List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Xin Yu, Qi Yang, Yucheng Tang, Riqiang Gao, Shunxing Bao, Leon Y. Cai, Ho Hin Lee, Yuankai Huo, Ann Zenobia Moore, Luigi Ferrucci, Bennett A. Landman
Abstract
2D low-dose single-slice abdominal computed tomography (CT) slice enables direct measurements of body composition, which are critical to quantitatively characterizing health relationships on aging. However, longitudinal analysis of body composition changes using 2D abdominal slices is challenging due to positional variance between longitudinal slices acquired in different years. To reduce the positional variance, we extend the conditional generative models to our C-SliceGen that takes an arbitrary axial slice in the abdominal region as the condition and generates a defined vertebral level slice by estimating the structural changes in the latent space. Experiments on 1170 subjects from an in-house dataset and 50 subjects from BTCV MICCAI Challenge 2015 show that our model can generate high quality images in terms of realism and similarity. External experiments on 20 subjects from the Baltimore Longitudinal Study of Aging (BLSA) dataset that contains longitudinal single abdominal slices validate that our method can harmonize the slice positional variance in terms of muscle and visceral fat area. Our approach provides a promising direction of mapping slices from different vertebral levels to a target slice to reduce positional variance for single slice longitudinal analysis.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16449-1_20
SharedIt: https://rdcu.be/cVRU0
Link to the code repository
https://github.com/MASILab/C-SliceGen
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper is dealing with a novel application, which is to reduce positional variance in cross-sectional abdominal CT slices by generating subject-specific target vertebral level slice given an arbitrary abdominal slice as input. This paper proposes C-SliceGen to capture positional variance in the same subject with conditional generative models. Experiments show the effectiveness of the proposed 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.
- The application is interesting. As the paper mentioned, this is the first method proposed to tackle the 2D slice positional variance problem.
- Since this is the first attempt to handle such a problem, the proposed method can be treated as a pioneer for the subsequent studies.
- This paper proposes a new way to evaluate the position variance.
- 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 purpose of reducing position variance is a little bit vague. In the abstract and introduction, the paper just emphase tacking the 2D slice positional variance problem. But what are the direct clinical application is not clearly written. It is not friendly for readers who are not experts in cross-sectional abdominal CT slices.
- I do not know why it is necessary to synthesize an image at a pre-defined vertebral level. I would like to see an explanation why not conducting registration to find a pre-defined vertebral level slice. Or use organ/vertebral information as reference to find the pre-defined vertebral level slice (see the last column in Fig. 3, it is not difficult to localize the red line according to this view).
- The synthesized CT slice seems very different from the target (see Fig. 3). Is it really helpful for clinical applications? The author should discuss more on this.
- Why do we need to reduce positional variance for single slice longitudinal analysis instead of 3D patches? Since this paper proposes a new application, it is important to illustrate more.
- Muscle area and visceral fat area may change during time. Does the results in Fig. 4 verify the effectiveness of the proposed method in reducing positional variance?
- 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 author says the code will be published upon the acceptance of the paper. Since this is a pioneer work on the new application. Publishing code is important for other researchers to follow.
- 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
This paper should illustrate more on the new clinical application and technical motivation. I also have a question regarding the experiment part. Fig. 4 shows spaghetti plot of muscle and visceral fat area longitudinal analysis. But muscle and visceral fat may change overtime. Some important detailed illustration is missing here, or in the introduction. See detailed comments in the weakness of the paper.
- 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?
The application seems novel. But the paper misses a detailed illustration on the clinical potential of the application. The author should answer my concerns in the weakness part.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #2
- Please describe the contribution of the paper
The paper aims to reduce the positional variance in cross-sectional abdominal CT scans by extending the conditional generative models to C-SliceGen that takes an arbitrary axial slice in the abdominal region as the condition and generates a vertebral level slice. Experiments are performed on 1170 subjects from an in-house dataset and 50 subjects from BTCV MICCAI Challenge 2015 dataset.
- 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 addresses an important problem of reducing the positional variance in cross-sectional abdominal CT scans.
-
The qualitative and quantitative results are quite impressive and show the superior performance of the proposed 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.
-
I’m concerned about the clinical use of this work as this cross-sectional 2D scans are generally done to assess body composition. I doubt that a generative approach would be the best way to solve this problem, given the fact that it doesn’t take into account shape, boundary information and heterogenous soft tissues.
-
The backbone of the proposed approach is VAEGAN, so it can’t be considered a novel approach from technical perspective.
-
- 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 seems 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/2022/en/REVIEWER-GUIDELINES.html
-
The colors in figure 2 are quite faint.
-
The paper lacks extensive quantitative evaluations from other registration and generative model based baselines.
-
- 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?
Interesting and important problem. Some novelty in the formulation.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Very 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 #3
- Please describe the contribution of the paper
Authors propose conditional SliceGen (C-sliceGen) to synthesizing slices to target vertebral level (axial position) to reduce the positional variance problem. Authors extend classic conditional generative generative method and input random axial slices.
They use one private dataset for generating images and two dataset such as BTCV MICCAI Challenge 2015) and BLSA dataset for external validation. They also report their results with SSIM, PSNR and LPIPS GAN metrics.
- 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.
I believe that there is not enough novelties except proposing dealing with positional variance problem.
- 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.
Although i appreciate authors’ effort, some words used in the current manuscript are not clear. For example, harmonizing “ By computing body composition metrics on synthesized slices, we are able to harmonize the longitudinal muscle and visceral fat area fluctuations brought by the slices positional variation.” or “we demonstrate that the proposed method can consistently harmonize the body composition metrics for longitudinal analysis.”
I expect that authors would show the difference between VAE and their proposed method because they claim that they extend VAE method in their paper. It is not clear how authors extend current method in the literature.
I would also expect to see CycleGAN and Pix2pix for fair comparison.
- 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
Work can 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/2022/en/REVIEWER-GUIDELINES.html
Authors may use other methods for comparison such as CycleGAN and Pix2pix.
- 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
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Application seem somehow new but there is not enough experiments for fair comparison such as popular defacto GAN methods CycleGAN or Pix2Pix.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
3
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
3
- [Post rebuttal] Please justify your decision
Although i appreacite authors’ work and feedback, i believe that authors are supposed to show comparison with popular baseline methods such as CycleGAN and Pix2pix for showing where those methods fail and where authors’ method improves even if it is not very applicable. Additionally, there are applications where CycleGAN and Pix2pix are applicable to medical datasets.
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 authors have deminstrated a novel application for mapping slices from different vertebral levels. However, I have a few questions and expect the authors could clarify them in the rebuttal, as well address the concerns from reviewers.
- The motivation of the task. Why a generarive model is needed here? Why a mapping or registration method won’t provide sufficient results for this method?
- The model is able to generate target slices from any conditional slice. But would a slice from similar verteebral level is able to generate higher-quality target slice? How can you ensure the correctness of the generated model?
- Figure 4 shows less variance after the harmonization. But that is not the ground truth of the muscle and fat area. Would a 3D method evaluating the muscle and fat area give more accurate results other than generating 2D slices and evaluating muscle and fat in mm^2?
- 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 reviewers and AC for the insightful comments. We appreciate that reviewers and AC agree on our work’s pioneering and effectiveness. This paper is the first exploration using AI to reduce cross-sectional variation for longitudinal body composition analysis. Our results show a promising direction for handling imperfect single slice CT with a conditional generative model. In the following we address the major concerns/points:
1) Clarity of motivation (Meta reviewer, Reviewer #1) Thanks to the reviewers’ comments. Our motivation starts from the longitudinal single slice CT. A wide age range of volunteers visited the Baltimore Longitudinal Study of Aging (BLSA) clinical unit every 1-4 years for abdominal scans over the last 20 years to study the aging effect on the body composition. A single 2D axial CT slice instead of the 3D whole-body CT scan is acquired during each visit to reduce unnecessary radiation exposure. However, it is challenging to acquire subject slices in the same vertebral level across all the visits (Fig1). The positional difference leads to significant variation in the organs and tissues captured and makes the longitudinal analysis invalid. Our motivation is to reduce slice cross-sectional variation induced by positional difference and allow the longitudinal body composition analysis using 2D single slices.
2) Why generative model instead of registration (Meta reviewer, Reviewer #1&2) Thank you for pointing this out. Registration will fail if structures (e.g., organs/tissues) in the fixed scan do not appear in the moving scan. However, in our context, the conditional slice and target slice (i.e., moving, and fixed slices in the registration context) may contain different structures. The generative model can solve this limitation by learning the joint distribution between the conditional and target slices. In the testing phase, given the conditional slice, the model can generate the target slice even if the structure changed.
3) Evaluation of generative model (Meta reviewer, Reviewer #1) We thank the AC for agreeing that our model can generate target slices from any conditional slice. Our model is intended to be applied to datasets like BLSA which only have single 2D axial slices. To ensure the correctness of our model, we train and validate our method first with datasets that have 3D volumetric CT data where target slices are acquired and used as ground truth to compare with generated slices. SSIM, PSNR, and LPIPS are used as evaluation metrics. The qualitative results (Fig3) and quantitative results (Table1) show the effectiveness of our proposed method (pointed out by Reviewer #2). In our experiments, however, we do not observe that conditioning on similar vertebral level slices can lead to higher-quality generated target slices. The reason behind this observation could be further investigated in future work.
The AC highlights that variance is reduced in Fig4. Indeed, Fig4 is not showing the ground truth for muscle and fat since the target slice is not acquired in the BLSA dataset in contrast to Fig3. As a result, 3D evaluation methods of muscle and fat are not applicable. In Fig 4, we show the applicability for longitudinal studies where the positional variation is reduced by the proposed model.
4) Method novelty (Reviewer #2&3) Thanks for pointing this out. Our method is an extension of the conditional VAEGAN. Our main contribution is how we embed the arbitrary abdominal slice as a condition and learn the joint distribution to generate the target slices. Other GAN methods were explored without success such as style transfer methods (pix2pix & CycleGAN). However, style transfer methods aim to preserve the image content while changing the image style. In our case, image content (organs/tissues) is also changing so the style transfer methods are not applicable.
5) Unclear words and Fig2 colors will be changed (Reviewer #2,3)
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 rebuttal addresses some of my concerns. Although, there are still some unclear questions, it seems that we can construct the whole 3D longitudinal CT scan given just one 2D slice. The strengths of this paper are the novel application and reasonably well results. I would say the strengths outweigh the weaknesses, such as the metrics are not entirely appropriate to demonstrate the correctness of the generated slices. If the paper is accepted eventually, the authors are expected to include the suggestions and comments during the review process.
- 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).
8
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.
The authors have provided a reasonably clear rebuttal. My main remaining concern for this work is the clinical applicability of the proposed GAN approach. The main focus of this work is to identify changes in body composition for characterizing tissue changes between healthy vs diseased tissue by using longitudinal CT scanning. However, the target image generation is related to a single time point, and the image synthesizing approach will change the appearance of the tissue to this target slice. I believe a vertebra level localization approach would be more suitable for this application. By using this identified level the CT image could be oriented to that level during follow-up scanning. In short, I was not convinced with the proposed methodological solution for the important specific clinical application.
- 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/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).
NR
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.
Despite weakness on experimental design, all reviewers agree that this is a novel application of cVAE-GAN. Although this AC personally did not think that the metrics such as SSIM, PSNR or LPIPS were appropriate here considering the clinical application which the method was applied to. The paper provides a first step towards addressing the problem in the target application with AI-based method.
- 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