List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yuan Bi, Zhongliang Jiang, Ricarda Clarenbach, Reza Ghotbi, Angelos Karlas, Nassir Navab
Abstract
Generalization capabilities of learning-based medical image segmentation across domains are currently limited by the performance degradation caused by the domain shift, particularly for ultrasound (US) imaging. The quality of US images heavily relies on carefully tuned acoustic parameters, which vary across sonographers, machines, and settings. To improve the generalizability on US images across domains, we propose MI-SegNet, a novel mutual information (MI) based framework to explicitly disentangle the anatomical and domain feature representations; therefore, robust domain-independent segmentation can be expected. Two encoders are employed to extract the relevant features for the disentanglement. The segmentation only uses the anatomical feature map for its prediction. In order to force the encoders to learn meaningful feature representations a cross-reconstruction method is used during training. Transformations, specific to either domain or anatomy are applied to guide the encoders in their respective feature extraction task. Additionally, any MI present in both feature maps is punished to further promote separate feature spaces. We validate the generalizability of the proposed domain-independent segmentation approach on several datasets with varying parameters and machines. Furthermore, we demonstrate the effectiveness of the proposed MI-SegNet serving as a pre-trained model by comparing it with state-of-the-art networks.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43901-8_13
SharedIt: https://rdcu.be/dnwCX
Link to the code repository
https://github.com/yuan-12138/MI-SegNet
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a mutual information based segmentation approach for unseen domain generalization in ultrasound images. Utilizing two encoder networks anatomical and domain features are extracted. The approach is evaluated on a combination of public and closed data sets and reportedly shows competitive values when compared to the state of the art.
- 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.
• Addresses a novel way to increase the segmentation performance in the very challenging domain of US imaging by utilizing anatomical as well as domain information in a disentangled way. • The paper addresses the very important topic of domain shift by applying a feature disentanglement-based approach where mutual information is applied as metric. • Results are demonstrated also on publicly available data sets, are accompanied by an ablation study and are compared to different levels of contender methods
- 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.
Actually, the quality and content of the proposed work does not show major weaknesses. What I’m missing is a more in-depth discussion on the choice of mutual information as a decoupling metric.
- 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 work was evaluated partly on publicly available datasets. Code will be available after acceptance.
- 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
• Although the application of mutual information in the context of this work is legit, a comparison or at least a discussion on different applicable metrics is missing. There might be other metrics more suitable for feature disentanglement in the US setting e.g., Wasserstein or CMD [1,2]. • US images have very distinct noise patterns which differ from gaussian noise. What kind of noise was applied in the context of this work? • In the SOTA comparison it would have been nice to see the performance of the nnUNet. [3,4] • A discussion on the limitations of the proposed approach would be appreciated. • A comparison in terms of computation cost (number of parameters, flops) would be interesting to see. • Will the curated data set be publicly available in the future? Please consider providing the data for the research community, as especially high quality US data is scarce.
- 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?
The overall quality of the paper is very good. Furthermore it addresses a general problem, the topic of domain shift in the challenging domain of ultrasound imaging with an interesting feature disentanglement approach.
- 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 #4
- Please describe the contribution of the paper
Ultrasound (US) imaging suffers from significant domain shift issues that includes acquisition quality and anatomy being imaged. The paper aims to introduce robustness in US image segmentation through a novel mutual information (MI) based framework with two encoders that are penalized when they share MI. This process explicitly disentangles the anatomical and domain feature representations.
- 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 manuscript focuses on the development of a segmentation algorithm for ultrasound images by training it on the appearance of the target/shape of the target vs. the background which is affected by the complex imaging variables. They use the multual information strategy for this. The work is clearly described and the results are impressive.
- 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.
There are no significant weaknesses in the work - as described within the scope of the paper.
- 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 have provided code and pointed out source of datasets. The description also appears to be fairly clear
- 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 paper describes the goals very clearly. Easy to understand and meaningful work.
- 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?
Clarity, significance of the problem (domain shift, ultrasound images, combining / disambiguating anatomical aspects with technical imaging parameters)
- 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 authors proposed the mutual information-based MI-SegNet to focus on the domain generalization tasks for carotid ultrasound segmentation. Extensive experiments validate 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.
Relatively complete literature review. Explicit writing logic. The motivation of the method is clear, and it is easy to realize, and the effect is good.
- 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.
Lack of visualization results. Lack of some key literature about mutual information for domain shifts in medical images. Details refer to Q9.
- 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
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
Images in Fig. 1 is too small. The author should better use the space.
Mutual information is not the first time to be introduced for the tasks of tackling the domain shifts in medical images. Please discuss with the related papers (e.g., [1]-[2]) in the related work.
The authors should consider more evaluation metrics for comprehensive analysis, including HD, IOU, etc,, not just the DICE score.
Authors should consider providing some visualization results in the main text, not just put the in the supplementary materials.
[1] Cha, Junbum, et al. “Domain generalization by mutual-information regularization with pre-trained models.” Computer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXIII. Cham: Springer Nature Switzerland, 2022. [2] Huang, Yuhao, et al. “Online Reflective Learning for Robust Medical Image Segmentation.” Medical Image Computing and Computer Assisted Intervention–MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part VIII. Cham: Springer Nature Switzerland, 2022.
- 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?
Moderate novelty, good motivation, and easy to implement.
- 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.
All reviewers agreed on the contributions of this study and gave positive rates. Most of reviewers found no significant weakness on this study. Nevertheless, the comments raised by R2 should be carefully revised in the final version, such as “fig 1 should be larger for visualization’.
Author Feedback
We are grateful to receive the approval from all three expert reviewers. Thanks for the time and valuable comments. We are committed to addressing any remaining points. In this work, we presented a mutual information-based ultrasound image segmentation network for unseen domain generalization. By excluding the domain features from the segmentation network, the model can achieve high generalization capability. The codes and the dataset will soon be publicly available to facilitate the reproducibility of the work.
There are three main suggestions from the three expert reviewers’ feedback; details responses are listed below:
Reviewers 1 and 2 mentioned that additional discussion about a few more recent studies can be added to further enhance the clarification of the start-of-the-art. We agree with the reviewers and the suggested studies are highly related to our work. In the iterated manuscript, these papers have been properly discussed.
Another common suggestion from the reviewers is that more information should be demonstrated in the experiment results, e.g., utilizing a second evaluation metric other than DICE, including nnUNet in SOTA comparison, presenting the computational cost, etc. We absolutely understand the concerns of the reviewers. Considering the strict page limit and the relatively rich content, we only reported the most representative results to the readers when we submitted our first draft. Now, given that we have an extra half-page space in the revised manuscript, we are committed to demonstrate more experiment information without changing the fundamental content of the original paper.
The third point emphasized by the senior reviewer is that the authors should better visualize the figures in the manuscript. 1) Fig. 1 should be zoomed in to provide better visualization and 2) the visualization results in the supplementary material should be moved to the main paper. We highly appreciate this feedback from the reviewers. These concerns will be taken into consideration when preparing the revised manuscript.