List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Wenhui Zhang, Surajit Ray
Abstract
The use of functional imaging such as PET in radiotherapy (RT) is rapidly expanding with new cancer treatment techniques. A fundamental step in RT planning is the accurate segmentation of tumours based on clinical diagnosis. Furthermore, recent tumour control techniques such as intensity modulated radiation therapy (IMRT) dose painting requires the accurate calculation of multiple nested contours of intensity values to optimise dose distribution across the tumour. Recently, convolutional neural networks (CNNs) have achieved tremendous success in image segmentation tasks, most of which present the output map at a pixel-wise level. However, its ability to accurately recognize precise object boundaries is limited by the loss of information in the successive downsampling layers. In addition, for the dose painting strategy, there is a need to develop image segmentation approaches that reproducibly and accurately identify the high recurrent-risk contours. To address these issues, we propose a novel hybrid-CNN that integrates a kernel smoothing-based probability contour approach (KsPC) to produce contour-based segmentation maps, which mimic expert behaviours and provide accurate probability contours designed to optimise dose painting/IMRT strategies. Instead of user-supplied tuning parameters, our final model, named KsPC-Net, applies a CNN backbone to automatically learn the parameters and leverages the advantage of KsPC to simultaneously identify object boundaries and provide probability contour accordingly. The proposed model demonstrated promising performance in comparison to state-of-the-art models on the MICCAI 2021 challenge dataset (HECKTOR).
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43901-8_51
SharedIt: https://rdcu.be/dnwDZ
Link to the code repository
N/A
Link to the dataset(s)
https://www.aicrowd.com/challenges/miccai-2021-hecktor
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents a method to segment tumors in PET images using a neural network with a gaussian kernel (usefully in the case of fuzzy objects) and a particular emphasis on probability contours. The method is validated on HEKTOR 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.
- To my opinion the major strength of this paper is the probability contours part. Indeed, in the case of fuzzy objects such as in PET images, I find this part very interesting.
- 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 evaluation part is too poor. in the case of fuzzy segmentation, we cannot say that a dice is better than an other one with only 0.002 of difference, it is not significant.
- the choice of the methods of the SOTA is not really justified, and their use is not clear especially for the top methods of HEKTOR challenge
- 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 link of the code is not available, but I think it will be added. The dataset is public and parameters of the method are described.
- 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 global idea of the article is good. This article suffers from some weaknesses as mentioned above. The evaluation part needs to be clearer. In the case of fuzzy objects, 0.002 of difference in terms of dice is not significant. The choice of the methods of the state of the art is good except for the top methods of Hektor Challenge. it is not clear at all if the 3D-Unet is the method of the winner of the 2022 challenge or an other 3D-Unet. The justification of the SOTA choice has to be written. When a recent dataset such as Hektor is used, it is mandatory to compare the method with the top-methods. The way the evaluation is performed is also not clear. Are the SOTA methods re-trained? For MSA and CCUT, what are these results? On the test set? The evaluation part should be reworked.
- 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?
There are some good points in this article, but too many points remains not clear.
- 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
The paper presents a new method for detecting tumor contours in PET imaging. The method extends a previous study by the author using a kernel smoothing-based (KsPC) method to extract the 3D contour of a grayscale image. In this study, the author proposes to further develop by combining it with a Unet model to optimize the input parameters of KsPC. Additionally, the author proposes a method for detecting the contour of isovalue of the intensity of a PET image. To validate the method, the paper uses the dataset from the HECKTOR Challenge and compares it to different models, including two from the challenge itself. The method proposed provides better results for the dice score, Hausdorff distance, and recall.
- 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 introduction is well developed, and the problem and contribution of the article are easily understood. The KsPC method is well described and clear. The validation method is well defined and highlights the added value of the 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.
For the introduction, it is unfortunate that there is no real state-of-the-art of other possible methods for extracting a 3D contour. Fortunately, the use of the challenge and associated methods (MSA-Net and CCUT-Net) highlights the method. The presentation of the KsPC method may be a bit too detailed, considering that it is not the added value of the paper and that it comes from a previous publication. It would have been interesting to push the description of KsPC-Net a bit further. To be able to reproduce the study, it would be interesting to provide the code as open source.
- 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 uses a public dataset to validate the proposed method. The method is presented through multiple papers, one of which defines the KsPC method, while the other presents the KsPC-Net. The paper can be considered reproducible; however, providing the code would enhance its reproducibility.
- 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
Thank you for this work,
The overall structure of the article is good, and we understand the issue, the application case, as well as the added value of the paper’s method.
However, some points could be improved:
A brief introductory paragraph presenting the various existing 3D contour extraction methods. The presentation of the KsPC-Net model deserves to be further elaborated, particularly the estimation of the contour probability.
- 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 application case should be accompanied by a clear explanation of its usefulness and a comparison with reference methods such as Unet and methods from the HECKTOR challenge.
- 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 #3
- Please describe the contribution of the paper
This paper proposed a kernel smoothing-based probability contour approach to produce contour-based segmentation maps for dose painting in radiotherapy planning. The proposed method can output accurate and reproducible object boundaries.
- 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 proposed a novel contour prediction pipeline in PET image segmentation, incorporating the kernel smoothing-based probability contour into segmentation network training.
- The proposed method is able to provide probability contours instead of a fixed one like in traditional U-Net.
- The proposed method achieved SOTA performance in comparison to other methods in comparison.
- 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 clinical benefit of the probability contour was not sufficiently motivated and elaborated.
- Some notations in the loss function part (Sec. 2.3) are not clearly explained, for example K_KsPC, L_CNN.
- 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
No concerns here because the authors will make the code available.
- 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
- Using U-net for segmentation and then KsPC for probability contour generation is probably also doable, it might be interesting to show the performance in this setting.
- The presentation of loss functions needs to be improved as some notations are not clearly explained.
- 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?
This paper presents a practical PET image segmentation method which can output probability contours. Overall, the paper is easy to follow and the proposed pipeline can easily be reproduced. However, presentation of the method and discussion about clinical impact could be improved.
- 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
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.
Strength: 1) Technical contributions are good by using probability contours for tumor segmentation on PET images. 2) Motivation is strong, excellent introduction to the problem statement, and validation experiments are sufficient.
Weakness: 1) As pointed by reviewer 1, there is no way to claim significant with only 0.002 improvement on Dice coefficient. 2) Reduce the context of KsPC method, while highlighting KsPC-Net more. 3) Explain more about loss functions.
Author Feedback
We thank all reviewers and the area chair for their invaluable comments and supportive feedback. All general clarification requests will be fully and thoroughly handled in the revised version, including (1) the addition of the open source code to assure the paper’s reproducibility [R1, R2, R3]. (2) Focus more on the selection and specifics of KsPC-Net [R1,R2,AC] and lessen the context of KsPC. (3) For the evaluation part, indeed, we acknowledge there was no significant improvement (0.004 dice) over other SOTA methods. However, our goal here is to show that probability contours can be obtained as a natural byproduct while preserving state-of-art equivalent accuracy level of the contour. In the updated version, the explanation will be re-illustrated [R1, R2, AC]. (4) The definition of the loss function is not adequately explained due to space constraints; this will be fixed and further discussed in the improved edition [R3,AC].