List of Papers Browse by Subject Areas Author List
Abstract
Ultrafast echocardiography based on diverging wave (DW) imaging enables extremely high frame rates, unlocking breakthrough insights into cardiac mechanics. However, high-quality DW cardiac imaging typically relies on motion compensation (MoCo) compounding from multiple transmissions, which leads to a fundamental trade-off between image quality and frame rate. In this study, we introduce a novel SpatioTemporal COnstrained Diffusion Echocardiography (ST-CODE) for high-quality cardiac imaging at high frame rate using only three-angle steered DW transmissions. ST-CODE formulates reconstruction as a dynamic subspace-guided and temporally adaptive diffusion process that embeds structural and temporal constraints into reverse denoising. A dynamic subspace guidance module is designed to capture sequence-level anatomical structure and enforce structural consistency across frames. In parallel, a temporal correlation sampling strategy is introduced to regulate the diffusion trajectory to propagate inter-frame dependency and suppress stochastic variations during inference. Extensive experiments conducted on in-vivo cardiac data demonstrate that ST-CODE achieves image quality comparable to 32-angle MoCo-compounding reference and consistently outperforms existing reconstruction methods, leading to anatomically consistent and temporally coherent reconstructions without sacrificing ultrafast frame rates.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0123_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LuJin_Ultrafast_MICCAI2026,
author = { Lu, Jingfeng AND Millioz, Fabien AND Porée, Jonathan AND Provost, Jean AND Zhang, Yi},
title = { { Ultrafast Echocardiography Based on Dynamic Subspace-Guided and Temporal Adaptive Diffusion } },
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
This paper proposes ST-CODE, a diffusion-based framework for reconstructing ultrafast cardiac echocardiography from only three-angle diverging-wave transmissions. The main idea is to improve both anatomical consistency and temporal coherence by introducing Dynamic Subspace Guidance (DSG) and Temporal Correlation Sampling (TCS). Experiments on in-vivo cardiac data show clear improvements over several learning-based baselines and conventional CDWC reconstruction.
- 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 problem in ultrafast echocardiography, namely improving image quality without losing temporal resolution. The method is well motivated, and the combination of DSG and TCS is reasonably novel and well aligned with the goal of preserving spatial structure and temporal coherence. The quantitative results are strong: ST-CODE achieves the best performance across all reported metrics, and the ablation study supports the contribution of both modules. The paper is also generally easy to follow, with a clear framework figure and a logical organization.
- 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.
My main concern is that the experimental details are still insufficient for a top-tier conference paper. The manuscript does not clearly report the number of subjects, the train/validation/test split, or whether the evaluation is subject-independent. Reproducibility is also limited because several implementation details are missing, such as the exact denoiser design and how the DSG condition is injected. In addition, the comparison with the closest recent diffusion-based ultrasound methods is not fully convincing, and the paper does not provide runtime or efficiency analysis despite claiming ultrafast reconstruction. Finally, while the temporal metrics are promising, the paper would be stronger with a clearer link to downstream cardiac analysis or clinical relevance.
- 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?
This is a promising paper with a meaningful problem setting and strong initial results. I would encourage the authors to improve the experimental reporting, include more implementation details, strengthen the comparison with the most relevant recent methods, and provide runtime analysis. A clearer discussion of subject split, ethics approval, and downstream clinical relevance would also make the paper more convincing.
- 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 methods section details a diffusion-based framework with dynamic subspace guidance (DSG) on the Grassmann manifold for anatomical consistency and temporal correlation sampling (TCS) via NCC for inter-frame coherence, trained frame-wise but inferred sequence-wise on 3-angle DW data from Verasonics. It compares favourably to baselines like UNet, AUGAN, and DDPM in visual quality (Fig. 2), using CW-MoCo (32 angles) as reference. Impressive that ST-CODE matches 32-angle MoCo quality with only three angles-those in-vivo cardiac clips look smooth and artifact-free. Outperforms baselines consistently, which is solid. But when considering the database, the sample size is low. Another interesting thing is frame rates stay ultrafast without quality drops
- 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 methods section details a diffusion-based framework with dynamic subspace guidance (DSG) on the Grassmann manifold for anatomical consistency and temporal correlation sampling (TCS) via NCC for inter-frame coherence, trained frame-wise but inferred sequence-wise on 3-angle DW data from Verasonics. It compares favourably to baselines like UNet, AUGAN, and DDPM in visual quality (Fig. 2), using CW-MoCo (32 angles) as reference.
- 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.
Major weaknesses of this paper are:
Poor quality image at low angles. Small scale in-vivo database. Temporal correlation sampling (TCS) is sensitive to spatial variances and image quality.
- 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
1.What’s the frame rate achieved in practice (fps) on in-vivo data with ST-CODE vs. 32-angle reference? 2.Did you compare your output with any standard output to measure its clinical reference? 3.What are methodological advancement of ST-CODE model over AUGAN and DDPM model? 4.What are your opinion about why UNet and CCGR models could not perform well compared to other training based method? Is there model improvement or lack of advancement in previous models? 5.How many in-vivo datasets were used in this study?
- 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 methods section details a diffusion-based framework with dynamic subspace guidance (DSG) on the Grassmann manifold for anatomical consistency and temporal correlation sampling (TCS) via NCC for inter-frame coherence, trained frame-wise but inferred sequence-wise on 3-angle DW data from Verasonics. It compares favourably to baselines like UNet, AUGAN, and DDPM in visual quality (Fig. 2), using CW-MoCo (32 angles) as reference. Poor quality image at low angles. Small scale in-vivo database. Temporal correlation sampling (TCS) is sensitive to spatial variances and image quality.
- 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 proposes ST-CODE, a diffusion-based framework for reconstructing high-quality ultrafast echocardiographic sequences from only three-angle steered diverging wave (DW) transmissions. The key contributions are twofold: (1) a Dynamic Subspace Guidance (DSG) module that models sequence-level anatomical structure on the Grassmann manifold and injects it as a structural condition into the reverse diffusion process, and (2) a Temporal Correlation Sampling (TCS) strategy that introduces spatially adaptive inter-frame correlation at the noise initialization stage during inference, propagating temporal coherence through the reverse diffusion trajectory without modifying the training procedure. The framework is evaluated on in-vivo cardiac data against several baselines (U-Net, AUGAN, CCGR, DDPM), using both spatial quality metrics (PSNR, SSIM, NMI) and temporal coherence metrics (MEPE), with 32-angle coherent-weighted motion-compensated compounding as reference.
- 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 addresses a genuine and clinically important trade-off in ultrafast echocardiography: achieving high image quality while maintaining high frame rates. Reducing the number of required DW transmissions from 32 to 3 while preserving image quality is a meaningful goal with direct clinical relevance for cardiac motion analysis. 2.The DSG module is a thoughtful contribution. By modeling the evolving cardiac anatomy as a trajectory on the Grassmann manifold, the method naturally captures basis-invariant subspace representations. The tangent-space update mechanism and projection-based structural conditioning are mathematically well-grounded and provide a principled way to inject spatiotemporal consistency into the diffusion process. 3.The TCS module is elegant in that it introduces inter-frame dependency without modifying the training procedure. The use of NCC-derived correlation maps to structure the initial noise and the first-order linear approximation argument for covariance propagation are well-reasoned. This design choice makes the method flexible and easy to integrate with existing diffusion training pipelines. 4.The paper evaluates both spatial quality and temporal coherence, the latter often neglected in ultrasound reconstruction studies. The inclusion of EPE and MEPE metrics, along with the cardiac-cycle-resolved MEPE curves, provides meaningful insight into temporal performance. The ablation study clearly demonstrates the individual and combined contributions of DSG and TCS.
- 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 evaluation is conducted on a single probe configuration (ATL P4-2), a single imaging view (apical four-chamber), and only healthy subjects. The paper does not report the number of subjects, the number of sequences, or the train/test split. Without this information, it is difficult to assess the statistical reliability of the results or the generalizability of the method to pathological hearts, where anatomical variability and motion patterns differ substantially. The authors should clarify the dataset composition and ideally evaluate on at least one pathological cohort. 2.The 32-angle CW-MoCo reconstruction is used as the ground truth, but this itself is an approximation. The paper acknowledges interleaving the 32-angle and 3-angle acquisitions to minimize temporal offsets, but does not quantify the residual temporal misalignment or its potential impact on the metrics. 3.The subspace dimension in the DSG module is a critical hyperparameter, yet the paper does not discuss how it is selected or provide sensitivity analysis. 4.Diffusion models are notoriously expensive at inference time. The paper mentions using 100 accelerated inference steps (reduced from T=1000), but provides no runtime comparisons against baselines. For a method motivated by ultrafast imaging, the computational cost of reconstruction is directly relevant. If reconstruction takes orders of magnitude longer than the acquisition itself, the practical utility is diminished. 5.The paper presents only successful cases. It would be valuable to show examples where the method struggles, e.g., frames with rapid valve motion, high inter-frame displacement, or severe acoustic shadowing.
- 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.
(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?
The paper addresses a clinically relevant problem with a novel and technically sound approach. The Grassmann-manifold-based dynamic subspace guidance is a principled and original contribution that goes well beyond standard conditioning strategies for diffusion models. The inference-time temporal correlation sampling is elegant, it introduces inter-frame coherence without modifying training, which is both practical and theoretically well-motivated. The quantitative improvements are large and consistent (+2.16 dB PSNR, SSIM 0.93 vs. 0.87, MEPE 0.09 vs. 0.11 mm), and the ablation study clearly demonstrates the value of each component. While the dataset description is insufficiently detailed (number of subjects, train/test split), computational cost is not reported, and some assumptions remain unvalidated (subspace dimension, first-order approximation), these are common shortcomings in the ultrasound imaging literature and are addressable. The core methodological contribution is strong enough to merit acceptance.
- 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
We sincerely thank the reviewers for their positive and constructive feedback.
Regarding the dataset description, we clarify that the in-vivo cardiac dataset consists of subject-independent sequences, and the data splits were performed in a subject-independent manner. The details regarding the data scale and independence have been specified in the revised version for clarity. We further clarify that the current study represents a preliminary evaluation under a controlled experimental setting with a single acquisition setup and limited to healthy subjects. This aspect has been clarified in the revised version.
Regarding the runtime and the “ultrafast” claim, we clarify that “ultrafast” in this work refers specifically to the data acquisition stage, where only three-angle transmissions are required compared to conventional multi-angle coherent compounding, thereby significantly reducing acquisition time. The reconstruction runtime corresponds to the post-acquisition processing stage rather than the acquisition speed, and we acknowledge that diffusion-based inference introduces high computational cost. We have included this clarification on inference efficiency in the camera-ready version.
Regarding the implementation details, we have specified key implementation details in the revised version.
We thank the reviewers again for their valuable feedback, which has helped further improve the clarity and presentation of the paper.
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 paper proposes ST-CODE, a diffusion-based framework for reconstructing high-quality ultrafast echocardiographic sequences from only three-angle diverging-wave transmissions, leveraging a Dynamic Subspace Guidance (DSG) module to enforce anatomical consistency and a Temporal Correlation Sampling (TCS) strategy to enhance temporal coherence. All three reviewers agree that the work addresses an important and clinically relevant problem, and that the proposed method is technically sound with meaningful novelty, particularly in its integration of subspace modeling and inference-time temporal coupling. The experimental results are consistently strong across spatial and temporal metrics, and the ablation studies support the contributions of the proposed components. However, reviewers also raise several common concerns, including insufficient reporting of dataset details (e.g., number of subjects, train/test splits, and independence), lack of runtime or efficiency analysis despite the “ultrafast” claim, limited evaluation scope (single acquisition setting and healthy subjects only), and missing implementation details affecting reproducibility. These issues do not undermine the core contribution.