List of Papers Browse by Subject Areas Author List
Abstract
Mitral regurgitation is one of the most prevalent cardiac disorders. Four-dimensional (4D) ultrasound has emerged as the primary imaging modality for assessing dynamic valvular morphology. However, 4D mitral valve (MV) analysis remains challenging due to limited phase annotations, severe motion artifacts, and poor imaging quality. Yet, the absence of inter-phase dependency in existing methods hinders 4D MV analysis. To bridge this gap, we propose a Motion-Topology guided consistency network (MTCNet) for accurate 4D MV ultrasound segmentation in semi-supervised learning (SSL). MTCNet requires only sparse end-diastolic and end-systolic annotations. First, we design a cross-phase motion-guided consistency learning strategy, utilizing a bi-directional attention memory bank to propagate spatio-temporal features. This enables MTCNet to achieve excellent performance both per- and inter-phase. Second, we devise a novel topology-guided correlation regularization that explores physical prior knowledge to maintain anatomically plausible. Therefore, MTCNet can effectively leverage structural correspondence between labeled and unlabeled phases. Extensive evaluations on the first largest 4D MV dataset, with 1408 phases from 160 patients, show that MTCNet performs superior cross-phase consistency compared to other advanced methods (Dice: 87.30%, HD: 1.75mm). Both the code and the dataset are available at https://github.com/crs524/MTCNet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3656_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{CheRus_MTCNet_MICCAI2025,
author = { Chen, Rusi AND Yang, Yuanting AND Yao, Jiezhi AND Song, Hongning AND Zhang, Ji AND Zhou, Yongsong AND Huang, Yuhao AND Yang, Ronghao AND Jia, Dan AND Zhang, Yuhan AND Tao, Xing AND Dou, Haoran AND Zhou, Qing AND Yang, Xin AND Ni, Dong},
title = { { MTCNet: Motion and Topology Consistency Guided Learning for Mitral Valve Segmentation in 4D Ultrasound } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15962},
month = {September},
page = {407 -- 417}
}
Reviews
Review #1
- Please describe the contribution of the paper
Mitral regurgitation (MR) is a prevalent cardiac disorder, and precise segmentation of the MV is critical for diagnosis and treatment planning. This paper introduces MTCNet, a novel semi-supervised learning (SSL) framework designed for accurate 4D mitral valve (MV) segmentation in ultrasound images. MTCNet addresses these issues by leveraging motion and topology consistency to improve segmentation accuracy and temporal coherence. MTCNet was evaluated on a dataset of 1408 phases from 160 patients. It achieved the performance with a Dice score of 87.30% and a Hausdorff Distance (HD) of 1.75 mm, outperforming other SSL methods.
- 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.
Mitral regurgitation(MR) is a serious cardiac disease. This paper focuses on the segmentation of the mitral valve in echocardiograms. The authors plan to release an expert-annotated dataset, which will contribute to advancing the algorithm development for diagnosis and treatment of MR.
- 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 paper lacks strong innovation. The proposed method of using partially annotated data from ED and ES frames to achieve the segmentation across the entire cardiac cycle. This idea has already been reported in existing literature such as [14] for ultrasound images. The similar frameworks combining prior domain knowledge for cardiac cavities segmentation in 4D CMR images have reported [r1, r2]. It is suggested to clarify the differences between the proposed approach and the existing methods by providing comparative performance evaluations.
- [r1] Qiao Z et al. Explainable cardiac pathology classification on cine MRI with motion characterization by semi-supervised learning of apparent flow. Medical image analysis 56 (2019): 80-95.
- [r2] Qi XM et al. STANet: Spatio-Temporal Adaptive Network and Clinical Prior Embedding Learning for 3D+T CMR Segmentation. IEEE-JBHI, 2024, 28(2):881-892.
-
The paper mentions the challenges in mitral valve segmentation such as severe motion artifacts, and complex deformations. Are there any unique contribution specific to tackle these challenges? If so, the authors should describe the innovative solutions in detail and give corresponding results as well as the discussion.
-
In Figure 4, the GT image demonstrates that the morphology of valves appears similar in phase MD and MD-1, yet Table 1 shows significantly lower performance for MD compared to MD-1. The authors should discuss the specific reasons for this discrepancy.
- 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.
- 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?
The novelty of the proposed method is limited.
- Reviewer confidence
Very confident (4)
- [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.
The novelty of the proposed method is limited. While the authors plan to release a 4D TEE mitral valve dataset with segmentation annotations, this could still serve as an important factor for the paper’s acceptance.
Review #2
- Please describe the contribution of the paper
The author suggest neural network based reconstruction approach for the mitral valve in 4D TEE image data. They claim that their approach requires only sparse temporal annotations, i.e. at end-systole and end-diastole. In addition, they claim that their results are consistent between unseen phases. This is enabled by their network design in combination with a newly introduced loss function. Their training and test set consists of 160 patients with 1408 phases in total. The results seem to indicate that there is a small improvement compared to the state-of-the-art.
- 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 authors suggest new terms in the loss function to ascertain consistent surface area and segmentation volume throughout the segmented time series. The authors provide a convincing evaluation and illustrate their method and results with helpful figures of the architecture as well as segmentation examples
- 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 state of the art is incomplete (DOI: 10.1016/j.compbiomed.2024.109154 should be considered in discussion) The reasoning behind the choice of the model architecture as well as parameters settings e.g. for thresholds is missing. For the evaluation the information about the ground truth in mid-diastole is missing. The qualitative results seem to be only marginally better than state-of-the-art methods used for comparison. The visualizations are very small, so that the qualitative results are difficult to assess.
- 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.
- 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 authors use the term “topology” to give a name to one of their building blocks (TCR), this does not seem appropriate to talk about topology in this context, the term should be changed. Adjectives such as powerful/superior and so on should be omitted in descriptions. A longer version of the paper (submitted to a journal) would be interesting for the community.
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper presents an interesting new approach for the consideration of volume and surface consistency in the segmentation of the mitral valve, which is of interest for the MICCAI community and can inspire good discussions. The paper is acceptable if the authors agree to address the listed weaknesses.
- 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.
Trusting that the authors will make the announced changes to the paper and act openly, objectively and with the appropriate respect for critical questions and alternative approaches in the conference discussions, I propose that the paper be accepted.
Review #3
- Please describe the contribution of the paper
This paper proposes a semi-supervised approach for obtaining mitral valve segmentation from 4D transesophageal echocardiography (TEE). Main contributions include a novel neural network design with memory and topology modules, and strong performance vs. many relevant baselines on a sizeable real patient dataset.
- 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 paper is well-written, and includes very meticulous & helpful figures. The method includes many interesting ideas, including teacher-student learning, memory blocks for handling multiple phases, and topology regularization using simple shape metrics. A large number of reasonable baselines were chosen, and the proposed method showed significant improvements. Ablation studies are reasonable and helpful. Qualitative results are promising, and even 3D printing was performed.
- 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.
Symbol definitions and general explanation of Eq(3) is a bit lacking (A, script(A), delta, where the equation comes from / what it’s called) Methods seem to be missing citations sometimes (e.g. attention, surface area, etc.)
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is a very strong paper with very few notable weaknesses. Figures and written contents are clear, methods are interesting, results show improvements vs. many baselines, and experiments are quite exhaustive (baselines, ablations, 3D printing).
- 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.
I have read all of the reviews and responses, and I stand by my original review/score. This paper includes 3 well-thought-out methodological choices that can spark interesting conversations, and the evaluations are quite thorough for a conference paper. My feedback is mostly to add more clarifications for the method descriptions, e.g. how L_sup and L_consis are defined, how delta A(v) is calculated, etc. There are also minor mistakes “see Fig. 2 (c)” - there is no Fig. 2 (c) citations 16 and 17 seem repeated
Author Feedback
We appreciate all the reviewers for reviewing and recognizing our work. Clarifications have been provided to address the comments.
Q1. Novelty. (R2) Our work has remarkable novelty in methodology, results and clinical applications, enabled by our first and largest 4D mitral valve (MV) dataset (R1, R3). We acknowledge that [14] shares a similar task definition, using ED and ES as labeled phases to segment the full cardiac cycle. However, the novelty of our work lies not in this task setup, but rather in how we exploit the latent structural consistency between labeled and unlabeled phases. Our novel designs include: (1) Motion-guided Consistency Learning (MCL): Unlike [14], which ignores temporal coherence, we introduce the bi-directional memory mechanism to ensure motion consistency. (2) Topology-guided Consistency Regularization (TCR): we propose a novel morphological regularization that leverages surface and volume invariance to enforce anatomical plausibility across phases.
Q2. Open Resource (R1, R2, R3) We will release code, high-resolution test samples with label, and 3D printing visualizations up acceptance. We believe this work will benefit the cardiac community. (R1,R3)
Q3. Comparison with [Ref 1-2]. (R2) The methods in [Ref 1] and [Ref 2] are unsuitable for 4D MV ultrasound segmentation, because: (1) [Ref 1] struggles to capture the complex deformation of MV, as it models only planar and unidirectional motion based on 2D optical flow. Clinically, myocardial motion is primarily unidirectional, while the MV exhibits anisotropic longitudinal displacement (8–12 mm), annular contraction (15–25%), and rotation (5–10°). Our motion-morphology consistency learning is designed to handle such challenging deformations. (2) [Ref 2] is not applicable to our sparse 4D setting. It relies on motion smoothness priors, suited for densely sampled CMR sequences (~25 phases/cycle). Our 4D MV has only ~9 sparse phases, with inter-frame displacements up to 15–20 mm, making such priors unreliable. Instead, our TCR provides a more suitable prior embedding for sparse 4D MV data.
Q4. Comparison with MV-GNN. (R1) MV-GNN (R1, Q1) simplifies the MV as an idea tubular sheet, limiting its use in simulating the common organic mitral regurgitation disease with valve abnormalities. Our MTCNet can handle such cases. We will cite & compare with MV-GNN in our final version.
Q5. Solutions for Task Challenges. (R1, R2) Our MCL and TCR modules are specifically designed to address motion artifacts and complex deformations. (1) Quantitative Improvements: In Tab. 2, MCL raises the mean Dice from 85.51% to 86.66%, and TCR further to 87.30%. Overall, MTCNet outperforms ICT by 1.07% in dice, all with p<0.05. (2) Qualitative Improvements: Fig. 4 shows MTCNet gets more visual-plausible segmentations. Particularly in mid-diastolic (MD) phase where severe motion and deformation often cause holes (bule arrows).
Q6. Phase MD vs. MD-1. (R2) MD is peak diastolic opening. MD-1 is the beginning of closure. At MD phase, the leaflets are maximally separated and often hit the left ventricular wall, making the boundary more blurred and hence difficult for segmentation. Fig. 1 highlights morphological differences, especially at the valve tips. In Tab. 1, MD has a slightly lower Dice than MD-1 (82.69% vs. 87.14%), which is reasonable.
Q7. Model Architecture and Parameters. (R1) We adopt the Mean Teacher framework for its strong learning capacity under sparse annotations, aligning well with the ED and ES labels in 4D MV segmentation. All parameters are empirically tuned on the validation set. (e.g. σ=0.1, balancing losses).
Q8. Writing Details. (R1, R3) We will replace ‘topology’ with the more appropriate term ‘morphology’ and revise the manuscript for a more objective tone. (R1) In Eq. (3), A(x) and dA(x) denote the local surface area in continuous space, while ΔA(v) is their discrete voxel-level form. Definitions and relevant citations will be added in final version. (R3)
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.
N/A
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A