List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Richard McKinley, Christian Rummel
Abstract
The thickness of the cortical band is linked to various neurological and psychiatric conditions, and is often estimated through surface-based methods such as Freesurfer in MRI studies. The DiReCT method, which calculates cortical thickness using a diffeomorphic deformation of the gray-white matter interface towards the pial surface, offers an alternative to surface-based methods. Recent studies using a synthetic cortical thickness phantom have demonstrated that the combination of \mbox{DiReCT} and deep-learning-based segmentation is more sensitive to subvoxel cortical thinning than Freesurfer.
While anatomical segmentation of a T1-weighted image now takes seconds, existing implementations of DiReCT rely on iterative image registration methods which can take up to an hour per volume. On the other hand, learning-based deformable image registration methods like VoxelMorph have been shown to be faster than classical methods while improving registration accuracy. This paper proposes CortexMorph, a new method that employs unsupervised deep learning to directly regress the deformation field needed for DiReCT. By combining CortexMorph with a deep-learning-based segmentation model, it is possible to estimate region-wise thickness in seconds from a T1-weighted image, while maintaining the ability to detect cortical atrophy. We validate this claim on the OASIS-3 dataset and the synthetic cortical thickness phantom of Rusak et al.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43999-5_69
SharedIt: https://rdcu.be/dnwxm
Link to the code repository
https://github.com/SCAN-NRAD/CortexMorph
Link to the dataset(s)
brain-development.org/ixi-dataset
https://data.csiro.au/collection/csiro:53241v1
Reviews
Review #3
- Please describe the contribution of the paper
This paper explores wether an existing deep learning (DL) based registration method (Voxelmorph) can be used to speed up the DiReCT algorithm for cortical thickness estimation from MRI scans. Previously it has already been shown that combining DireCT with DL based segmentation network can be more sensitive to subtle changes in cortical thickness compared to FreeSurfer. This paper now focuses on the registration part within DireCT and speeds it up by using a DL approach. They show that their method correlates well with DL+DireCT on the large OASIS-3 data and a public synthetic cortical thickness benchmark.
- 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.
- this paper shows that the DireCT algorithm can be significally sped up using Voxelmorph, which could be very relevant for the neuroimaging community
- interesting discussion section, explaining their choices and exploring potential other options.
- the experiments are sound, the authors used multiple datasets and specifically different datasets for tuning their method and for evaluation
- 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 technical novelty of their method is limited to applying existing methods. DL+DireCT did already exist, the authors just exchanged the traditional registration algorithm in DireCT with an existing DL registration model (Voxelmorph)
- The authors compare only to Ants+DireCT, and not to FreeSurfer, or other cortical thickness computation methods (see detailed comments).
- 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 want to publish code and pre-trained models. Further the required code for DeepSCAN and Voxelmorph is also public.
- 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
-
what data was the DeepSCAN model trained on? The paper [1] states ADNI, IXI and in-house data as training sources, so this seems fine, but the authors should state this in the paper.
- as mentioned above it would be interesting to see how the method compares to FreeSurfer. The authors argue that it has been shown that FreeSurfer might not be the optimal silver standard to compare to, referencing the study by Rusak et al.[2], however in that work it is stated that “A possible reason for the worse FreeSurfer performance versus DL+DiReCT in detecting introduced atrophy is the higher similarity of the CTh definition used in DL+DiReCT to the CTh definition used in our method compared to the CTh definition used in FreeSurfer.” [2]. This suggests that the cortical thickness benchmark might be biased towards methods using a similar definition of CTh. Also, the work by Rebsamen et al. [1] mentions that DL+DiReCT has higher correlation with FreeSurfer than ANTS with FreeSurfer and that FreeSurfer seems to be more robust than ANTS.
- The resolution of Figure 1 should be improved
[1] – Rebsamen et al. : Direct cortical thickness estimation using deep learning-based anatomy segmentation and cortex parcellation [2] – Rusak et al.: Quantifiable brain atrophy synthesis for benchmarking of cortical thickness estimation methods.
-
- 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
leaning towards accept, well written and sound paper. Even if the technical novelty is low, it could be of interest for the neuroimaging community.
- 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 #2
- Please describe the contribution of the paper
Cortical thickness is an established parameter for the quantification of various neuro-degenerative diseases. This submission proposes a fast approach for thickness estimation. An unsupervised deep learning method estimates the deformation from the GM/WM to the pial interface. The proposed method is evaluated on the OASIS-3 data set and semi-synthetic 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 topic of this submission is within the scope of this conference and of potential interest to its audience. The text is well-written and without major issues, except where noted below. The proposed method is a useful extension of current technology, with a solid and convincing evaluation.
- 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.
On the other hand, thickness estimation is a well-researched topic, for which efficient conventional methods exist, in terms of speed and accuracy. There is some doubt that all conventional methods need an equivalent method based on deep learning. The enthusiasm for this work would be greater if authors would demonstrate that their method resolves some of the issues of conventional methods (e.g., the resolution of touching cortical banks, connecting vessels).
- 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
Methods are reproducible, with some guessing/experimentation 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://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
Some minor issues should be corrected:
- p.8: CT. [13] -> CTh. [13]
- p.8: Rebasmen -> Rebsamen
- p.9, Ref 11: Please, check for archival publication.
- p.9, ref 12: Remove n/a (n/a).
- 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?
Solid conference contribution.
- 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
A simple paper whereby a VoxelMorph type framework is used to achieve registration-based cortical thickness measurements. Validation used synthetic data, as well as a comparison on real data where results from the ANTS method were treated as “ground truth”.
- 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.
A well written manuscript about a relatively simple idea.
Because the proposed approach aligns two (almost) binary images, a pre-trained model is likely to work with tissue maps extracted from a wide variety of different MRI contrasts. This could make it generally useful to clinician scientists.
- 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.
Perhaps the scientific contribution in terms of technological advances is not sufficiently substantial, given that it is an implementation of a VoxelMorph type algorithm instead of ANTS, for doing what Das et al did in 2009.
- 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
Sufficient details of the architecture are not provided to re-implement the method, and no placeholder for a link to a repository is yet included in the manuscript.
- 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’d suggest refocusing the manuscript slightly to increase the interest to clinician scientists, many of whom do not have access to good GPUs. Reporting deployment runtimes (and memory footprint) on CPU would be useful, and allow a fair comparison against ANTS. Availability of an out-of-the-box pretrained model (say for isotropic 1 mm voxels), that is easy-to-use, would also add interest.
The CSF in narrow sulci is often not visible in MRI scans of typical resolution, which may cause problems for methods that try to determine cortical thickness. Some discussion of this would have been helpful.
I’m not entirely convinced by the use of “WM” and “WM+GM” to denote tissue maps in the equations. It might be better to use the usual I_0 and I_1 or possibly W and B.
- 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I’m a bit on the border with this one. While there is little in the way of a novel scientific contribution, the work could be of interest to clinician scientists.
- 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.
There is consensus among the reviewers that this is well written and sound paper with more strengths than weaknesses. The present work is of interest for the neuroimaging community and clinician scientists despite low technical novelty. Highlighting aspects of interest for clinicians and providing comparisons to FreeSurfer cortical thickness estimates may further strengthen the paper.
Author Feedback
N/A