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

Abstract

Accurate Couinaud segmentation of liver CT/MR is essential in helping surgeons perceive the positional relationship between liver anatomy and intrahepatic lesions to make surgical planning. Unfortunately, current conventional and deep-learning based methods remain challenges in accurate Couinaud segmentation since the segmentation boundaries of different categories depending on hepatic vascular information are hard to predict. This work proposes a new deeply learned framework called anatomy-aware frequency-attention transformer networks (AFATN) for Couinaud segmentation of liver anatomy which contains the hybrid anatomy-aware preprocessing and frequency-attention transformer networks (FATN). Specifically, our framework first uses hybrid anatomy-aware preprocessing to integrate the hybrid cues of liver contour and hepatic venous centerline, then effectively utilizing hybrid cues for accurate Couinaud segmentation through the frequency-attention transformer networks with omission re-detected loss function. Our segmentation model FATN uses transformers to extract local structure and global semantic features and further focus on the hybrid cues with frequency-attention mechanisms. The proposed method was evaluated on clinical CT data and compared with currently available deep learning approaches, with the experimental results demonstrating that our method outperforms other approaches especially in accurately segmenting the Couinaud boundaries.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3005_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{FanWen_AnatomyAware_MICCAI2025,
        author = { Fan, Wenkang and Fang, Hao and Li, Rui and Lin, Yanduan and An, Chao and Luo, Xiongbiao},
        title = { { Anatomy-Aware Frequency-Attention Transformer Networks for Liver Couinaud CT/MR Segmentation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15960},
        month = {September},
        page = {56 -- 66}
}

Reviews

Review #1

  • Please describe the contribution of the paper
    1. The paper introduces FATN, which employs a transformer-based encoder that integrates local texture and global semantic features through a frequency-attention mechanism.
    2. The authors propose an innovative loss function that addresses the challenge of omission in segmentation, allowing the network to effectively re-detect and penalize areas where segments are inaccurately predicted.
    3. The paper provides comprehensive experimental results demonstrating that the AFATN outperforms several existing state-of-the-art segmentation methods, including CNN-based models and other transformer-integrated approaches.
  • Please list the major strengths of the paper: you should highlight 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 application of frequency-attention mechanisms within transformer networks is a novel approach in medical image segmentation. This technique allows the model to enhance structural boundary information.

    2. The paper includes comprehensive experimental settings and results comparing the proposed method against several state-of-the-art models.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
    1. The paper does not adequately address the computational efficiency of the proposed method.
    2. The paper does not provide the full experiment results, such as the results of Hepatic Vessels.
  • Please rate the clarity and organization of this paper

    Poor

  • 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.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

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

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
    1. The paper does not adequately address the computational efficiency of the proposed method.
    2. The paper does not provide the full experiment results, such as the results of Hepatic Vessels.
    3. The paper concludes simply, without a comprehensive description of the drawn conclusions and potential future research directions.
  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.
    1. Although the authors provide a parameter count (120 million) and an inference time (10 seconds on an RTX 4070), they omit direct comparisons with competing methods (e.g., U-Net with approximately 30 million parameters and Swin-Unet with around 60 million parameters). In the absence of such context, it remains unclear whether FATN’s complexity is justified by its performance.
    2. Additionally, while the authors indicate that Hepatic Vessels results are presented in Table 1, they do not specify the evaluation metrics (such as the Dice score or Intersection over Union) nor do they provide visual examples.



Review #2

  • Please describe the contribution of the paper

    Coiunad segmentation in a liver image is segmenting the fully functional independent segments into eight. Each segment exhibits vascular inflow, outflow and biliary drainage. This paper proposes a DL-based method to segment these 8 segments from the liver CT image using Frequency-Attention Transformer Networks. The results are being compared with the published results.

  • Please list the major strengths of the paper: you should highlight 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. Paper is well written
    2. Methodology, results and discussion are convincing
    3. The work is clinically more relevant as it acts as an assistive technology for the Coiunad segmentation to segment 8 liver segments which may be a time-consuming process during manual assessment
  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Even though technically the paper is well written, the testing and validation looks insufficient for the model acceptance. Following are the major critical comments.

    1. If CT/MRI is used, it creates confusion whether it is CT+MRI registered.

    2. Can we use the same model for the other vascular structures like, collateral formation, coronary arteries and the retinal structures if I can train the model using the annotations of these cases as ground truth?

    3. It is confusing that what image type is this? is it CT or MRI. In the diagram it is shown as CT/MRI. Use either CT or MRI based on the image used

    4. Is the dataset is a Contrast enhanced multiphase image?

    5. Page 3: all points on the extracted centerline selection is based on the skeletonization and perpendicular vectors calculation till the vein boundary using Gram Shmidt orthogonalization?

    6. How is equation is still valid when we use image from multiple center and multiple image acquisition parameters

    7. The input image is 3D volume or a 2D slice. The image looks like a direct volume rendered image where liver is already segmented

    8. As these are multi-center data, please mention the image acquisition parameters in each database. This helps to understand the image content in terms of voxel intensities. Otherwise it is difficult to judge or make a guess that whether the model succeeds with multi center data or it fails to work with such data

    9. Apart from mentioning the databases used, there is no details of key image acquisition parameters. I want to know who the model behaves when it is exposed to multi center dataset (variation in Hounsfield units) within the same modality

    10. If you show the output on even MRI images also, then it will be convincing that the model works for multi-modality or else difficult to accept the model output.

    11. What software is used for the ground truths creation? how were the GT validated or corrected if any for under segmented areas? Did you interpolate the slices during volume creation before defining ground truth?

    12. Can you show one vein segmented and its corresponding centerline to understand and check whether the centerline extraction.

    13. Please mention the direction of results visualization in Fig 3 clearly as per the radiology image interpretation convention w.r.t patient coordinate system. Vertical view is not good to use

    14. Apart from the objective validation, did you get the feedback from the doctors on the result (subjective), any variations in the opinion (inter-observer variations)?

    15. Too less information in conclusion, should be rewritten

    16. Table 1 results from other papers are based on their dataset and results which is compared with your method. Did you reproduced the results from their models on their data and then listed the results in table 1 or it is just mentioned by you based on the paper content?

    17. The authors mention everywhere that they use CT and MRI modalities, but only the CT images are shown in the results

    18. Literature review is too shallow and only few papers are mentioned. Each paper is not critically evaluated and concluded.

    19. How is the work different from the FATN described and discussed in the following papers

    20. https://ieeexplore.ieee.org/document/10096538/
    21. https://www.sciencedirect.com/science/article/abs/pii/S0957417424025533
    22. https://www.sciencedirect.com/science/article/abs/pii/S0031320323000468
    23. https://dl.acm.org/doi/10.1145/3696409.3700232
  • 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 submission does not provide sufficient information for reproducibility.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

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

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

    The work looks interesting, but there are many questions from domain point of view which need to be addressed. Unless the convincing answers, the paper content which looks like a lab experiment on few samples which limits its acceptance.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    The clarifications are provided for the review comments. Carefully went through all rebuttals. Now the paper can be accepted



Review #3

  • Please describe the contribution of the paper

    The paper presents a new deep-learning framework for Couinaud segmentation that includes anatomy-aware frequency-attention transformer networks. The framework complements local with more glogal feature extraction to identify blood vessels within the liver and uses this information to guide the segments bondaries.

    The network and parameter descriptions are clear and complete. They could only be more detailed if the page-limit allowed, which is not the case.

    The experimental results show that the approach attains better accuracy for the dataset tested when compared with several other deep learning segmentation approaches in the literature. Ablation analysis details the gains with each major component of the architecture.

  • Please list the major strengths of the paper: you should highlight 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 framework proposed seems to be novel and targeted to the problem of blood vessels and related segments, with good results in comparison to other methods.

    The motivation and the network structure are sufficiently detailed.

    The authors clearly state the method limitations.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    In the experimental part, more detail is needed to fully understand the impact of the proposed architecture. First, to reproduce the experiment more detail is needed regarding data and parameters. Then, the analysis is a bit shallow, containing only the essential metrics (probably due to lack of space in 8 pages).

    Only one liver is used to illustrate the results. I’d like to see other cases including one where the technique fails.

    Table and figure labels are vague and need more detail to make sense.

  • 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.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    The references cited are appropriate and recent. However, I fell a few words should be spent to contextualize CNN based liver segmentation within the area where several annotation-based and semi-automatic segmentation are still considered more reliable.

    Consider the possibility to cite these papers on Couinaud segmentation:

    Han, X., Wu, X., Wang, S. et al. Automated segmentation of liver segment on portal venous phase MR images using a 3D convolutional neural network. Insights Imaging 13, 26 (2022). https://doi.org/10.1186/s13244-022-01163-1

    H. G. Debarba, D. J. Zanchet, D. Fracaro, A. Maciel and A. N. Kalil, “Efficient liver surgery planning in 3D based on functional segment classification and volumetric information,” 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, Buenos Aires, Argentina, 2010, pp. 4797-4800, https://doi.org/10.1109/IEMBS.2010.5628026.

    Chen Y, Yue X, Zhong C, Wang G. Functional Region Annotation of Liver CT Image Based on Vascular Tree. Biomed Res Int. 2016;2016:5428737. https://doi.org/10.1155/2016/5428737

    D. C. Le, J. Chansangrat, N. Keeratibharat and P. Horkaew, “Functional Segmentation for Preoperative Liver Resection Based on Hepatic Vascular Networks,” in IEEE Access, vol. 9, pp. 15485-15498, 2021, https://doi.org/10.1109/ACCESS.2021.3053384.

  • 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.

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

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

    The method has novel aspects and delivers higher performance than provious methods. The details in the descriptions are compatible with MICCAI page limits and the missing details in the results do not undermine the understanding of the technique capability. So, I recommend presentation in MICCAI.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Author Feedback

We thank the reviewers for their constructive and valuable comments to strengthen our submission. We are modifying our paper by all the comments.

Reviewer#1 Q1: More detail is needed on experimental settings and analysis. A1: We apologize for this confusion. We have described the data and parameters in the experimental settings in detail, and we will make the code publicly accessible soon. We will add more analysis in our revision, such as average symmetric surface distance (ASSD), Jaccard coefficient (JC), and Hausdorff distance (HD95).

Q2: More cases should be provided. Table and figure labels are vague. A2: We are showing more results in the Supplementary Material and rephrasing the table and figure labels in our revision.

Reviewer#2 Q1: The paper does not adequately address the computational efficiency. A1: We apologize for this confusion. The parameters of our model are about 120M. The inference time of one case is 10s using an RTX 4070 GPU. We will further incorporate lightweight models to optimize computational efficiency.

Q2: Do not provide the full experiment results, such as Hepatic Vessels. A2: Since our method focuses on accurate liver Couinaud segmentation, we mainly show the results of the liver segments due to space limitations. We show the segmentation results of hepatic vessels in Table 1, and we will add more qualitative segmentation in our revision.

Reviewer#3 Q1: About the CT/MRI question. (1, 3, 4, 10, 17) A1: We apologize for this confusion. Our model is designed to perform liver Couinaud segmentation on both CT and MRI modalities, as the anatomical structure and spatial distribution of the eight liver segments are generally consistent across the two modalities. Nevertheless, the experiments presented in this study are primarily conducted on CT data. We will further discuss this in our revision.

Q2: Can we use the same model for the other vascular structures? (2) A2: Yes, we tested our model on renal vessel extraction from CTU data and achieved good results because the modules are designed for effectively extracting and enhancing vascular features.

Q3: How to extract the centerline of vessels? (5, 12) A3: The centerline extraction process in VTK is typically based on geodesic path computation using a fast marching method. A distance field is computed from defined inlet and outlet points on the vessel surface, and the centerline is extracted as the shortest path within this field. We will show one vein segmented and its corresponding centerline in our revision.

Q4: About the data preprocessing. (6, 7, 8, 9) A4: We apologize for this confusion. For each CT dataset from multiple centers, we first extract the liver boundary, then set the vessel regions to the maximum value of the liver region of the original CT, and finally perform normalization for training. By extracting the liver region for normalization, we can effectively avoid the interference from high-intensity non-liver structures, such as bones or imaging artifacts. This data preprocessing method is effective across multiple datasets.

Q5: How to obtain the ground truth? (11) A5: Couinaud labels were annotated manually by three surgeons through the 3-D slicer software.

Q6: Did you get the feedback from the doctors on the result? (14) A6: Yes, we asked five surgeons from different hospitals to evaluate the segmentation results produced by the model. Typically, they used these segmented results to guide liver tumor ablation, and generally believed these segmentation results were accurate and helpful to guide the ablation in the operating room.

Q7: About the results of Table 1. (16) A7: We reproduce the segmentation results of these models on the experimental data.

Q8: About the literature. (18, 19) A8: This paper mainly introduces work related to medical image segmentation. We are going to introduce latest CNN- or Transformer-based liver Couinaud segmentation models related to this work.



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • 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



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’

    Both R1 and R3 are relatively satisfied with the rebuttal and provide positive feedback. Overall, the AC recommends acceptance, but the authors need to address R2’s concerns in the final version.



Meta-review #3

  • 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



back to top