Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Tubular structures with tree topology such as blood vessels, lung airways, and more are abundant in human anatomy. Tracking these structures with correct topology is crucial for many downstream tasks that help in early detection of conditions such as vascular and pulmonary diseases. Current methods for centerline tracking suffer from predicting topologically incorrect centerlines and complex model pipelines. To mitigate these issues we propose Trexplorer, a recurrent DETR based model that tracks topologically correct centerlines of tubular tree objects in 3D volumes using a simple model pipeline. We demonstrate the model’s performance on a publicly available synthetic vessel centerline dataset and show that our model outperforms the state-of-the-art on centerline topology and graph-related metrics, and performs well on detection metrics. The code is available at https://github.com/RomStriker/Trexplorer.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2112_paper.pdf

SharedIt Link: https://rdcu.be/dV59n

SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72120-5_69

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/2112_supp.pdf

Link to the Code Repository

https://github.com/RomStriker/Trexplorer

Link to the Dataset(s)

https://github.com/giesekow/deepvesselnet/wiki/Datasets/

BibTex

@InProceedings{Nae_Trexplorer_MICCAI2024,
        author = { Naeem, Roman and Hagerman, David and Svensson, Lennart and Kahl, Fredrik},
        title = { { Trexplorer: Recurrent DETR for Topologically Correct Tree Centerline Tracking } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15011},
        month = {October},
        page = {744 -- 754}
}

Reviews

Review #1

  • Please describe the contribution of the paper
    1. For the task of 3D tree-like centerline tracking, it can prioritize the integrity and correctness of the tree-like topology.
    2. The iterative pipeline of the constructed model is simple and efficient, requiring no additional postprocessing/auxiliary detectors, and allows for end-to-end training.
    3. The proposed algorithm framework demonstrated excellent performance on the synthetic vessel dataset, proving its effectiveness.
  • 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.

    Addressing issues in previous work such as incorrect topology, numerous object queries, and a complex algorithm pipeline, the iterative pipeline of multi-object tracking in TrackFormer is innovatively modified and applied to the centerline tracking task. This approach enables the iterative generation of tree-like centerline topology, and experiments have proven the feasibility and effectiveness of this architecture and 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.
    1. The Introduction of the paper is too detailed.
    2. Fixed bifurcation degree 2 may be appropriate for the synthetic vessel dataset, but improper to other 3D arteries or airways.
    3. For bifurcations, eight-copy object queries will be classified as background in the next step. There are more details that has not been explained such as how to avoid overlapping with exsiting vessel centerlines. In 3D space, centerline connectivity is usually described using 26-connected neighbors.
    4. Experiments were conducted on only one synthetic vessel dataset. It needs to be validated on real vascular dataset. And explanations on ablation experiment were too simple.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • 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?
  • 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 author should simplify the introduction and expand the experimental analysis.
    2. The author could try other public available dataset, such as arteries, airways and retinal vessels through some challenges to validate the method.
    3. The author could appropriately discuss the limitations of the proposed method and future research directions.
  • 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 proposed model can effectively extract the centerlines of tree-like topological blood vessels and ensure the integrity and correctness of the topology, and it constructs a simple and effective iterative pipeline, which has a certain degree of innovation. But the paper has too lengthy Introduction and is lack of experimental workload and analysis.

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

  • [Post rebuttal] Please justify your decision

    The proposed model can effectively extract the centerlines of tree-like topological blood vessels and ensure the integrity and correctness of the topology. But the paper has too lengthy Introduction and is lack of experimental workload and analysis. Thank you for the rebuttal, but it didn’t address my concerns. Hence, I maintain my decision.



Review #2

  • Please describe the contribution of the paper

    This paper proposes a method to track vessels centerlines, and to keep the vessels topology, using a recurrent DETR, and consider them as graphs.

  • 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 problem is difficult and very interesting
    • the method is simple and does not require any pre-processing
  • 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 validation of the method is incomplete, there is no statistical analysis and the state of the art methods considered is incomplete (the method of ref 19, the providers of the database, should be compared)
    • the authors only test their method on synthetic dataset so we don’t know how their method works with real data
  • 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 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
    • there are some typos: in the bibliography (ref 10), in the text p7 “Trexpolorer” instead of “Trexplorer”, in Table 1 Trexplorer is named “Vexplorer” etc.
    • the authors should put in bold every best result in table 1, not only their results
    • a statistical analysis is missing here
    • the authors should explain why they only use a synthetic dataset and not real images.
    • the authors should compare their method with the one of Tetteh (ref 19, the provider of the dataset used by the authors)(or at least explain why it is not the case)
    • considering real databases (with segmentations associated in the ground truth) will add clDice as metric to evaluate the author’s work
  • 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?

    The article is interesting, but some points are missing, making this article a little bit too light for a MICCAI publication without modifications. However the method is interesting.

  • 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 answered to our remarks and questions, if they do the required changes this article can be published



Review #3

  • Please describe the contribution of the paper
    1. The paper introduces a novel method for centerline tracking, ensuring the generation of a tree topology without the need for any preprocessing steps.
    2. The proposed model is evaluated on a publicly available synthetic vessel dataset, demonstrating its state-of-the-art performance.
  • 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. Trexplorer guarantees the generation of a tree topology without the need for any preprocessing steps. This ensures the accuracy and reliability of the centerline graph in medical volumes, distinguishing it from many existing methods that may require post-processing to correct topology errors.
    2. Trexplorer offers a simple pipeline that tracks the centerline graph while attributing essential information such as radius and class for each point. This streamlined approach combines the simplicity of transformer-based models with the effectiveness of dynamic programming, enhancing efficiency and interpretability in medical imaging analysis tasks.
  • 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 presentation of synthetic vascular data is insufficient.
    2. There are insufficient comparative experiments, and some experimental indexes are not optimal.
    3. The details of the method are not very clear, the specific input/output tensor shapes and network layers need to be elaborated.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • 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?

    The paper claimed the code will be made available upon acceptance.

  • 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 paper needs to add more synthetic vessel maps and comparative experiments to show the results.
    2. There are only 2 comparative experiments used in the paper, and there is a need to add further updated comparative methods.
    3. The methodological details of the paper are too scanty, and the paper does not provide any further explanation of three figures in the appendix. For example, the input parameters, network details, etc.
  • 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?

    Methodological innovativeness, elaboration of details and experimental credibility

  • 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 have responded point by point to the questions raised and promise to present the detailed network structure in the supplementary material.



Author Feedback

We thank the reviewers for their valuable feedback and constructive criticism, and acknowledging the simple innovative design of our model to solve this difficult problem while addressing the limitations of previous works. We will fix the minor issues, and the major issues are addressed below.

  1. Testing on real datasets: (R1, R3) We agree that it is important to assess the performance of our model in real applications, but as we state in our manuscript in Sec. 3.1 “We evaluate Trexplorer on the synthetic vessel dataset [19], which is the only publicly available vessel centerline dataset to the best of our knowledge.”. We have semi-automatically generated centerline graphs for the Hepatic Venous Tree in the IRCADB-01 dataset using the provided vessel segmentation masks and trained our model on these centerline graphs. We observed similar performance gains as on the synthetic data, even though the datasets vary greatly (fewer samples, higher degrees of bifurcation, lower resolution, higher tortuosity, diverse background). However, it is difficult to verify the correctness of the generated centerlines, as it requires radiologist expertise. Due to these doubts, we decided not to include those results. In addition, the license of this dataset and others like ATM’22 and Parse2022 Challenge does not allow us to share derivatives preventing us from sharing these centerlines. We hope this clarifies our approach.
  2. Further comparative experiments: (R1, R4) We aimed to include recent state-of-the-art models that take a 3D volume as input and output a centerline graph with published results on a public dataset. Only Vesselformer met this criterion. The method of ref. 19 predicts a segmentation mask of the centerline and uses Dice score (a segmentation metric) for evaluation, making fair comparison difficult. The clDice metric relies on segmentation masks, which our model neither requires for training nor predicts.
  3. Statistical analysis: (R1) Our test set consists of 166 vessel trees spread across 10 volumes, totalling 654387 nodes and 654221 edges, providing a statistically sound estimate of the model’s performance. We did not report the means and variances of our results to stay consistent with the Vesselformer results and due to the large compute required for running multiple trainings. However, we can add them if the area chairs find it appropriate.
  4. Long introduction: (R3) We intended for Sec. 1 to cover both the introduction and relevant works, essential for establishing the need for our model. However, we will trim it and add more detailed explanations in the Experiments section. We will also expand the conclusion with the limitations of our model such as premature tracking termination and potential duplicate tracking, along with future directions such as utilizing DETR-variants with stronger priors and advanced tree matching algorithms.
  5. Fixed bifurcation degree and overlapping centerlines: (R3) It is correctly pointed out that setting the number bifurcation object queries (nb) to 2 only works for the synthetic dataset where the max bifurcation degree is 2. We got similar results for higher nb values. Generally, nb should be set to at least the maximum expected bifurcation degree in the data. The bifurcation object queries attend to each other through self-attention and are penalized for tracking the same branch in the Hungarian loss, which discourages overlapping centerlines. We will mention these points in Sec. 2.2.
  6. More methodological details and synthetic vessel maps: (R4) We will add the input/output shapes in Fig. 1. We did not include details about the encoder and decoder, since they are identical to DETR and SwinUNETR unless mentioned otherwise, however, we will add them in the supplementary material figures. Adding more vessel map examples is a good idea. We will modify Fig. 2 to fit a 3x4 grid of images, using the extra column for an extra example and the extra row for corresponding Vesselformer outputs.



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’

    The reviewers evaluated the novelty of the method, which performs centerline tracking considering the topological property of tubular structures (the main application target is the blood vessels). The main weakness of this paper is that only the synthetic dataset was used in the evaluation. While this paper has some weaknesses, the methodology is novel and will contribute to the progress of our research field (anatomical structure understanding, segmentation, etc.). Therefore, I recommend accepting 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).

    The reviewers evaluated the novelty of the method, which performs centerline tracking considering the topological property of tubular structures (the main application target is the blood vessels). The main weakness of this paper is that only the synthetic dataset was used in the evaluation. While this paper has some weaknesses, the methodology is novel and will contribute to the progress of our research field (anatomical structure understanding, segmentation, etc.). Therefore, I recommend accepting this paper.



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’

    N/A

  • 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).

    N/A



back to top