Authors

Long Chen, Han Li, S. Kevin Zhou

Abstract

In computational pathology, nuclei segmentation from histology images is a fundamental task. While deep learning based nuclei segmentation methods yield excellent results, they rely on a large amount of annotated images; however, annotating nuclei from histology images is tedious and time-consuming. To get rid of labeling burden completely, we propose a label-free approach for nuclei segmentation, motivated from one pronounced yet omitted property that characterizes histology images and nuclei: intra-image self similarity (IISS), that is, within an image, nuclei are similar in their shapes and appearances. First, we leverage traditional machine learning and image processing techniques to generate a pseudo segmentation map, whose connected components form candidate nuclei, both positive or negative. In particular, it is common that adjacent nuclei are merged into one candidate due to imperfect staining and imaging conditions, which violate the IISS property. Then, we filter the candidates based on a custom-designed index that roughly measures if a candidate contains multiple nuclei. The remaining candidates are used as pseudo labels, which we use to train a U-Net to discover the hierarchical features distinguish nuclei pixels from background. Finally, we apply the learned U-Net to produce final nuclei segmentation. We validate the proposed method on the public dataset MoNuSeg. Experimental results demonstrate the effectiveness of our design and, to the best of our knowledge, it achieves the state-of-the-art performances of label-free segmentation on the benchmark MoNuSeg dataset with a mean Dice score of 79.2%.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43987-2_65

SharedIt: https://rdcu.be/dnwKo

Link to the code repository

N/A

Link to the dataset(s)

https://monuseg.grand-challenge.org/Data/

Reviews

Review #1

Please describe the contribution of the paper

In this paper, the authors propose the SSimNet for unsupervised nucleus segmentation. First, the authors design a pipeline to produce the initial instance labels using the color/stain information. Second, the authors propose the USMI for training data selection.
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.

First, the authors design a pipeline to produce the initial instance labels using the color/stain information. Second, the authors propose the USMI for training data selection. According to Table 1, the proposed method achieves promising results.
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.

Since the proposed method relies on channel decomposition to produce pseudo labels, the different staining types may affect the performance and the robustness of the proposed method. Therefore, it is recommended to provide the variation of the model performances with the distribution of the stain color matrix.
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 authors claim to release the codes.
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

1) As illustrated in Section 2.2, the proposed algorithm only uses 70% patches for training. The author can draw the curve of the variation of the USMI values with the data ratios, which can better illustrate the importance of data purification. 2) It is recommended to conduct experiments to evaluate the effects of the ratios of the selected patches.
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?

In this paper, the proposed patch-selection strategy is promising. And the experimental results verify the effectiveness of the proposed method.
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 #3

Please describe the contribution of the paper

This paper proposes a new method for nuclei segmentation that does not require any manual labeling. The approach is based on intra-image self similarity and combines traditional image processing techniques with machine learning methods. The method also includes filters that are designed to detect multiple nuclei in a single candidate. The experimental results demonstrate that this approach achieves state-of-the-art performance in label-free segmentation and is comparable to fully supervised methods.
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 makes an interesting contribution to the field of nuclei segmentation by presenting a novel label-free approach based on intra-image self-similarity. The approach incorporates traditional image processing techniques and machine learning methods, including filters designed specifically to detect multiple nuclei in a candidate. The paper also introduces new methods for data purification, clustering, and active contour that are specifically tailored for this task. The use of channel decomposition to enhance the clarity of the Hematoxylin channel image is also a notable contribution.

The authors’ strong evalution of their approach is a major strength of the paper. They compare their work to both label-free and fully supervised methods, and provide both mean and standard deviation of the experimental results. This level of detail allows for fair and accurate comparison of the performance of different methods, making the paper a valuable contribution to the field.
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.

This paper lacks a detailed explanation of the data purification step, especially regarding the usage of Voronoi diagram. It would be helpful to provide more justification for this method, as using the Voronoi edge as 0 supervision may negatively affect the training process since the boundary of the nuclei should not be straight lines.

The necessity of using SSimNet is not clearly explained in the paper. It seems that traditional clustering and Voronoi separation methods could already separate the nuclei effectively, and the contribution of SSimNet in this regard is not clear,

Furthermore, the authors could have compared their proposed approach with more state-of-the-art nucleus segmentation methods, such as HoVer-Net, which is mentioned as the first reference in this paper, to demonstrate the superiority of their approach more comprehensively.
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

The paper seems to have a good reproducibility. The authors promise that they will release the training code and evaluation 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://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

1.It would be helpful if the authors could provide more explanation on the necessity of using the voronoi diagram and SSimNet in their approach. The current paper does not provide enough justification for why these methods were chosen and how they contribute to the overall performance. Adding more details and insights into these choices would make the paper more informative and helpful for future research. 2.The quality of Fig. 1 needs to be improved as it currently appears to be a low-quality image. It should be made more presentable to meet the high standards of a MICCAI paper. 3.It would be beneficial if the authors could compare their proposed approach with more state-of-the-art methods for nucleus segmentation, such as HoVer-Net. This comparison would help demonstrate the superiority of their method. 4.To better illustrate the performance of the proposed approach in separating the nuclei that are overlapped, it may be helpful to use differnt colors for the different instances in Fig. 4. The current visualization does not clearly show how well the method performs in this main challenge in nuclei segmentation.
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?

Although this paper presents an interesting new method that could provide new insights into the task of nuclei segmentation, the paper’s clarity and organization could be improved. The authors could provide more justification for why they chose certain steps in their approach. Additionally, the illustrations in the paper are not visually appealing and the resolution is too low, making it difficult to fully understand the approach. Overall, with some improvements in clarity, organization, and visualization, this paper could be a valuable contribution to the field of nuclei segmentation.
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 article proposes a label-free approach for segmenting nuclei from histology images. The method uses intra-image self-similarity (IISS) to generate pseudo-labels for training a U-Net, achieving state-of-the-art performance on the benchmark MoNuSeg dataset with a mean Dice score of 79.2%.
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. SSimNet is an effective method for label-free nuclei segmentation that employs the IISS property to capture prior knowledge and generate pseudo-labels as a supervision signal for learning the network.
2. The experiments are comprehensive and convincing.
3. Unsupervised nuclei segentation is practically significant to relieve the need for annotation.
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 robustness of the intra-image self-similarity needs to be further evaluated. For example, for one pathology image where multiple types of nuclei are included, or nuclei exist with diverse appearances, how is the performance of the intra-simage self-similarity?
2. How is the hyperparameter sentivity, e.g, alpha and lambda? More ablation studies are desired to evaluate this.
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

Authors do not mention whether to make the code publicly 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

The motivation is interesting and the method and experiments are convincing. I mainly wonder the proposed IISS is robust to intra-diversity and how is the hyperparameter sensitivity.
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?

Based on the above strengths and weaknesses, the paper is rated.
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 proposes a novel method for label-free cell segmentation in pathology images. Reviewers acknowledged the novelty of the paper, the outstanding experimental results compared with existing work, and the importance of the research study. Although the reviewers are generally positive about the paper, they also raised some concerns regarding the unclear method motivation and insufficient structure details. These should be updated in the final version.

Author Feedback

N/A

back to top

Label-Free Nuclei Segmentation Using Intra-Image Self Similarity