List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Chloé Gomez, Antoine Grigis, Lynn Uhrig, Béchir Jarraya
Abstract
Decoding the levels of consciousness from cortical activity recording is a major challenge in neuroscience. The spontaneous fluctuations of brain activity through different patterns across time are monitored using resting-state functional MRI. The different dynamic functional configurations of the brain during resting-state are also called “brain states”. The specific structure of each pattern, lifetime, and frequency have already been studied but the overall organization remains unclear. Recent studies showed that low-dimensional models are adequate to capture the correlation structure of neural activity during rest. One remaining question addressed here is the characterization of the latent feature space.
We trained a dense Variational Auto-Encoder (dVAE) to find a low two-dimensional representation that maps dynamic functional connectivity to probability distributions. A two-stage approach for latent feature space characterization is proposed to facilitate the results’ interpretation. In this approach, we first dissect the topography of the brain states and then perform a receptive field analysis to track the effect of each connection. The proposed framework instill interpretability and explainability of the latent space, unveiling biological insights on the states of consciousness.
It is applied on a non-human primate dataset acquired under different experimental conditions (awake state, anesthesia induced loss of consciousness).
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16431-6_40
SharedIt: https://rdcu.be/cVD6W
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #3
- Please describe the contribution of the paper
VAE applied to dFC
- 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 further associated the encoded information with different brain states
- 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.
No quantitative comparison to other methods
- 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
no concern
- 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
-
It seems reasonable by visualizing how different the consciousness state and how different brain states separated in the embedded space. But what is missing here is some quantitative analysis
-
Why did the authors choose to apply VAE to dFC? There are a large amount of debates regarding the sliding window based estimate, e.g. window size, reproducibility, etc. Is it possible to work on the original time series?
-
- 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?
A solid paper but missing some quantitative results
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
2
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #4
- Please describe the contribution of the paper
In this paper, the authors proposed a dense variational auto-encoder (dVAE) to generate low-dimensional representations which map dynamic functional connectivity to probably distributions. The dVAE model was validated on dataset with five rhesus macaques in different arousal states. A connection-wise receptive field (FR) analysis is then to visualized and interpret encoded trajectories between states of consciousness. The VAE-VIENT framework provides a complete definition of the latent space in terms of wakefulness status and dynamic brain trajectories.
- 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 provides an excellent framework to instill interpretability and exploitability of the latent space, which unveil the biological insights on the states of consciousness. In this model, receptive field analysis leverages the regions to perturb for inflecting trajectories and perhaps restoring wakefulness. The VAE framework for Visualizing and Interpreting the ENcoded Trajectories (VAE-VIENT) is then used to describe the latent space dynamic.
- 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.
Firstly, defining more equations are necessary, such as the loss function for dVAE, receptive field analysis. Secondly, the paper introduces too many contents but fails to explain them clearly. And the writing of the paper is poor. For example, the order for the subfigures of Fig.3 is a mess and Fig. 4 have few explanations in context. Thirdly, the sample size and the splits of training and testing dataset are not clear. 5-fold cross validation is used on training dataset. What is the test set?
- 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 paper meet the crieria on the checklist.
- 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
Firstly, defining more equations are necessary, such as the loss function for dVAE, receptive field analysis. Secondly, the paper introduces too many contents but fails to explain them clearly. And the writing of the paper is poor. For example, the order for the subfigures of Fig.3 is a mess and Fig. 4 have few explanations in context. Thirdly, the sample size and the splits of training and testing dataset are not clear. 5-fold cross validation is used on training dataset. What is the test set?
- 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The novelty of this paper focuses on the description of the latent space. However, more detailed information about the methods and experiments are required.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #5
- Please describe the contribution of the paper
This paper proposes characterization of brain activity patterns across states of consciousness based on variational auto-encoders, via analysis of the resulting low-dimensional latent feature space. The proposed method is experimented on a primate dataset and the results show that it can reconstruct the average pattern of each brain state with high accuracy, generate a latent feature space stratified into a base of brain states, and reconstruct new brain states coherently and stably.
- 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.
- Writing is generally concise and clear.
- The topic is of interest to the sub-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.
- Contributions of the paper are unclear. It seems the dVAE has already been applied to generative embeddings of brain collective dynamics in [21]. The authors should be more specific about their contributions.
- Fig. 2: the performance on state 2 is bad compared to others. Please comment.
- Experiments: “…, we select the trained weights associated with fold 3.”: given similar performance with fold 2 and fold 3, why the latter is selected?
- Effect of the β regularization parameter: please provide more details about the grid (interval, log scale if employed, etc.). It is better to also present the search results.
- More description and explanation are needed for Fig. 4(B).
- The manuscript does not follow the MICCAI format, e.g., the first lines of paragraphs are not indented in Introduction.
- 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
Reproducibility seems OK.
- 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
- Page 2 “A small number of animals was investigate as it is advised in nonhuman primate studies.”: investigate -> investigated
- Page 5 bottom “The highest the difference, …”: highest -> higher
- 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 contributions are unclear.
- Number of papers in your stack
6
- 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
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 paper uses a dense VAE-based model to learn a low-dimensional feature representation of functional connectivity. The application is interesting and some aspects of the paper seem novel. The proposed method promotes interpretability of the latent space. Authors are required to address the missing technical details and clarification questions (including clearly stating the paper’s contribution and improving figures and their descriptions) raised by the reviewers. Authors are also encouraged to present quantitative description of the results and how these results can be contrasted/compared with existing methods. The camera-ready, if accepted, should follow the MICCAI format.
- 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).
3
Author Feedback
We thank the reviewers for their insightful feedback on the paper.
We agree with reviewer #3 that no quantitative comparison to alternative methods is offered. However, as specified by reviewer #5 and stated in section 2.2 of the paper, the implemented dVAE architecture was described in [21] and validated in data acquired from healthy volunteers during wakefulness and sleep. We purposely applied this validated model, focusing our approach on a new framework to instill latent space interpretability: the VAE-VIENT framework described in section 2.4. As suggested by reviewer #5, we would like to summarize the original contributions of our work, which are three-fold: 1- the framework VIENT pathway introduces a simulation setting to perturb connections, allowing dynamic characterizations of the latent space in terms of induced trajectories. It brings new insights into the existing methods by adding to the static modeling a dynamical description. 2- the introduction of novel descriptors like dense inclination spreading, velocity, and confidence maps. 3- as far as we know, this is the first application of dVAE to non-human primate functional neuroimaging, with a specific application to the signatures of consciousness. We believe the clinical and scientific applications are numerous. First, this approach allows the description of new biomarkers of consciousness and anesthesia-induced loss of consciousness. Furthermore, it is a unique tool to simulate the consequence of targeted modulation of specific brain regions to restore consciousness loss in disorders of consciousness.
We agree with reviewer #3 that there is a debate about whether to work directly with time series or with sliding windows-based dFCs. The latter introduces hyperparameters that are not always easy to optimize. Nevertheless, from a methodological aspect, we believe that working with sliding windows acts as a natural augmentation scheme, helping during the training on our limited dataset (5 monkeys - 156 runs - 72384 dFCs). From a neuroscientific angle, we aim to capture dynamic configurations of the brain initially described with dFC (see section 1 or [1, 2, 3]). dFC patterns are effective in capturing strong signatures of consciousness. Interestingly, identical conclusions were drawn in humans, using a phase-based dynamic functional coordination analysis, suggesting a low bias (if it exists) induced by sliding windows [9].
As highlighted by reviewers #4 and #5, some technical details are missing, and some figures request further explanations. To study the effect of the β regularization, we use an [0.5, 20] interval (as depicted in section 3.1) with the following custom steps [0.5, 1, 4, 7, 10, 20]. As presented in section 3.1, monitoring the reconstruction quality with the structural similarity (SSIM) gives relatively stable performances except when β is over 10 (10% drop in the metric). The classification accuracy promotes the use of β=7 or β=10 (13% drop in the metric otherwise). As for the cross-validation, we forget to mention that a leave-one-subject-out strategy enables the creation of the test set. Stratification further enforces class distribution in each training split. We optimize the ELBO objective coupling mean-squared error loss and the β-weighted Kullback-Leibler divergence. To clarify figure 4, in subplot A, the proposed connection-wise receptive field analysis is applied to one dFC matrix (left). The covariance structure of the resulting latent variables presents an ellipse shape (middle). The resulting trajectories follow lines parametrized by their mean inclinations in degrees. The standard, max, and median inclination errors are plotted for each connection (right). In subplot B, we focus on two specific connections (top and bottom). We display inclinations across the whole latent space (left), and associated standard error (right). The line hypothesis holds for most latent locations, and besides, the inclination map appears smooth and regular.
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.
Authors have addressed missing technical details and clarified the paper’s contribution. Authors are encouraged to add these clarifications to the camera-ready version.
- 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 #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 original contributions of the work is clear (VAE-VIENT framework, use of novel descriptors such as dense inclination spreading, velocity, and confidence maps, and application of dVAE to non-human primate functional neuroimaging). The rebuttal address the technical details that the reviewers asked for (e.g. effect of beta regularization, description and explanation for some of the figures, etc.)
- 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).
NR
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 authors point out the main contributions more clearly in the rebuttal. With this update in the paper and the additional technical details provided, I think this paper can be accepted.
- 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