List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Thomas F. Kirk, Martin S. Craig, Michael A. Chappell
Abstract
Surface-based analysis methods for functional imaging data have been shown to offer substantial benefits for the study of the human cortex, namely in the localisation of functional areas and the establishment of inter-subject correspondence. A new approach for surface-based parameter estimation via non-linear model fitting on functional timeseries data is presented. It treats the different anatomies within the brain in the manner that is most appropriate: surface-based for the cortex, volumetric for white matter, and using regions-of-interest for sub-cortical grey matter structures. The mapping between these different domains is incorporated using a novel algorithm that accounts for partial volume effects. A variational Bayesian framework is used to perform parameter inference in all anatomies simultaneously rather than separately. This approach, called hybrid inference, has been implemented using stochastic optimisation techniques. A comparison against a conventional volumetric workflow with post-projection on simulated perfusion data reveals improvements parameter recovery, preservation of spatial detail and consistency between spatial resolutions. At 4mm isotropic resolution, the following improvements were obtained: 2.7% in SSD error of perfusion, 16% in SSD error of Z-score perfusion, and 27% in Bhattacharyya distance of perfusion distribution.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43993-3_39
SharedIt: https://rdcu.be/dnwNK
Link to the code repository
https://github.com/tomfrankkirk/svb_evaluation
Link to the dataset(s)
N/A
Reviews
Review #4
- Please describe the contribution of the paper
Developing improved methods for detecting functional brain activity from fMR images is an on-going topic in medical image analysis. This submission proposes a “mixed-domain” approach, using a surface-based domain for the cortex and a volume-based domain for white matter and basal ganglia. A variational Bayesian framework was employed for parameter optimization. Semi-synthetic data were used for evaluation, with a demonstrated reduction of the reconstruction error.
- 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 of this submission is within the scope of this conference and of potential interest to its audience. The text is (mostly) well written, and understandable for a reader with a moderate background in recent methods of medical image analysis.
Similar approaches were tested in the “early days” of fMRI data analysis but conceptually incomplete because only the cortex was considered. This method offers the interesting option that different compartments can be modeled with different priors for the parameters. The included evaluation on semi-synthetic data is convincing. Interesting conference contribution.
- 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.
None for a conference contribution.
- 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
Some details are sketchy, but methods should generally be reproducible.
- 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
See #5 above.
- 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?
Interesting conference contribution with some novel ideas.
- 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 #3
- Please describe the contribution of the paper
The authors propose a surface-based parameter estimation approach with non-linear model fitting on functional timeseries data. The proposed method treats the different anatomies within the brain with different manners: surface-based for the cortex, volumetric for white matter, and using regions-of-interest for sub-cortical grey matter structures. An existing algorithm that accounts for partial volume effects incorporate the mapping between these different regions. A variational Bayesian framework is used to perform parameter inference in all anatomies simultaneously rather than separately. They demonstrate that it outperforms a conventional volumetric workflow with post-projection on simulated perfusion 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.
- The problem statement is good to me.
- A variational Bayesian learning-based framework is used to perform parameter inference in all anatomies simultaneously rather than separately.
- 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.
Unfortunately, I cannot judge the novelty of this work in neuroscience application field. Regarding the technical part, the weaknesses are mainly in the formulation and a lack of details about the proposed methods. I simply list the specific questions below:
Formulation.
- It is challenging to understand the notation. For example, how the generative model M is constructed (parameterized by theta in Equation 1). What is L and l in Equation 3.
- How variational inference is performed formally given the perfusion / fMRI dataset?
- What are the input, output and the loss function objective to optimize?
- The prior distribution (Normal, Laplacian, etc.) is not formulated.
Details.
- How complex the generative model is?
- What is the optimization process for the variational inference?
- How the final result is obtained since it is variational?
- How is the parameter (epistemic) uncertainty? Or is it important?
Minors:
- Why ‘many-to-one’ is emphasized in Section Method and how it relates to generative modeling? I only know that the scenario of ‘one-to-many’ can be well formulated in a Bayesian fashion.
- 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
Reproducibility does not look good.
- 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
I mainly listed the small questions here when I read through the manuscript. I am not an expert in neuroscience but try to interpret the algorithm as far as I can. Thus, I cannot judge the novelty of this work in neuroscience application field.
Formulation.
- It is difficult for me to understand the notation. For example, how the generative model M is constructed (parameterized by theta in Equation 1). What is L and l in Equation 3.
- How variational inference is performed formally given the perfusion / fMRI dataset?
- What are the input, output and the loss function objective to be optimized?
- The prior distribution (Normal, Laplacian, etc.) is not formulated.
Details.
- How complex the generative model is? Model architecture/what mapping to learn.
- What is the optimization process for the variational inference?
- How the final result is obtained since it is variational?
- What is the result of parameter (epistemic) uncertainty mentioned in the Abstract/Introduction? Or is it important in this work?
Minors:
- Why ‘many-to-one’ is emphasized in Section Method and how it relates to generative modeling? I only know that the scenario of ‘one-to-many’ can be well handled with a Bayesian modeling.
- 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?
Maybe I am not an expert in this topic, it is not easy to follow the paper because of the notations, incomplete formulation and a lack of details.
- Reviewer confidence
Somewhat 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
The authors constructed a hybrid stochastic variational Bayes (hSVB) model for parameter inference, which allowed them to perform both volumetric and surface operations simultaneously on functional MRI. This technique is then tested on simulated 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.
- The paper gives a great overview of the problem and a good background of the algorithm.
- With the hybrid approach, the authors are able to mitigate the issues other algorithms may have of being aware of the underlying anatomy, which include understanding which voxels are associated with multiple tissue types with the help of priors.
- This work is reproducible, as the methods section is a clear walk-through of the algorithm and steps necessary to complete the study
- great reporting of when hSVB performs better and worse when compared to BASIL-projection with respect to SNR.
- Great discussion section.
- 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.
- Overall, great paper. This paper is just missing the imaging parameters used for single-subject anatomical scan and the perfusion (ASL) scan. Specifically TR, TE, flip-angle. The paper doesn’t mention what was used for the anatomical scan.
- 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
- This work is reproducible as equations are provided, and the additional software that is used are all open source software tools.
- Information about the simulated data that was used is provided.
- Mentions the models used including the hardware used to run the method.
- 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
- The only thing missing is information about the single subject dataset used. For example, TR, TE, and the flip angle would be useful metrics to have. Also, since the software package FreeSurfer was used for the anatomical scan, this assumes that a T1 scan was used. However, this is not mentioned. This should also be mentioned in the methods, along with imaging parameters TR, TE, and flip angle.
- 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?
This paper was well written, well organized, and clearly detailed their approach and the reasoning behind their approach. This work is also highly reproducible.
- 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 #2
- Please describe the contribution of the paper
The paper describes a variational Bayes approach for estimating the parameters of arterial spin labeling experiments. The paper builds on an earlier work by Chappell et al., with an important difference that spatial priors in the model are computed on the cortical surface for gray matter structures and volumetrically for white matter and subcortical structures. The approach is evaluated using simulated 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.
The paper builds on existing theory and appears to be mathematically solid. The idea of introducing priors that follow cortical architecture is sound and is shown to improve parameter estimation compared to state of the art methods. The evaluation strongly favors the proposed method.
- 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.
Two areas of improvement are that the methods are presented rather opaquely, for example, the precise nature of how q() is modeled is not clear, and even the concept of inference vs. training is not clearly explained, as solving for optimal F involves an optimization problem and there does not appear to be a separate training set of which the parameters of q() are estimated and an inference set on which q is applied to unseen data.
The second limitation is that the evaluation is only in simulated data. I would have like to see experiments in real ASL experiments showing clinically significant findings with the proposed method compared to the alternative. It is possible that the way that the simulation was conducted, with a simple pattern placed on the cortical surface, is biased towards the proposed technique; whereas on real data the advantage of the hybrid surface/volume prior may be less evident.
- 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
I would not be able to reproduce the paper based on the methods section; too many details are unclear.
- 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
Please see comments above.
- 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?
I think the overall paper is solid but the weaknesses above lower my enthusiasm, particularly only simulated validation data.
- Reviewer confidence
Not 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.
- With the hybrid approach, the authors are able to mitigate the issues other algorithms may have of being aware of the underlying anatomy. Great contribution.
- Very well written as mentioned by reviewers.
- evaluation on real data is lacking as mentioned by reviewer 2. Please clarify if synthetic evaluation would differ from real data evaluation or not.
- image scan parameters should be added
- some writing details about notation and forumation could be improved.
Author Feedback
N/A