List of Papers Browse by Subject Areas Author List
Abstract
Uncertainty in echocardiography segmentation is predominantly geometric: speckle, shadowing, and inter-scanner variation create boundary ambiguity that propagates to clinical measurements.
We propose WaCT-US, a contour-native uncertainty representation that outputs a calibrated prediction set (``contour tube’’) around a single predicted boundary.
Cyclic landmark alignment resolves closed-curve indexing, and a learned heteroscedastic scale profile specifies where the boundary is intrinsically ambiguous.
Split conformal calibration on held-out data yields one global scalar that meets a user-chosen coverage level with finite-sample, distribution-free marginal validity.
Inference requires one forward pass with light post-processing.
We evaluate on CAMUS and EchoNet-Dynamic across three seeds at nominal $0.9$ conformal coverage.
While matching Dice, WaCT-US achieves the best mean calibrated uncertainty scale (representing conformal efficiency and tube thickness) while holding comparable coverage and improving uncertainty-guided error localization. This information can estimate how wrong the underlying segmentation model is on each input image. The code will be publicly available at \url{https://github.com/MohammadMahdiKazemi/WaCT-US}.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/6356_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/MohammadMahdiKazemi/WaCT-US
Link to the Dataset(s)
N/A
BibTex
@InProceedings{KazMoh_WaCTUS_MICCAI2026,
author = { Kazemi Esfeh, Mohammad Mahdi AND Gholami, Zahra AND Liao, Renjie AND Luong, Christina AND Tsang, Teresa S. M. AND Abolmaesumi, Purang},
title = { { WaCT-US: Cyclic-Alignment Scaled Contour Tubes for Efficient, Calibrated Uncertainty in Echocardiography } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16882},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper addresses uncertainty estimation in image segmentation aimed at echocardiography using conformal predictions with a focus on boundary rather than pixelwise probabilities.
A contour tube is estimated around a single contour with varying thickness indicating uncertainty. Cyclic alignment is performed to have a consistent indexing of the contour. The method is evaluated on public data with comparison to alternatives and promising 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.
The boundary-centric conformal approach is interesting and timely. This is a nice adaptation of the conformal approach to boundaries.
Calibration is appropriate.
The method is computationally efficient.
The evaluation is thorough and the results are nicely interpretable.
- 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.
It is unclear how this method could generalize to 3D surfaces.
No discussion of the sensitivity to distribution shift.
A qualitative illustration of the advantage of the tube approach over pixelwise analysis would be useful.
- 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 provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Of interest to the MICCAI community and well-formulated.
- 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
Review #2
- Please describe the contribution of the paper
The authors propose WaCT-US, a contour-native uncertainty representation that outputs a conformal calibrated prediction set (“contour tube”) around a predicted boundary. They introduce a second prediction head to predict the local scale for the confidence region around the predicted landmark, driven by a pinball loss. The authors evaluate on CAMUS and EchoNet-Dynamic.
- 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 methodology is novel and addresses an important problem (deriving efficient confidence regions around predictions with adequate coverage of the ground truth).
The framework is sound.
- 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.
- I am not certain that the paper conclusively demonstrates the superiority of the method over the SOTA. The methods seem to all have very similar performance with respect to all metrics, with a slight edge for the proposed method w.r.t. Scale, that may only be statistically significant for one of the two datasets (no direct significance test is performed).
Part of my unease comes from the fact that the methods to derive the Scale metric for the proposed method and for the baselines might be fairly diverse and this might impact the reported numbers. I would be grateful if the authors can comment on this.
- One weakness is that despite deriving confidence sets that are geometrically more constrained (as contour tubes), the main segmentation head itself is unchanged. As in, it is still based on predicting a mask, trained with Dice+BCE, which for echo data will often lead to predictions that are both topologically incorrect and geometrically unconstrained. In my opinion this is a strong limitation for the practical usefulness of the work.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/en/REVIEWER-GUIDELINES.html
- I am unsure why all metrics are bolded for the WaCT-US lines in Table 1.
-
The Dice performance of the trained networks seems to be a bit below standard numbers reported for CAMUS. Although this is not the main aim of the paper, I would be more at ease analyzing confidence estimates on baselines that are fully on par with the literature.
-
Baselines specific to echocardiography could be relevant additions, for instance Judge et al., Asymmetric Contour Uncertainty Estimation for Medical Image Segmentation, MICCAI 2023
- My understanding is that there is no particularly beneficial interaction between the segmentation head and the scale head during training. Is that correct?
- 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 — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
At this point of time, it seems to me the weaknesses of the paper slightly outweigh its strengths, but I am willing to adjust my rating after the rebuttal.
- 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.
Many thanks to the authors for taking the time to respond to the reviewers’ comments and questions.
Final assessment: For me, the paper is entirely borderline. Given a choice between accept and reject, I am choosing accept because I believe the paper could be interesting to the UQ community. Strengths:
- a boundary-centric conformal approach with good novelty Weaknesses:
- Limited evidence for the relevance and effectiveness of the approach, given limited improvement across datasets with no statistical significance test
- One could have favored exactly the opposite approach of demonstrating the relevance of the framework by demonstrating its applicability on a natively contour-based segmentation method, and demonstrating both superior segmentation quality and sharper uncertainties.
- The benefit of having calibrated prediction sets is reduced when the returned contour is as likely to be topologically and geometrically incorrect.
Review #3
- Please describe the contribution of the paper
The proposed provides ~sota uncertainty estimates in echocardiography segmentation at significant speedup to competing non-deterministic methods. The paper includes some detail on comparing against competing methods fairly.
- 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.
significant speedup and reduced uncertainty with locality
- 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 “competing” baselines noted are older, and it’s unclear why several already-cited recent works would not make more meaningful comparison.
- 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 mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
approximately sota segmentation performance while providing calibrated uncertainty scale and error localization. Improvements noted seem marginal and comparison is against older works.
- 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.
Given rebuttal, decision remains weak accept.
Author Feedback
We thank the reviewers and meta-reviewer (MR) for constructive feedback. We focus on fair comparison/statistical interpretation, recent echo baselines, and scope/practicality.
Fair comparison, Scale, significance (R2/R3/MR). We do not claim Dice/HD superiority; the segmentation protocol is intentionally kept standard/comparable to isolate UQ. Bolding WaCT-US rows in Table 1 was intended only to mark the proposed method, not metric-wise dominance. The claim is coverage-efficiency: at nominal 0.9 conformal coverage, WaCT-US gives the smallest calibrated tube scale with one forward pass. Scale is comparable because all methods are converted to the same tube family: sample the method’s local uncertainty on the predicted contour, compute the same Eq.(3) score, calibrate a method-specific λ̂ by the same quantile rule on the same validation split, evaluate on the same test frames, and report Scale=mean_i[λ̂ mean_j s_ij]. Raw entropy/vacuity/std units are not compared; a global rescaling of any raw map is absorbed by λ̂, leaving λ̂s and Scale unchanged. Table 1 reports comparable coverage with exact 95% CIs; WaCT-US has the lowest mean Scale (CAMUS 8.09 vs 8.98 next; EchoNet 5.10 vs 5.27 next). We did not include a direct Scale significance test, and the EchoNet gap is modest; thus the claim is an observed Pareto advantage in Scale and compute, not formal statistical dominance.
Recent/relevant baselines (R2/R3/MR). The baselines cover UQ regimes under protocol parity: post-hoc deterministic maps, evidential/temperature maps, and sampling/Bayesian approximations (MC dropout, Laplace, SWAG), all conformalized identically. Judge et al. is relevant and cited; however, it produces asymmetric contour-regression uncertainty, whereas WaCT-US produces a split-conformal prediction set around a standard segmenter with finite-sample marginal coverage. A fair comparison would require an adapter/conformalized contour-set variant, so we do not claim dominance over that line. Similarly, recent conformal segmentation works cited in the paper are mask/morphology based or contour objects in another setting; they do not evaluate the same closed-contour, learned-local-scaling, single-pass echo setting.
Segmentation/topology and scale head (R2). We agree topology-constrained echo segmenters are valuable. WaCT-US is architecture-agnostic (Sec.2.3) and can wrap stronger/topology-aware backbones. The Dice+BCE U-Net keeps the center-contour model unchanged across UQ methods. R2 is correct that the scale head is not designed to improve Dice/topology or feed back to correct the center contour; it learns residual-dependent local scale so the output reports where/how much the contour may be wrong. Limitations of the chosen segmenter, including the CAMUS Dice concern, therefore affect absolute segmentation quality, but not the within-paper UQ comparison: WaCT-US is not given a stronger center-contour model than the baselines.
Shift, 3D, practical advantage, reproducibility (R1/MR). The paper states the exchangeability requirement: Theorem 1 gives P_te(d_s≤λ̂)≥1−α−δ under TV shift δ, so we do not claim robustness to arbitrary shift; weighted conformal alternatives are noted for drift. In 3D, split-conformal validity could apply to a chosen surface score under exchangeability, but cyclic shift would need surface correspondence/alignment; our empirical claim is limited to 2D echo contours. Practical value is shown in Fig.2: WaCT-US has the lowest AURC and deferring the top 20% most uncertain CAMUS samples reduces retained contour MAE by 15.7%; Fig.2b shows contour-tube overlays where uncertainty tracks image quality/contour mismatch. For reproducibility, Secs.2.3/3.1-3.3 specify public datasets/splits, preprocessing, model, losses, optimizer/epochs, M=64, τ=0.9, scale weight, calibration quantile, baselines, and metrics; we intend to release code after acceptance.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
Please prioritize responding on comparisons with more recent and relevant baselines and clarify whether the observed performance differences are statistically significant. Additionally, address concerns regarding the evaluation of the Scale metric and ensure fair comparability across methods. Finally, respond to limitations and generalizability concerns, including applicability to 3D settings, robustness to distribution shift, and evidence of the practical advantages of the proposed approach.
- 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.
The paper presents WaCT-US, a contour-native uncertainty representation that outputs a conformally calibrated prediction set. The strengths and limitations of the work were both clearly discussed during the review process, and the benefits appear to outweigh the weaknesses. In particular, the boundary-centric conformal formulation is novel and practically appealing, especially given the significant speedup reported by the authors. The questions regarding relevant baselines were also adequately addressed in the rebuttal. Overall, the paper is recommended for acceptance.
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.
All reviewers agreed that this paper is well prepared and of interest to MICCAI. R2 originally recommended borderline reject, but raised their score to accept after authors’ rebuttal. AC recommends acceptance.
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.
This paper proposes a novel efficient uncertainty quantification in echocardiography segmentation around boundaries through scaled contour tubes. It adopts cyclic landmark alignment to address index inconsistency and split conformal calibration to ensure marginal coverage. Although there are some shortcomings in terms of generalizability, sensitivity analysis, and performance advantages, these do not undermine the contributions and potential impact of this paper’s methodological innovation. The method will provide a new perspective on uncertainty quantification in medical image segmentation and is expected to enhance the reliability and accuracy of diagnoses in clinical practice.