List of Papers Browse by Subject Areas Author List
Abstract
Myocardial infarction (MI) remains a leading cause of mortality worldwide. Echocardiography (Echo) is a widely available modality for MI assessment, where regional wall motion abnormality is a key indicator. Prior learning based methods for myocardial motion analysis often use handcrafted descriptors or densely supervised estimation, but the need for extensive annotation limits applicability. Foundation models have recently improved vision-based Echo analysis; however, most methods operate on single views and segment-level localization remains unreliable under view-dependent ambiguity, especially in apical views. To address this, we propose MCF-Net, a novel motion-guided multi-view fusion framework that fuses myocardial motion cues with foundation model representations to localize infarction. Visual features are extracted using EchoPrime, a pretrained Echo foundation model shared across dual views. Cardiac motion is modeled with extremely sparse supervision: a single annotated template frame is transferred across videos to initialize point tracking, avoiding dense labels. Motion-derived segment-aware soft masks provide coarse spatial priors that selectively enhance features for challenging myocardial segments. A motion-conditioned fusion mechanism then integrates motion and vision across views, refining predictions without overriding strong appearance cues. On segment-level MI localization, MCF-Net achieves 72.4% F1 and 84.9% accuracy, outperforming state-of-the-art motion-only, vision-only, and fusion baselines. Code is available at https://github.com/GYang1619/MCF-Net-MICCAI2026.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3435_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/GYang1619/MCF-Net-MICCAI2026
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YanGua_MotionConditioned_MICCAI2026,
author = { Yang, Guang AND Xu, Wentian AND Wang, Siyu AND Raman, Betty AND Li, Lei AND Grau, Vicente},
title = { { Motion-Conditioned Multi-view Fusion for Myocardial Infarction Localization from Echocardiography } },
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 addresses a clinically relevant task and presents a fairly complete methodological framework. The proposed framework utilizes a pretrained foundation model to extract rich visual representations and introduces Motion Guided Soft Refinement to focus feature extraction on infarction-relevant regions. Motion Conditioned Fusion further modulates appearance features using view-specific motion and enables cross-view interaction between motion dynamics and visual semantics, thereby improving fine-grained localization.
- 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.Unlike methods that only determine whether MI is present, this work tackles segment-level localization, which is more consistent with the clinical logic and is also more challenging than coarse-grained classification. The problem setting is meaningful and clinically relevant, as it goes beyond simple classification to a finer-grained task. 2.The authors emphasize that the entire dataset requires only a single manually annotated template frame, which is then transferred to all videos through template matching and point tracking. This design indeed reduces the burden of dense annotation.
- 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 sparse motion initialization relies on a single annotated template, NCC-based matching, and a fallback-to-prior strategy. This may introduce structural bias or propagate initialization errors, especially when transferring across A4C and A2C views. 2.According to the sensitivity curves on page 8, the soft refinement strength λ has a narrow effective range, and performance drops when masking becomes too strong. This suggests that the refinement mechanism may be heuristic and not inherently robust. 3.The dataset is relatively small, no external validation is provided, and no subgroup analysis is conducted. For a localization task, this is a substantial limitation. 4.The paper verifies that MCF and MSR help, but does not sufficiently dissect why they help, under what conditions they fail, or how robust the sparse motion pipeline is.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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 motivation is clear, but the innovation is limited. 2.The model design is reasonable, but the experiment is not enough.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The authors introduce MCF-Net, a framework for localizing myocardial infarction in echocardiography using two different cardiac views. It combines pretrained visual features from EchoPrime with motion information obtained from tracking a small number of points. A key idea is to rely on very minimal motion supervision—starting from just a single manually annotated frame for the entire dataset—to estimate motion patterns of the heart. These motion cues are used to generate soft masks that help refine the visual input and fuse information across the two views (A2C and A4C).
- 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 topic is very interesting, and the paper is well written. It also presents a clear methodological contribution, with two key components: motion-guided soft refinement and motion-conditioned cross-view fusion.
- 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 using only a single annotated template frame for the entire dataset is an interesting idea, the approach relies on transferring one A4C template to all videos—including A2C views—using patch matching, with a fallback to atlas coordinates when the matching confidence is low. This raises concerns about how robust the method is, especially in cases with severe pathology or poor image quality. In addition, the paper does not explain whether the method can handle cases where only one view is available. There is also no statistical significance analysis for the quantitative results, and the authors do not discuss the limitations of their approach or potential failure cases.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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, technically novel, and the topic is clinically important and interesting; however, it lacks statistical significance analysis and discussion of failure cases.
- Reviewer confidence
Somewhat confident (2)
- [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
Within the context of myocardial infarction localization in A2C/A4C multi-view echocardiography, authors propose (action) co-leveraging existing motion tracking and EchoPrime echo/text foundation model embedding for improved 12-segment MI classification. CoTracker3 with limited supervision informs both a localized video-brightening (MSR) before the foundation model and modulation of the foundation model output after (MCF) before multi-head attention. Experimental results show improved MI localization over baseline and competing methods, and an ablation study provide additional evidence.
- 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.
Their model shows improved (and sota) results over the baseline EchoPrime-only model, including repeated performance testing against several baselines and competing methods, and ablation study on the specific MSR/MCF additions. An informed approach to injecting prior knowledge (specific motion tracking of the myo wall) into the otherwise fixed cross-purpose foundation model for MI detection.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
the improvement is modest (though persistent), unclear whether MSR (and all of 2.1) adds anything that some simpler contrast enhancement would not.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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
how different are naive fusion and EchoPrime+MCF to have such different performance, is it just the FiLM conditioning?|| The MSR colormap in Fig. 1 begs the question whether simply brightening the LV (which can be localized blindly on such curated datasets) produces better EchoPrime output, and if so by how much; it seems likely that MCF with its additional model complexity is still worth doing, but it would be disturbing if the foundation model output could be so vastly improved with simple image processing. (Not impugning the validity of Fig 3 and discussion)|| The method seems to require temporal and spatial correspondence across views; is temporal provided implicitly by the ED-ES videos? And when the initial manual fiducials are propagated to the other view through NCC, are you assuming a point correspondence, because A2C/A4C are intended to be on orthogonal planes? Also, the fiducials appear scattered in inner and outer walls, but the paper does not mention a particular protocol. Lastly, patch matching through NCC has plenty of priors to cite in registration, leaving space for discussion on more advanced concepts.|| Though [6] is cited, it should be again in the first mention of CoTracker in text, at the end of 2.1.|| Good non-EchoPrime models are tested, but Gomez 10.1016/j.medengphy.2025.104400 (lesser journal than [5]) reports excellent results and it is more recent.
- 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?
Improved SOTA results with reasonable testing against baseline and competing methods, some novelty of application to MI detection of the MCF main concept (modulate FM output with complementary info, cross attention).
- 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.
This reader appreciates the authors’ responses to the issues raised.
Author Feedback
We thank reviewers/AC for the assessment. We are encouraged that the paper was recognized as clinically relevant MI localization (R1), well written and technically novel (R2), and supported by SOTA results and baselines (R3). The AC noted a technically sound approach with smart motion integration. Our key clarification is that sparse motion is not a fragile pipeline, but a low-supervision physiological prior, quality controlled and injected softly into EchoPrime.
1) Motion robustness and A4C/A2C transfer (R1,R2,R3,AC). The concern is whether one-template NCC initialization propagates errors; our response is that A4C was chosen to target an identified failure mode. Per-view analysis showed EchoPrime is stronger on A2C than A4C: AUROC/PR-AUC 0.880/0.745 vs. 0.796/0.644, with SEG3/SEG6 near-random (AUROC 0.683/0.635). These results, omitted due to space, identify A4C as the main failure mode. Thus, single annotation in A4C makes within-view transfer anatomically consistent while targeting the harder view. For A2C, robustness comes from localized NCC in a 41x41 window (<=20 px drift), fallback to atlas prior (Eq.2), and simpler anatomy. Each view is initialized/tracked independently; no physical A2C-A4C correspondence is assumed. QC removes unreliable trajectories, and substantial visual inspection suggested retained points are mostly on/near the LV wall. We will clarify this protocol and failure modes.
2) MSR is not heuristic and not simple brightening (R1,R3). The concern is whether MSR is generic LV enhancement; our clarification is that it is a dynamic, trajectory-derived guidance for A4C attention failure. EchoPrime can be distracted by non-LV chamber structures in difficult A4C cases, so MSR uses a per-frame Gaussian mask from tracked myocardial points and boundedly modulates the clip as in Eq.3.Thus MSR is trajectory-dependent, time-varying, wall-localized and bounded, while preserving raw Echo context. This also explains Fig.3: EchoPrime is trained on natural Echo clips, so strong masking/augmentation or excessive lambda distorts its expected input distribution; moderate lambda provides a weak anatomical cue.
3) MCF is beyond naive fusion/FiLM (R1,R3). The concern is how naive fusion and MCF differ. Naive fusion mixes heterogeneous spaces: 512-d EchoPrime embeddings and 128-d CoTracker descriptors. Direct fusion becomes EchoPrime-dominated or harmful, as EchoPrimeNF drops to F1 59.3 vs. EchoPrime 64.3 (Table 1). MCF maps motion to view-specific conditioning, reweights visual embeddings by scale/shift, then uses joint motion as query Q against visual tokens as K,V in cross-view attention (Eq.4/Sec.2.3). Motion neither replaces vision nor merely concatenates; it asks which visual features, for this patient/view, are relevant. This explains Table 2: MCF alone reaches F1 70.3 and full model 72.4. 4) Validation, statistics and single-view use (R1,R2,AC). We agree external validation and subgroup/failure analysis would strengthen generalization. We are limited by available public data: to our knowledge, HMC-QU is the only public benchmark with paired A2C/A4C Echo and segment-wise MI labels. Its size supports methodological validation, but clinical deployment requires larger external validation. We repeated experiments over 3 seeds with mean+/-std and observed consistent gains over motion-only, vision-only and fusion baselines (Table 1), with coherent ablations (Table 2). Paired patient-level bootstrap confirms F1 improvement over EchoPrime: Delta=0.08, 95% CI=[0.01,0.15], p<0.05.For single-view use, Fig.1 has self-contained branches; given 1 view, the corresponding branch produces its 6 AHA-segment predictions. The main contribution remains dual-view localization covering 12/17 AHA segments.
5) Minor points/ethics. We will cite CoTracker at first mention, add NCC/registration discussion, and cite Gomez et al. as related cascaded LV-segmentation/feature-classification work. We will clarify use of public de-identified data.
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 presents a technically sound approach to a challenging clinical problem, with a smart integration of motion cues into a strong visual backbone. However, it is held back by a lack of robustness analysis for its core motion pipeline and a validation strategy that is not deep or rigorous enough.
- 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 MCF-Net, a sparse motion-guided multi-view fusion framework for myocardial infarction (MI) localization in echocardiography, integrating pre-trained visual models with cardiac motion cues. The main concerns include the motion robustness, the effectiveness of the MSR and MCF, and statistical significance of the results. The authors clarified these concerns well and acknowledged the limitation of the single dataset. Given the clinical importance, the whole methodology design, and the promising results, I would like to recommend the work to be accepted.
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 authors proposed MCF-Net, a Motion Conditioned Fusion Network for segment-level myocardial infarction localization from dual-view echocardiography. MCF-Net utilizes a pretrained foundation model to extract rich visual representations and introduces Motion Guided Soft Refinement to focus feature extraction on infarction-relevant regions. Motion Conditioned Fusion modulates appearance features using view-specific motion and enables cross-view interaction between motion dynamics and visual semantics, improving fine-grained localization.
The method is technically solid. Though the novelty might be limited as an integration to the current FM, the motion vector feature injection is clearly motivated. The validation is further strengthened through the authors’ response, though the rigorousness is still somewhat limited due to the restriction of rebuttal. I am giving a weak acceptance considering all these.
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.
The motion-guided fusion with sparse tracking is a reasonable and clinically motivated contribution, and the rebuttal adequately addressed the main concerns arouns robustness and the transfer strategy