List of Papers Browse by Subject Areas Author List
Abstract
Deep learning models for electrocardiogram (ECG) classification often suffer from significant performance degradation when deployed in unseen domains due to shifts in acquisition devices and patient populations. Test-time adaptation (TTA) offers a practical solution by adapting models using only unlabeled data at inference time. However, existing TTA methods often underperform on ECG tasks, since naive online updates ignore the hierarchical beat-rhythm structure of cardiac cycles and are vulnerable to signal artifacts, which leads to unstable adaptation and model drift. We propose BeatRhythm-TTA, an ECG-tailored TTA framework that explicitly accounts for ECG’s noisy observations and structured beat-rhythm semantics under domain shift. First, to handle pervasive ECG artifacts, we introduce a Signal Quality Index (SQI)-gated adaptation scheme that selectively filters out low-quality signals to prevent harmful updates. Second, to leverage ECG’s beat-rhythm semantics, we enforce dual-level consistency so the model preserves beat morphology and rhythm dynamics while adapting to shifted acquisition conditions. Extensive experiments on multi-label ECG diagnosis across three adaptation protocols, using PTB-XL as the source domain and CPSC2018/Georgia as two target domains, demonstrate the effectiveness of our method, yielding an average +2.70% relative improvement in Macro-F1 over the best competing method. The code is available at https://github.com/Vincent-Wenhan/BeatRhythm-TTA.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2304_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Vincent-Wenhan/BeatRhythm-TTA
Link to the Dataset(s)
PTB-XL ECG dataset (Public 12-lead ECG dataset used as the source domain in this paper): https://physionet.org/content/ptb-xl/
CPSC2018 ECG dataset (Public 12-lead ECG dataset from the PhysioNet/Computing in Cardiology Challenge 2020, used as a target domain): https://physionet.org/content/challenge-2020/
Georgia 12-lead ECG Challenge dataset (Public 12-lead ECG dataset from the PhysioNet/Computing in Cardiology Challenge 2020, used as a target domain): https://physionet.org/content/challenge-2020/
BibTex
@InProceedings{JiaWen_TestTime_MICCAI2026,
author = { Jiang, Wenhan AND Deng, Zhipeng AND Zhou, Jiale AND Wang, Haolin AND Ou, Yafei AND Zheng, Yefeng},
title = { { Test-Time Adaptation for ECG Classification via SQI-Gated Self-training and Beat-Rhythm Consistency } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16896},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes BeatRhythm-TTA, an ECG-specific test-time adaptation framework that adapts pretrained models to unseen target domains using only unlabeled target data at inference time. The key idea is to combine beat-wise SQI-gated self-training, which reduces harmful updates from low-quality ECG segments, with a dual-level consistency regularization that preserves both local beat morphology and global rhythm dynamics under augmentation. Experiments on public ECG datasets under three adaptation protocols demonstrate the improvements over representative test-time adaptation baselines.
- 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 paper is generally well organized and clearly presents the proposed framework, with Fig. 2 effectively illustrating the overall pipeline and workflow. 2.The motivation is well introduced. The authors show that generic TTA methods are suboptimal for ECG classification because they are sensitive to noisy signals and do not capture ECG-specific beat-rhythm semantics. To address these gaps, the paper proposes an ECG-tailored adaptation framework for more reliable cross-domain classification.
- 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 performance gains are consistent, but the absolute improvements remain relatively modest. For example, in the offline setting, the average Macro-F1 increases from 60.64% to 62.42%, and Macro-AUC from 88.13% to 88.84%. The paper would be stronger if the authors could better contextualize the practical significance of these gains, particularly with respect to their potential clinical relevance. 2.While the results on the two target datasets provide a useful starting point, the experimental scope remains relatively narrow. Restricting evaluation to a shared 6-class label space substantially simplifies the adaptation problem compared with real-world ECG diagnosis. To better support the paper’s claims of practical utility, the authors should provide a more thorough discussion of how the proposed method may generalize to broader diagnostic categories and more diverse clinical settings beyond the current evaluation.
- 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?
The problem is relevant, and the proposed method is reasonably aligned with the characteristics of ECG signals. The overall idea is sensible, the paper is readable, and the ablation study helps support the design choices. My main reservation is that the empirical gains remain modest, and the evaluation is limited to only two target domains and a shared 6-class label space. Without validation on a broader diagnostic range or more diverse institutional data, it is difficult to assess whether the method truly generalizes to real-world distribution shifts or is mainly effective for this specific experimental setting.
- 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 address my concerns.
Review #2
- Please describe the contribution of the paper
The paper proposes BeatRhythm-TTA, a framework for test time adaptation in ECG classification. It addresses domain shift by evaluating cross-domain generalization (Germany to the USA and China). The authors introduce an SQI-gated adaptation mechanism tailored to test time adap. and a dual level regularization to align representations at different levels.
- 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 addresses an important issue. Domain shift is a practical problem that directly affects the usability of models in clinical applications.
- It applies the TTA to a relevant domain and proposes the use of meaningful features aligned with application (e.g., beat structure and rhythm-level).
- The incorporation of SQI into TTA to handle noise is a practical and efficient design choice.
- The experimental setup and execution appear adequate, with evaluations across different settings and comparisons against a diverse set of methods from the literature.
- The use of multiple datasets is appropriate for assessing domain shift.
- The evaluation includes different scenarios (offline, continual online, and independent online).
- The structure and organization of the paper are clear and easy to follow.
- 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 ideas are relevant and coherent, the level of innovation does not constitute a major methodological advance. The method mainly provides incremental improvements through a combination of established techniques, with novelty relying more on their application to TTA rather than on fundamentally new concepts.
- The reported performance gains are modest. The paper lacks statistical significance testing and confidence intervals across models and settings. While improvements appear consistent, they are relatively small. Additional evaluation metrics (e.g., balanced accuracy or sensitivity) would be valuable, especially given their importance in clinical contexts.
- Although the ablation study demonstrates the contribution of each module, the paper lacks deeper analysis. In particular, there is no per-class performance evaluation, qualitative examples comparing methods, and no discussion of how the improvements translate into clinical settings.
- The discussion section is relatively superficial. The paper does not sufficiently analyze failure cases or conditions under which drift affects performance. Additionally, without confidence intervals, the robustness of the method under different scenarios remains unclear.
- Some implementation details are missing or insufficiently described. These include augmentation strategies, certain model parameters, and aspects of the loss formulation. The SQI thresholds are not fully justified. Although code is provided, it is not straightforward to follow due to limited documentation. All key details affecting reproducibility should be clearly described in the paper itself.
- 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
The paper is well motivated and addresses a relevant problem, but it would benefit from deeper discussion on the clinical impact of the model and a more rigorous statistical evaluation.
- 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?
The idea is solid and paper is organized, but it does not clearly stand out in terms of innovation and lacks strong evidence to support the robustness of its results.
- 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
The authors proposed a test time adaptation (TTA) method for ECG classification intended to overcome the model updation due to noisy artifacts and to use the beat rhythm semantic of cardiac cycles. 1.Based on the proposed Signal Quality Index (SQI), the beats are either rejected or used to update the model and pseudo labelling. 2.For capturing rhythm irregularity across multiple beats, cosine alignment loss (L_rhythm) is calculated between augmented and clean beat features. For capturing the beat morphology(ST segment, T-wave inversion), the cosine distance is calculated, and the beat loss(L_beat) is calculated. Both losses are combinedly optimised.
- 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 proposed method is evaluated in publicly available standardized datasets in three protocols - offline, continual online and independent online and compared with state-of-the art methods.
- 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 performance comparison between the proposed method and existing state-of-the-art techniques is inherently biased. Specifically, the authors adopted the optimization strategy of disabling dropout and updating only BatchNorm affine parameters which was derived from [17] for their own method and SAR [17]. Omitting this strategy for the remaining baselines results in an unfair evaluation framework.
[17]. Niu, Shuaicheng, et al. “Towards stable test-time adaptation in dynamic wild world.” arXiv preprint arXiv:2302.12400 (2023).
- 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
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 — 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?
1.The proposed work properly addressed the existing research gap in Test Time Augmentation using SQI gated pseudo labelling and dual level consistency. However, the non ideal evaluation made the reviwer to opt for Weak Reject.
- Reviewer confidence
Somewhat confident (2)
- [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.
While the authors explained the ambiguity about the “biased evaluation and comparison with state of the art methods” , the justification regarding BatchNorm affine parameters is not convincing. Hence the decision is made.
Review #4
- Please describe the contribution of the paper
This paper proposes BeatRhythm-TTA, a test-time adaptation framework for multi-label ECG classification under cross-domain shifts. The method argues that generic TTA methods are not well suited for ECG because they ignore two modality-specific properties: ECG signals are frequently corrupted by artifacts, and diagnostic information exists at both beat-level morphology and rhythm-level temporal dynamics.
- 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.TTA has been studied heavily in vision, but ECG-specific TTA remains less explored. The paper identifies a real gap: ECG adaptation should account for signal quality and physiological temporal structure, rather than directly importing image-based TTA objectives. 2.The two proposed ingredients: SQI gating and beat/rhythm consistency are well aligned with ECG interpretation. Gating low-quality segments is clinically intuitive, and separating beat morphology from rhythm dynamics is a reasonable way to encode ECG structure. 3.The authors test offline TTA, continual online TTA, and independent online TTA. This is a strength because TTA behavior can differ substantially depending on whether target samples are processed in batch, sequentially with accumulated updates, or independently with per-sample resets. 4.The method achieves the best average Macro-F1 and Macro-AUC in offline, continual online, and independent online settings. The gains are not huge, but they are stable, which supports the usefulness of the ECG-specific adaptation design.
- 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 reports consistent improvements, the margins are modest. For example, in offline TTA, the average Macro-F1 improves from 60.64% to 62.42%; in continual online TTA, from 60.62% to 61.74%; and in independent online TTA, from 60.66% to 62.63%. These are useful gains, but the paper should avoid overstating them. 2.The three components in Eq.3 may have different scales and distributions, and the paper does not specify normalization, calibration, or validation of this SQI against known signal-quality labels. The choice of multiplicative combination is intuitive but not well justified. 3.Beat-level SQI and beat consistency depend on detected R-peaks from WFDB. Under strong domain shift, noise, arrhythmia, or low-quality recordings, R-peak detection itself may fail. The paper does not analyze sensitivity to R-peak errors, which is a key dependency of the method. 4.The paper motivates beat-level consistency using ST-segment changes and T-wave inversion, but the actual label space includes AF, AV block, bundle branch blocks, PAC, and normal sinus rhythm. Some of these are more rhythm-dependent than morphology-dependent. The paper should more carefully connect each class to the proposed beat/rhythm consistency mechanisms. 5.PTB-XL → CPSC2018 and PTB-XL → Georgia are useful cross-domain settings, but still limited. ECG domain shift can vary widely across devices, hospitals, filtering pipelines, and patient populations. More target datasets or additional source-target combinations would strengthen the claim of general ECG TTA robustness. 6.The method is motivated by beat/rhythm semantics, but the paper does not show examples of ECG traces where the method improves predictions, nor does it analyze class-wise behavior in detail. Per-class F1/AUC or examples of noisy signals filtered by SQI would make the physiological argument stronger.
- 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?
The paper is not conceptually groundbreaking, but it is a well-motivated and reasonably executed modality-specific TTA method. Its consistent gains across multiple protocols make it stronger than a simple engineering tweak. With additional analysis on SQI robustness, class-wise behavior, and statistical variance, it would become substantially more convincing.
- Reviewer confidence
Somewhat confident (2)
- [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 addresses several of my concerns, especially by providing per-class results, statistical testing, and additional degraded-signal experiments. However, the response only partially resolves the dependence on R-peak detection, since the added noise experiments do not directly test peak detection failures. The SQI formulation remains heuristic, and the limited closed-set 6-class evaluation is still a scope limitation rather than a resolved issue. Overall, I would maintain my Weak Accept score, but I would not increase it.
Author Feedback
We thank the reviewers and ACs for the constructive comments. We are encouraged that reviewers recognized the clinical motivation, ECG-specific design, and clear organization of our work. We are committed to revising our manuscript accordingly and address the main concerns below.
1.Fairness of adaptation settings (R4). We thank R4 for noting this ambiguity. The manuscript described the update scope of our method, which may have implied that this setting was applied only to our method and SAR. This is a presentation issue, not an experimental bias. All methods used the same source checkpoint, target stream, preprocessing, and protocol-specific adaptation budget, while following their official/default implementation protocol. We used no privileged dropout or extra trainable parameters compared to baselines. We will explicitly list the dropout handling and adaptation scope of each baseline.
2.Per-class performance (R2/R5). Due to page limits, we did not include the full per-class table, but we examined class-wise performance. Under independent online TTA setting, compared with SAR, our method improves IAVB/LBBB/RBBB/PAC/NSR F1 from 62.48/68.97/85.95/10.06/52.78 to 67.57/70.32/86.22/14.39/54.91, while AF slightly decreases from 83.72 to 82.40.Thus, the gain is not driven by a single class but we will avoid claiming uniform per-class improvement.
3.Statistical testing (R2). To quantify the consistency of observed improvements (averaged over three runs), we performed paired t-tests against the strongest baseline per protocol. The resulting p-values for Macro-F1 are p=1.356e-05, p=7.216e-05, and p=3.911e-03 for offline, continual online, and independent online TTA. Macro-AUC improvements are smaller and will be interpreted cautiously.
4.Robustness under degraded signals and R-peak dependence (R5). We additionally tested ECGs degraded by Gaussian noise and baseline wander. In offline TTA, Ours achieves Avg AUC/F1 88.64/60.89, outperforming UCon 88.06/59.97 and Source Only 88.03/59.38.In independent online TTA, Ours achieves 88.07/59.54, compared with SAR 87.82/58.04 and Source Only 87.77/57.81.This suggests SQI-gated adaptation remains stable under common waveform degradations. While this cannot fully isolate R-peak errors, future work will examine detector choices, peak detection errors, and noise severities.
5.Limited evaluation scope and generalizability (R1/R5). We agree that clinical ECG diagnosis covers more conditions. Using PTB-XL (source) and CPSC2018/Georgia (targets) provides a controlled cohort and acquisition shift setting. The shared 6-class space is required for closed-set TTA and enables fair comparison across datasets with inconsistent diagnostic taxonomies. Handling unseen classes is an open-set problem beyond current scope. Future work will explore unknown-class rejection and broader label spaces.
6.SQI design, thresholds, and implementation details (R2/R5). SQI is a lightweight unsupervised reliability gate for updates, not a clinically validated quality classifier. It reflects general ECG waveform properties: QRS energy concentration, QRS sharpness, and baseline-wander suppression. After robust normalization, these dimensionless terms are combined as a soft gate to down-weight unreliable samples. The thresholds tau_c=0.2 and tau_q=0.05 are fixed across all target datasets/protocols without target-label tuning. We will also specify the strong-view augmentations: Gaussian noise, gain jitter, and one temporal mean-mask, without time shift, crop, or warping to preserve beat-window alignment.
7.Modest absolute gains and contribution (R1/R2/R5). The gains should be viewed in the difficult label-free, source-free TTA setting, where avoiding model collapse during online updates is challenging. Our framework yields consistent improvements across each protocol by suppressing artifact-driven updates via SQI gating. We will emphasize this deployment robustness while avoiding overstatement.
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.
This paper proposes an ECG-tailored Test-Time Adaptation (TTA) framework utilizing SQI-gated self-training and beat-rhythm consistency. While reviewers appreciate the modality-specific clinical motivation, the consensus leans borderline due to concerns regarding evaluation fairness, statistical rigor, and edge-case robustness. Specifically:
-R4’s concern regarding biased optimization settings (BatchNorm/dropout updates) applied to your method versus the baselines.
-per-class performance breakdowns (R2, R5).
-Analyze the framework’s sensitivity to R-peak detection failures under severe noise or domain shifts (R5).
-Justify the limited evaluation scope (2 target datasets, shared 6-class space) and discuss generalizability (R1).
- 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 presents a clinically well-motivated test-time adaptation framework tailored to ECG multi-label classification, which effectively integrates physiological structure via SQI gating and dual-level consistency. Although concerns regarding baseline optimization fairness and modest absolute gains were raised, the authors’ post-rebuttal clarifications, additional statistical tests, and noise-degradation experiments successfully satisfied the majority of the review panel. Balancing the methodological suitability for clinical deployment against the remaining minor baseline ambiguities, a Weak Accept is recommended.
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.
After reviewing the manuscript, the reviews, and the authors’ rebuttal, I recommend acceptance.
The rebuttal adequately addresses most of the major concerns raised during review. In particular, the additional analyses, statistical testing, and degraded-signal experiments strengthen the empirical evidence supporting the proposed approach. This is reflected in the updated assessments from Reviewers 1 and 5. I acknowledge the remaining concerns regarding the interpretation of certain design choices and the limited evaluation scope. However, these issues do not undermine the core contribution or the validity of the reported results.
Overall, I believe the paper makes a solid contribution and that the remaining weaknesses are not sufficient to outweigh its strengths. Therefore, I recommend 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.
Reject
- Please justify your recommendation.
I would lean toward rejection. Although the paper addresses a relevant ECG test-time adaptation problem and the proposed SQI-gated beat/rhythm consistency design is reasonable, the contribution appears incremental and the reported gains are relatively modest. More importantly, several concerns remain insufficiently resolved, including the fairness of baseline comparisons, the heuristic SQI formulation, dependence on reliable R-peak detection, and the limited closed-set 6-class evaluation. Therefore, I am not fully convinced that the current evidence supports acceptance at MICCAI.