Pith. sign in

REVIEW 5 major objections 6 minor 31 references

Generating Realistic Multi-Beat ECG Signals

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A three-layer framework generates multi-minute synthetic ECG sequences by diffusing single beats, synthesizing a per-beat feature scaffold, and stitching beats onto it via weighted feature matching; the paper claims the result preserves…

desk verdict A genuinely new three-layer stitching approach for long ECG, but the claim that it preserves essential diagnostic characteristics is under-supported because the matcher only sees hand-curated beat features. read the letter →

arxiv 2505.18189 v1 pith:CX6TQJQM submitted 2025-05-19 eess.SP cs.LG

Paul Pöhl , Viktor Schlegel , Hao Li , Anil Bharath This is my paper · ORCID
classification eess.SPcs.LG
keywords ECGsynthesisdiffusionmodelslong-formtimeseriesarrhythmiaclassificationR-Rintervalsfeature-guidedmatchingTSTRevaluationsyntheticmedicaldata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Generative models for electrocardiograms tend to produce convincing single beats but fall apart over the multi-beat stretches that clinicians actually read. This paper tries to fix that by splitting the problem into three layers: a diffusion model generates a large store of high-fidelity single beats, a separate multivariate model generates long sequences of per-beat features (wave locations, amplitudes, intervals, R-R timing), and a matcher stitches the best-fitting beats onto that feature scaffold. The paper claims that the assembled long-form ECGs preserve both single-beat morphology and inter-beat dependencies, and that classifiers trained on them outperform classifiers trained on end-to-end diffusion baselines while approaching classifiers trained on real data. The paper argues this makes synthetic ECGs usable at the minute scale for privacy-preserving data augmentation and arrhythmia model development.

What carries the argument

The machinery is the per-beat feature vector and the weighted nearest-neighbour matcher of Eq. (1): for each target feature vector $f_n$ generated by the continuous feature model, the matcher selects the beat $b_m$ from a store of diffusion-generated beats (10,000 normal and 10,000 abnormal) that minimizes $\sum_k w_k (f_{n,k} - g_k(b_m))^2$, where $g_k(b_m)$ reads the same descriptor off a candidate beat. The feature vector encodes P/Q/R/S/T wave locations and amplitudes plus derived intervals and R-R timing. The matcher's output is then aligned at the R-peak and stitched with small boundary smoothing to form the long signal. This object carries the argument because whatever diagnostic information the final ECG contains must be present either in the diffusion beat store or in the feature scaffold; assembly only recombines those two sources.

What would settle it

Use the pipeline to generate long synthetic ECGs from a dataset annotated for ST-segment changes or QT prolongation, then measure whether ST-segment deviation and QT intervals in the synthetic signals reproduce the real distributions. ST-segment and QT dynamics are not in the paper's feature list, so their systematic absence or collapse would show that 'essential diagnostic characteristics' are not preserved.

Watch

Extended reading notes

Core claim

The central claim is that realistic long-form ECG does not require a long-context generative model. By separating local waveform fidelity from inter-beat dynamics, the paper argues, each part can be generated with a simpler model and recombined without losing diagnostic content. The evidence is the assembled signals' beat-level morphology metrics, preserved R-R interval and amplitude distributions, and train-on-synthetic-test-on-real (TSTR) arrhythmia classification: on the supraventricular database, classifiers trained on the synthetic long ECGs achieve overall accuracy up to about 0.948, close to the 0.933 of classifiers trained on real data, while end-to-end diffusion baselines drop to 0.283-0.764; the paper presents this as evidence that 'essential diagnostic characteristics' survive assembly. The abnormal-class numbers are weaker than the normal-class numbers, yet the paper's stated conclusion is that the framework outperforms end-to-end generation and approaches real-data performance.

Load-bearing premise

The load-bearing premise is that the hand-picked feature set—P/Q/R/S/T wave locations and amplitudes plus derived intervals—captures the clinically relevant information in an ECG, so a long signal assembled to match those features has the same diagnostic content as a real recording.

Editorial extensions

If this is right

  • Long synthetic ECG can be produced at minute scale without training a long-context generative model, removing the memory and compute bottleneck that limits end-to-end diffusion baselines.
  • Because classifiers trained on the synthetic long ECGs transfer to real data in the paper's evaluations, synthetic data could augment small or imbalanced arrhythmia datasets without sharing patient records.
  • The modular design means the feature scaffold can be extended with additional descriptors (for example, QT interval or ST-segment measures) to adapt the pipeline to diagnostic tasks beyond the normal/abnormal split tested here.
  • The same beat-store-and-stitch assembly could be applied to other quasi-periodic biosignals, such as pulse oximetry or respiration, whenever a single-cycle generator and a feature scaffold can be defined.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct consequence the paper does not test: if a clinically relevant marker is absent from the feature scaffold, the final long-form ECG cannot contain it even when single beats look perfect; an obvious check is generating ST-elevation or QT-prolongation cases and measuring those intervals in the output.
  • The reported overall accuracies hide a weaker abnormal-class signal (for example, abnormal-class F1 of 0.479 versus 0.778 on the supraventricular database), so a practical extension would be oversampling abnormal beats in the store or reweighting the matching cost toward rare morphologies.
  • The matching cost in Eq. (1) treats each beat independently; adding a transition penalty between consecutive beats could reduce stitching artifacts and improve rhythm-level consistency beyond what R-peak alignment alone provides.
  • One could test the framework's claim directly on a held-out diagnosis whose markers are not in the feature set: if performance matches real-data performance anyway, the feature set is more complete than it appears, and if not, the missing-feature diagnosis is the boundary of the method.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes a three-layer framework for generating long-form ECG signals: (1) a diffusion model trained on single-beat segments from the MIT-BIH Supraventricular Arrhythmia Database produces a store of synthetic beats; (2) a multivariate model trained on extracted beat-level features (P, Q, R, S, T locations and amplitudes plus derived intervals) synthesizes feature trajectories; and (3) a matching algorithm selects beats from the store by minimizing a weighted squared distance to the target features, then stitches them into a continuous ECG via R-peak alignment and smoothing. Evaluation includes beat-level morphology metrics, feature-distribution distances, Train-on-Synthetic-Test-on-Real (TSTR) arrhythmia classification on single beats and assembled five-beat signals, and comparison against two end-to-end baselines (BRIDGE and TimeVQVAE), with an additional MIT-BIH Arrhythmia Database experiment in the appendix. The paper claims generation of unprecedented multi-minute ECG sequences while preserving essential diagnostic characteristics.

Significance. If the claims are validated, the paper offers a practical alternative to long-sequence diffusion models by decomposing ECG synthesis into single-beat generation, inter-beat feature synthesis, and feature-guided assembly. The multi-tiered evaluation and the inclusion of two end-to-end baselines are appropriate first steps, and the beat-level morphology results are encouraging. However, the central claim that the assembled signals preserve 'essential diagnostic characteristics' is not currently established: the feature vector used for matching is not shown to be sufficient for clinically relevant content, the long-form evidence is limited to a single 100-beat example, and the classification results lack error bars and subject-disjoint validation.

major comments (5)
  1. [§III-B(a), §III-C(a), Eq. (1)] The matching objective in Eq. (1) selects beats by minimizing weighted squared deviations in the hand-curated feature set only (P, Q, R, S, T peak locations and amplitudes plus derived intervals). Any diagnostic content outside this vector—ST-segment deviation, QT-interval dynamics, T-wave morphology, baseline wander, or noise texture—has no mechanism to be preserved, because the matcher can at best reconstruct the features it is given. The paper does not test whether this feature set is complete for any clinically meaningful task. Table III is consistent with missing diagnostic information: the synthetic feature-based long ECGs achieve abnormal-class F1 of 0.479 versus 0.778 on original data, despite near-perfect normal-class F1. I recommend adding explicit evaluation of unmodeled features on the assembled signals (e.g., ST-segment level, QT interval, T-wave area/morphology) or using a downstream task that is not defined by the same feature family used for matching.
  2. [§IV-E, Table V, Eq. (1)] The claim that assembled signals preserve R-R and amplitude structure is partly a check of the construction itself: the matcher explicitly minimizes the distance to the target features, so the close agreement in Table V and the corresponding statement in Section IV-E are expected by design. This does not establish that the inter-beat dynamics are realistic. An independent validation is needed, for example comparing heart-rate variability metrics, spectral measures, transition probabilities between beat classes, or performance on a rhythm-classification task using the assembled signals against real long-form records.
  3. [§III-B(b)] The multivariate feature generator is load-bearing because it produces the inter-beat feature trajectories that determine rhythm, but the manuscript does not specify the model family, architecture, training objective, hyperparameters, or sampling procedure. Without this information, the feature-level results in Table II and Figure 4 cannot be reproduced, and the reader cannot assess whether the reported distributional distances are dominated by the generator quality or by the evaluation setup.
  4. [§IV-C, §IV-E, Tables I and III] All classification results are reported as point estimates without error bars, confidence intervals, or significance tests, and the manuscript does not describe the cross-validation scheme or the subject-wise train/test split. Because the SVDB contains multiple beats per subject, leakage across beats from the same subject could inflate the reported accuracies. Reporting standard deviations over folds or subjects and using a subject-disjoint split is necessary to support the downstream-utility claim.
  5. [§IV-E, Table V] The abstract claims multi-minute ECG generation, but the only long-form quantitative evidence is a single 100-beat assembly in Table V (at 128 Hz and roughly 0.8 s per beat this is about 80 seconds), and the downstream comparison is limited to five-beat windows because of baseline memory constraints. The paper should either provide quantitative evaluation of multi-minute assemblies—covering continuity, rhythm statistics, and diagnostic content—or qualify the claim to the length actually demonstrated.
minor comments (6)
  1. [Table III] The column headers in Table III are ambiguous; the reader cannot map the entries to precision, recall, and F1 for normal and abnormal classes. The header should be expanded to name each column explicitly.
  2. [§III-B(b)] There is a typo in Section III-B(b): 'W are thus able' should read 'We are thus able'.
  3. [Appendix B] Appendix B contains the duplicated article in 'validates the the single beats quality'; it should be corrected to 'validates the single-beat quality'.
  4. [References] Reference [17] is an incomplete placeholder ('arXiv:2304.XXXX') and should be completed or removed.
  5. [Table I] In Table I, the Synthetic Data rows for Balanced SVM and SVM are numerically identical; if this is a genuine coincidence, clarify it, otherwise it may be a copy-paste error.
  6. [Figure 1] The caption of Figure 1 lists items (a), (b), and (c) inconsistently with how the sub-panels are referenced in the text; the caption and in-text references should be aligned.

Circularity Check

1 steps flagged · score 5.0 of 10

Matcher makes R-R/amplitude 'preservation' check tautological; the stronger realism claim is only partially supported.

  1. self definitional [Section IV-E (Evaluation of Long-Form ECG Assembly); Eq. (1) in Section III-C(a)]
    "for each feature vector fn we select a beat bm from the synthetic beat data B with minimal deviations from fn: Match(fn,B) = argmin m∈{1...|B|} Σ wk (fn,k − gk(bm))^2 (1) ... quantitative checks confirm that the distribution of R-R intervals in the assembled ECG preserves the synthetic feature trajectory (as can be seen in the Appendix Table V). Such alignment suggests that both short-range and long-range ECG characteristics are realistically captured."

    The matcher's objective (Eq. 1) directly minimizes the squared deviation between each synthesized feature vector fn (which includes R-R interval and R-amplitude descriptors, per Section III-B(a)) and the corresponding descriptor gk(bm) of the selected beat. Therefore the assembled long-form signal's R-R interval and R-amplitude distributions are, by construction, the synthetic feature trajectory plus matching residuals, not an independently observed property of the assembled signal. Reporting that the assembled ECG 'preserves the synthetic feature trajectory' is a check that the argmin succeeded; it cannot validate that real inter-beat dynamics are captured.

full rationale

The paper has substantial non-circular content: single-beat morphology is evaluated against real beats (Table IV), the feature generator's distributions are compared to real feature distributions (Table II), and downstream TSTR classification on real data is an external test. The end-to-end baseline (BRIDGE, ref [1]) is the authors' own prior model, but using it as a baseline is a legitimate ablation rather than load-bearing circularity, since the proposed pipeline is compared against its own component. The main circular element is the long-form assembly validation: because Eq. (1) selects beats by minimizing deviation from the synthesized feature vector, the assembled signal's R-R intervals and R amplitudes are forced to match that vector. The paper's statement that this 'suggests ... realistically captured' overstates the evidence. The central claim that 'essential diagnostic characteristics' are preserved is only as strong as the hand-curated feature set; the large abnormal-class F1 gap (0.479 synthetic vs 0.778 original) is consistent with missing diagnostic content, but that is a correctness limitation rather than a circularity. Overall, one important evaluation step reduces by construction, so partial circularity is present, but the core generation pipeline still has independent content.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on four domain assumptions (feature sufficiency, feature-generator fidelity, beat-generator realism, concatenation integrity) and one ad hoc reliance on the authors' own BRIDGE soft-prompt diffusion framework. The only free parameters reported are design choices such as store size and segment length, plus the under-specified matching weights w_k.

free parameters (3)
  • Matching weights w_k
    In Eq. (1), weights w_k scale the contribution of each feature dimension; the paper neither fixes them to 1 nor reports learning them, leaving the matching objective under-specified.
  • Synthetic beat store size = 10,000 per class
    The synthetic beat repository size per class is a user-chosen capacity parameter; it caps morphological diversity, yet no ablation studies its effect.
  • Beat segment length = 100 to 200 samples at 128 Hz
    The one-cycle window around the R-peak is a design choice; very short windows may truncate P or T waves, and very long ones add noise to the diffusion target.
assumptions (5)
  • domain assumption The set of hand-picked features (P/Q/R/S/T locations and amplitudes, intervals) is sufficient to capture clinically relevant ECG content.
    Invoked in Section III-B(a); if the feature set is incomplete, the matching step cannot recover missing diagnostic information.
  • domain assumption The multivariate feature generator learns the joint distribution of features accurately enough for long horizons.
    Section III-B(b); only distributional metrics are reported, with no test of long-range temporal coherence of generated feature sequences.
  • domain assumption The single-beat diffusion model produces realistic and sufficiently diverse beats.
    Section III-A; beat quality is evaluated with aggregate metrics, and the paper itself admits subtle distributional differences at high-variance timesteps.
  • domain assumption Concatenation with cross-fading does not distort beat morphology.
    Section III-C(b); only qualitative inspection is reported for boundary artifacts.
  • ad hoc to paper The soft-prompt learning bases from BRIDGE (ref [1]) apply to single-beat ECG generation.
    Section III-A relies on the authors' own prior BRIDGE framework without a standalone description or external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generating Realistic Multi-Beat ECG Signals." pith.science (2026). https://pith.science/paper/CX6TQJQM

@misc{pith2026250518189,
  author       = {Pith},
  title        = {Pith review of: Generating Realistic Multi-Beat ECG Signals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CX6TQJQM}},
  note         = {Machine review of arXiv:2505.18189}
}
read the original abstract

Generating synthetic ECG data has numerous applications in healthcare, from educational purposes to simulating scenarios and forecasting trends. While recent diffusion models excel at generating short ECG segments, they struggle with longer sequences needed for many clinical applications. This paper proposes a novel three-layer synthesis framework for generating realistic long-form ECG signals. We first generate high-fidelity single beats using a diffusion model, then synthesize inter-beat features preserving critical temporal dependencies, and finally assemble beats into coherent long sequences using feature-guided matching. Our comprehensive evaluation demonstrates that the resulting synthetic ECGs maintain both beat-level morphological fidelity and clinically relevant inter-beat relationships. In arrhythmia classification tasks, our long-form synthetic ECGs significantly outperform end-to-end long-form ECG generation using the diffusion model, highlighting their potential for increasing utility for downstream applications. The approach enables generation of unprecedented multi-minute ECG sequences while preserving essential diagnostic characteristics.

Figures

Figures reproduced from arXiv: 2505.18189 by the authors.

Figure 1
Figure 1. High-level methodology overview: (a) Beat-level ECG generation using a diffusion model; (b) Feature extraction and generation of synthetic features; (c) Long-sequence assembly via feature-beat matching. We propose a three-layer synthesis framework for gener￾ating long-form ECG signals: We first learn to generate high￾quality beats with a state-of-the-art diffusion model S (Section A). We then extract advanced per-be… view at source ↗
Figure 3
Figure 3. Heatmap of the difference in probability density (Gen [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. A comparison of the mean of individual generated beats [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Overview of pairwise correlations of features from the [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: Examples of individual beats and their averages, [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 24 canonical work pages

  1. [1]

    Bridge: Bootstrapping text to control time-series generation via multi-agent iterative optimization and diffusion mod- elling,

    H. Li, Y .-H. Huang, C. Xu, V . Schlegel, R.-H. Jiang, R. Batista-Navarro, G. Nenadic, and J. Bian, “Bridge: Bootstrapping text to control time-series generation via multi-agent iterative optimization and diffusion mod- elling,” arXiv preprint arXiv:2503.02445 , 2025

    arXiv 2025

  2. [2]

    Compatibility of a security policy for a cloud-based healthcare system with the EU general data protection regulation (GDPR),

    D. Georgiou and C. Lambrinoudakis, “Compatibility of a security policy for a cloud-based healthcare system with the EU general data protection regulation (GDPR),” Inf., vol. 11, no. 12, p. 586, 2020

    work page 2020

  3. [3]

    Team: PULSAR at probsum 2023: PULSAR: pre- training with extracted healthcare terms for summarising patients’ problems and data augmentation with black-box large language models,

    H. Li, Y . Wu, V . Schlegel, R. Batista-Navarro, T. Nguyen, A. R. Kashyap, X. Zeng, D. Beck, S. Winkler, and G. Ne- nadic, “Team: PULSAR at probsum 2023: PULSAR: pre- training with extracted healthcare terms for summarising patients’ problems and data augmentation with black-box large language models,” in BioNLP@ACL. Association for Computational Linguist...

    work page 2023

  4. [4]

    PULSAR at mediqa-sum 2023: Large language models augmented by synthetic dialogue convert patient dialogues to medical records,

    V . Schlegel, H. Li, Y . Wu, A. Subramanian, T. Nguyen, A. R. Kashyap, D. Beck, X. Zeng, R. T. Batista-Navarro, S. Winkler, and G. Nenadic, “PULSAR at mediqa-sum 2023: Large language models augmented by synthetic dialogue convert patient dialogues to medical records,” in CLEF (Working Notes) , ser. CEUR Workshop Proceed- ings, vol. 3497. CEUR-WS.org, 2023...

    work page 2023

  5. [5]

    Repurposing foundation model for generalizable medical time series classification,

    N. Huang, H. Wang, Z. He, M. Zitnik, and X. Zhang, “Repurposing foundation model for generalizable medical time series classification,” CoRR, vol. abs/2410.03794, 2024

    arXiv 2024

  6. [6]

    Sleepfm: Multi-modal represen- tation learning for sleep across brain activity, ECG and respiratory signals,

    R. Thapa, B. He, M. R. Kjær, H. E. M. IV , G. Ganjoo, E. Mignot, and J. Zou, “Sleepfm: Multi-modal represen- tation learning for sleep across brain activity, ECG and respiratory signals,” in ICML. OpenReview.net, 2024

    work page 2024

  7. [7]

    LLMs are not Zero-Shot Reasoners for Biomedical Information Extraction

    A. Nagar, V . Schlegel, T. Nguyen, H. Li, Y . Wu, K. Binici, and S. Winkler, “Llms are not zero-shot reasoners for biomedical information extraction,” CoRR, vol. abs/2408.12249, 2024

    work page Pith review arXiv 2024

  8. [8]

    AI in health- care: Time-series forecasting using statistical, neural, and ensemble architectures,

    S. Kaushik, A. Choudhury, P. K. Sheron, N. Dasgupta, S. Natarajan, L. A. Pickett, and V . Dutt, “AI in health- care: Time-series forecasting using statistical, neural, and ensemble architectures,” Frontiers Big Data, vol. 3, p. 4, 2020

    work page 2020

Show all 31 references
  1. [9]

    Real-valued (medical) time series generation with recurrent condi- tional GANs,

    C. Esteban, S. L. Hyland, and G. R ¨atsch, “Real-valued (medical) time series generation with recurrent condi- tional GANs,” 2017

    2017

  2. [10]

    Time-series GAN: Generative time-series modeling,

    J. Yoon, D. Jarrett, and M. van der Schaar, “Time-series GAN: Generative time-series modeling,” in NeurIPS, 2019

    2019

  3. [11]

    Synthesis 2DOI: https://doi.org/10.14469/hpc/2232 of realistic ECG using generative adversarial networks,

    A. M. Delaney, E. Brophy, and T. E. Ward, “Synthesis 2DOI: https://doi.org/10.14469/hpc/2232 of realistic ECG using generative adversarial networks,” 2019

    2019 doi

  4. [12]

    Electro- cardiogram generation with a bidirectional LSTM-CNN generative adversarial network,

    F. Zhu, Y . Fei, Y . Fu, Q. Liu, and B. Shen, “Electro- cardiogram generation with a bidirectional LSTM-CNN generative adversarial network,”Scientific Reports, vol. 9, 2019

    2019

  5. [13]

    Generative adversarial networks in electrocardiogram synthesis: Re- cent developments and challenges,

    L. Berger, M. Haberbusch, and F. Moscato, “Generative adversarial networks in electrocardiogram synthesis: Re- cent developments and challenges,” Artificial Intelligence in Medicine, vol. 143, p. 102632, 2023

    2023

  6. [14]

    Generating electrocardiogram signals by deep learning,

    N. Wulan, W. Wang, P. Sun, K. Wang, Y . Xia, and H. Zhang, “Generating electrocardiogram signals by deep learning,” Neurocomputing, vol. 404, pp. 122–136, 2020

    2020

  7. [15]

    Variational autoencoder–based neural electrocardiogram synthesis trained by FEM-based heart simulator,

    R. Nishikimi, M. Nakano, K. Kashino, and S. Tsukada, “Variational autoencoder–based neural electrocardiogram synthesis trained by FEM-based heart simulator,” Cardio- vascular Digital Health Journal, vol. 5, no. 1, pp. 19–28, 2024

    2024

  8. [16]

    TransFusion: Generating long, high fidelity time series using diffusion models,

    M. F. Sikder et al. , “TransFusion: Generating long, high fidelity time series using diffusion models,” arXiv preprint arXiv:2307.12667, 2023

    2023 arXiv

  9. [17]

    EHRDiff : Exploring realistic EHR synthesis with diffusion models,

    S. Y . Hongyi Yuan, Songchi Zhou, “EHRDiff : Exploring realistic EHR synthesis with diffusion models,” arXiv preprint arXiv:2304.XXXX, 2023

    2023

  10. [18]

    Meddiff: Generat- ing electronic health records using accelerated denoising diffusion model,

    H. He, S. Zhao, Y . Xi, and J. C. Ho, “Meddiff: Generat- ing electronic health records using accelerated denoising diffusion model,” CoRR, vol. abs/2302.04355, 2023

    2023 arXiv

  11. [19]

    Synthetic ECG signal genera- tion using probabilistic diffusion models,

    A. S. F. Edmondb Adib, “Synthetic ECG signal genera- tion using probabilistic diffusion models,” IEEE, 2023

    2023

  12. [20]

    Diffecg: A versatile probabilistic diffusion model for ECG signals synthesis,

    N. Neifar, A. Ben-Hamadou, A. Mdhaffar, and M. Jmaiel, “Diffecg: A versatile probabilistic diffusion model for ECG signals synthesis,” in SERA. IEEE, 2024, pp. 182– 188

    2024

  13. [21]

    Ecg synthesis via diffusion-based state space augmented transformer,

    M. H. Zama and F. Schwenker, “Ecg synthesis via diffusion-based state space augmented transformer,” Sen- sors, 2023

    2023

  14. [22]

    TSGM: A flexible framework for generative modeling of synthetic time series,

    D. Nikitin, L. Iannucci, and S. Kaski, “TSGM: A flexible framework for generative modeling of synthetic time series,” arXiv preprint arXiv:2305.11567, 2023

    2023 arXiv

  15. [23]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” neurips, 2014

    2014

  16. [24]

    Advancing retail data sci- ence: Comprehensive evaluation of synthetic data,

    A. Chin-Cheong et al. , “Advancing retail data sci- ence: Comprehensive evaluation of synthetic data,” arXiv preprint arXiv:2406.13130, 2023

    2023 arXiv

  17. [25]

    NeuroKit2: A python toolbox for neurophysiological signal processing,

    D. Makowski, T. Pham, Z. J. Lau, J. C. Brammer, F. Lespinasse, H. Pham, C. Sch ¨olzel, and S. H. A. Chen, “NeuroKit2: A python toolbox for neurophysiological signal processing,” Behavior Research Methods , vol. 53, no. 4, pp. 1689–1696, feb 2021

    2021

  18. [26]

    Improving explainability of deep neural network- based electrocardiogram interpretation using variational auto-encoders,

    R. van de Leur, M. Bos, K. Taha, A. Sammani, M. W. Yeung, S. van Duijvenboden, P. Lambiase, R. Hassink, P. van der Harst, P. Doevendans, D. Gupta, and R. van Es, “Improving explainability of deep neural network- based electrocardiogram interpretation using variational auto-enc...

    2022

  19. [27]

    Leveraging an ECG beat diffusion model for morphological reconstruction from indirect signals,

    L. Bedin, G. Cardoso, J. Duchateau, R. Dubois, and E. Moulines, “Leveraging an ECG beat diffusion model for morphological reconstruction from indirect signals,” in NeurIPS, 2024

    2024

  20. [28]

    Transfer function gain between heart period and QT interval vari- ability decreases at a 10-year follow-up in half-marathon runners,

    B. D. Maria, V . Bari, B. Cairo, F. Gelpi, D. Lucini, M. Pa- gani, M. Malacarne, A. M. Catai, M. de Oliveira Gois, F. Perego, L. A. D. Vecchia, and A. Porta, “Transfer function gain between heart period and QT interval vari- ability decreases at a 10-year follow-up in half-mar...

    2022

  21. [29]

    Accurate predictions on small data with a tabular foundation model,

    N. Hollmann, S. M ¨uller, L. Purucker, A. Krishnakumar, M. K ¨orfer, S. B. Hoo, R. T. Schirrmeister, and F. Hut- ter, “Accurate predictions on small data with a tabular foundation model,” Nature, 2025

    2025

  22. [30]

    Vector quantized time series generation with a bidirectional prior model,

    D. Lee, S. Malacarne, and E. Aune, “Vector quantized time series generation with a bidirectional prior model,”

  23. [2023]

    Available: https://arxiv.org/abs/2303

    [Online]. Available: https://arxiv.org/abs/2303. 04743 APPENDIX A. Single Beat Generation Analysis The synthesis framework relies on generating high-fidelity individual ECG beats, which are subsequently matched. Gen- erating a large variety of high quality, accurate synthetic ...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.