Back to top List of papers List of papers - by topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Sahar Ahmad, Ye Wu, Pew-Thian Yap
Abstract
Human brain templates are a basis for comparison of brain features across individuals. They should ideally capture anatomical details at both coarse and fine scales to facilitate comparison at varying granularity.
Brain template construction typically involves spatial normalization and image fusion. While significant efforts have been dedicated to improving brain templates with sophisticated spatial normalization algorithms, image fusion is typically carried out using intensity-based averaging, causing blurring of anatomical structures.
Here, we present an image fusion method that exploits cortical surfaces as guidance to help preserve details in brain templates. Our method encodes cortical boundary information given by a cortical surface mesh in a signed distance function (SDF) map. We use the SDF map to help determine localized contributions of the individual images, especially at cortical boundaries, in image fusion. Experimental results demonstrate that our method significantly improves the preservation of fine gyral and sulcal details, resulting in detailed brain templates with good surface-volume agreement.
Link to paper
DOI: https://doi.org/10.1007/978-3-030-87234-2_37
SharedIt: https://rdcu.be/cyl8x
Link to the code repository
N/A
Link to the dataset(s)
https://www.humanconnectome.org/study/hcp-young-adult/data-releases
Reviews
Review #1
- Please describe the contribution of the paper
In the context of brain template construction, this paper introduces an image fusion method that exploits cortical surface information through signed distance map to preserve details when averaging brain images.
- 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 and the topic address is of great interest for the neuroimaging community. The proposed approach addresses the problem at hand in an efficient way.
- 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 weakness of this work is related to the reference method considered in the evaluation part, which relies on a global image averaging based only on temporal weighting. A comparison with a local image fusion strategy based on both temporal and intensity patch based weighting would have been more suitable to highlight the benefit of the proposed surface-guided image fusion.
- 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
the paper meets the standard criteria of 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://miccai2021.org/en/REVIEWER-GUIDELINES.html
I would suggest the authors to consider as a reference method a local image fusion strategy based on both temporal and intensity patch based weighting. I would also suggest the authors to comment about the potential impact of the proposed fusion scheme on neuroimaging study. Is this just limited to enhance the visual appearance of the average brain representation ?
- 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?
This paper provides a nice and efficient framework to improve brain template construction but the lack of comparison with a local image fusion strategy based on both temporal and intensity patch based weighting does not allow to evaluate the real added value of the proposed surface-based fusion strategy.
- What is the ranking of this paper in your review stack?
3
- Number of papers in your stack
5
- Reviewer confidence
Very confident
Review #2
- Please describe the contribution of the paper
A novel image fusion strategy to achieve more clear cortical local details for brain template construction. Signed distance function (SDF) maps are proposed to represent brain cortical surface. Image fusion process is weighted by the SDF difference between individual image and the template. Results show that the image fusion strategy achieves better contrast between cortical GM, WM and CSF.
- 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 proposed a novel and efficient strategy for image fusion step in brain template generation. The authors define the Signed distance function (SDF) from pial and WM cortical surface template, which is accurate to estimate difference between template surface and individual surface.
- 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 is not clear if the time weighting in eq. (8) is still necessary. It seems to work as the weighing of “global” contribution of each involved image, but I after adding surface guidance, will this function still work?
- 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
Easy to reproduce.
- 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
Motivation is not clear enough for people who don’t familiar with brain template construction.
- 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?
Novelty is clear, improvement of result is sufficient, evaluation is plausible. However, motivation is not clear enough for people who don’t familiar with brain template construction. Contribution only focus on image fusion, seems to be a bit limited.
- What is the ranking of this paper in your review stack?
1
- Number of papers in your stack
4
- Reviewer confidence
Very confident
Review #3
- Please describe the contribution of the paper
The authors proposed an image fusion method by using signed distance function maps to help determine voxel-wise contributions instead of intensity averaging during brain MRI template construction.
- 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 topic is important and timely. The article is well organized. And their approach performs better than AKR method on boundary sharpness and consistency to the cortical surface.
- 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.
a) The major issue is that these is no evaluation on the application effects of the templates generated by AKR and SGIF. Maintaining cortical surface details seems good for the group templates. But how does this benefit the accuracy of individual image normalization or voxel-based statistics?
b) Another concern is that some outliers are introduced in the boundaries of gray matter, which is quite obvious in the frontal part of brain template at 30 yr. Further improvement should be made to avoid this.
- 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
Some proprecessing steps are easy to repeated. But the core code of template construction via Surface Guide Image Fusion is needed.
- 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
a) A flowchart should be given for the entire process of brain template construction. b) Please explain how the cortical surfaces being subsequently availed to compute SDF maps of each subject. c) Please give a clear definition of SDF maps rather than show the web address of connectome-workbench.
- 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?
Building high quality brain template is an on-going goal in neuroscience studies and this is essential for the application studies on brain MRI.
- What is the ranking of this paper in your review stack?
2
- Number of papers in your stack
3
- Reviewer confidence
Very confident
Review #4
- Please describe the contribution of the paper
Brain templates are often used to capture morphometric variation at the cohort level. This submission describes an approach for template generation that uses the signed distance from surfaces to guide the registration process from individual information to the template. Results based on anatomical data of a publicly available data set (HCP) are included. The performance of the approach is compared with another approach (“adaptive kernel regression”, ref. 17) and demonstrates a slight but statistically significant improvement.
- 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.
Topics of this submission are within the scope of this conference, and of potential interest to its audience. The text is mostly well-written (except where noted below), without major errors, and readily understandable for a reader with a moderate background in medical image 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.
Authors may be reminded that signed distance maps were used for surface-based registration procedures before volume-based methods became viable. Thus, the advance described here is minor, but leads to some improvement.
The major concern raised against this work is a conceptual one: While it is viable to construct cohort-wise average maps at the gross (e.g., lobar) anatomical level, constructing detailed maps (at ~cm level) is arguable, because it is increasingly recognized that sub-groups with distinct cortical configurations exist. As demonstrated here, an average can be computed on any measure, but there is some doubt that it may be helpful. The manuscript does not discuss this issue.
- 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
Authors provided a fair assessment on the checklist. Apart from a minor issue (see 7.2), methods are reproducable.
- 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
- Please, see answer to Q4 above.
- More information than just the reference to the connectome workbench is necessary to understand “SDF maps” here, e.g., are distances (thresholds) on the mm or voxel level?
- Please, add bibliographic detail to ref. 18.
- 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?
Minor improvement, arguable concept.
- What is the ranking of this paper in your review stack?
2
- Number of papers in your stack
2
- Reviewer confidence
Very 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.
This paper presents a new fusion strategy based on signed distance function (SDF) for brain MRI template creation. Large publicly datasets are used to create wide age range of templates (from 16 to 60 yo). Results are compared with a baseline with a temporal weighting method based on adaptive kernel regression (AKR) and are qualitatively evaluated and also quantitatively compared with sharpnes mesures.
The topic tackled in this work is highly relevant for neuroimaing studies. I think though that reviewers overrated this work that in my opinion does not demonstrated any added/practical value. My major concerns that I would like to see discussed in the rebuttal are listed here below:
- Baseline for comparison: why the authors did not consider a local weighting fusion strategy (local intensity patches for instance as suggested by R1). I think this could be included in eq 9 and thus compare if the SDF is really outperforming another local-spatial prior. I did understand if AKS has temporal weights here are also global and local? Overall, the methodological contribution of SDF seems still limited as raised by R3.
- The authors claim their approach generates better templates, in the sense of finer anatomical details and improved surface-volume consistency. I acknowledge some finer details in the SDF templates but I am also noticing these weird effects mostly withing the cortex of “noisy/whitish” voxels appearing, also noticed by one reviewer. Could the authors discuss/justify this? Fig 2 30yo bottom (back of the brain) also visible in many other as 20,25,40yo, 60 also visible Fig4a.
- The results remain overall at a preliminary stage as authors do not demonstrated their SDF is helpful in a template-based application. So how the proposed template is better in practice (subject based or group-wise analysis) is not shown. Could the authors develop on this, maybe since the submission there is no new evidence of added practical value for template-based application?
Thus overall, I think that this work presents a minor technical novelty and though it seems well executed evaluation is too qualitative, and comparison with a baseline that is not completely fair in my opinion (though maybe the authors can justify this). Still, results are not convincing of the improvements the SDF can bring in a further analysis.
Please remind that the main purpose of this rebuttal is to provide clarification or to point out misunderstandings, and include new details that can better highlight the value of this work. I will not consider any promise of adding future experiments and results.
- 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).
9
Author Feedback
Comparison with Local Weighting Fusion: We tested local fusion strategy based on tissue segmentation maps. We computed the voxel-wise weights using the difference between the segmented tissues atlases and the individual subjects’ warped segmented tissue maps in Eq. 7. We call this method as tissue-guided image fusion (TGIF); the intensity-based brain templates for TGIF were constructed using both the temporal and spatial weights — like SGIF. The mean (std) of MHD for TGIF is 0.67 (0.05) and 0.58 (0.08) for MAD, whereas SGIF achieved 0.55 (0.08) for MHD and 0.36 (0.04) for MAD. The improvements given by SGIF are statistically significant for both distance measures (p<0.01). We also computed the frequency domain image blur measure and standard deviation of the TGIF-based templates and found them to be lower than SGIF-based templates; the mean (std) is 0.00053 (0.00019) and 195.62 (2.34) for frequency domain image blur measure and standard deviation, respectively.
These results confirm that SDF based image fusion method provides better alignment of the volumetric cortical structures with the cortical surface template. These results can be incorporated in the camera-ready version without exceeding the page limit.
Noisy Voxels: We adjusted the thresholds in Eqs. 5 & 6 and employed an outlier rejection technique to remove voxels with ambiguous intensities in the 30-year-old brain template. The updated brain template now does not show any noisy voxels and preserves the anatomical details.
Applications of Brain Templates: Brain templates with more anatomical details and distinct structural boundaries help in better spatial normalization. Since, SGIF-based templates show greater structural features and have sharp boundaries at GM/WM and GM/CSF interface, therefore, these templates will facilitate in improving the image normalization accuracy. Moreover, we have demonstrated that the volumetric templates are well-aligned with the cortical surface templates, thus, allowing consistent surface-volume analyses in a common space. Note that most existing fusion methods have either focused on volume or surface atlases, but not joint volume-surface atlases and the consistency of the two types of atlases. Generating volume-surface consistent atlases is an application of our fusion method.
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.
This paper presents a new fusion strategy based on signed distance function (SDF) for brain MRI template creation. Large publicly datasets are used to create wide age range of templates (from 16 to 60 yo). Results are compared with a baseline with a temporal weighting method based on adaptive kernel regression (AKR) and are qualitatively evaluated and also quantitatively compared with sharpnes mesures.
I think that reviewers overrated this work that in my opinion does not demonstrated any added/practical value.
One of my major concerns (related to the local weighting) has been very well addressed: authors provided now new evaluation with a local weighting fusion strategy that the proposed approach seem to “outperform” as with the given measures provided. I did really appreciate this additional experiments.
However, the authors did not explain why outliers occur in their generated templates. They applied thresholds as outlier rejection and cleaned 30yo template, but what are these thresholds about, why the artefacts occur initially and the same values work well for the other template ages where I clearly see the outliers 20,25,40yo, 60 also visible Fig4a. ?
Generating finer anatomical details in an atlas is a first step, and the hypothesis that they can support better anatomical normalization is valid. But the authors are not demonstrating this at all, the practical value (this better normalization) provided by these specific templates is not really proven. It would have been easy to simple evaluate the non-rigid registration of the atlas with single subject like to perform atlas-based segmentation and demonstrated that segmentation is better with this atlas than another one. Or even simply by comparing similarity metrics of this atlas with another one after registration of a subject. I think it would be also worth to try if VBM analysis with these templates show more precise results than with other templates for spatial normalization, though the blurring step might be a strong confounding. Ideally, the demonstration of volume-surface consistency should be also given practically, though this would already be for an extended journal version of this work.
Given that the outperformance of using these atlases is not prove, I consider the current evaluation of this work still very preliminary as relying only on metrics of the generated atlas. Also the origin of the generated artefacts is unclear and the I am not sure the proposed thresholds for correcting the 30yo generalize easily to any other age range, so really solve the problem. As such I would not recommend this paper for acceptance.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Reject
- What is the rank of this paper among all your rebuttal papers? Use a number between 1 (best paper in your stack) and n (worst paper in your stack of n papers).
17
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.
Authors present a new fusion strategy for cortical brain templates, resulting in sharper templates. The rebuttal discussed artifacts observed by reviewers and mentioned these to be fixed by changing parameters - which may raise some questions on robustness and systematic validation. There is some limited novelty, moreover as signed distance maps were used extensively in the past in regard to shape alignment and shape statistics. This reviewer still sees this paper at a preliminary stage where there is a need for much more experimental, quantitative evaluation. This paper may make the cut if there is room for more accepted papers.
- 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 (best paper in your stack) and n (worst paper in your stack of n papers).
9
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 meta reviewer raised several important issues, which were partially addressed by the rebuttal. The concern on lack of demonstration on practical applications is valid, but in my opinion not essential for a conference paper if the paper shows interesting and promising work, which seems to be the case as acknowledged by all reviewers.
- 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 (best paper in your stack) and n (worst paper in your stack of n papers).
8