List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Zhenxi Zhang, Ran Ran, Chunna Tian, Heng Zhou, Xin Li, Fan Yang, Zhicheng Jiao
Abstract
Consistency learning plays a crucial role in semi-supervised medical image segmentation as it enables the effective utilization of limited annotated data while leveraging the abundance of unannotated data. The effectiveness and efficiency of consistency learning are challenged by prediction diversity and training stability, which are often overlooked by existing studies. Meanwhile, the limited quantity of labeled data for training often proves inadequate for formulating intra-class compactness and inter-class discrepancy of pseudo labels.
To address these issues, we propose a self-aware and cross-sample prototypical learning method (SCP-Net) to enhance the diversity of prediction in consistency learning by utilizing a broader range of semantic information derived from multiple inputs. Furthermore, we introduce a self-aware consistency learning method which exploits unlabeled data to improve the compactness of pseudo labels within each class. Moreover, a dual loss re-weighting method is integrated into the cross-sample prototypical consistency learning method to improve the reliability and stability of our model. Extensive experiments on ACDC dataset and PROMISE12 dataset validate that SCP-Net outperforms other state-of-the-art semi-supervised segmentation methods and achieves significant performance gains compared to the limited supervised training.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43895-0_18
SharedIt: https://rdcu.be/dnwx0
Link to the code repository
https://github.com/Medsemiseg/SCP-Net
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This work studied prototype-based semi-supervised segmentation techniques, and proposed self-aware prototype prediction and cross-sample prototype prediction to enhance the robustness.
- 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.
1.the paper is well written and easy to follow. 2.the proposed cross-sample prototype prediction is interesting, and experiments showed its effectiveness. 3.the experiments is valid to prove the benefit of the proposed modules.
- 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 motivation of key idea of the proposed method is not clear, and the description is a bit confusing.
- 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 mathematical formulation of the proposed method is clear, and thus could be reproduced.
- 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
- on page 2, it said ‘minor perturbations that have no discernible effect on the predicted result’, to my knowledge, VAT can produce the adversarial noise that have high influence on the prediction.
- on page 2, it said ‘, prior research has neglected to consider the robustness and variability of prediction results in response to perturbations’. I think previous works aims to improve the robustness of model against perturbations.
- on page 2, it said ‘to ensure appropriate prediction diversity in consistency learning’, how is the diversity guaranteed?
- please keep consistent symbols for p_{ki}^{c} in eq.(1) 5.for Eq.(6), what if e_k is larger than 1, and w_{1ki} is then a negative value? Eq.(9) would be an unstable constraint.
- for comparison study, please analyse with p-value for statistical test
- in Table 1, why the assd of URPC is the best while their DICE is the worst.
- Does all the compared method have been tuned for hyper-parameters with the validation set?
- 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 think the idea about using cross-sample prototype prediction is quite interesting.
- 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
This paper presents a new semi-supervised segmentation method based on prototype comparison. to enhance the diversity of prediction in consistency learning by utilizing a broader range of semantic information derived from multiple inputs.
- 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 novel angle to use the prototype in semi-supervised segmentation.
- Validation on two public datasets show the effectiveness of the proposed method.
- Well-written.
- 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.
- Each operation needs more detailed motivation and explanation, otherwise it’s kind of hard to follow. For example, how the binary represented mask is generated ?
- Some definitions are not well justified. E.g., what is the reason to normalize entropy by C?
- There are many bells and whistles which make the real contribution less obvious. The complexity makes it hard to track which parts of the design exactly contribute to the improved performance.
- 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
un-known
- 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
to improve the readability.
- 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?
A quite engineering-intensive approach, and the narrative need improvement for the readability. An interesting angle to the semi-supervised segmentation.
- 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 #3
- Please describe the contribution of the paper
This paper makes use of unlabeled data to enhance semi-supervised segmentation performance. It proposes self-aware prototype prediction and cross-sample prototype prediction to ensure sufficient diversity in consistency learning. Moreover, to improve the stability and reliability of cross-sample prototypical consistency learning, the paper introduces a dual loss re-weighting method, which effectively mitigates the negative impact of noisy pseudo labels.
- 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 presents a network architecture that leverages unlabeled data to enhance semi-supervised segmentation performance. The proposed approach integrates self-aware and cross-aware prototypical consistency constraints, with the latter incorporating uncertainty estimation and self-aware probability prediction. The presented methodology demonstrates promise for advancing the field of semi-supervised segmentation.
- 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 distinction between the proposed method and commonly used prototypical learning is ambiguous. The Introduction briefly introduced consistency constraint whereas prototypical learning was not, making the research gap unclear. In the method, which part is proposed by the author, and which part is introduced from other work is uncertain., leading to uncertainty regarding the originality of the work.
- 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 manuscript provides some information that could facilitate the reproducibility of the method, including the use of the UNet backbone, the connection of CPCC and SPCC to their respective equations, and the presentation of hyperparameters. However, the lack of access to the source code limits the extent to which the method can be reproduced, which raises concerns about the robustness and reliability of the reported results.
- 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
-
It would be beneficial to include a more comprehensive introduction to “prototypical learning” in the paper’s Introduction section. This would help readers to better understand the context of the proposed method and how it differs from existing techniques.
-
The paper mentions that the variability of prediction results in response to perturbations is significant and should be considered. However, it would be helpful to elaborate on why this is important and how it relates to the practical applications of the proposed method.
-
Please address the typo in clarifying Equation 6 “e_k \in”.
-
It would be helpful to clarify the difference between “cross-aware” and “cross-sample” and align the expression to avoid any confusion or misunderstanding.
-
The paper states that “All 3D scans are converted into 2D slices. Then, each slice is resized to 256 × 256 and normalized to [0, 1]”. However, it would be useful to explain why the authors chose to do this rather than build a 3D model.
-
To further evaluate the proposed method’s performance, please consider adding an experiment to Table 2 that illustrates the Lseg + Lcpcc + Lspcc with only w2.
-
- 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?
In my recommendation of Accepting this manuscript, I take into consideration several key factors, including the clear and well-written presentation of the research, the extensive experimental results that demonstrate the efficacy of the proposed method, and somewhat novelty of the 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
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 proposed a new prototype-based semi-supervised segmentation method. Reviewers were in consensus that the work was (a) interesting with a novel angle, and (b) convincing in experimental validation.
Authors should address reviewer feedback and suggestions to improve the introduction (esp on related work in prototypical learning), readability, and provide more details on the methods in the camera ready version.
Author Feedback
Dear Reviewers,
Thank you for your valuable comments and suggestions on our paper #503. We appreciate your time and effort in reviewing our work and are pleased to know that you found it interesting with a novel angle and convincing in experimental validation. We have carefully considered your feedback and have made the following improvements to the camera-ready version:
(1) Introduction: We have revised the introduction to provide a more comprehensive and detailed overview of related work in prototypical learning. Additionally, we have analyzed the distinctions between our method and existing prototypical learning methods. (2) Readability: To address this concern, we have made several improvements. Firstly, we have ensured consistency in the expressions used throughout the paper, such as replacing “cross-aware” and “cross-sample” with appropriate terms. Secondly, we have carefully reviewed and corrected any typos in the equations. Additionally, we have provided detailed explanations for Equation 6 to enhance its clarity and understanding. Furthermore, we have analyzed the significance of prediction diversity in the context of semi-supervised consistency learning, highlighting its importance. (3) Methodology: In response to your suggestion, we have incorporated additional details on the proposed SCP-Net in the camera-ready version. Specifically, we have provided comprehensive explanations of the self-aware prototypical learning and cross-sample prototypical learning components of SCP-Net. Our aim is to enhance the reader’s understanding and motivation behind each step of SCP-Net. By supplementing the motivation behind these steps, we strive to provide a clearer and more comprehensive picture of the proposed approach.
Once again, we appreciate your constructive feedback, which has helped us improve the quality and presentation of our work. We believe that the revised camera-ready version of our paper addresses your concerns and aligns with the high standards set by MICCAI 2023. Thank you for your time and consideration.