List of Papers Browse by Subject Areas Author List
Abstract
Ventricular tachycardia screening is crucial for early intervention and prevention of life-threatening cardiac events. Myocardial scar topology on late gadolinium enhancement (LGE) MRI offers detailed structural insights that may be closely associated with the mechanisms underlying ventricular tachycardia. However, accurate characterization presents challenges due to the substantial shape variability of myocardium, indistinct boundaries, small scar volumes, and potential issues with image quality. In this study, we present PolarNet, a novel framework for automatic scar segmentation and topological pattern characterization in polar coordinates. The framework incorporates a boundary-aware segmentation branch that explicitly models boundaries essential for scar characterization (endocardium, scar-start, scar-end, and epicardium), ensuring geometric consistency and anatomical coherence. Our method outperforms nnU-Net in both scar segmentation and topological pattern characterization. Code will be available at https://github.com/Sheng-xc/VTS_PolarNet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2683_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Sheng-xc/VTS_PolarNet
Link to the Dataset(s)
N/A
BibTex
@InProceedings{SheXic_Automated_MICCAI2025,
author = { Sheng, Xicheng and Zhang, Yang and Li, Lei and Chen, Bailiang and Odille, Freddy and Zhuang, Xiahai},
title = { { Automated Characterization of Myocardial Scar Topological Patterns for Ventricular Tachycardia Screening } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15962},
month = {September},
page = {67 -- 77}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents a new approach for myocardial scar segmentation and characterization, motivated for VT screening. The key contribution of the paper boils down to a polar coordinate representation of the LV, combined with a topologically-constrained boundary detection. Experiments were conducted on 181 post-infarct patients across five centers, evaluating against manually segmented scar contours. Both the characterization of the scars (in 5 categories) and scar segmentation were evaluated, where the presented method demonstrated improvements over mmUNet. The effect of various methodological components was ablated.
- 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.
Myocardial scar characterization in the setting of VT screening is a challenging problem due to the heterogeneity of the myocardial scar.
The methodology presented is well motivated and clearly described. The incorporation of the topological constraints in the form of a recursive ordering is especially interesting.
The experiments, especially with detailed scar contouring and characterization for all patients, involved significant effort.
- 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.
The improvements obtained were in general quite marginal and 1-2 magnitudes smaller than the std reported. This raises questioning on the significance of the improvements reported. The visual results reported (e.g., in Fig 4) also did not provide convincing results that the presented method was more consistent with the ground truth.
While motivated for VT screening, the evaluation did not consider any aspects related to the scar characterization to the underlying VT risk. This further undermines the clinical relevance of the marginal improvements reported.
Risk of VT is heavily determined by the “heterogeneity” of the myocardial scar in terms of the distribution of “gray zone” within dense scars. These characteristics were not considered in the scar segmentation or characterization, which only focused on a binary contouring of the scar.
2.3 is very brief and it is not clear how the segmented scar is linked to the classification of the four categories.
Shouldn’t the boundary detection a function varying with r for each angle \theta? Why is it being aggregated across the radius to produce one prediction per \theta per label?
While the authors criticized existing methods for being image intensity based, it seems that the presented method is essentially also driven by intensity, except in a different coordinate space and with some additional topolocial constraints.
- 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.
- 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?
While this paper address an important and difficult problem with a clearly described methodology that has some interesting components, it is not clear if the marginal improvements reported were significant, nor did the evaluation consider clinically relevant metrics (e.g. VT risks, or scar gray zone / core heterogeneity) that are important for the motivating clinical problem.
- Reviewer confidence
Very confident (4)
- [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.
Thanks for the authors’ response to my earlier comments. With the additional results on statistical significance and the clarifications of how the boundary detection is defined as well as the clinical implications of the scar subtypes, I’m happy to change my earlier rating in support of the acceptance of this paper.
Review #2
- Please describe the contribution of the paper
This manuscript proposes a novel architecture to segment the left ventricle myocardium and scar tissue from LGE-MRI. The proposed framework includes a preprocessing step to convert cartesian 2d slices into a polar reference. These topological properties are then exploited to improve the prediction of the endocardial, epicardial, and scar boundaries. This paper shows scar segmentation improvement compared to nn-Unet.
- 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.
- Consistent good scientific style writing with a clear paper organization. The paper follows a logical storyline.
- Proper explanation of the methodological approach in terms of polar map calculation and loss definition.
- Strong choice of baseline network for comparison.
- Good design of ablation study.
- Appropriate figure design and presentation. The figures 3 and 4 are very well drawn and are very informative to the reader to understand the segmentation and label assignment approach.
- Qualitative assessment of results is properly depicted in figure 5, with comparison to the baseline nn-Unet and other evaluated network configurations.
- 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.
- Lack of technical details on the prediction of the LV centroid: How is the centroid predicted? What is the prediction accuracy?
- Lack of precise technical details for the network architecture: the name of the input polar map (G), output features (F), pixel-wise probabilities (Pix) and boundary probabilities (Q) could have been included Fig. 2. Additional details on the architecture of the 2 branches (e.g. number of convolutional layers, kernel size…) could have been provided.
- No discussion section: I believe that the discussion of the results should have been done in a dedicated discussion section. This issue is just minor in my opinion due to the length restrictions of the paper.
- 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.
- 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
- There are multiple datasets from previous MICCAI challenges that might be available for training and evaluation of the method that the authors could use to increase their dataset size.
- In many works that require LGE-MRI scar segmentation, threshold-based methods are still utilized. I believe it would have been interesting to see how the performance of these compares to the method proposed in this manuscript. However, I think that the choice of the nn-Unet as a baseline is appropriate.
- The network enforces that boundaries are ordered in the transmural direction, including the scar boundary. Although I believe this assumption is correct for the endocardial and epicardial boundaries, I do not agree for the scar. Is my understanding that scar discontinuities might appear in the transmural direction and that these might be relevant, since they might be substrates for the re-entrant activity that sustains the VT. How does the network deal with the appearance scar discontinuities in the transmural direction? Will not this lead to the network underestimating the appearance of anatomical channels which are “parallel” to the endocardial and epicardial contours? Did you observe any possible examples of this in your dataset?
- 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 paper proposes a novel method to segment the myocardium and the scar from LGE-MRI images, showing improvements with respect to nn-Unet. The paper is clearly written, easily understandable, and explains most technical and methodological details that needed to be included. The missing details can be easily included during rebuttal.
- 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.
This paper presents a novel and interesting methodology, which has potential to improve the segmentation of scar for post-MI VT, a quite challenging task. Although I only raised minor comments during review, the authors addressed these questions appropriately. For these reasons, I believe that the paper should be accepted.
Review #3
- Please describe the contribution of the paper
This paper proposes a topology-aware architecture for boundary-sensitive myocardial scar mapping, leveraging a novel polar transformation to directly target scar regions. Additionally, a dedicated boundary branch and tailored loss function are designed to preserve the correct boundary topology. On their dataset, the proposed method achieves state-of-the-art performance, outperforming existing 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.
-
The core concept of the polar transformation is a novel attempt.
-
The boundary branch with its recursive operation ensures the accurate alignment of fine structures.
-
- 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.
Although the use of polar coordinates is conceptually sound and the overall network design is novel, the performance shows limited improvement over nnU-Net and even falls short on some metrics, which diminishes the overall impact of the work. Moreover, the paper requires thorough proofreading to improve clarity and consistency. For example, the GDice metric is not properly introduced or described before and after its application.
- 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.
- 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.
(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?
Considering both its strengths and limitations, I believe this work meets the standards of MICCAI and recommend it for acceptance.
- Reviewer confidence
Very confident (4)
- [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 rebuttal addressed all of my concerns.
Author Feedback
Dear Area Chairs, We thank reviewers (R1–R3) for their constructive and thoughtful comments. We have summarized several main comments with corresponding responses. We also highlight that the reviewers acknowledged the study’s novelty (R1–R3), clear organization (R1–R3), and sound experimental design (R1, R2).
1.Experimental results concerns: -R1, R3 raised concerns that quantitative improvements over nnU-Net were limited; R1 specifically questioned the statistical significance. -R1 found the visual results insufficiently compelling to demonstrate superiority. 1.1(1)One-sided Wilcoxon signed-rank tests show that PolarNet significantly outperforms nnU-Net in key metrics: gDice (p=0.01), SEN (p=0.03) in classification; Dice (p=0.02), ASSD (p=0.03), SEN (p=0.002) in segmentation.Slightly worse HD/SPEs are not statistically significant and are mainly due to outliers. Std’s reflect varying difficulty across the test set. (2)We emphasize that our core contribution lies in boundary-sensitive scar segmentation that enables subtype classification, offering a topological view of arrhythmogenic substrate characterization, rather than in tuning for marginal gains. 1.2As shown in Fig.4, subtype classification is challenging, as it requires capturing scar–myo structure radially at each θ. Nevertheless, PolarNet yields more consistent predictions of scar presence (i.e., non-green regions) and better detection of intramural scars—key substrates for re-entrant VT. Fig.5 further shows improved scar localization (rows 1–2) and sharper boundaries (row 3).
2.Evaluation concerns: -R1 raised concerns about the lack of clinical metrics related to VT risk, such as scar heterogeneity (gray zone). -R2 suggested using more datasets. -R3 requested clarification of GDice. 2.1We acknowledge the importance of metrics such as gray zone/core heterogeneity. However, such measures require standardized intensity thresholding and outcome-linked cohorts, which are beyond the scope of this methodological study. Instead, we focus on topological heterogeneity via subtype classification, offering a complementary structural perspective. Future work may evaluate the predictive value of subtype volumes, analogous to gray zone metrics. This work contributes methodologically, as noted by R1–R3, aiming to bridge clinical insight and large-scale validation. 2.2As our study targets chronic-phase MI with detailed scar topology, suitable public data is limited. More data could be included from trials in future study. 2.3GDice=2∑k(Ak·Bk)/∑k(Ak+Bk) is a weighted Dice score accounting for all categories.
3.Methodology concerns: -R1 expected more clarity on subtype classification, boundary detection and its contribution beyond intensity features. -R2 asked for more architectural and centroid prediction details. -R2 questioned the assumption of transmural continuity for scar boundaries. 3.1Subtype classification assigns each θ on a slice to a category based on whether the radial line intersects scar and whether the scar contacts the endo/epi. Boundary_k detection predicts a probability map Q[:,:,k] where Q[r,θ,k] is the likelihood of boundary k at θ occurring at r. Following 6.He et al. (2021)(see app.), the expected value over r Sˆk[θ] yields a robust boundary estimate at θ.The boundary branch explicitly focuses on boundary positions, offering guidance beyond intensity-based pixel classification. 3.2The backbone is a 7-stage nnU-Net (ResEnc) with encoder block depths [1, 3, 4, 6, 6, 6, 6], decoder blocks 1/stage, feature sizes [32, 64, 128, 256, 512, 512, 512], kernel size (3, 3). The LV centroid is the center of mass of the LV predicted by an nnU-Net trained on 145 public LGE-CMRs from MICCAI challenges (MSCMR, MyoPS, MyoPS++), using LV, RV, and LV myo labels. 3.3Seg model does not enforce transmural continuity for scars—auxiliary envelope simply guides attention to boundaries. We acknowledge that rare cases of transmural discontinuities may lead to underestimation of risk.
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’
N/A