Paper Info Reviews Meta-Review Author Feedback Post-rebuttal Meta-Reviews

Authors

Qiankun Ma, Chen Zu, Xi Wu, Jiliu Zhou, Yan Wang

Abstract

Accurate segmentation of organs at risk (OARs) from medical images plays a crucial role in nasopharyngeal carcinoma (NPC) radiotherapy. For automatic OARs segmentation, several approaches based on deep learning have been proposed, however, most of them face the problem of unbalanced foreground and background in NPC medical images, leading to unsatisfactory segmentation performance, especially for the OARs with small size. In this paper, we propose a novel end-to-end two-stage segmentation network, including the first stage for coarse segmentation by an encoder-decoder architecture embedded with a target detection module (TDM) and the second stage for refinement by two elaborate strategies for large- and small-size OARs, respectively. Specifically, guided by TDM, the coarse segmentation network can generate preliminary results which are further divided into large- and small-size OARs groups according to a preset threshold with respect to the size of targets. For the large-size OARs, considering the boundary ambiguity problem of the targets, we design an edge-aware module (EAM) to preserve the boundary details and thus improve the segmentation performance. On the other hand, a point cloud module (PCM) is devised to refine the segmentation results for small-size OARs, since the point cloud data is sensitive to sparse structures and fits the characteristic of small-size OARs. We evaluate our method on the public Head&Neck dataset, and the experimental results demonstrate the superiority of our method compared with the state-of-the-art methods.

Link to paper

DOI: https://doi.org/10.1007/978-3-030-87193-2_34

SharedIt: https://rdcu.be/cyhMc

Link to the code repository

https://github.com/DeepMedLab/Coarse-to-fine-segmentation

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

The author presents a two-stage algorithm for multi-organ segmentation combined with target detection methods. In the first stage, the TDM model is used to locate the target, then coarse segmentation is conducted, and in the second stage, fine segmentation is conducted. Taking into account the difference in scale of different organs, large organs and small organs are segmented separately using different networks, shallow feature fusion is used for large organs to solve the problem of blurred organ edge, and small organs are segmented using a point cloud model. Experiments were conducted on a public dataset and compared with the SOTA networks. The experimental results proved the effectiveness of the proposed algorithm.
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.

Novelty: Use the combination of detection and segmentation for multi-organ segmentation, and design different networks for segmentation for organs of different sizes, which can effectively improve the segmentation results of different organs. Article structure: The principle is clearly expressed, the module description is detailed, and the experiment introduction is sufficient;
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.
Problems in expression: There are problems in expression:
1. oral expressions “The architecture of the proposed method is illustrated in Fig. 2, where we can find two stages.” “The dataset was split by the Challenge, where 33 subjects are used as training set and the remaining 15 subjects are used as the test set. Before input to our model, these image volumes need pre-processing to accommodate the maximum input size of our model, i.e., 240×240×112.”
2. too long sentences: “We train our TDM using a multi-task loss function including a classification loss 𝐿@ for OARs classification task and a regression loss 𝐿U for bounding box regression task, as formulated in Equation 2”:
3. unclear expressions: “Moreover, considering the insufficiency of Cartesian coordinate information, we additionally extract the image information by following [13] and concatenate the extracted image information with the point cloud 𝑃 as the input of our PCM”; Problem in model: When the feature map of the input data is small during ROI extraction, how to ensure accurate detection of small organs and ensure the subsequent segmentation results of small organs? The problem in result analysis: In the ablation experiment, it may be limited by space, but it did not show the performance improvement brought by the segmentation of different organs of different sizes.
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

According to the author’s description, the dataset is available, training codes and settings are available, and experiments can be reproduced. However, it is insufficient in the analysis of experimental parameters and scenarios where the algorithm fails.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

The author presents an effective multi-organ segmentation method, but there are something can be improved. The mentioned sentences need to be modified, and in the experimental analysis, the ablation segmentation results need to be displayed to show the performance gain of the proposed module for different organs segmentation.
Please state your overall opinion of the paper

accept (8)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Innovation， writing
What is the ranking of this paper in your review stack?

1
Number of papers in your stack

5
Reviewer confidence

Very confident

Review #2

Please describe the contribution of the paper
- The authors develop a corse-to-fine segmentation framework for organs-at-risks(OARs) segmentation.
- In the fine stages, the authors design two different modules for large and small OARs: an edge-aware module (EMA) for capturing the boundary difference for large organs; a point cloud module (PCM) to refine small organs.
- In the experiment, the authors demonstrate superior performance to SOTA methods, and in ablation study, clearly show the contribution of different components.
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 design of two different modules for organs with different sizes is proper and interesting.
- The EAM module also predicts the boundary of the organ, serving as an additional training signal.
- The use of point cloud module (PCM) to handle sparse data, which is effective in representing small organs.
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 coarse-to-fine segmentation framework itself, especially using detection to locate OARs, is not new. As in the reference [9,10,11], previous works have proposed to use detection to perform coarse level segmentation.
- Even though the authors demonstrate the contribution of each of the individual component by reporting the average of all organs, they did not show the improvement on large and small organs separately. This is important because the EMA and PCM modules are designed and used for different organ sizes. The authors need to confirm the improvement indeed comes from the organs that different modules are designed for.
- The authors need to add more details to the Method part, e.g., how do the authors decide which organs are considered as small and which are large.
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 authors used existing public available dataset.
- The authors responded in the reproducibility checklist that they will release their code.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html
- As mentioned in the weakness, the authors may want to report the DSC scores for small and large organs separately when demonstrating the contribution of EAM and PCM.
Please state your overall opinion of the paper

Probably accept (7)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The coarse-to-fine framework for OAR segmentation itself is not new in literature [9,10,11]. The design of EAM and PCM for organs with different sizes is interesting and somewhat new.
What is the ranking of this paper in your review stack?

1
Number of papers in your stack

2
Reviewer confidence

Very confident

Review #3

Please describe the contribution of the paper

This paper proposes a new end-to-end two-stage course-to-fine segmentation network. The target detection module is to locate and segment the target. Edge-aware module and point cloud module respectively refine the segmentation results of large organs and small organs.
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.

From the results, the method proposed in this paper is simple and effective. The improvement of the proposed modules can be reflected in the experiments. And I think it is meaningful to use the point cloud.
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.
1. The experiments lacks the display of the performance of large and small organs. In the comparison experiment with the SOTA method and the ablation experiment, I hope to see the dice of each organ.
2. Is the network experimented under nnUNet? If not, I hope to see the experimental results of nnUNet (3D UNet).
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 comments.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://miccai2021.org/en/REVIEWER-GUIDELINES.html

As mentioned above.
Please state your overall opinion of the paper

borderline accept (6)
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

I am familiar with this field.
What is the ranking of this paper in your review stack?

3
Number of papers in your stack

5
Reviewer confidence

Confident but not absolutely certain

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 propose a corse-to-fine segmentation framework for organs-at-risks (OARs) segmentation.

The strengths of this paper are:
- The design of specific modules (EAM and PCM) for organs with different sizes that refine the segmentation results is proper and interesting.
- Experiments propose a comparison to SotA methods and an ablation study, and show the contribution of different components.
Weakness: the reviewers have all noted that performance (such as Dice) should have been given for small and large organs separately for the newly designed modules, since they are specifically for different organ sizes.
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).

1

Author Feedback

As suggested, we have calculated the performance gain of the newly designed modules for small and large organs separately. Specifically, by incorporating the TDM, the dice metrics of segmentation results were improved for all organs. With the introduction of the EAM and PCM, the 3D Unet+TDM+EAM achieves the best dice results on three large organs, (i.e., brainstem, parotid, and SMG R, with 88.6, 89.5, and 81.5, respectively) and the 3D Unet+TDM+PCM achieves the best dice results on two small organs, (i.e., Optical Chiasm and Optical Nerve L, with 67.3 and 75.8, respectively). We will report the results in the final paper.

back to top

Coarse-to-fine Segmentation of Organs at Risk in Nasopharyngeal Carcinoma Radiotherapy