Abstract

The segmentation of cardiovascular arteries in 3D medical images holds significant promise for assessing vascular health. Despite the progress in current methodologies, there remain significant challenges, especially in the precise segmentation of smaller vascular structures and those affected by arterial plaque, which often present as disconnected in images. Addressing these issues, we introduce an innovative refinement method that utilizes a data-driven strategy to correct the appearance of disconnected arterial structures. Initially, we create a synthetic dataset designed to mimic the appearance of disconnected cardiovascular structures. Our method then re-frames the segmentation issue as a task of detecting disconnected points, employing a neural network trained to identify points that can link the disconnected components. We further integrate an open curve active contour model, which facilitates the seamless connection of these points while ensuring smoothness. The effectiveness and clinical relevance of our methodology are validated through an application on an actual dataset from a medical institution.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3954_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/3954_supp.pdf

Link to the Code Repository

https://github.com/YuliWanghust/CTA_repairment

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Wan_CarDcros_MICCAI2024,
        author = { Wang, Yuli and Hsu, Wen-Chi and Shi, Victoria and Lin, Gigin and Lin, Cheng Ting and Feng, Xue and Bai, Harrison},
        title = { { Car-Dcros: A Dataset and Benchmark for Enhancing Cardiovascular Artery Segmentation through Disconnected Components Repair and Open Curve Snake } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15001},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper describes an approach to segment coronary arteries and reconnect potential disconnected segments.

  • 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.
    • Intriguing idea with simulation disconnected vessels, I really like that!
  • 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 literature could be discussed in more details, especially regarding pros and cons of existing methods. A comparison with other literature is missing:
      • G2 Blending Ball B-Spline Curve by B-Spline: https://dl.acm.org/doi/10.1145/3585504
      • DeepVesselNet: Vessel Segmentation, Centerline Prediction, and Bifurcation Detection in 3-D Angiographic Volumes: https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2020.592352/full
    2. The results are not convincing. I would like to see more examples before and after reconnection. How many segments are reconnected? How many iterations are required? How many segment are reconnected per iteration?
    3. The evaluation could be improved by employing, e.g., established tracking algorithms for reconnection, and compare them with the proposed method.
  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Do you have any additional comments regarding the paper’s reproducibility?

    No.

  • 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

    Please see the strengths and weaknesses.

  • 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

    Reject — should be rejected, independent of rebuttal (2)

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

    Although the manuscript offers an interesting perspective on learning how to reconnect disconnected segments, it is not ready for publication in its current form. It needs a fundamental revision. I therefore argue that it should be rejected in its current form

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Weak Reject — could be rejected, dependent on rebuttal (3)

  • [Post rebuttal] Please justify your decision

    Despite the comments on my remarks, I am still not convinced. There are so many papers that deal with reconnecting disconnected components or identifying appropriate points (or even sub-vessel-trees) for reconnection. I don’t see where this manuscript stands out. Other papers that deals with that (although not with CNNs) are:

    • T. Pock: Robust Segmentation of Tubular Structures in 3D Volume Data, Master’s thesis, TU Graz, 2004
    • G. Mistelbauer et al.: “Semi-automatic vessel detection for challenging cases of peripheral arterial disease.” CIBM 2021, DOI: 10.1016/j.compbiomed.2021.104344



Review #2

  • Please describe the contribution of the paper

    The authors proposed a new framework for cardiovascular artery segmentation. The authors consider this problem as a segmentation+tracking problem, where they segment the 3D volume and detect disconnected points by CNN, then repair disconnection with an open curve snake. The proposed framework outperformed other methods on 2 public and 1 private dataset.

    Besides the method, the authors also synthesized a dataset for cardiovascular artery disconnection detection and made it publicly available.

  • 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. The authors proposed a new way to formulate this cardiovascular artery segmentation task, combining the advantages of CNN-based segmentation and active contour-based tracking. The new framework achieved the best performance on the public dataset.
    2. The authors make the synthesis dataset public available, which helps the entire community.
    3. The experiments are sound. The authors tested the methods on both public and private datasets and showed both quantitative and qualitative results. Besides, the authors also analyzed the impact of disconnected intervals in the training of detection networks.
  • 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 segmentation, detection, and snake models used in this manuscript are all off-the-shelf models.
    2. It is not clear how the framework distinguishes small disconnected components versus noise scatters.
    3. Computational cost. The computational cost and the inference time will be large if the segmentation results have multiple small disconnected components, which means the detection network needs to infer n times for n small components.

    minor: the input in Fig. 2 is not clear.

  • 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 submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • 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
    1. It would be great if the authors could describe how the framework deals with noise in the segmentation results.
    2. Is the detection network able to compute all P1, P2 pairs at one inference time?
  • 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?

    Pros:

    1. novel design
    2. public dataset Cons:
    3. missing some details
    4. computational cost
  • 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 authors has answered my questions. Although the method cannot perform well with multiple disconnected components, but the dataset and the framework still have their value



Review #3

  • Please describe the contribution of the paper

    This paper introduces Car-Dcros, a new dataset and benchmark for enhancing the segmentation of cardiovascular arteries in 3D medical images, with a focus on addressing the challenges of segmenting smaller vascular structures and those affected by arterial plaque, which often appear as disconnected components in images. The authors propose a refinement method that uses a data-driven strategy to correct the appearance of these disconnected arterial structures. The effectiveness of the methodology is validated on actual datasets from a medical institution, and the authors provide a comprehensive evaluation, which shows that the Car-Dcros model outperformed other state-of-the-art methods, demonstrating its potential for clinical application.

  • 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. This paper constructs a synthetic dataset that mimics disconnected artery structures, which are common in images of cardiovascular diseases.
    2. This paper re-frames the disconnected cardiovascular structures issue in segmentation as a task of detecting disconnected points and employs a neural network identify points that link the disconnected components by an open curve active contour model.
  • 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 application of the methodology is limited. As the authors assume only one disconnected component with two disconnected points exists in a network input, the network may not be usable if the disconnected component is long or multiple disconnected components exist.
  • Please rate the clarity and organization of this paper

    Excellent

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • 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
    1. The authors need to improve the method for more complex disconnected components or at least discuss it.
  • 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

    Accept — should be accepted, independent of rebuttal (5)

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

    The authors focus on a very important problem in the field of vessel segmentation, i.e., the vascular disconnection in segmentation results, and correct this problem from the point of view of post-processing the segmentation results by using disconnected point detection and disconnected component completion, and construct and open-source the corresponding dataset. The work is innovative.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • [Post rebuttal] Please justify your decision

    The authors’ rebuttal have answered my query, and although the current work fails to deal with the occurrence of multiple disconnected components, both the methodological design of this paper and the dataset provided are of value.




Author Feedback

We appreciate the reviewers’ (R1, R3, and R4) comments and insightful suggestions. The value of our proposed new approach to formulating the cardiovascular artery segmentation task as a disconnected components detection task was acknowledged. Our methodology’s effectiveness has been validated on two public and one private dataset, with a comprehensive evaluation (R3). Additionally, we have made the synthesized dataset public, benefiting the entire community (R1). Below, we address the reviewers’ concerns. R1: Small disconnected components versus noise scatters: To mitigate the impact of noise scatter, we implemented a noise filter (excluding components smaller than 4 voxels) in suppl. algorithm 1, line 2. If accepted, we will revise the algorithm in the final submission to further clarify this point. Additionally, our focus on the relatively large disconnected components for reconnection was driven by clinical considerations. The clinical impact of very small disconnected components on cardiac flow dynamic calculations and/or plaque detection might be very limited due to the uncertainties and low signal-to-noise ratios, so we excluded these small components. Fig. 2 revision: We thank R1 and will improve them in the final submission. R1 and R3: R1 (computation cost) or R3 (Multiple disconnected components): To bridge the gap between training and real-world inference scenarios and achieve relatively high detection and reconnection accuracy, we iteratively detected and repaired one pair of points at a time. We acknowledge that this is a limitation of our method due to the relatively high computation cost. While we did explore the method of detecting multiple disconnected components simultaneously, this resulted in severe missed detections. In the future, we will further our investigation to find a balance between detection accuracy and efficiency. In the final submission, we will add more discussions on this topic. R4: Comparison with other methods: We thank R4 for the suggestions and will include all recommended references in the Intro. of the final submission. The paper by Zhao et al. (PACMCGIT 2023) proposed a Ball B-Spline Curve method for modeling free tubular objects, which requires knowledge of the disconnected points’ location. In contrast, our method introduced an automated framework that includes both the detection of disconnected components and the connection using an open curve snake approach. Compared to DeepVesselNet (Tetteh et al., Front. Neurosci. 2020), which focuses on direct vessel segmentation and addresses bifurcation detection, our approach emphasized the post-processing step once the vessel tree with disconnections is available. Our method therefore has the potential to apply to all vessel segmentation tasks as an additional step to address topology concerns after leveraging any SOTA methods for the CNN-based segmentation. More results and technical details: Our current disconnection detection network identifies one pair of disconnected points at a time. Therefore, there is indeed a gap between this setup and real-world scenarios. To address this, we proposed Algorithm 1 in the suppl., which includes a step to iterative run the network until all disconnected components are reconnected. The number of required iterations depends on the number of disconnected components. Evaluation improvement: Our primary goal for this paper is to propose a novel framework to improve vessel segmentation and share our real data, the data synthesis pipeline, and the resulting dataset with the community. While our current disconnection detection method has achieved good performance (OKS = 0.859), we strongly agree with the reviewers’ comments regarding the need for further experiments and analysis on this algorithm. We will run more ablation experiments on these aspects in our future studies including comparisons of different methods. We will revise the final submission to point out these directions in the Discussion section.




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’

    After the rebuttal, two reviewers recommended acceptance. Considering the value of the research problem addressed in this paper, I believe it should be accepted. However, I have a suggestion: the methods used in this paper, including data generation, keypoint detection, and metrics, are highly similar to those in the MICCAI ‘23 paper on pulmonary tree repairing. This similarity was not mentioned in the paper, which, if not intentional, represents a significant omission in the literature review.

  • 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).

    After the rebuttal, two reviewers recommended acceptance. Considering the value of the research problem addressed in this paper, I believe it should be accepted. However, I have a suggestion: the methods used in this paper, including data generation, keypoint detection, and metrics, are highly similar to those in the MICCAI ‘23 paper on pulmonary tree repairing. This similarity was not mentioned in the paper, which, if not intentional, represents a significant omission in the literature review.



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’

    Most reviewers (2 of 3) 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).

    Most reviewers (2 of 3) agree to accept this paper.



back to top