List of Papers Browse by Subject Areas Author List
Abstract
In quantitative echocardiography, detecting end-diastole (ED) and end-systole (ES) frames is crucial for downstream clinical tasks. However, ECG synchronization and frame-level annotations are often unavailable in portable ultrasound devices. While recent self-supervised methods extract cardiac phase cues from latent motion trajectories, they suffer from view-specific reconstruction objectives, leading to unstable estimates across different views. To address this, we propose CardiacPULSE, a Cardiac Phase detection framework that achieves Ultralight inference by Localizing Signal Extrema from the inherent periodicity of cardiac motion. Our method constructs a physics-guided spatial prior via per-pixel temporal Fourier analysis to isolate rhythmically active regions. To enable efficient edge deployment, we train a lightweight auxiliary network to predict this Fourier-derived motion mask in a single forward pass under purely self-supervised losses. Our approach achieves a 34.0 ms mean absolute error on the EchoNet-Dynamic dataset, outperforming CardiacPhase by 31%.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1789_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026/supp/1789_supp.zip
Link to the Code Repository
https://github.com/xmed-lab/CardiacPulse
Link to the Dataset(s)
EchoNet-dynamic dataset: https://echonet.github.io/dynamic/
CAMEO dataset: https://ieee-dataport.org/documents/descriptor-cardiac-multi-view-echo-dataset-cameo
BibTex
@InProceedings{GuoJia_CardiacPULSE_MICCAI2026,
author = { Guo, Jiarong AND Yang, Jiewen AND Du, Yao AND Qin, Yi AND Li, Xiaomeng},
title = { { CardiacPULSE: Frequency-Aware Self-supervised Cardiac Phase Detection in Echocardiography } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16880},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The proposed method involves constructing a physics-guided spatial prior via per-pixel temporal Fourier analysis. The dominant cardiac frequency is identified from the spectrum and used to generate a motion-energy map. Next, a lightweight auxiliary network is trained to approximate the Fourier-derived prior by using only spectral concentration and phase coherence losses. During the inference stage, the auxiliary network predicts the motion mask in a single forward pass without view-specific reconstruction. The entire algorithm can operate in real time on standard hardware.
- 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 presents a new algorithm for phase detection in echocardiography that is self-supervised and uses a small number of trainable parameters.
- The algorithm performs equally well regardless of the specific ultrasound view.
- The results reported on public benchmarks are good.
- 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.
There are no significant weaknesses identified.
- 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.
(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 proposed algorithm is new and could be used in portable ultrasound devices.
- 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.
I stand by my initial evaluation of this submission, particularly with respect to its originality.
Review #2
- Please describe the contribution of the paper
This work proposes an annotation-free method that localizes key cardiac phase frames by learning the inherent periodic signals in cardiac videos. Compared with previous studies mainly focusing on the four-chamber view, this work further extends the framework to more additional cardiac views. The design also takes practical deployment into account by introducing a lightweight auxiliary network to learn periodic priors, which gives the method promising potential for real-world clinical application.
- 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.
A notable strength of this work is that the proposed method is able to learn cardiac periodicity without relying on any manual annotations. The design also takes practical motion characteristics into account. For instance, when extrema alone cannot reliably distinguish ES from ED, the method further introduces phase-specific frequency cues to separate the two key frames. Moreover, the use of a spatial mask to filter out irrelevant background noise while preserving rhythmically active regions is a thoughtful and effective design choice.
- 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 inference stage appears to rely on a full input video sequence rather than a short partial clip, since motion-energy analysis and harmonic reconstruction are both performed on the extracted temporal signal. It would be helpful if the authors could clarify whether the method can operate reliably on partial clips containing only limited cardiac cycles, and how sensitive it is to input length.
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This work is quite innovative, and many of the proposed design choices are well motivated by the specific characteristics of the cardiac phase detection task. The framework also explicitly considers lightweight deployment. The experimental results on two datasets and multiple cardiac views further support its effectiveness and highlight its strong potential for practical real-world application.
- 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
echocardiography ED/ES phase detection through unsupervised spectral learner, further distilled to function in real-time potentially on edge devices. Extensive testing shows excellent performance relative to competing methods at much less complexity.
- 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.
efficient and accurate phase detection in echocardiography leveraging Fourier analysis per pixel and a lightweight CNN learned on prior cardiac motion consequences in the frequency domain.
- 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.
This reader does not understand the clinical import: an edge device can now quickly label ED/ES candidates, but no downstream processing would take place on the edge device anyway. Also, the dominant cardiac frequency may be a weak indicator in pathologic conditions, or as noted non-ideal acquisition.
- 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
The echonet dataset also includes numerous non-protocol ED/ES labels such as ES preceding ED or ED/ES labeled on different beats. Plus the protocol itself (and downstream clinical indices) calls for finding the frame with the largest LV cavity; other than computation, explicit cavity estimates aren’t something one would wish to do without. Additionally, the validation is that a proposed candidate is nearby the relevant expert labeled frame, but there’s also no penalty for the number of candidates. Lastly, not clear how the supervised (and per-frame cavity-generating) competitors do not get just as close using local max volume.
- 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?
Interesting conceptualization of cardiac motion in echo and with clearly encouraging results, but the clinical import of the target application is unclear.
- 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 authors thoughtfully address concerns noted in the initial review. This reader appreciates that the authors will expound on noted issues in the camera-ready version.
Author Feedback
We thank reviewers and AC. R1 raised no major concerns. R2: input length/partial clips. R3: clinical value, weak/pathological cues, EchoNet label noise, cavity comparison, sparse candidates. AC: clinical utility, full-video reliance, pathological cues. A. Clinical role and cavity-based methods (AC, R3). CardiacPULSE is a lightweight temporal-indexing front-end, not an anatomical measurement method. It returns sparse ED/ES candidates to initialize LV-volume/EF/SV pipelines, reduce search from all frames to a few frames, and provide acquisition QC when ECG/cloud are unavailable. Even if final quantification is off-device, it can verify a usable cycle and pass the right frames to later analysis, useful for handheld/POCUS and emergency/rural workflows. We agree cavity extrema are clinically meaningful. Our claim is not to avoid cavity estimation, but to perform a different stage. A local maximum of predicted cavity volume is not available for all baselines: R3D-50 is an event classifier, not a volume estimator; MAEF uses frame-wise LV segmentation, where contour noise can make extrema unstable; and UVT is optimized for EF/volume rather than sparse ED/ES indexing. Stronger supervision does not remove per-frame speckle, contrast, or contour noise. CardiacPULSE instead imposes periodicity and harmonic consistency, trading per-frame anatomical precision for stable phase timing. B. Weak/pathological signals and label noise (AC, R3). We do not claim pathology-specific robustness without stratified labels. Our narrower claim is that the method does not rely on one raw cavity-amplitude cue: the mask is estimated over the whole frame, the band covers 0.5-4.0 Hz (30-240 bpm), and motion energy is mask-normalized. Ablations support this redundancy: Mcnn alone gives 34.8 ms vs. 34.0 ms full, Mfft alone 42.2 ms, and Fourier-only candidates 70.8 ms. EchoNet-Dynamic spans a wide EF range, including severely reduced function, so the test set is not restricted to ideal cases; severe arrhythmia, view drift, probe motion, or non-cardiac artifacts remain limitations. For EchoNet label noise, including ES preceding ED or ED/ES marked on different beats, we follow the established EchoNet/CardiacPhase protocol. Since our output is a sparse candidate set across visible cycles and MAE uses the closest prediction, evaluation does not require a fixed ED-before-ES order within one beat; it asks whether a plausible ED/ES candidate exists near the expert label. CAMEO over nine views reduces concern that the gains are EchoNet-specific. C. Input length and partial clips (AC, R2). Training uses 64-frame clips downsampled by 2, about 2.5 s and usually 2-3 cycles, for stable self-supervised spectra. At inference, EchoNet numbers use full benchmark videos at native rate for fair comparison. The method is window-based: reliable frequency estimation and harmonic anchoring need at least one stable cycle, preferably two. With very short clips, 1/T frequency resolution degrades, and rolling-window or multi-clip aggregation would be needed. Thus CardiacPULSE is not frame-instantaneous, but can operate on short cine windows. D. Sparse candidates (R3). The predictions are sparse, deduplicated candidate sets, not dense per-frame labels. Sec. 2.3 fuses motion-energy extrema with ED/ES harmonic anchors and removes near duplicates, so candidates are tied to cycle extrema rather than flooding the video. By construction, phase-specific anchors provide one ED and one ES candidate per detected cycle, followed by deduplication; the motion-energy pool is merged rather than sampled densely, so count scales with visible cycles, not video length. We use CardiacPhase closest-prediction MAE only for comparability. In the same submitted runs, our count after fusion/deduplication is close to CardiacPhase, so the gain is not from proposing more frames; counting uses existing outputs, not a new run. We will make these clarifications in the camera-ready.
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 a self-supervised, frequency-aware framework for cardiac phase detection in echocardiography. Reviewers appreciate the annotation-free design and view-agnostic performance. They also see clear potential for deployment on edge devices. Concerns include reliance on full video sequences, limited clarity on clinical utility, and weaker cues in pathological cases. Given these mixed assessments, the paper is invited to rebuttal for the authors to address the concerns.
- 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 effectively clarified the method’s role as a lightweight temporal-indexing front-end and addressed concerns around input length sensitivity and sparse candidate generation, leading all three reviewers to accept post-rebuttal.
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 paper proposes a self-supervised, frequency-aware framework for cardiac phase detection in echocardiography. Reviewers recognized the novelty of using frequency-domain cardiac motion cues and the potential for hand-held device. The concerns of R3 are adequately addressed. After rebuttal, all responding 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.
The paper presents a self-supervised framework for cardiac phase detection in echocardiography. Reviewers highlighted the annotation-free design, frequency-aware formulation, and lightweight inference as strengths. Concerns focused on clinical utility, reliance on video length, and performance in challenging cases. The rebuttal clarified these points, and the updated reviewer feedback supports acceptance.