pith. sign in

arxiv: 2411.11896 · v3 · submitted 2024-11-08 · 📡 eess.SP · cs.LG

HeartBERT: A Self-Supervised ECG Embedding Model for Efficient and Effective Medical Signal Analysis

Saedeh Tahery , Fatemeh Hamid Akhlaghi , Termeh Amirsoleimani This is my paper

Pith reviewed 2026-05-23 17:27 UTC · model grok-4.3

classification 📡 eess.SP cs.LG
keywords ECG analysisself-supervised learningRoBERTasleep stage detectionheartbeat classificationsignal embeddingsmedical machine learningbidirectional LSTM
0
0 comments X

The pith

HeartBERT adapts RoBERTa-style self-supervised pretraining to ECG signals so that downstream classifiers can reach strong performance with smaller labeled sets and fewer parameters.

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

The paper presents HeartBERT as a model that learns general embeddings from unlabeled ECG recordings by adapting the RoBERTa architecture and its self-supervised objectives. These embeddings are then paired with lightweight bidirectional LSTM heads for two concrete medical tasks: sleep stage detection and heartbeat classification. Experiments compare the resulting systems against rival models and report that HeartBERT versions maintain or exceed accuracy while using less training data and fewer trainable parameters. The central goal is to lower the barrier of labeled data and compute that currently limits machine-learning use on ECG signals in clinical settings. If the transfer works as claimed, the same pretrained embeddings could serve many additional ECG analysis problems without repeating the full supervised training cycle.

Core claim

HeartBERT is built on the RoBERTa architecture and trained with a self-supervised objective on ECG data to produce embeddings that, when combined with bidirectional LSTM heads, deliver competitive or superior results on sleep stage detection and heartbeat classification while requiring smaller training datasets and fewer learning parameters than competing models.

What carries the argument

HeartBERT embeddings generated by RoBERTa-style self-supervised pretraining on ECG waveforms, then fed to bidirectional LSTM heads for supervised downstream tasks.

If this is right

  • HeartBERT-based pipelines achieve effective performance on sleep stage detection and heartbeat classification with smaller training datasets than rival models.
  • The same pipelines require fewer learning parameters while matching or exceeding rival accuracy.
  • The pretrained embeddings support multiple downstream ECG tasks without repeating large-scale labeled-data collection for each task.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same self-supervised ECG embeddings could be tested on additional tasks such as arrhythmia subtyping or stress detection without new pretraining.
  • If the embeddings prove stable across recording hardware and patient populations, they could reduce the annotation burden in multi-center ECG studies.
  • Replacing the LSTM heads with even lighter classifiers might further cut inference cost while preserving the reported data-efficiency gains.

Load-bearing premise

Self-supervised pretraining on ECG data with a RoBERTa architecture produces embeddings that transfer effectively to new supervised tasks without needing large labeled datasets for those tasks.

What would settle it

A head-to-head experiment in which HeartBERT-based systems fail to match or exceed the accuracy of rival models on sleep stage detection or heartbeat classification when both are trained on the same reduced-size labeled sets.

Figures

Figures reproduced from arXiv: 2411.11896 by Fatemeh Hamid Akhlaghi, Saedeh Tahery, Termeh Amirsoleimani.

Figure 2
Figure 2. Figure 2: Training the Tokenizer. a) The vocabulary is obtained through tokenizer training. b) The trained tokenizer acts as a token segmenter. Our tokenization approach offers significant advancements over traditional ECG tokenizers used in other works [45, 51, 52]. Conventional methods often rely on fixed-size windowing or [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An overview of the HeartBERT model. The model is first pre-trained using textual representations of ECG signals as our training data, and subsequently utilized in fine-tuning for downstream tasks. Transformer Layers: HeartBERT employs six transformer blocks, each comprising a multi￾head self-attention mechanism [53] and a position-wise fully connected feed-forward network. The self-attention mechanism enab… view at source ↗
Figure 4
Figure 4. Figure 4: Histogram of token lengths. The low frequency of samples for lengths greater than 16 has been cropped [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Trends of loss over epochs. The decreasing trend of loss indicates improvement in model performance with training iterations. 4. Experimental Study Having completed the pre-training phase, we extract encoded embeddings enriched with contextual information, which serve as potent representations for downstream tasks. The sleep￾stage and heartbeat classification tasks are two downstream applications that will… view at source ↗
Figure 6
Figure 6. Figure 6: The hybrid architecture of our model combining HeartBERT and Bi-LSTM for two downstream tasks—sleep-stage and heartbeat classification. We evaluate different configurations by freezing various layers of HeartBERT [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Class-wise performance metrics (Precision, Recall, and F1) for sleep-stage classification. a) Three￾stage classification, b) Five-stage classification. 4.5.2 Heartbeat classification To evaluate the proposed model's performance in heartbeat classification, we first examine the effect of fine-tuning different numbers of HeartBERT layers. Based on these results, the best￾performing model is selected for the … view at source ↗
read the original abstract

The HeartBert model is introduced with three primary objectives: reducing the need for labeled data, minimizing computational resources, and simultaneously improving performance in machine learning systems that analyze Electrocardiogram (ECG) signals. Inspired by Bidirectional Encoder Representations from Transformers (BERT) in natural language processing and enhanced with a self-supervised learning approach, the HeartBert model-built on the RoBERTa architecture-generates sophisticated embeddings tailored for ECG-based projects in the medical domain. To demonstrate the versatility, generalizability, and efficiency of the proposed model, two key downstream tasks have been selected: sleep stage detection and heartbeat classification. HeartBERT-based systems, utilizing bidirectional LSTM heads, are designed to address complex challenges. A series of practical experiments have been conducted to demonstrate the superiority and advancements of HeartBERT, particularly in terms of its ability to perform well with smaller training datasets, reduced learning parameters, and effective performance compared to rival models. The code and data are publicly available at https://github.com/ecgResearch/HeartBert.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces HeartBERT, a RoBERTa-based self-supervised model for ECG signal embeddings, with the goals of reducing labeled data requirements, lowering computational costs, and improving performance on medical signal tasks. It evaluates the model with bidirectional LSTM heads on sleep stage detection and heartbeat classification, claiming superiority over rival models especially in small-dataset regimes and with fewer parameters. Code and data are released publicly.

Significance. If substantiated, the work could advance efficient ECG analysis by showing transfer from self-supervised pretraining to downstream tasks with limited labels. The public code release is a clear strength for reproducibility.

major comments (2)
  1. [Experiments] Experiments section: The reported comparisons of HeartBERT+LSTM systems against rival models on sleep staging and heartbeat classification do not include a control arm using an identically architected RoBERTa (or the same LSTM head) that is randomly initialized or trained only on the downstream task. Without this, performance gains on small datasets cannot be attributed to the self-supervised pretraining objective rather than architecture, head design, or parameter count differences.
  2. [Abstract] Abstract and results presentation: The abstract asserts superiority on small datasets with reduced parameters and effective performance, yet supplies no quantitative results, baselines, error bars, dataset sizes, or exclusion criteria to support these claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important aspects for strengthening the attribution of results to self-supervised pretraining and for improving the abstract's support of claims. We address each point below.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The reported comparisons of HeartBERT+LSTM systems against rival models on sleep staging and heartbeat classification do not include a control arm using an identically architected RoBERTa (or the same LSTM head) that is randomly initialized or trained only on the downstream task. Without this, performance gains on small datasets cannot be attributed to the self-supervised pretraining objective rather than architecture, head design, or parameter count differences.

    Authors: We agree this is a valid concern and that the current experimental design does not fully isolate the contribution of self-supervised pretraining. In the revised manuscript we will add the requested control arms: a randomly initialized RoBERTa model using the identical LSTM head, and the same architecture trained from scratch on the downstream tasks without pretraining. These results will be reported alongside the existing comparisons on both tasks, with particular attention to the small-dataset regimes. revision: yes

  2. Referee: [Abstract] Abstract and results presentation: The abstract asserts superiority on small datasets with reduced parameters and effective performance, yet supplies no quantitative results, baselines, error bars, dataset sizes, or exclusion criteria to support these claims.

    Authors: The abstract was written in a concise, high-level style typical for the venue. We acknowledge that adding quantitative anchors would better substantiate the claims. We will revise the abstract to include key performance metrics (e.g., accuracy or F1 on small-data splits), parameter counts relative to baselines, and brief dataset details while preserving brevity. revision: yes

Circularity Check

0 steps flagged

No circularity; purely empirical model introduction and evaluation

full rationale

The paper presents HeartBERT as a RoBERTa-based self-supervised pretraining approach for ECG signals, followed by empirical evaluation on downstream tasks (sleep staging, heartbeat classification) using bidirectional LSTM heads. No equations, derivations, or fitted quantities are described that reduce by construction to inputs, self-citations, or ansatzes. Superiority claims with smaller datasets rest on reported experimental comparisons rather than any self-definitional or load-bearing self-citation chain. The work is self-contained against external benchmarks via public code and data, with no mathematical prediction step that collapses to its own premises.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the model reuses the existing RoBERTa architecture and standard self-supervised objectives without introducing new postulated quantities.

pith-pipeline@v0.9.0 · 5718 in / 1271 out tokens · 35102 ms · 2026-05-23T17:27:48.888595+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages · 2 internal anchors

  1. [1]

    Review of complementary and alternative medical treatment of arrhythmias,

    A. Brenyo and M. K. Aktas, "Review of complementary and alternative medical treatment of arrhythmias," The American journal of cardiology, vol. 113, no. 5, pp. 897-903, 2014

    work page 2014

  2. [2]

    Heart failure classification using deep learning to extract spatiotemporal features from ECG,

    C.-J. Zhang, F.-Q. Tang, H.-P. Cai, and Y.-F. Qian, "Heart failure classification using deep learning to extract spatiotemporal features from ECG," BMC Medical Informatics and Decision Making, vol. 24, no. 1, p. 17, 2024

    work page 2024

  3. [3]

    Ensemble deep learning approach for ecg -based cardiac disease detection: Signal and image analysis,

    T. Mahmud, A. Barua, D. Islam, M. S. Hossain, R. Chakma, K. Barua, M. Monju, and K. Andersson, "Ensemble deep learning approach for ecg -based cardiac disease detection: Signal and image analysis," in 2023 International Conference on Information and Communication Technology for Sustainable Development (ICICT4SD) , 2023: IEEE, pp. 70-74

    work page 2023

  4. [4]

    Deep learning for ECG Arrhythmia detection and classification: an overview of progress for period 2017–2023,

    Y. Ansari, O. Mourad, K. Qaraqe, and E. Serpedin, "Deep learning for ECG Arrhythmia detection and classification: an overview of progress for period 2017–2023," Frontiers in Physiology, vol. 14, p. 1246746, 2023

    work page 2017

  5. [5]

    Early detection and prediction of Heart Disease using Wearable devices and Deep Learning algorithms,

    S. Sivasubramaniam and S. Balamurugan, "Early detection and prediction of Heart Disease using Wearable devices and Deep Learning algorithms," Multimedia Tools and Applications, pp. 1-15, 2024

    work page 2024

  6. [6]

    Deep learning models for predicting left heart abnormalities from single -lead electrocardiogram for the development of wearable devices,

    M. Sato, S. Kodera , N. Setoguchi, K. Tanabe, S. Kushida, J. Kanda, M. Saji, M. Nanasato, H. Maki, and H. Fujita, "Deep learning models for predicting left heart abnormalities from single -lead electrocardiogram for the development of wearable devices," Circulation Journal, vol. 88, no. 1, pp. 146-156, 2023

    work page 2023

  7. [7]

    Smart wearables for the detection of cardiovascular diseases: a systematic literature review,

    M. Moshawrab, M. Adda, A. Bouzouane, H. Ibrahim, and A. Raad, "Smart wearables for the detection of cardiovascular diseases: a systematic literature review," Sensors, vol. 23, no. 2, p. 828, 2023

    work page 2023

  8. [8]

    Electrocardiogram monitoring wearable devices and artificial -intelligence-enabled diagnostic capabilities: a review,

    L. Neri, M. T. Oberdier, K. C. van Abeelen, L. Menghini, E. Tumarkin, H. Tripathi, S. Jaipalli, A. Orro, N. Paolocci, and I. Gallelli, "Electrocardiogram monitoring wearable devices and artificial -intelligence-enabled diagnostic capabilities: a review," Sensors, vol. 23, no. 10, p. 4805, 2023

    work page 2023

  9. [9]

    Heart disease classification based on ECG using machine learning models,

    S. M. Malakouti, "Heart disease classification based on ECG using machine learning models," Biomedical Signal Processing and Control, vol. 84, p. 104796, 2023

    work page 2023

  10. [10]

    Self -supervised representation lear ning from 12 -lead ECG data,

    T. Mehari and N. Strodthoff, "Self -supervised representation lear ning from 12 -lead ECG data," Computers in biology and medicine, vol. 141, p. 105114, 2022

    work page 2022

  11. [11]

    Practical intelligent diagnostic algorithm for wearable 12-lead ECG via self-supervised learning on large-scale dataset,

    J. Lai, H. Tan, J. Wang, L. Ji, J. Guo, B. Han, Y. Shi, Q. Feng, and W. Yang, "Practical intelligent diagnostic algorithm for wearable 12-lead ECG via self-supervised learning on large-scale dataset," Nature Communications, vol. 14, no. 1, p. 3741, 2023

    work page 2023

  12. [12]

    Self-supervised contrastive learning for medical time series: A systematic review,

    Z. Liu, A. Alavi, M. Li, and X. Zhang, "Self-supervised contrastive learning for medical time series: A systematic review," Sensors, vol. 23, no. 9, p. 4221, 2023

    work page 2023

  13. [13]

    Applications of self -supervised learning to biomedical signals: A survey,

    F. Del Pup and M. Atzori, "Applications of self -supervised learning to biomedical signals: A survey," IEEE Access, 2023

    work page 2023

  14. [14]

    Self -supervised learning for label -efficient sleep stage classification: A comprehensive evaluation,

    E. Eldele, M. Ragab, Z. Chen, M. Wu, C. -K. Kwoh, and X. Li, "Self -supervised learning for label -efficient sleep stage classification: A comprehensive evaluation," IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 31, pp. 1333-1342, 2023

    work page 2023

  15. [15]

    PSNSleep: a self-supervised learning method for sleep staging based on Siamese networks with only positive sample pairs,

    Y. You, S. Chang, Z. Yang, and Q. Sun, "PSNSleep: a self-supervised learning method for sleep staging based on Siamese networks with only positive sample pairs," Frontiers in Neuroscience, vol. 17, p. 1167723, 2023

    work page 2023

  16. [16]

    In -distribution and out-of-distribution self- supervised ecg representation learning for arrhythmia d etection,

    S. Soltanieh, J. Hashemi, and A. Etemad, "In -distribution and out-of-distribution self- supervised ecg representation learning for arrhythmia d etection," IEEE Journal of Biomedical and Health Informatics, 2023

    work page 2023

  17. [17]

    RoBERTa: A Robustly Optimized BERT Pretraining Approach

    Y. Liu, "Roberta: A robustly optimized bert pretraining approach," arXiv preprint arXiv:1907.11692, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  18. [18]

    Application of the Max –Lloyd quantizer for ECG compression in diving mammals,

    M. Rodrı́guez, A. Ayala, S. Rodrı́guez, F. Rosa, and M. Dı́az-González, "Application of the Max –Lloyd quantizer for ECG compression in diving mammals," Computer methods and programs in biomedicine, vol. 73, no. 1, pp. 13-21, 2004

    work page 2004

  19. [19]

    Quantizing for minimum distortion,

    J. Max, "Quantizing for minimum distortion," IRE Transactions on Information Theory, vol. 6, no. 1, pp. 7-12, 1960

    work page 1960

  20. [20]

    Neural machine translation of rare words with subword units,

    R. Sennrich, B. Haddow, and A. Birch, "Neural machine translation of rare words with subword units," in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, 2016, pp. 1715–1725

    work page 2016

  21. [21]

    A review of automated sleep stage based on EEG signals,

    X. Zhang, X. Zhang, Q. Huang, Y. Lv, and F. Chen, "A review of automated sleep stage based on EEG signals," Biocybernetics and Biomedical Engineering, 2024

    work page 2024

  22. [22]

    Does REM sleep contribute to subjective wake time in primary insomnia? A comparison of polysomnographic and subjective sleep in 100 patients,

    B. Feige, A. Al‐Shajlawi, C. Nissen, U. Voderholzer, M. Hornyak, K. Spiegelhalder, C. Kloepfer, M. Pe rlis, and D. Riemann, "Does REM sleep contribute to subjective wake time in primary insomnia? A comparison of polysomnographic and subjective sleep in 100 patients," Journal of sleep research, vol. 17, no. 2, pp. 180-190, 2008

    work page 2008

  23. [23]

    Sleep -related breathing disorders and cardiovascular disease,

    F. Roux, C. D’Ambrosio, and V. Mohsenin, "Sleep -related breathing disorders and cardiovascular disease," The American journal of medicine, vol. 108, no. 5, pp. 396 - 402, 2000

    work page 2000

  24. [24]

    Sleep disorders: causes, effects, and solutions,

    G. M. Tibbitts, "Sleep disorders: causes, effects, and solutions," Primary Care: Clinics in Office Practice, vol. 35, no. 4, pp. 817-837, 2008

    work page 2008

  25. [25]

    Obesity, obstructive sleep apnea and type 2 diabetes mellitus: Epidemiology and pathophysiologic insights,

    S. Jehan, A. K. Myers, F. Zizi, S. R. Pandi-Perumal, G. Jean-Louis, and S. I. McFarlane, "Obesity, obstructive sleep apnea and type 2 diabetes mellitus: Epidemiology and pathophysiologic insights," Sleep medicine and disorders: international journal, vol. 2, no. 3, p. 52, 2018

    work page 2018

  26. [26]

    International classification of sleep disorders,

    M. J. Sateia, "International classification of sleep disorders," Chest, vol. 146, no. 5, pp. 1387-1394, 2014

    work page 2014

  27. [27]

    Development of the polysomnographic database on CD‐ ROM,

    Y. Ichimaru and G. Moody, "Development of the polysomnographic database on CD‐ ROM," Psychiatry and clinical neurosciences, vol. 53, no. 2, pp. 175-177, 1999

    work page 1999

  28. [28]

    An Introduction to Convolutional Neural Networks

    K. O'Shea, "An introduction to convolutional neural networks," arXiv preprint arXiv:1511.08458, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  29. [29]

    Introduction to convolutional neural networks,

    J. Wu, "Introduction to convolutional neural networks," National Key Lab for Novel Software Technology. Nanjing University. China, vol. 5, no. 23, p. 495, 2017

    work page 2017

  30. [30]

    Long Short-term Memory,

    S. Hochreiter, "Long Short-term Memory," Neural Computation MIT-Press, 1997

    work page 1997

  31. [31]

    ICENTIA11K Single Lead Continuous Raw Electrocardiogram Dataset,

    S. Tan, S. Ortiz-Gagné, N. Beaudoin-Gagnon, P. Fecteau, A. Courville, Y. Bengio, and J. Cohen, "ICENTIA11K Single Lead Continuous Raw Electrocardiogram Dataset," PhysioNet, 2022

    work page 2022

  32. [32]

    Icentia11k: An unsupervised representation learning dataset for arrhythmia subtype discovery,

    S. Tan, G. Androz, A. Chamseddine, P. Fecteau, A. Courville, Y. Bengio, and J. P . Cohen, "Icentia11k: An unsupervised representation learning dataset for arrhythmia subtype discovery," in Computing in Cardiology Conference (CinC), 2021

    work page 2021

  33. [33]

    ECG -based heartbeat classification for arrhythmia detection: A survey,

    E. J. d. S. Luz, W. R. Schwartz, G. Cámara -Chávez, and D. Menotti, "ECG -based heartbeat classification for arrhythmia detection: A survey," Computer methods and programs in biomedicine, vol. 127, pp. 144-164, 2016

    work page 2016

  34. [34]

    Heart arrhythmia diagnosis based on the combination of morphological, frequency and nonlinear feat ures of ECG signals and metaheuristic feature selection algorithm,

    V. Mazaheri and H. Khodadadi, "Heart arrhythmia diagnosis based on the combination of morphological, frequency and nonlinear feat ures of ECG signals and metaheuristic feature selection algorithm," Expert Systems with Applications, vol. 161, p. 113697, 2020

    work page 2020

  35. [35]

    DeepArr: An investigative tool for arrhythmia detection using a contextual deep neu ral network from electrocardiograms (ECG) signals,

    W. Midani, W. Ouarda, and M. B. Ayed, "DeepArr: An investigative tool for arrhythmia detection using a contextual deep neu ral network from electrocardiograms (ECG) signals," Biomedical Signal Processing and Control, vol. 85, p. 104954, 2023

    work page 2023

  36. [36]

    A deep learning framework for ECG denoising and classification,

    H. Peng, X. Chang, Z. Yao, D. Shi, and Y. Chen, "A deep learning framework for ECG denoising and classification," Biomedical Signal Processing and Control, vol. 94, p. 106441, 2024

    work page 2024

  37. [37]

    ECG Language processing (ELP): A new technique to analyze ECG signals,

    S. Mousavi, F. Afghah, F. Khadem, and U. R. Acharya, "ECG Language processing (ELP): A new technique to analyze ECG signals," Computer methods and programs in biomedicine, vol. 202, p. 105959, 2021

    work page 2021

  38. [38]

    A transformer-based deep neural network for arrhythmia detection using continuous ECG signals,

    R. Hu, J. Chen, and L. Zhou, "A transformer-based deep neural network for arrhythmia detection using continuous ECG signals," Computers in Biology and Medicine, vol. 144, p. 105325, 2022

    work page 2022

  39. [39]

    A novel time representation input based on deep learning for ECG classification,

    Y. Huang, H. Li, and X. Yu, "A novel time representation input based on deep learning for ECG classification," Biomedical Signal Processing and Control, vol. 83, p. 104628, 2023

    work page 2023

  40. [40]

    ECGformer: Leveraging transformer for ECG heartbeat arrhythmia classification,

    T. Akan, S. Alp, and M. A. N. Bhuiyan, "ECGformer: Leveraging transformer for ECG heartbeat arrhythmia classification," in 2023 In ternational Conference on Computational Science and Computational Intelligence (CSCI), 2023: IEEE, pp. 1412- 1417

    work page 2023

  41. [41]

    ECGTransForm: Empowering adaptive ECG arrhythmia classification framework with bidirectional transformer,

    H. El-Ghaish and E. Eldele, "ECGTransForm: Empowering adaptive ECG arrhythmia classification framework with bidirectional transformer," Biomedical Signal Processing and Control, vol. 89, p. 105714, 2024

    work page 2024

  42. [42]

    Bidirectional gated recurrent unit with auto encoders for detecting arrhythmia using ECG data,

    R. Sarankumar, M. Ramkumar, K. Vijaipriya, and R. Velselvi, "Bidirectional gated recurrent unit with auto encoders for detecting arrhythmia using ECG data," Knowledge-Based Systems, vol. 294, p. 111696, 2024

    work page 2024

  43. [43]

    Unsupervised Pre-Training Using Masked Autoencoders for ECG Analysis,

    G. Wang, Q. Wang, G. N. Iyer, A. Nag, and D. John, "Unsupervised Pre-Training Using Masked Autoencoders for ECG Analysis," in 2023 IEEE Biomedical Circuits and Systems Conference (BioCAS), 2023: IEEE, pp. 1-5

    work page 2023

  44. [44]

    A lightweight SelfONN model for general ECG classification with pretraining,

    K. Qin, W. Huang, T. Zhang, H. Zhang, and X. Cheng, "A lightweight SelfONN model for general ECG classification with pretraining," Biomedical Signal Processing and Control, vol. 89, p. 105780, 2024

    work page 2024

  45. [45]

    Biot: Biosignal transformer for cross-data learning in the wild,

    C. Yang, M. Westover, and J. Sun, "Biot: Biosignal transformer for cross-data learning in the wild," Advances in Neural Information Processing Systems, vol. 36, 2024

    work page 2024

  46. [46]

    Resampling Strategies and their Influence on Heart Rate Variability Features in Low Sampling Rate Electrocardiogram Data,

    M. Zakariyah, U. Zaky, M. Nurjaman, A. G. Istikmal, and H. A. Widianto, "Resampling Strategies and their Influence on Heart Rate Variability Features in Low Sampling Rate Electrocardiogram Data," JURNAL INFOTEL, vol. 16, no. 1, pp. 17-31, 2024

    work page 2024

  47. [47]

    NeuroKit2: A Python toolbox for neurophysiologica l signal processing,

    D. Makowski, T. Pham, Z. J. Lau, J. C. Brammer, F. Lespinasse, H. Pham, C. Schölzel, and S. A. Chen, "NeuroKit2: A Python toolbox for neurophysiologica l signal processing," Behavior research methods, pp. 1-8, 2021

    work page 2021

  48. [48]

    Least squares quantization in PCM,

    S. Lloyd, "Least squares quantization in PCM," IEEE transactions on information theory, vol. 28, no. 2, pp. 129-137, 1982

    work page 1982

  49. [49]

    Improving language understanding by generative pre-training,

    A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever, " Improving language understanding by generative pre-training," 2018

    work page 2018

  50. [50]

    Fusion Token: Enhancing Compression and Efficiency in Language Model Tokenization,

    R. Kwiatkowski, Z. Wang, R. Giaquinto, V. Kumar, X. Ma, B. Xiang, and B. Athiwaratkun, "Fusion Token: Enhancing Compression and Efficiency in Language Model Tokenization," 2024

    work page 2024

  51. [51]

    ECGBERT: Understanding hidden language of ECGs with self-supervised representation learning,

    S. Choi, S. Mousavi, P. Si, H. G. Yhdego, F. Khadem, and F. Afghah, "ECGBERT: Understanding hidden language of ECGs with self-supervised representation learning," arXiv preprint arXiv:2306.06340, 2023

    work page arXiv 2023

  52. [52]

    Generative pre -trained transformer for cardiac abnormality detection,

    P. L. Gaudilliere, H. Sigurthorsdottir, C. Aguet, J. Van Zaen, M. Lemay, and R. Delgado-Gonzalo, "Generative pre -trained transformer for cardiac abnormality detection," in 2021 Computing in Cardiology (CinC), 2021, vol. 48: IEEE, pp. 1-4

    work page 2021

  53. [53]

    Attention is all you need,

    A. Vaswani, "Attention is all you need," Advances in Neural Information Processing Systems, 2017

    work page 2017

  54. [54]

    Bert: Pre -training of deep bidirectional transformers for language understanding,

    J. Devlin, "Bert: Pre -training of deep bidirectional transformers for language understanding," in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), 2019, pp. 4171–4186

    work page 2019

  55. [55]

    PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals,

    A. L. Goldberger, L. A. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and H. E. Stanley, "PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals," circulation, vol. 101, no. 23, pp. e215-e220, 2000

    work page 2000

  56. [56]

    The impact of the MIT-BIH arrhythmia database,

    G. B. Moody and R. G. Mark, "The impact of the MIT-BIH arrhythmia database," IEEE engineering in medicine and biology magazine, vol. 20, no. 3, pp. 45-50, 2001

    work page 2001

  57. [57]

    PTB -XL, a large publicly available electrocardiography dataset,

    P. Wagner, N. Strodthoff, R. -D. Bousseljot, D. Kreiseler, F. I. Lunze, W. Samek, and T. Schaeffter, "PTB -XL, a large publicly available electrocardiography dataset," Scientific data, vol. 7, no. 1, pp. 1-15, 2020

    work page 2020

  58. [58]

    The European ST -T database: standard for evaluating systems for the analysis of ST-T changes in ambulatory electrocardiography,

    A. Taddei, G. Distante, M. Emdin, P. Pisani, G. Moody, C. Zeelenberg, and C. Marchesi, "The European ST -T database: standard for evaluating systems for the analysis of ST-T changes in ambulatory electrocardiography," European heart journal, vol. 13, no. 9, pp. 1164-1172, 1992

    work page 1992

  59. [59]

    A deep learning approach to detect sleep stages,

    K. Stuburić, M. Gaiduk, and R. Seepold, "A deep learning approach to detect sleep stages," Procedia Computer Science, vol. 176, pp. 2764-2772, 2020

    work page 2020

  60. [60]

    Ensemble computational intelligent for insomnia sleep stage detection via the sleep ECG signal,

    P. Tripathi, M. Ansari, T. K. Gandhi, R. Mehrotra, M. B. B. Heyat, F. Akhtar, C. C. Ukwuoma, A. Y. Muaad, Y. M. Kadah, and M. A. Al-Antari, "Ensemble computational intelligent for insomnia sleep stage detection via the sleep ECG signal," IEEE Access, vol. 10, pp. 108710-108721, 2022

    work page 2022

  61. [61]

    Generative adversarial network with transformer generator for boosting ECG classification,

    Y. Xia, Y. Xu, P. Chen, J. Zhang, and Y. Zhang, "Generative adversarial network with transformer generator for boosting ECG classification," Biomedical Signal Processing and Control, vol. 80, p. 104276, 2023

    work page 2023

  62. [62]

    Testing and Reporting Performance Results of Cardiac Rhythm a nd ST Segment Measurement Algorithms, Standard ANSI/AAMI EC38, Association for the Advancement of Medical Instrumentation, 1998

    work page 1998

  63. [63]

    The AASM manual for the scoring of sleep and associated events: rules, terminology, and technical specification,

    C. Iber, "The AASM manual for the scoring of sleep and associated events: rules, terminology, and technical specification," 2007

    work page 2007

  64. [64]

    A review of methods for imbalanced multi-label classification,

    A. N. Tarekegn, M. Giacobini, and K. Michalak, "A review of methods for imbalanced multi-label classification," Pattern Recognition, vol. 118, p. 107965, 2021

    work page 2021

  65. [65]

    ECG heartbeat classification using multimodal fusion,

    Z. Ahmad, A. Tabassum, L. Guan, and N. M. Khan, "ECG heartbeat classification using multimodal fusion," IEEE Access, vol. 9, pp. 100615-100626, 2021

    work page 2021

  66. [66]

    Deep convolutional recurrent model for automatic scoring sleep stages based on single-lead ECG signal,

    E. Urtnasan, J.-U. Park, E. Y. Joo, and K.-J. Lee, "Deep convolutional recurrent model for automatic scoring sleep stages based on single-lead ECG signal," Diagnostics, vol. 12, no. 5, p. 1235, 2022. This paper is currently under investigation/review, and further changes may be applied in the future

    work page 2022