List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
William Consagra, Martin Cole, Zhengwu Zhang
Abstract
This work considers a continuous framework to characterize the population-level variability of structural connectivity. Our framework assumes the observed white matter fiber tract endpoints are driven by a latent random function defined over a product manifold domain. To overcome the computational challenges of analyzing such complex latent functions, we develop an efficient algorithm to construct a data-driven reduced-rank function space to represent the latent continuous connectivity. Using real data from the Human Connectome Project, we show that our method outperforms state-of-the-art approaches applied to the traditional atlas-based structural connectivity matrices on connectivity analysis tasks of interest. We also demonstrate how our method can be used to identify localized regions and connectivity patterns on the cortical surface associated with significant group differences.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16452-1_27
SharedIt: https://rdcu.be/cVRY9
Link to the code repository
https://github.com/sbci-brain/SBCI_Modeling_FPCA
Link to the dataset(s)
https://db.humanconnectome.org
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a spatially-continuous connectivity model for white-matter connections, to analyse dMRI data for possibly salient connections. The method does not need pre-existing atlases of connectivity, giving it more freedom to find new relevant inter-region connections.
- 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 exposition is good, the method is well-motivated and laid out, and the writing is clear (up to where the math lost me around theorem 1, but that is usual for me). So the paper is a pleasure to read. The approach is creative and compelling, one of those papers that make you smile and say “wow, that’s neat”. The removal of the atlas constraint is valuable.
- 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.
In the results (Fig 1), it’s unclear to me if the higher rate of “significant” discoveries is simply due to CC’s higher resolution, or if CC has more significant findings at the resolution of current methods.
- Please rate the clarity and organization of this paper
Excellent
- 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 indication in main paper about reproducibility.
- 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
Methodology: -Would there be any advantage to defining U_i as bounded (vs potentially infinite)? In practice of course it is bounded. -Is the Borel-measurable a required condition? All the regions E_i are closed sets in practice. If not required, why is it mentioned? -“for any pair (omega_1, omega_2)”: Is this a point, or a small discrete region with dimension dx, dy (eg a pixel)? -Re “underlying random function”: Is it practical to encode known biophysical connectivity information about the cortex as a prior (a function with pre-defined shape), instead of assuming no information? -I was not clear about the interaction between the non-discrete assumptions in the math and the discrete reality of the data: Is the infinite dimensional, continuous surface assumption necessary to develop the method? If no, would it be clearer to define the problem over discrete space (eg a certain sized voxel/pixel)?
Reduced-Rank Embedding: “infinite dimensionality of the continuous connectivity”: Is this due to a continuous (ie not discretized) assumption on the surface? If yes, is the continuous assumption necessary? Definition of V_K: Is this effectively imposing a discretization on the 2-spheres?
Misc items: “assume the U has been centered”: you could just state this, plus “WLOG” Fig 1: Bigger axis labels would help readability Define “HCP” Square brackets are clearer for references.
A few typos: -diffomorph -> diffeomorph -for for -> for -it’s -> its -analysis are known -> analyses are known -analysis N -> analyis of N -Between subject -> add hyphen, or say “Inter-subject”
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The creativity of the proposed solution to the atlas problem. The clarity of the exposition.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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
Review #2
- Please describe the contribution of the paper
This paper proposes to use low-rank embedding of continuous random function to represent the connectivity intensity matrix of white matter fibers. The paper is written in excellent English. The introduction part is educative and clearly summarizes the limitations of state-of-the-arts methods. The methodology section gives a good mathematical description of the algorithm, and the experiments clearly demonstrates the merit of this method.
- 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.
Compared to the exiting methods, this work extends the classical discrete representation to continuous space and develops mathematics schemes to solve the computation task. Experimental results prove the advantage of this new method in group-wise inference from SC data as well as in localizing group differences to brain regions and connectivity patterns on the cortical surface. Overall, this paper contributes a novel method to the field of brain functional connectivity analysis.
- 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 method and experiment sections needs a few more details, as I list in the detailed comments below.
- 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
The method is novel, the availability of the author’s original code may greatly help with the reimplementation. The authors admits promised the availability of the code, but I did not see the download link in the paper or in the supplementary file.
- 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
- The choice of algorithm parameters needs more explanations, e.g., alpha 1, alpha 2 in page 6 and the thresholds for subnetwork discovery in page 7. Why did you choose those specific values? Is the algorithm sensitive to the parameter choice?
- Page 6, from the implementation aspect, you gave the computer configuration but still need to report the computation time.
- Page 7, $200, $40 and $160 seems typos or latex error.
- Page 8, the color bar of fig.2 B is too small and numbers on it are vague. Please enlarge it.
- Page 6, ‘the fours panels’ should be ‘the four panels’.
- 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?
The methodology novelty is good.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
6
- [Post rebuttal] Please justify your decision
The rebuttal further clarified the novelty of the paper. I suggest acceptance of this paper. It has good contributions in methodology.
Review #4
- Please describe the contribution of the paper
This work theoretically analyzes brain structural connectivity as a continuous function.
- 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). In this work, the authors comprehensively employ real analysis and functional analysis to prove the brain structural connectivity as a continuous function. 2). This paper is written well.
- 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). Topic Issue From the reviewer’s perspective, this work concentrates on real analysis, functional analysis, and statistical Theory to discuss the continuous brain structural connectivity. This is a theoretical research work that probably does not match the topics of MICCAI. Furthermore, there are fewer medical, clinical translational, and imaging analytics provided in this work. Therefore, the reviewer would suggest submitting this work to NeurIPS, ICML, or COLT. 2). Inconsistent Format This paper is organized inconsistently. At first, the reference citation is not consistent with the original template. The authors utilize (1) as a citation of reference #1, but this results in the confusion as an equation (1). Furthermore, the Bibliography should be replaced as a Reference. There are many inconsistencies in the format of this work. 3). Theoretical Issues/Questions Although authors provide many theoretical analytics to brain structural connectivity, some theoretical issues lead to confusion among reviewers. At first, the authors proposed the concept as Borel measurable regions, but the authors utilized these regions to denote a Lebesgue calculus. What is the definition of Boreal measurable regions? Are these regions could be covering via Boreal sets? Or what is the relation between Lebesgue and Boreal measure? Why cannot authors adopt the Lebesgue measure to denote these regions? Unfortunately, the authors do not provide more details and explanations. Furthermore, in Section Reduced-Rank Embedding of a Sample of Continuous Connectivity, it seems that the authors try to conduct a novel space. This novel space is defined based on a standard Euclidean metric and finite dimensionality. Why can the authors directly definite all operators in a Banach space? Is the novel space complete? Can authors prove the completeness of the proposed new space? Finally, the authors denote the target function as Eq. (1). However, there is no detailed description to introduce the optimizer and the convexity of the target function. Can authors prove the proposed target function a convex or non-convex problem?
4). Technical Issue. In section Algorithm, the authors employed SVD to perform the matrix decomposition. However, there are several shortcomings of SVD. For example, SVD cannot implement the decomposition of a sparse matrix; SVD is difficult to solve the over-complete problem; if the input matrix is not square, SVD could be time-consuming. Therefore, the reviewers provide the following research works for authors as references: [1] Wen, Z., Yin, W., & Zhang, Y. (2012). Solving a low-rank factorization model for matrix completion by a nonlinear successive over-relaxation algorithm. Mathematical Programming Computation, 4:333-361. [2] Shen, Y., Wen, Z., & Zhang, Y. (2014). Augmented Lagrangian alternating direction method for matrix separation based on low-rank factorization. Optimization Methods and Software, 29:239-263.
- Please rate the clarity and organization of this paper
Poor
- 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 reproducible experiments provided in this paper. No source code is released.
- 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
Validation Issues: The authors only validate proposed embedding methods with the other three peer algorithms. Nevertheless, further validation is required. The reviewers hope that the authors can validate the proposed technique with current machine learning embedding techniques such as word2vec.
- 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
2
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
-
Authors need to clarify their theoretical analytics and proof with more details.
-
Authors need to validate their proposed method with peer machine learning algorithms.
-
- Number of papers in your stack
6
- What is the ranking of this paper in your review stack?
8
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
2
- [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 ideas in the paper are novel and relevant. The idea of looking at connectivity in a continuous sense is new (although a variation of this idea has been proposed before). The authors have cited the work by Moyer et al. The authors should discuss how this approach is different from Moyer et al. Further in the experimental evaluations, they should also compare their results with that of Moyer et al. since the idea of modeling end points of fiber tracts as point processes are also borrowed from Moyer et al.
In addition to the above comments, the authors should respond to the reviewer about organization about the paper, but particular reviewer #3 who outlines several weaknesses.
- 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).
6
Author Feedback
We thank the reviewers for their many positive comments and helpful suggestions. The importance of this work is highlighted by the reviewers: we introduce a “novel” (R2 Meta-R) and “creative and compelling” (R1) approach for analyzing structural connectivity that importantly “removes the atlas constraint” (R1) for multi-subject analysis and provides “experiments [that] clearly demonstrates the merit of this method” (R2). Below, we summarize and respond to the main questions and criticisms: Comparing our method to Moyer et. al.[15] (Meta-R): [15] proposed the important concept of continuous connectivity (CC) and an estimation method for a single subject. As discussed in the second paragraph on page 3, our framework extends their per-subject formulation to model replicated CCs, in which the per-subject CCs from [15] are assumed to be realizations of an underlying latent random field. The per-subject CCs are represented using a discretization over a high-dimensional grid. Direct joint analysis of such data is computationally problematic due to the enormous data sizes, as detailed in the last paragraph of the Algorithm subsection. As shown in Section 2, our method facilities efficient joint analysis by i) constructing a novel function space for the embedding of a sample of CCs into a common K-dimensional Euclidean space ii) proposing an efficient algorithm to estimate the reduced-rank space with complexity independent of grid resolution. To summarize, our method proposes a rigorous framework to analyze a sample of high-resolution CCs. As such, it is not appropriate to compare our results with [15]; our method is to be used in tandem with it. Concern of topic fit in MICCA (R4): Rather than purely theoretical, our work is foremost motivated by an important problem in medical imaging analysis: how to best model and analyze structural brain connectivity. Inconsistency in format (R4): We thank R4 for identifying the inconsistency in reference citation and bibliography naming. These will be fixed for the final version. Theoretical questions and concerns (R4 R1): R4 (and R1) expressed confusion about the Borel measurability condition. As this is a minor technical point, to avoid confusion and improve readability, following R1s suggestion we will replace “Borel” with “closed and bounded” in the final copy. In regard to completeness (R4), as discussed on page 3, the “novel space” V_K is a finite dimensional inner product space and thus complete. In regard to the target function Eq. (1) (R4), the derivation of the discretized optimization problem and algorithm are detailed in Section 2. The problem is non-convex. As noted in the first line of the Algorithm subsection, the proposed AO scheme forms an approximate (read, local) solution to the global problem. We will clarify this point in the final copy. Technical issues with SVD (R4): As outlined in the Algorithm subsection, our method leverages the local smoothness of the CC and uses spherical spline basis expansion on the marginal spaces to facilitate a reduction transformation based on the SVD. The SVD need only be computed once prior to the AO updates. The indicated references discuss computational issues resulting from performing SVDs of large matrices iteratively, which is not the case here. Add/clarify comparisons with competing methods (R1 R4): R4 asks for additional performance comparisons, particularly with word2vec. In the standard implementation, word2vec-based approaches learn a latent space for a single network, rather than a sample of networks, so a comparison to word2vec is inappropriate here. Addressing R1s question about the interpretation of Figure 1, the identification of more significant discoveries by our method vs. the competitors is precisely due to the higher resolution of the CC. A key advantage of our approach is the ability to analyze the sample of connectivity at high resolution. Reproducibility (R2 R4): The code and data will be shared via SBCI-Brain Github.
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.
The papers makes a sound theoretical contribution. All reviewers commented on the novelty of the ideas. The authors responded well to the rebuttal comments.
- 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).
6
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.
- I vote to accept this paper. I didn’t find the argument by the reviewer that this paper should be send to COLT convincing. Those conferences don’t focus on this kinds of problem.
- yes, the paper might be theoretical and its usefulness is yet to be seen but the miccai is the place such work can be published
- 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).
na
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 reviewers appreciate this paper, which take a continuous, atlas-free approach to detecting salient structural connectivity features from diffusion MRI.
The main objection against the paper seems to be that of R4, who finds the paper out of scope for MICCAI. I disagree with this assessment: MICCAI is a methodological medical imaging conference, and this is a methodological medical imaging paper. If the paper requires math that is not well known in the community, this might just as well indicate that the community needs to expand its mathematical versatility. In this particular case, though, the math used (such as Borel sets) is standard undergraduate probability and statistics.
- 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