List of Papers Browse by Subject Areas Author List
Abstract
Coronary artery stenosis is a common cardiovascular disease, with severe, untreated cases posing significant risks of heart attack. Although coronary (X-ray) angiograms remain the standard for stenosis diagnosis, they are invasive, time- and resource-intensive, and therefore only performed on patients with a high probability of disease based on symptoms and prior clinical tests. However, a subset of patients, especially those without symptoms, may remain undiagnosed. Detecting indications of stenosis from ECGs, which are fast, cheap, non-invasive, and thus routinely acquired even in asymptomatic patients, would support early diagnosis. However, as no reliable stenosis-specific signal has been identified in ECGs, they can not currently be used for stenosis risk stratification. To address this, we introduce StenCE, a pretraining framework, allowing stratification of patients based on features derived directly from ECGs. Evaluations across varying stenosis severity thresholds and additional ECG disease classification tasks demonstrate consistent performance improvements across different ECG encoders, outperforming previous work. The obtained models successfully detect signals for stenosis diagnosis in ECGs and are the first to achieve high performance in severe stenosis classification. The source code is available at https://github.com/NikolaCenic/ecg-stenosis-cls.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1945_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/NikolaCenic/ecg-stenosis-cls
Link to the Dataset(s)
N/A
BibTex
@InProceedings{CenNik_CrossModal_MICCAI2026,
author = { Cenikj, Nikola AND Turgut, Özgün AND Müller, Alexander AND Steger, Alexander AND Kehrer, Jan AND Brugger, Marcus AND Rueckert, Daniel AND Martens, Eimo AND Müller, Philip},
title = { { Cross-Modal Contrastive Learning of ECG and Angiography Representations for Severe Stenosis Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16885},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces the first cross-model contrastive pretraining framework between ECG and coronary X-ray angiography. This enables an ECG encoder to detect coronary artery stenosis signals.
- 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.
For the first time, stenosis-specific diagnostic signals latent in ECGs can be unlocked through contrastive pretraining with paired coronary angiography data.
- 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.
From the previous study, the datasets were not available, so comparison is not possible. External validation is needed. As the paper mentioned, tranining population is different from the real clinical target population.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
This is clinically meaningful but with weakness that I mentioned above, I recommend weak accept.
- 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
a constructive pretraining framework that aligns ECG representations with features from a multi-view angiography stenosis classification model, thereby enabling an ECG encoder to detect stenosis signals from ECGs alone.
- 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 study is supported by a reasonably robust cohort size, with 3,001 cases for training, 204 for validation, and 209 for testing. This split provides a solid foundation for model development and internal evaluation, and suggests that the authors have taken care to separate data appropriately to mitigate overfitting. It would be helpful to further clarify whether the data were collected across multiple centers or scanners, and how class balance (particularly for severe stenosis cases) was handled to ensure generalizability. The proposed StenCE framework demonstrates promising performance, achieving an AUC of 0.82 in detecting the most severe cases. This indicates good discriminative capability, particularly in clinically critical scenarios where accurate identification of high-risk lesions is essential. To strengthen the impact, the authors could compare this performance against existing state-of-the-art methods or clinical benchmarks, and provide additional metrics (e.g., sensitivity, specificity, calibration) to better contextualize the model’s clinical utility.
- 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.
Based on American College of Cardiology (ACC) guidelines, stenosis severity is typically categorized as: minimal (<25%), mild (25–49%), moderate (50–69%), severe (70–99%), and total occlusion (100%). It would significantly strengthen the manuscript to provide a detailed breakdown of the dataset according to these categories, including both the number of subjects and the number of vessels in each severity group. This information is important to assess class distribution, potential imbalance, and the representativeness of clinically meaningful categories—particularly for moderate-to-severe lesions, which are most relevant for decision-making. The current evaluation focuses primarily on identifying the most severe stenosis cases. However, from a clinical perspective, it is equally important to assess the model’s ability to classify stenosis extent across all severity levels. The authors are encouraged to report performance at both the per-patient level and per-vessel level, including metrics such as accuracy, sensitivity, specificity, and confusion matrices for multi-class classification. This would provide a more comprehensive understanding of the model’s utility in routine clinical workflows, where grading stenosis severity—not just detecting extremes—is essential. External validation is lacking and represents a key limitation. Validation on independent datasets from different institutions, scanner vendors, or patient populations is critical to demonstrate the generalizability and robustness of the proposed framework. Without such validation, it is difficult to assess how the model would perform in real-world, heterogeneous clinical settings. A detailed failure analysis should be included to better understand the limitations of the model. Specifically, the authors should characterize cases where the model underperforms (e.g., heavy calcification, motion artifacts, poor image quality, or borderline stenosis). Identifying systematic error patterns would not only strengthen confidence in the model’s robustness but also provide insights for future methodological improvements and clinical deployment.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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
NA
- 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?
Based on American College of Cardiology (ACC) guidelines, stenosis severity is typically categorized as: minimal (<25%), mild (25–49%), moderate (50–69%), severe (70–99%), and total occlusion (100%). It would significantly strengthen the manuscript to provide a detailed breakdown of the dataset according to these categories, including both the number of subjects and the number of vessels in each severity group. This information is important to assess class distribution, potential imbalance, and the representativeness of clinically meaningful categories—particularly for moderate-to-severe lesions, which are most relevant for decision-making. The current evaluation focuses primarily on identifying the most severe stenosis cases. However, from a clinical perspective, it is equally important to assess the model’s ability to classify stenosis extent across all severity levels. The authors are encouraged to report performance at both the per-patient level and per-vessel level, including metrics such as accuracy, sensitivity, specificity, and confusion matrices for multi-class classification. This would provide a more comprehensive understanding of the model’s utility in routine clinical workflows, where grading stenosis severity—not just detecting extremes—is essential. External validation is lacking and represents a key limitation. Validation on independent datasets from different institutions, scanner vendors, or patient populations is critical to demonstrate the generalizability and robustness of the proposed framework. Without such validation, it is difficult to assess how the model would perform in real-world, heterogeneous clinical settings. A detailed failure analysis should be included to better understand the limitations of the model. Specifically, the authors should characterize cases where the model underperforms (e.g., heavy calcification, motion artifacts, poor image quality, or borderline stenosis). Identifying systematic error patterns would not only strengthen confidence in the model’s robustness but also provide insights for future methodological improvements and clinical deployment.
- Reviewer confidence
Very confident (4)
- [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 #3
- Please describe the contribution of the paper
This paper introduces StenCE, a pretraining framework, allowing stratification of patients based on features derived directly from ECGs. The evaluations are performed across varying stenosis severity thresholds and additional ECG disease classification tasks.
- 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-This paper proposes a contrastive pretraining framework that aligns ECG representations with features from a multi-view angiography stenosis classification model.
2-The paper employs multi-modal contrastive learning between an ECG encoder and an angiography encoder.
3-The dataset, consisting of 3,414 patients, is worthy of the community.
- 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-Although the paper introduces the clinical motivation, the underlying technical challenges are not clearly presented. In particular, the difficulties of cross-modal alignment between ECG and angiography are not explicitly discussed. 2-The proposed framework largely follows existing paradigms of cross-modal contrastive learning (e.g., CLIP-style alignment) combined with teacher–student knowledge transfer. The technical components, including feature projection, contrastive loss, and fine-tuning, appear to be straightforward adaptations of prior work, with limited methodological innovation. 3-The overall method diagram (Fig. 1) does not clearly convey the core technical contributions of the paper. 4-The overall presentation could be significantly improved. Some sections lack logical flow, and the formatting is not always clean, which negatively affects readability.
- 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 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?
None
- 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
None
- 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 limited methodological innovation.
- Reviewer confidence
Very confident (4)
- [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
Thank you for the review. We acknowledge the comments regarding the lack of external validation. We currently do not have access to datasets collected at different institutions or from diverse patient populations, which limits our ability to perform such evaluations at this stage. We, however, highlight that we have a large internal data cohort captured using different scanners, providing strong evaluation, and as part of future work, aim to evaluate it on multisituational datasets We also agree that the dataset is subject to selection bias, which we explicitly acknowledge in the limitations section. Constructing a fully unbiased dataset would require angiograms from asymptomatic individuals. However, acquiring such data is inherently challenging, as it would involve exposing healthy patients to an invasive angiographic procedure without a clinical indication. Regarding the comments on evaluating different severity thresholds, Table 1 and Figure 3 present both the data distribution and model performance across multiple thresholds. While we do not explicitly adopt the threshold defined by the American College of Cardiology guidelines, the evaluated thresholds closely align with those recommendations. Lastly, the novelty of our work does not stem from the contrastive pretraining methodology itself, but rather from the adaptation of the CLIP framework to paired angiography and ECG data. Specifically, our contribution lies in extending a multi-view encoder architecture for contrastive learning by introducing separate CLS and CLIP tokens, enabling effective alignment between angiography and ECG modalities within the contrastive framework.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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.
This submission presents a clinically relevant and technically sound study on transferring angiography-derived information to ECG representations for severe coronary stenosis classification. The main strengths are the clear clinical motivation, the new angiography-ECG cross-modal pretraining setting, the reasonably large internal paired cohort, and the comparative experimental validation. After checking the manuscript, the core contribution is supported: the proposed StenCE framework improves ECG-based severe stenosis classification, with the strongest result reported for the most severe threshold, and the paper also includes baseline comparisons, multi-threshold evaluation, and an ablation supporting the main design choices. The decisive concerns are the absence of external validation, the selected angiography-referred population that introduces acknowledged selection bias, and the fact that performance declines substantially for less severe stenosis levels. These are real limitations. However, some reviewer criticisms are only partially supported. After checking the manuscript, the concern that evaluation focuses only on the most severe cases is overstated because the paper already evaluates several stenosis thresholds and discusses the resulting drop in performance. Similarly, the criticism that the contribution is too limited methodologically to merit consideration gives insufficient weight to the new modality pairing, the comparative evidence, and the ablation study. Accordingly, I assign limited weight to those criticisms in the overall assessment.