List of Papers Browse by Subject Areas Author List
Abstract
Left atrial (LA) segmentation is a crucial technique for irregular heartbeat (i.e., atrial fibrillation) diagnosis. Most current methods for LA segmentation strictly assume that the input data is acquired using object-oriented center cropping, while this assumption may not always hold in practice due to the high cost of manual object annotation. Random cropping is a straightforward data pre-processing approach. However, it 1) introduces significant irregularities and incompleteness in the input data and 2) disrupts the coherence and continuity of object boundary regions. To tackle these issues, we propose a novel Dynamic Position transformation and Boundary refinement Network (DPBNet). The core idea is to dynamically adjust the relative position of irregular targets to construct their contextual relationships and prioritize difficult boundary pixels to enhance foreground-background distinction. Specifically, we design a shuffle-then-reorder attention module to adjust the position of disrupted objects in the latent space using dynamic generation ratios, such that the vital dependencies among these random cropping targets could be well captured and preserved. Moreover, to improve the accuracy of boundary localization, we introduce a dual fine-grained boundary loss with scenario-adaptive weights to handle the ambiguity of the dual boundary at a fine-grained level, promoting the clarity and continuity of the obtained results. Extensive experimental results on benchmark dataset have demonstrated that DPBNet consistently outperforms existing state-of-the-art methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0950_paper.pdf
SharedIt Link: https://rdcu.be/dZxdn
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72111-3_20
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0950_supp.pdf
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Xu_Dynamic_MICCAI2024,
author = { Xu, Fangqiang and Tu, Wenxuan and Feng, Fan and Gunawardhana, Malitha and Yang, Jiayuan and Gu, Yun and Zhao, Jichao},
title = { { Dynamic Position Transformation and Boundary Refinement Network for Left Atrial Segmentation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15008},
month = {October},
page = {209 -- 219}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents an approach to Left Atrial Segmentation (LAS) through the implementation of a composed network architecture. The proposed network incorporates both random and centered preprocessing techniques aimed at adjusting the position of the target. Additionally, a fine boundary search strategy is employed, leveraging an introduced boundary loss function to enhance segmentation accuracy. Authors show improvements in segmentation performance compared to existing approach UNetR.
- 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 adeptly elucidates the clinical problem, offering a clear and accessible overview, while also providing an initial classification of existing state-of-the-art methodologies. • The paper exhibits robustness in its evaluation, conducting thorough comparisons with previously published algorithms within the chosen database. Authors claim to achieve superior segmentation indices, including Dice, Jacqard, HD95, and ASSD, showcasing the efficacy of its proposed methods. • The evaluation encompasses two distinct approaches, denoted as R-DPBNet and C-RPBnNet, contributing to the comprehensive assessment of segmentation performance.
- 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 paper is not completely understood in the methodology. It needs a review of grammar as well as style, for example in the abstract the sentence: “The core idea is dynamically…” is only one examples. There are sentences that are not understood until after having read the entire article, and that cause confusion.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
Greater attention should be devoted to grammar and typographical errors. For instance, in Figure 1, the type of images being visualized is not clearly delineated until the experiments section. Additionally, other issues, such as the mislabeling of “DOBNet” instead of “DPBNet” in the supplemental material, hinder the evaluation process. These errors complicate the understanding of the methodology. A meticulous revision of style and grammar is warranted to elucidate the methodology. Specific details, such as the meaning of terms like C, H, W, D, and additional indices used in the mathematical description, should be provided. Furthermore, clarification regarding the location of components within the pipeline depicted in Figure 2 is essential for comprehensibility. Addressing these concerns will enhance the clarity of the paper.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Reject — could be rejected, dependent on rebuttal (3)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method lacks sufficient explanation or background information necessary for reproducibility. Absence of code further hinders reproducibility. Grammar and stylistic errors are particularly concerning, especially for a conference paper like this.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
It is definitely worth to further explain the bases of the proposed methodology. Many important issues remain unexplained, as well a lack of implementation details that could help to reproduce the paper and add value to the scientific community. Perhaps if the code is shared it can be clearer.
Review #2
- Please describe the contribution of the paper
An effective single-stage LA segmentation method and an novel solution to the problem of Rand-Crop is proposed.
- 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 motivation is clear and logical. The proposed methods exhibit a certain level of innovation
- 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.
Not much higher accuracy compared to SOTA. The detail of method is not clear.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
This paper points out the benefits that rand-crop can bring to LA segmentation compared to center-crop, and identifies two current problems with rand-crop: 1) region of interest (ROI) irregularity 2) disruption of boundary continuity. To address these two problems, the authors propose an attention-based mechanism to make the deep network pay more attention to the hard boundary pixels in order to solve the problem of image irregularity in the target region. In addition, the authors design a new boundary loss function. The article is very well motivated and the experimental results show the effectiveness of the method. Problems
- Is there any damage to the structural information of the image itself by shuffle operation? Can you include some images to show how the feature map being disrupted solves the problem you’ve expressed?
- As far as I know, there are a number of improved versions of Rand-Crop in the medical image processing field, does this other category of Rand-Crop address the issue you raised? Or what are the shortcomings of these methods for LA segmentation? I think adding an explanation of these methods would make your method look better. You can look for these Rand-Crop methods in the encapsulated code provided by MONAI(Medical Open Network for AI).
- It seems like your code is not available. If the ProblemⅠis not easy to solve, available code is a strong proof to demonstrate your work.
- You mentioned that you were inspired by the SwinTransformer, so it might be better to include experiments with SwinUNetr in the comparison experiments.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the code is not publicly available and there are some problems with the main methodology, the motivation is clear and logical.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
The rebuttal has addressed my concerns
Review #3
- Please describe the contribution of the paper
This paper proposed a shuffle-then-reorder attention module to adjust the position of disrupted objects in the latent space for capturing dependencies among random cropping targets. Besides, authors introduced a boundary weight loss to handle the boundary in fine-grained level.
- 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 is well-organized. It firstly investigate random cropping input for one-stage framework.
- 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.
6.1) The paper mentioned that its idea was inspired by the Transformer model, but there is no comparative experiments executed on the Transformer in the experimental section. Authors should illustrate formula (5) in details. What’s the meaning of K^3 and why ground truth g_i is processed by 3D-CNN layer?
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
Authors should illustrate formula (5) and demonstrate the strengths and weakness between proposed model and Transformer.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Critical formula need reinterpretation.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
Authors had solved the puzzel of mine.
Author Feedback
We thank reviewers (R1, R3, & R4) for their valuable feedback, describing our approach as “novel” and “very well motivated” (R1), “ robustness” (R3), “well-organized” and “first investigation” (R4). We sincerely appreciate their insightful comments and address their concerns as below.
Q1: Image shuffling influence and visualization. (R1) A1: Thanks. The shuffling process employed in our method does not compromise the structural information of the image. This is ensured by the subsequent reorder operation, which can accurately reconstruct the original positions. To provide a more intuitive visualization, we have utilized Grad-CAM [1] to visualize the intermediate feature results on DPBNet. Despite the random cropping that generates separate targets, DPBNet can effectively focus on those targets by shuffle-and-reorder attention mechanism. We will incorporate the visual results and discussions in the final version.
Q2: Limitations of the Rand-crop methods and more clarifications. (R1) A2: Thanks. The limitation of Rand-Crop methods in preserving integrous targets within medical images [2] is well-known, as the lack of predetermined target regions results in disruptions regardless of the random cropping method employed. We will provide detailed explanations of the Rand-Crop in the final version. to enhance readability.
Q3: Presentation. (R3) A3: Thanks. We have made significant improvements to enhance the presentation of this paper in the following ways. 1) We have carefully double-checked the manuscript, including grammatical and typographical errors. and improving clarity. 2) The symbols C, H, W, and D denote the number of channels, height, width, and depth within a feature map, respectively. To avoid confusion for readers, we will add a notation table in the final version, elucidating the frequently used symbols. 3) Inspired by your comments, we have carefully revised the notations, annotations, and sub-figures in Fig. 2. This includes providing detailed depictions of the convolution block, sigmoid function, shuffle ratios, and a more distinct illustration of the DFB Loss. Due to display limitations, the revised Fig. 2 will be included in the final paper.
Q4: Comparison with Transformer structure. (R1 & R4) A4: Thanks. Firstly, in our previous manuscript, we have compared our method with an advanced transformer-based method termed CANet in Table 1, demonstrating the superiority of our method. Following your suggestion, we have conducted comparisons between SwinUNETR and DPBNet, empirically demonstrating the superiority of our method on the LA dataset. These findings highlight the inherent challenges faced by Transformer-based networks in LA segmentation, which is characterized by small-scale data and significant variability in atrial structures. Due to the rebuttal policy restrictions, these results and discussions will be included in the final version.
Q5: Formula (5). (R4) A5: Thanks. Formula 5 is designed to assign weights based on the contrasting values surrounding the central point. We have revised this formula to make it clearer: w_i={k^3 - f^3d(g_i) + 1, i=1; f^3d(g_i) + 1, i=0}. where k^3 represents the maximum possible count of foreground points within the k-nearest neighbors. Moreover, we employ a 3D-CNN to accurately determine the actual number of foreground points (with a value of 1) within the k-nearest neighbors of a central point. For instance, when the central point is classified as a foreground point, we determine the count of surrounding background points by subtracting the output of the 3D-CNN from k^3.
Q6: Reproducibility and code sharing (R1 & R3). A6: Thanks. The complete source codes and pre-trained models will be released for reproducibility if the paper is accepted.
[1] Selvaraju, R. R., et al. Grad-cam: Visual explanations from deep networks via gradient-based localization. ICCV, 2017. [2] Shorten, C., et al. A survey on Image Data Augmentation for Deep Learning. Journal of big data, 2019.
Meta-Review
Meta-review #1
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All the reviewers agree to accept it potentially, and I also believe that itis a good article.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
All the reviewers agree to accept it potentially, and I also believe that itis a good article.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All reviewers agree to accept this paper.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
All reviewers agree to accept this paper.