Paper Info Reviews Meta-review Author Feedback Post-Rebuttal Meta-reviews

Authors

Yanshuai Tu, Duyan Ta, Zhong-Lin Lu, Yalin Wang

Abstract

The mapping between visual inputs on the retina and neuronal activations in the visual cortex, i.e., retinotopic map, is an essential topic in vision science and neuroscience. Human retinotopic maps can be revealed by analyzing the functional magnetic resonance imaging (fMRI) signal responses to designed visual stimuli in vivo. Neurophysiology studies summarized that visual areas are topological (i.e., nearby neurons have receptive fields at nearby locations in the image). However, conventional fMRI-based analyses frequently generate non-topological results be-cause they process fMRI signals on a voxel-wise basis, without considering the neighbor relations on the surface. Here we propose a topological receptive field (tRF) model which imposes the topological condition when decoding retinotopic fMRI signals. More specifically, we parametrized the cortical surface to a unit disk, characterized the topological condition by tRF, and employed an efficient scheme to solve the tRF model. We tested our framework on both synthetic and human fMRI data. Experimental results showed that the tRF model could remove the topological violations, improve model explaining power, and generate biologically plausible retinotopic maps. The proposed framework is general and can be applied to other sensory maps.

Link to paper

DOI: https://doi.org/10.1007/978-3-030-87234-2_60

SharedIt: https://rdcu.be/cyl86

Link to the code repository

N/A

Link to the dataset(s)

N/A


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper introduces a method for mapping visual input at the retina to the brain’s cortex for interpreting fMRI data. The outstanding characteristic of the proposed methodology is that the mapping is topology-preserving, both in terms of the large area covered by the visual field as well as sub-regions (extracted from a topology prior). The model was verified against a gold standard using simulated data, and demonstrated in the Human Connectome Dataset retinotopy dataset. Both of these instances verify that the method indeed works as designed.

  • 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 method is innovative and includes a comparison to other sensible approaches used in the interpretation of fMRI data.
    • The paper describes theoretical underpinnings, which makes the method transparent in the sense that that humans are able to understand how it works. This is in contrast to perhaps having used the Connectome data to train a black-box deep network enforcing some sort of topology loss function. Perhaps this will be the direction eventually, but the initial understanding of the problem feels valuable at this point.
    • There is a clear gain in terms of performance compared to registration and in the sample dataset (and also with respect to the other mapping techniques). The method’s superior performance is appreciable both visually and using the metrics assigned to quantify the differences.
  • 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.
    • It would had been nice to see it the method accomplished a task beyond working as it should, such as classification of quantification of amblyopia patients and controls or the like.
    • The author’s could had made the paper more succinct and provide us with some conclusions indicating the limitations or any potential downsides of the method.
  • 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

    It is not immediately obvious that another group of experts could readily reproduce this paper. The method was quite involved, even including an additional Supplementary Information page (which I have not seen in MICCAI—then again, I have only been participating in this conference for only a handful of years). The software packages, languages, and hardware used are not clearly listed. Some of the data used is available online, and the variables are clearly defined, which is good. But, again, it is hard to say how difficult it would be to reproduce this method and the results.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

    This manuscript is a very good methodological paper with a logical theoretical foundation, verification on synthetic data, and demonstration on real-word measurements. Although a bit long winded and missing a Conclusion, the paper is clear and well-organized. The mathematical variables are defined and given real values. The figures and tables are easy to read (colors and fonts stand out). The Results include the desired characteristics in the ideal setting, a description of other sensible approaches (including registration, which makes the most sense to this reviewer), and details on why the numerical results are important and challenging to obtain if it were not for the method.

    This paper can improve by making some details more succinct (in the Introduction and Method Sections) and provide some discussion ideally supporting the concluding ideas in the Abstract. That is, why exactly is meant by improving ‘model explaining power’ and ‘biologically plausible’? It seems that this simply means that the prior is preserved in the end. But then, by design, of course it is. If there are any specific deviations from the prior that may serve as clinical markers, it seems that the proposed method would enforce its built-in normal condition, lessening the perceived effect of the clinical condition. Perhaps this can be done by tuning the method, but then this would require some a priori information on what one is looking for. This seems like a major problem if the ultimate goal is to monitor the progress of a condition. This is not necessarily a bad thing, because the method is now a tool to help sorting all this out. But the authors should had mentioned this in a Conclusion, containing also other specific ideas as to how to use their new tool.

    The study would had been stronger if the proposed method would be had been used to support a neurophysiological conclusion, perhaps including controls and patients, or in an auditory map (if the point is to demonstrate the method generalizes). One can perhaps extrapolate the usage and generality of the methods, but actual evidence is superior than relying on the imagination of the readers. As an additional minor note, the differences provided in Fig. 4 are quite stark, to the extent that one wonders if smoothing was implemented correctly (the sign changes immediately superior of the 90 degree zone are particularly worrying). If the results in Fig. 4 indeed reflect the state of the art, then other studies may be making serious mistakes on the interpretation of these type of results. Again, it would had been nice to be shown what these potential mistakes may be, or what exactly one can gain by adding a topology preservation step.

    The points above feel minor, and this is a very good paper, which nevertheless did not qualify exactly as excellent.

  • Please state your overall opinion of the paper

    Probably accept (7)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Adding topological soundness to retinal-cortical fMRI makes intuitive sense, and this group has done it, although (from the paper) it is hard to go beyond intuition as justification.

  • What is the ranking of this paper in your review stack?

    2

  • Number of papers in your stack

    4

  • Reviewer confidence

    Confident but not absolutely certain



Review #2

  • Please describe the contribution of the paper

    The paper proposed a topological receptive field (tRF) framework that decodes retinotopic mapping from noisy fMRI signals with topological constraints. The paper tested the method on both synthetic data and real human fMRI data.

  • 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. The paper proposed a method that introduces topological constraints for computing retinotopic mapping from fMRI signals, which avoids topological violations in retinotopic mapping.
    2. The paper used synthetic data to validate the method. Given no ground truth is available for fMRI data, using synthetic data gives readers confidence in the method.
    3. The paper compared their method to several widely used retinotopic mapping methods. Their method outperforms others in terms of the proposed evaluation metrics.
  • 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 proposed method relies on registering a prior segmentation atlas to the subject’s parametrization disk. The authors could discuss the effect of misregistration on their results and whether such process weakens individual difference.
    2. The authors could discuss how a zero topological violation in retinotopic mapping benefits (or hurts) neuroscience research (depending on different applications). Similar to point 1, a hard constraint may cover useful information in retinotopic maps. In other words, some downstream applications of retinotopic mapping can be included as evaluation methods for the baseline and proposed methods.
  • 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

    Public data were used. Method description is clear. As long as source code is released after acceptance, the reproducibility of the paper should be high.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
    1. Description of black lines in Fig. 3 would be beneficial.
    2. Though I get the general idea of Tn in Table 1 and 2, I cannot find a clear definition of Tn in the paper.
    3. Please see Question 4 for more comments.
  • Please state your overall opinion of the paper

    accept (8)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The proposed framework that incorporates topological constraints into retinotopic map decoding is innovative and significant.

  • What is the ranking of this paper in your review stack?

    2

  • Number of papers in your stack

    6

  • Reviewer confidence

    Confident but not absolutely certain



Review #3

  • Please describe the contribution of the paper

    This paper introduces a method that projects a 3D surface onto a 2D disk space such that neuroscientific knowledge (topological organization) can be incorporated into the RF modeling process and regularize the visual maps.

  • 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 main strength of this paper is to propose a method that creates more reasonable early visual cortical maps. The key point here is to include two aspects of neuroscientific knowledge (1) the visual coordinates are topological with respect to the surface parameterization within each area; (2) the areas hold the same organization.

  • 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.

    As a neuroscientist, I did not see clear drawbacks of this method. I did not quite get the math part. But I only have one question. Does this method attenuate individual differences in cortical maps as strong priors are included?? Can you quantify this? My point here is that a regularizor is certainly good but there must exists trade off between smoothness versus uniqueness. I am not sure whether this methods can be used to reveal some individual or even group differences (e.g., patients versus healthy controls)

  • 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

    I am not sure why the authors marked traning code and evaluate code as “not applicable”. If I understand it correctly, the codes should be published along with this paper.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

    This paper introduces a method that projects a 3D surface onto a 2D disk space, and a new mathematical method that incorporates the neuroscientific knowledge (topological organization of the human visual cortex) into the RF modeling process and regularizes the visual maps.

    As a neuroscientist focusing on the applications, this method looks promising as it does create nicer retinotopic maps as compared to conventional methods. I understand the long-standing issue when we project data from the volume space to the surface space or even transform between different forms (i.e., inflated/flattened) of the surface space. All these transformations will somehow distort data.

    To be honest, I did not quite get math of the Beltrami equation part. This part should be evaluated by other experts.

    Another good point of this paper is that the authors evaluate their results on a large dataset (i.e., HCP 7T database).

    I only have one question regarding its clinical applications. Does this method somehow reduce the individual differences of visual cortical maps after you add strong regularization? This is important for clinical applications, especially for the scenarios where we want to quantify some metrics of visual areas, e.g., surface area. If I understand it correctly, there exists a trade-off between homogenous versus individually unique maps.

  • Please state your overall opinion of the paper

    borderline accept (6)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    overall model performance

  • What is the ranking of this paper in your review stack?

    1

  • Number of papers in your stack

    1

  • Reviewer confidence

    Not Confident



Review #4

  • Please describe the contribution of the paper

    The paper proposes a topological receptive field (tRF) model that can reveal retinotopic maps in visual cortex from fMRI signals. The Beltrami coefficient is used to enforce the resulting retinotopic map to be topological within each visual area. The performance of the proposed method is evaluated on synthetic data and a 7T fMRI dataset.

  • 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.
    • Novel solution to an important problem
    • solid methods
  • 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.
    • computational cost not discussed
  • 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

    Study cohort not explained although it was checked as [Yes] It would be valuable if authors chose to release the code upon acceptance

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
    • p.2: The term “The topological condition” is used a lot here, but only gets a proper definition in Section 2.4. It would help flow & understanding if the authors gave a more specific definition up front, especially since the term appears often throughout the paper
    • p.3: “The pRF predicts perception parameters” - not sure what is meant here by “perception”. pRFs are usually fit on neural activity, so there is no perceptual task
    • p.3: “Given the receptive field model […], the predicted fMRI is:” - what is s in this equation? Also, this is the predicted BOLD response, not the “predicted fMRI”
    • p.4 Please give a ref for visual field sign - I believe it is Warnking et al. (2002)
    • p.7 Please briefly describe the fMRI dataset
    • p.8 Please add a Conclusion paragraph
  • Please state your overall opinion of the paper

    Probably accept (7)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The paper presents a novel solution to an important problem. The solution is well motivated and the math is rigorous (though I’m not 100% certain since this is not my main field). Although the qualitative improvement can be appreciated in Fig. 4, the raw numbers in Table 2 suggest there is only an incremental improvement compared to the “vanilla” pRF approach. So I am wondering how much more computationally expensive the proposed method is compared to the other methods. A Discussion/Conclusion section is missing, but I think the paper would benefit from a benefit of the advantages/disadvantages of the proposed method.

  • What is the ranking of this paper in your review stack?

    2

  • Number of papers in your stack

    6

  • Reviewer confidence

    Somewhat confident




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 proposes an innovative method that creates more reasonable early visual cortical maps. There is a clear gain in terms of performance compared to registration and in the sample dataset (and also with respect to the other mapping techniques).

  • 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).

    3




Author Feedback

N/A



back to top