Paper Info Reviews Meta-Review Author Feedback Post-rebuttal Meta-Reviews Back to top
List of Papers By topics Author List
Paper Info Reviews Meta-review Author Feedback Post-Rebuttal Meta-reviews

Authors

Anne-Marie Rickmann, Fabian Bongratz, Christian Wachinger

Abstract

Mesh-based cortical surface reconstruction is a fundamental task in neuroimaging that enables highly accurate measurements of brain morphology. Vertex correspondence between a patient’s cortical mesh and a group template is necessary for comparing cortical thickness and other measures at the vertex level. However, post-processing methods for generating vertex correspondence are time-consuming and involve registering and remeshing a patient’s surfaces to an atlas. Recent deep learning methods for cortex reconstruction have neither been optimized for generating vertex correspondence nor have they analyzed the quality of such correspondence. In this work, we propose to learn vertex correspondence by optimizing an L1 loss on registered surfaces instead of the commonly used Chamfer loss. This results in improved inter- and intra-subject correspondence suitable for direct group comparison and atlas-based parcellation. We demonstrate that state-of-the-art methods provide insufficient correspondence for mapping parcellations, highlighting the importance of optimizing for accurate vertex correspondence.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43993-3_31

SharedIt: https://rdcu.be/dnwNw

Link to the code repository

https://github.com/ai-med/V2CC

Link to the dataset(s)

N/A

Reviews

Review #2

  • Please describe the contribution of the paper

    Utilization of the L1-loss instead of the Chamfer loss in order to perform vertex-wise correspondence mapping of a brain scan to a common space for accurate cross-sectional group anatomical comparison, such as with cortical thickness.

  • 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 has a very solid and well-written introduction to the problem with background information, previous solutions, and their method.

    2. Created a model, based on Vox2Cortex, that can perform vertex correspondence instead of using a common mesh and utilize an L1-loss instead of the commonly used Chamfer loss for subject scan wise cortical reconstruction.

    3. Multiple open source datasets were used for training, testing and validation of the model. This, along with the detailed model description, allows for high reproducibility of the study. The code used is also publicly available.

    4. Utilization of multiple metrics for testing and comparison between current state-of-the-art measurements.

    5. Comparison of V2CC to V2C to reveal smaller boundary parcellation errors using V2CC rather than V2C. Since the authors’ model builds on the V2C framework, it shows that the model modifications used to create V2CC were necessary to further minimize parcellation error.

    6. The Alzheimer’s disease use case was a good choice as many studies rely on cortical thickness, and other anatomical metrics for staging and group classification, such as determining the level of degeneration.

  • 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 authors should discuss why different epochs were used in different algorithms. The authors should have used the same amount of epochs across algorithms, and then introduce early stopping conditions. This way no user bias is introduced in terms of epochs used for training and you can truly visualize the models that are trained quicker.

    2. Although it may be obvious that the T1 was used, it is still necessary to include that T1w images were used, including scan parameters (TR, TE, and flip angle), scanners, and associated field strengths.

  • 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

    Multiple open-source datasets were used for training, testing, and validation of the model. This, along with the detailed model description, allows for the high reproducibility of the study. The code used is also publicly available. There is a clear description of the mathematical framework, with the exception that model weights were not included in the equation/ calculations written in the 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://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

    1. This is just a minor detail, but make sure to define your model’s name before using the acronym. In the paper, you just refer to it as V2CC without ever using the complete name. I think it’s Vertex2Cortex Correspondence, however, this needs to be introduced. The best place for you to introduce your model with the full name and acronym would be the last paragraph of your ‘Introduction’ section. You talk about your model, however you don’t bring up the name.

    2. The comma is in the wrong place for the number 163842, referred to in the ‘Experiment and Results’ section for fsaverage

    3. Instead of using the term white surface, it would be more correct to use the term “white matter surface”. You should also briefly describe what the white matter and the pial surfaces represent.

    4. Is preprocessing necessary when using this method? Why or why not? These questions should be addressed in methods. Especially since these data probably come from different scanners with different field strengths, which could introduce issues such as intensity bias field inhomogeneity. FreeSurfer takes care of this, however, how is this handled for V2CC? Is it not necessary or are there a series of preprocessing steps that happen prior to introduction to the model?

  • 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?
    1. The authors produced a highly reproducible study.
    2. Their model produced comparable results to other state-of-the-art methods, with the exception that they were able to perform better in inter and intra -subject correspondence with respect to cortical thickness.
    3. The paper was well written.
    4. This could be used in longitudinal studies where it is crucial to understand the amount of cortical degeneration within a subject’s time points.
  • Reviewer confidence

    Very confident

  • [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



Review #1

  • Please describe the contribution of the paper

    In this work, the authors present a deep learning approach that reconstructs brain surfaces with vertices that are in correspondence with those of a template surface, thus removing the need to align surfaces as a post-processing step that can be computationally costly. To do so, the authors propose to replace the Chamfer loss that is used in existing methods with the L1 loss and adding a regularisation term to enforce consistency. They extensively evaluate their approach, for example by assessing inter- and intra-subject vertex correspondence, and by performing group comparison and diagnosis prediction.

  • 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.
    • Replacing the Chamfer loss with the L1 and a regulariser seems novel and appears to be effective.
    • The approach is extensively evaluated, first using various metrics to assess the quality of the reconstructed surfaces, and then to assess whether the new surfaces lead to coherent group comparisons and accurate diagnosis prediction.
    • The evaluation is performed using multiple and rather large data sets.
    • The results are convincing. The metrics do not necessarily show a better performance but the fact that an equivalent performance is reached without the burden of spherical registration is valuable.
    • The paper is clear and easy to follow.
  • 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.
    • Taking into account the 8-page constraint, the paper does not seem to present major weaknesses.
  • 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

    The answers given by the authors seem consistent with what is present in the paper. Code will be made available on GitHub, and the methods and results are well described for an 8-page 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://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
    • Table 1 takes some effort to be understood properly (particularly regarding the results obtained for the TRT dataset) but I am not sure this can be improved while still respecting the 8-page limit.
    • Some typos could be corrected e.g. 16,3842 -> 163,842 vertices in fsaverage; Vox2cortex -> Topofit in ref [2], ref [12] and [13] are the same.
  • 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 method is innovative, well-validated and leads to convincing results.

  • 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

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #3

  • Please describe the contribution of the paper

    The authors proposed a cortical surface reconstruction method based on the template-deformation based approach used in Vox2Cortex. Here, as the main contribution, the authors propose using the L1 loss to perform registration as opposed to the Chamfer loss to maintain one-to-one vertex correspondence between the deformed template and the final reconstructed surface. This is done by first registering the the ground truth surface to the template for computing the L1 loss between corresponding vertices.

  • 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 paper is well written. They include a good discussion of previous work and the methods are explained well with helpful figures.
    • The strength of the proposed method is that it allows cortical surface reconstruction while maintaining correspondence with a population-level template. This facilitates group-level analysis of cortical structure without having to perform registration to an atlas as a separate step.
    • The authors perform several evaluation experiments at both the inter and intra- subject level using 3 different datasets. They also show its utility on the downstream task of disease classification relative to the traditionally used surface-based registration approach.
  • 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 main limitation of the proposed method seems to be the need for registration and resampling between the ground truth mesh and the template, and the introduction of self-intersections during this step. Even though this method provides vertex correspondences, further steps are needed to ensure topological correctness of the reconstructed surface.

  • 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

    The authors use publicly available datasets, and build on the V2C code which is also publicly available.

  • 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/2023/en/REVIEWER-GUIDELINES.html
    • It would be helpful if the authors provided more quantitative details regarding the number of self-intersections in the fsaverage templates, and the resampled ground truth meshes for reference when evaluating the %SIF metric.
    • Why were the other two regularization terms used in V2C not included here?
    • Was the number of epochs/training time for V2CC same as that of V2C?
  • 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 paper is well written and several evaluation experiments were done to demonstrate the performance of the proposed method. The combined approach for vertex correspondence/reconstruction would be useful in research practice.

  • Reviewer confidence

    Very confident

  • [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



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.
    • good solid paper, well written
    • Reviewers have pointed out some typos.
    • as reviewer pointed out, enforcing same connectivity in the mesh across different subjects may reduce mesh quality and produce self intersections. How do you address this or is this a problem?



Author Feedback

N/A



back to top