, Xiaowei and Zhang, Lu and Cao, Chao and Chen, Tong and Lyu, Yanjun and Zhang, Jing and Liu, Tianming and Zhu, Dajiang},
title = { { Gyri vs. Sulci: Core-Periphery Organization in Functional Brain Networks } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15012},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces a novel framework called Twin Transformer, comprising a Spatial Transformer and a Temporal Transformer, designed to factorize fMRI signals and reveal their intrinsic core-periphery relationship. Validation is performed using a substantial adult fMRI dataset, the Human Connectome Project (HCP), providing compelling evidence for the effectiveness of the proposed approach.
- 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 paper introduces a novel scope for analyzing fMRI data, specifically focusing on investigating the relationship between gyri and sulci.
- An intriguing conclusion is drawn from extensive experiments on a large dataset, highlighting the core-periphery organization between gyri and sulci. This conclusion is poised to have a significant impact on subsequent research in the field.
- 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.
-
Some critical details are missing, which may lead to the confusion of readers, e.g., how to calculate the relationship matrix based on the functional brain network.
-
Lack of experiments. It is necessary to compare with the traditional factorization methods, and provide some discussion on why the proposed method could unveil the special organization between gyri and sulci, while the other methods could not?
-
Some network designs require further elucidation. For instance, it would be beneficial to provide additional explanation on how the voxels are partitioned into patches in the spatial transformer. Clarifying this aspect is crucial as it directly influences the interpretation of the final conclusion. Specifically, understanding how the proposed method treats the output of each patch as a distinct brain functional network would enhance the comprehensibility and applicability of the findings.
-
- 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
-
It is recommended to provide a detailed explanation of the methodology used to compute the relationship matrix.
-
It would be particularly intriguing if the authors could specify the number of functional brain networks (FBNs) generated and offer visualizations of these FBNs.
-
A critical discussion is warranted regarding why the Twin Transformer can discern the core-periphery organization between gyri and sulci, whereas traditional factorization methods cannot. Did the authors introduce additional constraints to guide the network, such as distinguishing between sulci and gyri?
-
The method by how voxels were rearranged for patch partitioning in the spatial transformer is not clearly elucidated.
-
Please provide the precise values of hyperparameters, such as the number of patches/FBNs (P) and the weight for the second loss term (alpha).
-
It would be beneficial to explore potential applications of the significant conclusions drawn in this paper. For instance, could these findings facilitate improved individualized functional parcellation?
-
- 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?
I made the decision because this paper lacks some key details, which impede readers’ understanding and evaluation of its contribution, despite its interesting conclusion.
- 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
Really appreciate for the authors to provide detailed response. My remaining concern is that this work does not include any comparison methods or ablation studies which may hinder the readers to evaluate its technical contribution. I understood that MICCAI rules prohibit new experiments. Therefore, based on this submission, I gave my recommendation of weak accept. If this paper is finally accepted, it would be really helpful if they can use some simple sentences to briefly describe how to decide the vertices order when they conduct patch division in the Twin-Transformer and the potential effect of other hyperparameters.
Review #2
- Please describe the contribution of the paper
Using the HCP task-fMRI dataset, and using a twin-transformer network, the authors disentangle spatial and temporal patterns of the fMRI signal and they separate gyral and sulcal regions. Their results suggest that gyri are more interconnected than sulci, thus that gyri and sulci work in a core-periphery network manner.
- 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 use state-of-the-art method (spatial and temporal twin transformers)
- The authors properly separate gyri and sulci in their fMRI analysis and ask an innovative question about gyri and sulci networks using fMRI datasets
- 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.
- There is more fMRI signal on gyri than on sulci as cortical thickness on gyri is higher. This could induce a signal-to-noise ratio smaller from sulci than from gyri. The authors should prove or discuss that such potential higher noise on the sulci doesn’t change their main result, namely that sulci are less interconnected than gyri
- The authors have used the task MRIs rather than resting state MRIs for the training of their network
- 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?
The authors should give, at least in appendix, the hyperparameters of their networks
- 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 authours should provide a rationale or discuss if the core-periphery network structure that they found could come from a poorer signal-to-noise ration on sulci
- The authors should explain the rationale why using task MRIs, and not resting state MRIs, for training
- 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 proposed method to seperate sulci and gyri to analyze fMRI signal is absolutely intriguing. The authors used adapted state-of-the-art methods to combine and separate temporal and sptial patterns. A stronger discussion about the other interpretations than the core-periphery network would be needed.
- 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
In their rebuttal, the authors stressed again, rightly I think, that their method is robust to the difference of signal strength between gyri and sulci. They promised to add further clarification on this based on their data. I believe that it can be a method with potential interesting applications, hence my score update
Review #3
- Please describe the contribution of the paper
Authors proposed a novel Twin-Transformer framework to disentangle the spatialtemporal patterns of the functional brain networks to analyse 4D signal located in gyri and sulci. Their results suggest that gyri and sulci could work together in a core-periphery network manner. These findings have shed new light on our fundamental understanding of the brain’s basic structural and functional mechanisms.
- 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 topic of structure-function coupling is important in the field.
-
Being the first work to analyse the relationship between spatial-temporal feature located in gyri and sulci.
-
Presenting some findings of core-periphery organization in brain network.
-
- 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 related works of spatial-temporal Transformer are not mentioned. Why you need to propose a new model, twin Transformer, to capture the spatial-temporal feature?
-
No color bar in Fig 3. Do those heat maps have the same scale level? If so, the five subjects are not very consistent under the same task.
-
Although you tested three thresholds and get the same conclusion, is it possible to test more thresholds?
-
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
No
- 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
See weakness.
- 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?
Good paper overall.
- 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 the reviewers for their valuable comments and their recognition of the strengths in our work:
- Innovative question about the relationship between gyri and sulci (R1, R4, R5).
- The advanced method for analyzing the spatial-temporal features of fMRI data (R1, R4).
- The importance of the topic and the significant impact of the conclusions in the field (R4, R5). Here, we address the main concerns. We will incorporate all other suggestions in the revision and publish our code (R4).
Main questions:
More fMRI signals on gyri than on sulci. Discuss potential differences in signal-to-noise ratios (SNRs) between gyri and sulci and their impact on the results (R1). We will add a discussion on this. The potential different SNRs do not influence our conclusion since we set the same threshold for gyri and sulci and we report the experimental results under different thresholds in the paper.
Use task fMRI instead of resting fMRI (R1). For task fMRI, we have ground truth of activated brain areas to verify and support our conclusions. However, in resting fMRI, activated brain regions may vary across different subjects.
The hyperparameters of their networks. (R1/R4). Due to MICCAI policy, only figures, tables, and proof of equations are allowed in the appendix. We will list the hyperparameters appropriately in the paper in accordance with the MICCAI submission guidance. We will also release/publish our code upon acceptance.
Lack of experiments (R4) / Test more thresholds (R5) MICCAI rules prohibit new experiments.
More explanation of technical details (R4). Due to the rich content and MICCAI’s policy regarding appendices, we did not include all technical details. However, we will address this concern appropriately by publishing our code.
Visualizations of functional brain networks (FBNs) (R4). Please see the appendix since we presented the visualizations in the appendix.
The advantage of the proposed Twin-Transformer over traditional methods (R4). As recognized by Reviewers 1 and 5, our method can simultaneously extract temporal and spatial features from fMRI data, and the design of the Twin-Transformer can process signals from gyri and sulci concurrently. Additionally, the Transformer architecture, with its self-attention mechanism, has been well-proven in the machine learning community to have better feature extraction ability than traditional methods such as PCA, dictionary learning, and sparse coding.
Related works of Twin-Transformer (R5). We will include a discussion of related works on spatial-temporal Transformers.
The color bar in Fig 3 (R5). Thanks for your reminder. We will add the color bar next to the figures. Not all heat maps have the same scale.
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’
All reviewers are convinced. It also addresses interesting neuroscientfic question.
- 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).
All reviewers are convinced. It also addresses interesting neuroscientfic question.
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’
The authors’ rebuttal has comprehensively addressed reviewers’ concerns within the limited space.
- 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 authors’ rebuttal has comprehensively addressed reviewers’ concerns within the limited space.