Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Arrhythmias often violate the periodicity assumed in segmented cardiac cine-MRI, producing blurring and ghosting that degrade functional assessment. Due to ECG signal irregularities that cause temporal misalignment, arrhythmia-related artifacts impose more stringent requirements on denoising methods than conventional artifacts. We propose a Latent-Dictionary Encoding framework for Motion-Artifact Reduction (LDE-MAR). Using a latent image animator (LIA) formulation, LDE-MAR learns orthogonality-constrained sparse dictionaries to separate physiological motion into region and direction components, and adds a noise dictionary to capture artifact-related deviations in latent motion trajectories. Training is staged: motion dictionaries are pretrained on clean cine sequences, then only the noise dictionary is fine-tuned with paired clean-corrupted data to suppress irregular dynamics while preserving anatomy and plausible myocardial motion. This staged, multi-dictionary decomposition explicitly isolates arrhythmia-induced deviations from physiological dynamics, improving robustness under large R-R variability. We evaluate our method on a private dataset of 398 subjects and the public ACDC dataset. Our experimental results demonstrate that LDE-MAR outperforms state-of-the-art methods in artifact removal and in restoring motion and structural details. The code is available at https://github.com/gaoningn/LDE-MAR.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3222_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/gaoningn/LDE-MAR

Link to the Dataset(s)

N/A

BibTex

@InProceedings{NinGao_ArrhythmiaRobust_MICCAI2026,
        author = { Ning, Gaoning AND Yu, Chengjin AND Pu, Cailing AND Gao, Zhifan AND Xu, Chenchu AND Liu, Huafeng},
        title = { { Arrhythmia-Robust Cine-MRI via Latent Motion Artifact Characterization } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16888},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    In this paper, the authors propose LDE-MAR, which is a latent-dictionary encoding framework for motion artifacts reduction for cardiac cine-MRI that are affected by arrhythmia. Using three dictionary modules, their method decouples cardiac motion into region, direction, and noise components to explicitly characterize arrhythmia-induced artifacts. Experimental results demonstrate that the method achieves good results and relatively clean cardiac cine images. The framework was compared to several artifact suppresion methods and achieved the highest performance.

  • 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 major strength of the paper is a novel application of LDE for arrhythmia artifact suppresion in segmented cine images. By pretraining using clean images, the direction and regional dictionary modules can learn the cardiac motion patterns; then using supervised learning, noise reduction can be achieved. Another strength is the relatively high performance of the framework, and the ablation studies, which demonstrate the level of importance of different modules.

  • 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 paper has a couple of strengths, I found more weaknesses of the paper.

    First, how was the reference image generated or was it necessary to formulate the image wrapping/motion as two steps? Seems that the authors directly used LIA (latent image aminator) without reconsidering this step.

    Second, what do the authors mean by “further encode Wt->r into Wr->t”? How was this done and was it truly reversable?

    Third, I am not sure whether Equation 4 is correct or alpha is necessary there, because delta already describes the direction and magnitude of the motion, if there is no motion for a region, the magnitude should just be 0.This is probably also the reason that Model 4 has very similar performance against the full model in the ablation studies. Also, the authors should give more brief explanation of why using softmax and tanh in Equations (2) and (6).

    Fourth, the authors pretrained the motion dictionary modules first using clean images then trained the noise module. However, in reality, artifacts caused by arryhthmia can also affect/blur the cardiac motion. What if the training order switched, i. e. , to train the denoising/deartifact module first, then train the motion module? Furthermore, an iterative training may be more reasonable.

    Fifth, this is more of a big picture. The framework can be useful to suppress artifacts that are caused by occasional arrhythmia. However, for patients who may encounter more arrhythmia events, how would the method perform? On the other hand, if it still gives “high quality” image, the true arrhythmic cardiac motion will be suppressed, and are these images useful?

  • 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

    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?

    Given the strengths and weaknesses listed above, the current recommendation is 2.Although the images look nice and the metrics are high, I think there are more weaknesses and flaws than strengths in this paper, and I am not sure whether the weaknesses can be sufficiently addressed.

  • 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.

    Thank you for the clarification. The rebuttal is helpful to further understand the work, and most of my concerns are addressed. After reading the paper and inspecting the images again I think this method has potential to be applied towards enhancing cardiac cine images for subjects who may have occasional arrythmia.



Review #2

  • Please describe the contribution of the paper

    This paper addresses motion artifacts in cardiac cine MRI caused by arrhythmias. The authors propose a latent motion representation framework, LDE-MAR, that separates plausible physiological motion and artifact-related components in latent space using separate dictionaries. The training is multi-stage: the model first learns clean physiological motion dictionaries and then fine-tunes the noise dictionary. The method is evaluated on a private dataset of 398 subjects and on the public ACDC dataset, with both baseline comparisons and ablation studies. Based on the reported results, the framework appears to preserve structural patterns and reduce arrhythmia-induced artifacts relative to state-of-the-art image-domain 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.
    • Clear motivation and clinical relevance: The paper addresses a clinically meaningful problem in cardiac cine MRI, namely arrhythmia-related motion artifacts, and explains why these artifacts are more challenging than conventional noise and may require motion-aware modeling.
    • Technical innovation: The proposed LDE-MAR framework separates physiological motion and artifact-related deviations using distinct latent dictionaries. The decomposition of physiological motion into direction and region components is also an interesting design choice.
    • Comprehensive evaluation: The paper includes comparisons against six deep learning baselines as well as ablation studies and evaluates the method on both a 398-subject in-house dataset and the public ACDC dataset.
    • Good presentation: The paper is well organized and visually clear, with strong supporting figures/tables that make the experimental setup and results easy to interpret.
  • 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.
    • Missing ablation studies: Please clarify why the following two ablation experiments were not performed: 1.using only the region dictionary, and 2.using both the region and noise dictionaries.
    • Lack of limitations discussion: The manuscript would be stronger if it included a short discussion of the limitations of the proposed LDE-MAR framework, including experimental assumptions.
    • Limited temporal visualization of results: Since this is a cine MRI problem, showing only selected frames makes it harder to assess motion preservation over the full cardiac cycle. If feasible within space constraints, additional temporal visualization or supplemental cine sequences would strengthen the evidence that the method preserves beat-to-beat motion while removing arrhythmic artifacts.
    • Limited reproducibility detail: The paper would benefit from clearer reporting of implementation details, including optimizer, learning-rate schedule, batch size, training duration, hardware, and other key hyperparameters. Code is also not provided.
  • 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

    Address weaknesses:

    • Report the missing training details.
    • Add a brief paragraph discussing the study limitations.
    • Provide a clearer illustration of how temporal motion is preserved.
    • Include supplemental cine sequence visualizations.
    • Provide justification for why the ablation experiments were not performed.
  • 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?

    Overall, this paper presents a technically interesting and clinically relevant approach to arrhythmia-related motion artifact reduction in cine MRI. Its main strengths are the latent-space formulation and the breadth of experimental evaluation. My main concerns relate to ablation studies and the limited temporal visualization of motion preservation. These concerns do not necessarily undermine the central contribution, but they do affect how easy it is to assess, reproduce, and interpret the method.

  • 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.

    The rebuttal adequately addresses my initial concerns by clarifying the ablation ambiguities and acknowledging the missing details of reproducibility and limitations that need to be included in the final version. Since these issues appear resolvable through the promised revisions and do not undermine the paper’s main contribution, I maintain my recommendation to accept.



Review #3

  • Please describe the contribution of the paper

    -This paper proposeds a denoising moethod for arrhythmic cardiac cine-MRI to mitigate irregular rhythm-induced artifacts. -Three spasrse dictionaries are specifically designed for motion region, motion direction, and motion noise.

  • 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.
    • This paper addresses an important issue in imaging patient with irregular heart beating (arrhythmias). The clinical significance is high.
    • The denoising module is novel, with specific design on motion region, motion direction, and motion noise.
    • A two stage is proposed to learn normal pattern from self-supervised framework and learng arrhythmic cine from supervised training.
    • The proposed method is validated on both private dataset and public dataset.
  • 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.

    -Most of the comparison methods are based on CNN or diffusion model. The paper can be improved by comparing with other motion artifacts removal approaches using dictionary learning.

    • The evaluation is based on pixel-wised (PSNR, SSIM). The evaluation can be much stronger if involving perception level evaluation or downstream applications.
  • 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 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?

    The paper is well written and there are novelty in method design.

  • 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 paper is good for acceptance after the rebuttal.



Author Feedback

We thank the meta-reviewer and all reviewers for valuable comments. Special thanks to M1 for noting our work “addresses an important clinical problem”. We are encouraged by R1’s “novel application of LDE”; R2’s praise on “clear motivation and clinical relevance” and “comprehensive evaluation”; R3’s comments on “clinical significance” and “novel denoising module”. Main concerns cover 3 aspects: methodology, experimental design and evaluation scope & clinical relevance. Methodology: 1)reference image & two-step formulation[R1]; 2)“encode wt→r into wr→t” meaning/reversibility[R1]; 3)necessity of α and softmax/tanh[R1] A: 1)Reference image xr is a hypothesized canonical-phase cardiac image; its latent code zr is an implicit normalization anchor. We retain two-step formulation because E encodes each frame independently, avoiding a pairwise encoder on (xs,xt), S needs stable latent coordinate to define ηt, otherwise motion/artifacts entangle. Thus we adapt LIA’s latent decomposition to nonlinear cardiac motion by using D for signed direction/magnitude, R for activated latent regions and S for artifacts; Tabs.1-2 validate this. 2)Apologizes for misleading “encode wt→r into wr→t”. It is one-way dictionary recomposition, not inversion: wt→r=E(xt) feeds MLPs for Ut,Vt, αt=RUt, δt=DVt, wr→t=αt⊙δt. This extracts motion from image embedding without mixing xs/xt appearance features. 3)Both α/δ are necessary. LIA’s single dictionary jointly encodes region/direction/magnitude, causing entanglement; our preliminary tests showed linearized motion and lost physiological details. We use D&R with ⊙ applying δ in α-selected regions. R yields modest SSIM/PSNR gain but improves myocardial texture/local contraction. Unlike LIA’s linear LeakyReLU, softmax provides normalized nonnegative sparse region weights while tanh gives bounded signed direction/noise coefficients, their advantage on cardiac motion is validated by our pre-experiments. Experimental design: 1)reverse/iterative training[R1]; 2)R-only/R+S ablations[R2]; 3)dictionary-learning baselines[R3] A: 1)Reverse training may mismatch: S corrects artifact offsets relative to latent motion space formed by R&D. If trained first, S would absorb physiological motion as artifacts (cf. Model 3, Tab.2). Iterative training is inspiring; alternately updating motion/noise dictionaries refines each other, also supported by our experiments. 2)Since R complements D and cannot encode direction or represent motion independently, we used D-based ablations to show R’s added value and additionally conducted R-specific ablations. 3)We’ll add dictionary-learning baselines in journal version since its importance. Evaluation scope & clinical relevance: 1)perception/downstream metrics & temporal viz[R3,R2]; 2)true-motion preservation[R1]; 3)limitations[R2] A: 1)Our task is motivated by real clinical problems such as EF assessment. Preliminary results show better segmentation robustness and EF/volume consistency, enabling more accurate risk stratification. We’ll report these findings and full-cycle cine viz in the journal version. 2)Artifacts and true arrhythmic motion differ: artifacts arise from R-R-induced k-space misregistration/mixing and are nonphysical modulations, whereas premature/post-compensatory contraction or dyssynchrony are real biomechanical deformations constrained by myocardial fibers and electromechanical coupling. Our design separates them by inductive bias: R&D are orthogonal sparse bases for low-dim physiological motion, while S is unconstrained, capturing high-entropy residuals from k-space phase perturbation (Fig.3b). Fig.3a shows S does not remove biomechanically plausible motion variations. Clinical results suggest generalizability to more severe arrhythmias, though applicability needs further study. 3)We’ll add limitations on clinical validation/applicability in revision. Reproducibility[R2] A: We have clarified implementation details. Code uploaded to GitHub will be released upon 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.

    This paper proposes LDE-MAR, a latent-dictionary encoding framework for reducing arrhythmia-induced artifacts in cardiac cine MRI. The reviewers agree that it addresses an important clinical problem and shows promising results, but they also raise several concerns about the clarity of the method, the completeness of the ablations, the scope of the evaluation, and the strength of the evidence for temporal motion preservation. These concerns appear addressable through clearer explanation, better justification of design choices, and more careful discussion of limitations and applicability, rather than requiring a fundamentally different method. I therefore recommend invite for rebuttal. In the rebuttal, the authors should focus on clarifying the methodology, justifying the current experimental design, and strengthening the discussion of evaluation scope and clinical relevance.

  • 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 LDE-MAR, a latent-dictionary encoding framework for reducing arrhythmia-induced artifacts in cardiac cine MRI, aiming to separate physiological cardiac motion from artifact-related components through region, direction, and noise dictionaries.

    The reviewers recognized the clinical significance of arrhythmia-robust cine MRI, the novelty of using latent dictionaries to model motion and artifacts, and the comprehensive evaluation on both private and public datasets. They also appreciated the reported reconstruction quality and the ablation studies supporting the main components.

    The reviewers raised concerns about methodological clarity, missing ablations, temporal motion preservation, and reproducibility details. In the rebuttal, the authors clarified the role of the reference latent image, the one-way dictionary recomposition process, the necessity of separate region and direction dictionaries, and the rationale for the staged training strategy. They also addressed concerns about true arrhythmic motion preservation and acknowledged the need to include additional implementation details and limitations.

    Overall, the rebuttal sufficiently addressed the main reviewer concerns, and all reviewers supported acceptance after rebuttal. While further temporal visualization, downstream validation, and reproducibility details would strengthen the final version, these issues do not undermine the main contribution. Therefore, I recommend 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.

    The rebuttal addressed the main methodological concerns raised by R1, including the two-step formulation rationale, the directionality of dictionary recomposition, and the necessity of the α and softmax/tanh design choices. After rebuttal, all reviewers recommend acceptance. Congratulations!



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 makes a valuable exploration in addressing motion artifacts in cardiac MRI images, proposing a novel LDE-MAR method and demonstrating its effectiveness through experiments. Most of the previous concerns regarding methodology, experimental design and evaluation were clearly clarifed in the rebuttal, and all reviewers agreed to accept it.



back to top