List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Hengxu Zhang, Pengpeng Liang, Zhiyong Sun, Bo Song, Erkang Cheng
Abstract
Both CNN-based and Transformer-based object detection with bounding box representation have been extensively studied in computer vision and medical image analysis, but circular object detection in medical images is still underexplored. Inspired by the recent anchor free CNN-based circular object detection method (CircleNet) for ball-shape glomeruli detection in renal pathology, in this paper, we present CircleFormer, a Transformer-based circular medical object detection with dynamic anchor circles. Specifically, queries with circle representation in Transformer decoder iteratively refine the circular object detection results, and a circle cross attention module is introduced to compute the similarity between circular queries and image features. A generalized circle IoU (gCIoU) is proposed to serve as a new regression loss of circular object detection as well. Moreover, our approach is easy to generalize to the segmentation task by adding a simple segmentation branch to CircleFormer. We evaluate our method in circular nuclei detection and segmentation on the public MoNuSeg dataset, and the experimental results show that our method achieves promising performance compared with the state-of-the-art approaches. The effectiveness of each component is validated via ablation studies as well.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43993-3_48
SharedIt: https://rdcu.be/dnwNT
Link to the code repository
https://github.com/zhanghx-iim-ahu/CircleFormer
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper develops a CircleFormer for Circular Nuclei Detection in Whole Slide Images with Circle Queries and Attention.
- 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 paper has some noverlity.
- the simulations can verify the effectivenss of the proposed work.
- 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 main contributions are missing.
- lack of qualitative results.
- 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
It can be reproducable.
- 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 main contributions are missing.
- lack of qualitative results.
- 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?
the quantitative is sufficient but qualitative results is missing.
- 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
This work presents a transformer based circular object detection models with dynamical circle, and experiments on the public MoNuSeg dataset shows its superior performances.
- 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 main strength of the work is that it presents a circle object detection framework based on transformer which bring a large margin of performance boosting comparing to other works.
- 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 main weakness of the work is that it only test on the nuclei detection task which the size of objects are small with regular shape. And also, detection nuclei is not a hard work in clinical practice.
- 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 didi not provide their source code, and thus the reproducibility is not guaranteed.
- 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
I would recommend the authors to. test on some other tasks where the objects are more diverse according to the shape and size.
- 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?
The authors provide a general framework for circular object detection, although the performance is significant, they need more test on. diverse tasks to demonstrate. the generalizability of the new method.
- 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 proposes a new approach for circular object detection in medical images using a Transformer-based method called CircleFormer. The approach uses queries with circle representation in the Transformer decoder to refine circular object detection results, and introduces a circle cross attention module to compute similarity between circular queries and image features. Additionally, the paper proposes a new regression loss, called generalized circle IoU (gCIoU), for circular object detection.
- 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 is well-written and clearly structured. The paper’s contribution is further enhanced by the use of a new regression loss called generalized circle IoU (gCIoU), which improves the accuracy of circular object detection. Additionally, the paper includes an ablation study to validate the effectiveness of each component of the proposed method, which adds credibility to the proposed approach.
- 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.
Although the paper presents promising experimental results for circular nuclei detection and segmentation, a visual comparison would have added more clarity to the performance comparison. Another weakness of the paper is that the choice of hyperparameters is not well-explained, particularly with regard to the proposed loss function.
- 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
To ensure the reproducibility of the results, I recommend that the authors release the code and any relevant details about the implementation.
- 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 Fig.2 focuses on the network details rather than the method overview. It would have been beneficial to have a more detailed description of the selection process and how different choices of hyperparameters affect the performance of the proposed method. A more detailed discussion of the hyperparameter selection process and a visual comparison of the proposed method against state-of-the-art approaches would have improved the paper’s overall contribution and credibility.
- 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?
The idea of anchor circles is not novel, the combination with the ad-hoc loss function and Transformer shows improvement.
- 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
6
- [Post rebuttal] Please justify your decision
According to the feedback, the author has answered questions raised by reviewers and promised to revise the paper to add qualitative results and to make the code available.
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 introduces CircleFormer, a novel Transformer-based approach for circular object detection in medical images. The reviewers recognized the innovation of the proposed method, the improved performance achieved, and the thorough ablation study conducted. However, they expressed concerns regarding the limited datasets used, the absence of qualitative results, and insufficient technical details provided. For detailed feedback, please refer to the reviewer comments. It is crucial for the author to address these major concerns.
Author Feedback
We thank the reviewers (R1, R2, R3 and Meta) for their valuable suggestions and would like to make the following clarifications.
Q1 (R1): Main contributions are missing. Thanks for this meaningful suggestion. In this paper, we design a transformer-based method for circular object detection in medical images. The main contributions include: (1) Queries based on dynamic anchor circle are designed for the transformer decoder. (2) Circle cross attention is proposed to capture the appearance features of a given circle. A circle matching loss with a generalized circle IoU is proposed as well. (3) Superior performance of our algorithm over CNN-based circle detection methods and transformer-based rectangle detection approaches.
Q2 (R2, Meta): Limited datasets. We understand the concern regarding the limited datasets used in our experiments. It is worth noting that the MoNuSeg dataset we use is publicly available and consists of over 50,000 nuclei with some images containing close to 1,000 objects. The dataset, originally introduced by [16], is specifically created for nuclei detection and includes challenging cases with significant object overlap. As described in [16], detecting a large number of overlapping objects is indeed a challenging task. Therefore, MoNuSeg dataset provides a substantial number of target objects to demonstrate the effectiveness of our algorithm. We totally agree that the importance of testing our approach on more diverse datasets. We have verified that our method is also effective on the detection of spheres in 3D images in an ongoing work, and we promise to add the results on 3D images in the revised version.
Q3 (R1, R3 and Meta): Absence of qualitative results and visual comparison. Thanks for raising this question, and qualitative results and visual comparisons are indeed very important in medical image analysis. Due to space limitations, we were unable to present visual comparisons in our manuscript. In the camera-ready version of the paper, upon acceptance, we will address this concern by adjusting the structure of the paper and adding visual results.
Q4 (R3): Lack of method overview figure. We appreciate your feedback regarding the absence of an overview figure of our network in the manuscript. Similar to [10], as our proposed module primarily focuses on the decoder part of the Transformer architecture, an overview picture of the network is not provided. We understand the value of having an overview figure to aid in understanding the method. In the revision, we will add an overview figure to show the scheme of our approach.
Q5 (R3, Meta): Insufficient technical details. We understand the importance of providing detailed technical information. In our experiments, we directly use the hyperparameters of the DAB-DETR model for fairness and reproducibility. We will add more detailed technical information about our implementation and experimental setup in the revised version to address this concern. Also, we promise to make our code publicly available.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
This paper presents CircleFormer, a novel approach based on Transformers for circular object detection in medical images.
During the first round of review, the reviewers acknowledge the innovation of the proposed method, the achieved performance improvements, and the thorough ablation study conducted. However, they raise concerns regarding the limited datasets used, the absence of qualitative results, and insufficient technical details provided. The author responds with a comprehensive rebuttal, summarizing and addressing these concerns. As a result, the paper receives two positive reviews and one negative review.
The incorporation of circle representation, which integrates domain knowledge into object detection, is an intriguing aspect that warrants presentation at MICCAI. The evaluation process demonstrates rigor by employing comprehensive benchmarks. I believe the rebuttal effectively addresses the concerns raised during the review process.
Based on these reasons, I am inclined to recommend accepting the paper.
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
This manuscript presents a Transformer-based object detector, which is specifically designed for circular objects (e.g., nuclei) in histopathological images. Compared with box representation-based detectors, this method can produce better nuclei detection performance. In addition, the rebuttal has addressed the major concerns from the reviewers, such as the limited dataset used in the experiments, missing technical details and lack of qualitative results. Thus, an acceptance is suggested.
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The authors have adequately responded to the reviewer concerns and provided a plan to significantly revise the paper to add necessary qualitative results and release code. I recommend acceptance.