List of Papers Browse by Subject Areas Author List
Abstract
Functional magnetic resonance imaging (fMRI) analysis models the detected temporal signals as a superposition of linear hemodynamic responses (HDR) to task-related stimuli, yielding spatial maps of brain function. However, recent studies have demonstrated that neural responses exhibit significant nonlinearity, challenging the validity of such linear models. In this work, we propose a novel mathematical framework, Regional Synchronization based on Graph Eigenmodes (RS-GEm), to analyze fMRI data and localize brain activation without relying on the linear assumptions of traditional models. Using Laplacian Eigenmaps (LEM), we capture the graph structure of the brain and derive its eigenmodes. These eigenmodes characterize possible spatial organizations of neural activity across different hierarchical levels of the human brain. By computing the regional synchronization of fMRI signals embedded in the eigenmode space and employing clustering metrics, we extract task-relevant eigenmodes to identify task-evoked activation regions. Validations on the Human Connectome Project (HCP) dataset demonstrate that our method can map task-evoked brain activations without the linear assumptions. The proposed approach offers a novel methodological framework for elucidating understudied aspects of brain function featured with nonlinear HDRs, thereby facilitating a more complete understanding of brain dynamics.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1638_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{TanZhe_Brain_MICCAI2025,
author = { Tang, Zhenyu and Zhao, Yu and Yang, Yang and Liu, Xiaoyu and Su, Jingyong},
title = { { Brain activation mapping based on Regional Synchronization of fMRI signals embedded in Graph Eigenmodes } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15971},
month = {September},
page = {119 -- 128}
}
Reviews
Review #1
- Please describe the contribution of the paper
As linear models fail to capture the nuances of the neural responses to task related stimuli, the authors propose a mathematical framework using Graph Eigen Modes to capture graph structure of hierarchical regions of the brain. To identify task evoked activation maps of the brain, regional synchronization of the fMRI signals were computed from the embedded eigenmode space using different clustering metrics.
- Please list the major strengths of the paper: you should highlight 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.
On the whole, the method described is simple and elegant.
The identification of relevant voxels/regions via synchronization metric is well expounded. The decomposition of the signal into task-wise eigenmodes is neat.
The clear dominance of a subset of eigenmodes associated with the movement of different limbs is extremely interesting.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The major weakness I see is that the results presented in this paper is not put in context of other works. HCP itself has extensive documentation of task-activation for different task conditions. How do your results compare with them?
Figure 2D shows the functional map with GLM, but how does that compare with your result?
What are the benefits of your method over existing methods?
- 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.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Please respond to the questions listed under weakness.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper describes a methodology for generating maps representing contributions to observed fMRI brain signal under a task condition. These maps are derived by measuring the synchrony of brain vertices relative to their functional and spatial neighbors. The resulting maps under a motor condition are consistent with known motor mapping locations.
- Please list the major strengths of the paper: you should highlight 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 major strength of this paper is its novel approach to deriving eigenmaps of brain signal and clustering vertices based on the synchrony of the eigenmaps. The resulting maps look similar to well established motor mapping locations. The maps also appear to have less noise compared to a GLM approach. The authors clearly did a lot of work to generate their proposed framework.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The validation appears to be mainly based on qualitative visual inspection. The methodology requires many steps and the rationalization for any given step as well as the interpretation of the outputs of any given step are not always clear.
- 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 provide sufficient information for reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
What is the modified spatial filter that was used? (page 4) Is Xwi a r by n matrix where r is the number of neighbors and n is the number of timepoints? Do you compute a different Xwi for each eigen mode? What is the intuition behind what Xwi represents? Equation 5 shows clustering score to be a scalar value that is summed over all the vertices, but Figure 1F shows clustering score being used to generate an activation map. How do you sum over all the vertices and then get a vertex wise measurement as the output? Is 1F just showing the eigenmap with the greatest clustering score? How are the activation maps in Figure 2B computed? Are those the clustering scores using only the data for a given task block for all 100 unrelated subjects? Do you have the copyright to Figure 2C? Small comments: I found the use of the term “voxels” to describe vertices confusing. Voxels imply volumetric elements were used, while the methodology appears to only use data that has been mapped onto mesh vertices. Please correct me if I am wrong and voxels were indeed used in the analysis.
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The analysis method appears to be novel, and the results look promising. The paper is interesting as it is and this may be enough to get it accepted. However, the intuition behind each step in the methodology is confusing. The utility appears to be limited to generate group activation maps.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper introduces Regional Synchronization based on Graph Eigenmodes, an unsupervised learning method to derive brain activation patterns from fMRI data without relying on the traditional linear hemodynamic response model. The proposed approach applied graph Laplacian eigenmodes on functional connectivity networks to characterize spatial synchronization among voxels. By computing eigenmode-specific regional synchronization and using a clustering score to identify task-relevant modes, RS-GEm generates task-specific activation maps. The approach is evaluated using motor task fMRI data from the Human Connectome Project.
- Please list the major strengths of the paper: you should highlight 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 approach made use of a data-driven, model-free approach that is a fresh alternative from generalized linear models traditionally used in fMRI data analysis.The proposed use of Laplacian Eigenmaps, regional synchronization via PCA are well-motivated.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- The primary issue with the paper is an overly strong claim on the proposed approach’s advantage over GLM. For example, this claim in page 7 stood out “The results show that, compared to the GLM, the RS-GEm method demonstrates superior spatial consistency between test and retest data, underscoring its enhanced robustness and reliability in identifying neural activations.”. The only support for this claim is the visual comparison offered in Fig 2D, which itself raises questions such that (1) are both the plotted RS-GEm and GLM make use of 59k surface template and are they group-level or individual-level activation patterns? (2) if so, the RS-GEm result seems to be a lot more “smoothed-out”, which is understandable given RS-GEm seems to intrinsically optimize for smoothness of spatially neighboring voxels. As such, on a cursory glance, Fig 2D might give the impression of “reliability” rather than a smoothness artifact. If the GLM maps are smoothed to the same level as RS-GEm maps, would RS-GEm still quantitatively yield better test-retest consistency? (3) There are activations in the parietal lobes present in the GLM results that are not at all present in RS-GEm. Can the authors comment on this?
- In a similar vein, there is no rigorous comparison of the proposed method with GLM and existing atlases. For example Fig 2B can be updated to show the extent of overlap, quantitatively and qualitatively with traditional GLM results and other Motor maps. Fig 2A and 2C are rather redundant and yield little usefulness in assessing the results.
- The evaluation is restricted to motor tasks only. How good are the proposed method when applying on other cognitive tasks (e.g., working memory, language)? Similar to point 2, quantitative comparison would be compact and helpful to evaluate the generalizability of RS-GEm beyond one single task. Interpretation of the eigenmodes: can the authors please clarify what the scale in Fig 3B is referring to (-0.2 to 0.2)? Are there a consistent set of eigenmodes across all tasks in the HCP dataset? Visualizing this set of eigenmodes (if they are of a manageable numbers, e.g. <= 20) and how they share and diverge across tasks would be a great way to interpret the eigenmodes.
- Definition of the clustering score: how are the neighbors defined? For example, are they the immediately adjacent vertices on the surface template?
- 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 provide sufficient information for reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The lack of rigorous comparison with other methods makes the paper’s strong claim rather unsubstantiated.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
Thank all reviewers for your positive feedback and valuable comments. We will incorporate your suggestions in the final version. Our responses are presented as follows.
R1-Q3, R1-Q1/R4-Q3: Novelty and Task Selection We propose a new analytical paradigm that does not rely on traditional linear model assumptions, allowing us to capture nonlinear brain activations and offering a novel perspective for fMRI research. We have conducted experiments on the emotion task and also observed activations in relevant regions. However, we only report the results of motor tasks due to page limit. The motor tasks have specific activation regions, making them ideal for validating our method.
R3-Q1, R4-Q2, R4-Q3: Qualitative and Quantitative Results Currently, due to individual variability and complexity of brain functions, there is no baseline for brain activation at the individual level. Thus, this study does not employ specific quantitative metrics but instead uses activation maps to visually demonstrate accurate detection. As suggested, quantitative metrics such as overlap rates with functional atlases will be explored to further validate the effectiveness in future. We believe both qualitative and quantitative analyses will provide a more comprehensive evaluation of our method.
R1-Q2, R4-Q1: Comparison with GLM The comparison is conducted by analyzing the performance of our method with GLM in detecting activations in task-relevant regions across test-retest datasets. We evaluate the consistency of results across sessions. The results have shown that our method performs better in terms of consistency and activation region. The comparison with GLM primarily serves to validate the effectiveness of our method and to highlight its potential advantages in detecting activation patterns. Our goal is not to directly replace GLM but to propose a novel analytical framework designed to explore nonlinear activations.
R3-Q2: Rationality and Explanation of Methodological Steps We will include explanations for the construction of key steps and further elaborate on their physiological significance in the final version.
R4-Q3: Explanation of Eigenmode Scale The primary purpose of the scale in Fig. 3 is to clearly visualize spatial patterns of eigenmodes, facilitating intuitive observation of functional region distributions within the modes. The absolute values themselves do not hold specific physiological significance.
R4-Q4: Definition of Clustering Score Neighborhood The neighborhood is defined as spatially nearby vertices on the surface template.
R3-Additional comments: The modified spatial filter was designed by calculating functional connectivity strength between a voxel and its spatial neighbors. This approach helps reduce interference from functionally unrelated signals in neighboring regions. For Xwi, it is a r × n matrix, where r is the number of neighbors and n is the number of timepoints. A separate Xwi is computed for each eigenmode, and the intuition behind Xwi is that it enhances local signals so that signals from voxels with similar functions are also amplified in the corresponding eigenmode. The clustering score is calculated for each synchrony map, and the most significant synchrony maps are selected. These maps are then averaged to produce the final activation maps. The activation maps in Fig. 2B were computed using individual-level data. Lastly, the term “voxels” was used to emphasize the spatial nature of the data but essentially refers to vertices in brain surface representations. No volumetric voxel data was used in the analysis.
We sincerely thank all the reviewers once again for your valuable insights, which will greatly help to improve our paper.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A