Pith. sign in

REVIEW 4 major objections 5 minor 6 cited by

xECG, a recurrent xLSTM model trained with self-distillation, achieves the highest BenchECG score (0.868) and is the only publicly available ECG foundation model that performs strongly across all benchmark datasets and tasks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

xECG, an xLSTM-based ECG foundation model pretrained with SimDINOv2, achieves the best average performance on BenchECG, a new benchmark spanning eight datasets and ten tasks.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection A useful ECG benchmark plus a strong open baseline, but the CODE-15% pretraining overlap contaminates two of the ten tasks and should be fixed before the headline ranking is taken at face value. the 4 major comments →

arxiv 2509.10151 v1 pith:S2RAEY4J submitted 2025-09-12 cs.LG cs.AI

BenchECG and xECG: a benchmark and baseline for ECG foundation models

Riccardo Lunelli , Angus Nicolson , Samuel Martin Pr\"oll , Sebastian Johannes Reinstadler , Axel Bauer , Clemens Dlaska This is my paper
classification cs.LG cs.AI
keywords ECG foundation modelBenchECGxECGxLSTMself-supervised learningECG benchmarkSimDINOv2long-context ECG
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper introduces BenchECG, a standardised benchmark for ECG foundation models spanning eight public datasets, about 1.67 million recordings, and ten tasks that range from 12-lead diagnostic classification to R-peak detection, sleep-apnea segmentation, PPG-based atrial-fibrillation detection, age regression, blood-test abnormality prediction, and ICU mortality survival analysis. Against this benchmark, the paper claims that xECG—a bidirectional xLSTM-based encoder pretrained with a SimDINOv2 teacher–student objective adapted to ECG—achieves the best overall score (0.868 ± 0.003), with an average rank of 1.50 under finetuning and 1.20 under linear probing, and is the only public model with strong results on all tasks. The broader claim is that recurrent architectures (xLSTM) combined with non-contrastive self-supervised pretraining give more transferable ECG representations than the prevailing transformer-based SSL models, especially for long-context signals, while being roughly 5–10× cheaper to finetune. If true, this gives the field a reproducible baseline: a single benchmark, a single scoring metric, and a model and code release that future ECG foundation models can be compared against directly.

Core claim

On the paper's own terms, the central discovery is that xECG, a bidirectional xLSTM with alternating scalar-memory sLSTM and matrix-memory mLSTM blocks, trained via SimDINOv2 self-supervised learning on roughly eight million ECG recordings, outperforms all publicly available ECG foundation models on the new standardised BenchECG benchmark. The reported BenchECG score is 0.868 ± 0.003, with an average rank of 1.50 across ten tasks under finetuning and 1.20 under linear probing; the paper states that xECG is the only publicly available model to perform strongly on all datasets and tasks. The paper also establishes that xLSTM-based models dominate on long-context tasks (sleep-apnea segmentation

What carries the argument

The load-bearing pieces are (1) BenchECG, a benchmark with a single normalised 'BenchECG score' computed as the mean across tasks of task-appropriate metrics (AUROC for classification and segmentation, F1 for R-peak detection, 1−SMAPE for age regression, C-index for survival analysis), and (2) xECG, a bidirectional stack of alternating sLSTM and mLSTM blocks that processes raw signals as 250 ms temporal patches, enabling linear-time and linear-memory scaling with sequence length. Its pretraining uses SimDINOv2, a non-contrastive teacher–student self-distillation objective with an exponential-moving-average teacher, multiple global and local temporal views, patch masking with reconstruction i

Load-bearing premise

The benchmark's fairness rule—that evaluation models must not have been pretrained on evaluation data—is not fully met, because xECG was pretrained on CODE, of which CODE-15% is a subset, and CODE-15% is exactly the finetuning set for the age and mortality tasks; if pretraining on those same unlabeled recordings inflates downstream accuracy, the ranking on those tasks is not a clean test of representation quality.

What would settle it

Train a version of xECG with CODE entirely excluded from pretraining (keeping only INCART, Chapman, and Ningbo), then re-run the age-regression and mortality tasks that are finetuned on CODE-15%; if xECG's advantage over ST-MEM and ECG-JEPA on those tasks shrinks or disappears, the pretraining overlap with CODE-15% is materially responsible for the reported lead.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Future ECG foundation models can be compared head-to-head on the same ten tasks and metric, making claims of 'state of the art' reproducible rather than dataset-specific.
  • On long-context ECG tasks such as sleep apnea and ambulatory arrhythmia, transformer-based SSL models are not just slower but also worse; the results imply that recurrence may be a better inductive bias for minute-scale physiological dependencies.
  • xECG's strong linear-probing performance (average rank 1.20) indicates that its frozen representations transfer well without task-specific adaptation, supporting use in low-label clinical settings.
  • Because the xLSTM backbone scales linearly with input length, xECG can be finetuned or applied to longer recordings at a fraction of the memory and time cost of transformer ECG foundation models, making continuous-monitoring deployment more feasible.

Where Pith is reading between the lines

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

  • The two cross-population tasks (age and mortality) are finetuned on CODE-15%, which is a subset of CODE used in xECG pretraining (and CODE-15% itself was used to pretrain ST-MEM and ECG-JEPA). The paper discloses this but does not quantify how much overlap inflates the ranking; a clean test would ablate CODE from pretraining and re-run those tasks.
  • If the reported linear-probing advantages hold, BenchECG could be extended to zero-shot or few-shot protocols such as frozen-feature logistic regression or k-nearest-neighbour probing, giving a more direct measurement of representation quality.
  • The simple mean over disparate metrics (AUROC, F1, SMAPE, C-index) makes the aggregate BenchECG score sensitive to task composition; weighting by clinical relevance or reporting task-family subscores might change which model leads.
  • The recurrent backbone's linear scaling suggests an immediate testable extension: evaluate xECG on continuous multi-day Holter monitoring or ICU telemetry, where even 30-minute recordings are short by clinical standards.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces BenchECG, a standardized benchmark for ECG foundation models built from eight public datasets and ten tasks spanning classification, segmentation, detection, regression, and survival analysis. It also proposes xECG, a bidirectional xLSTM encoder pretrained with SimDINOv2-style self-supervised learning, and evaluates it against publicly available foundation models (ST-MEM, ECG-JEPA, ECGFounder) plus two in-house ablations. Across five-seed finetuning and linear-probing runs, xECG is reported to achieve the highest aggregate BenchECG score (0.868 ± 0.003) and the best mean ranks (1.50 finetuning, 1.20 linear probing), and is claimed to be the only publicly available model that performs strongly on all datasets and tasks.

Significance. If the comparison is accepted, BenchECG addresses a genuine gap: ECG foundation models are currently evaluated on heterogeneous task selections, making cross-paper ranking difficult. The design is broad and clinically varied, including OOD tasks, long-context recordings, and a non-ECG modality (PPG). The paper's strengths include multi-seed evaluation with Welch's t-tests, per-task result tables, detailed hyperparameter documentation, computational cost comparisons, and public release of code and weights. The xECG architecture itself is a plausible and interesting alternative to quadratic-attention transformers for long ECG sequences. However, the central comparison rests on the fairness of the benchmark and on a small number of benchmark-construction decisions; several load-bearing points need to be addressed before the headline ranking can be considered established.

major comments (4)
  1. [Methods, Age and Mortality tasks; Table 1] The stated fair-comparison rule says models should not be pretrained on evaluation datasets, and CODE-15% is formally outside the evaluation set. However, Age and Mortality are finetuned on CODE-15%, and CODE-15% is a subset of CODE, which was used for xECG pretraining; ST-MEM and ECG-JEPA were pretrained directly on CODE-15%. These models therefore saw the exact unlabeled finetuning recordings during SSL pretraining, while ECGFounder and Supervised xLSTM did not. The paper discloses this overlap but does not quantify its effect. Because Age and Mortality are two of the ten tasks feeding the BenchECG score and the 'only publicly available model to perform strongly on all tasks' claim, please add a control (e.g., an xECG variant pretrained without CODE, or an exclusion analysis recomputing scores and ranks without these two tasks) and report whether the ranking persists.
  2. [Methods, Exercise task] The Exercise-ECG split is described as 'patients 10-13 for validation and 13-20 for testing.' Patient 13 is included in both validation and test. Since validation is used for early stopping and checkpoint selection, the Exercise R-peak results in Supplementary Table 6 are potentially optimistically biased for all models, and the task is not a clean generalization test. Please make the patient-level validation and test sets disjoint and re-run the five-seed evaluation for this task.
  3. [Results, BenchECG score; Methods, BenchECG Metrics] The aggregate BenchECG score is the central ranking instrument, but it is not defined precisely enough to be reproduced. The text says it is the 'mean performance of a model across all tasks' but does not specify how the Age task's three test sets are combined into one task score, whether all ten tasks are weighted equally, or how metrics with different chance levels (AUROC and C-index at 0.5, F1, and SMAPE) are normalized before averaging. Please provide an explicit formula and a table listing each task's normalized score and contribution so that the reported 0.868 ± 0.003 and mean ranks can be verified.
  4. [Results, linear probing paragraph] The text states that under linear probing 'xECG ranking first in all tasks (p<0.05), except mortality risk prediction and PTB-XL, where it ranked second.' This is contradicted by the Age task results: Supplementary Tables 9 and 10 show ECGFounder achieves lower mean MAE (8.402 vs 8.659) and lower mean SMAPE (0.078 vs 0.081) than xECG under linear probing. Please correct the claim or state the ranking criterion explicitly. Relatedly, the claim that xECG is 'the only publicly available model to perform strongly on all datasets and tasks' needs a quantitative threshold, since xECG is not the best finetuned model on PTB-XL (Table 2), MIT-BIH R-peak (Table 5), Exercise (Table 6), or Age (Tables 9-10).
minor comments (5)
  1. [Methods, MIT-BIH] Typo: 'excude' should be 'exclude' in the description of pacemaker patients.
  2. [Methods, BenchECG Metrics] Typo: 'wich' should be 'which' in the sentence introducing the SMAPE formula.
  3. [Methods, Blood test] The list of blood markers contains duplicates: 'Acetaminophen' appears twice and 'Urea Nitrogen' appears twice. Please clean up the list and ensure the number of markers is consistent with the experiments.
  4. [Methods, Age task] Minor grammar: 'we did not considered patients' should be 'we did not consider patients.'
  5. [Results, fair-comparison note] The sentence in Results saying CODE-15% 'is not used in BenchECG evaluation' is confusing because Table 1 lists CODE-15% as a BenchECG dataset and the Age and Mortality tasks train on it. Clarify that CODE-15% is a training-only dataset, not an evaluation test set.

Circularity Check

1 steps flagged

No formal derivation circularity; the main issue is an unquantified pretraining/finetuning data overlap for two BenchECG tasks that weakens the benchmark's independence premise.

specific steps
  1. other [BenchECG fairness rule / Age and Mortality tasks]
    "For a fair comparison, models that wish to be evaluated on BenchECG should not be pretrained on any of the evaluation datasets (Table 1). Note that the CODE-15% dataset is not used in BenchECG evaluation, but is used during finetuning for the age estimation and mortality prediction tasks. This is because it is commonly used in pretraining of self-supervised methods [8, 9] and is a subset of the CODE dataset [36], used in xECG pretraining."

    xECG is pretrained on CODE, which includes CODE-15%; ST-MEM and ECG-JEPA are pretrained directly on CODE-15%. The Age and Mortality tasks are finetuned on CODE-15%. Thus, for two of the ten BenchECG tasks, the compared models were pretrained on the exact recordings used to fit the downstream heads, while ECGFounder and Supervised xLSTM were not. The paper's own fairness rule excludes models pretrained on evaluation datasets, and CODE-15% is listed in Table 1 as a BenchECG dataset. This is a disclosed but unquantified overlap that makes the ranking on those tasks partly self-referential (pretraining input equals finetuning input), weakening the 'only model to perform strongly on all tasks' claim. It is not a formal derivation cycle, but it is a circularity in the evaluation's independence p

full rationale

The paper is an empirical benchmark study, not a mathematical derivation: there is no equation whose output reduces to a fitted value or to a self-citation. The core comparison uses external public datasets with held-out test folds, and the reported BenchECG score is a mean of standard task metrics. The most significant concern is the CODE-15% overlap: xECG pretraining on CODE includes the exact dataset used for finetuning the Age and Mortality heads, and ST-MEM/ECG-JEPA were pretrained directly on CODE-15%. The authors disclose this but do not quantify its effect, and the paper's limitation section admits that models pretrained on evaluation datasets cannot be fairly compared. This is a benchmark-validity confound rather than a derivation-level circularity, so it contributes only a modest score. No load-bearing self-citations, no imported uniqueness theorems, and no ansatz-smuggling were found.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

This is an empirical benchmark paper, so the central claim rests on modeling and evaluation choices rather than derived constants. The listed free parameters are benchmark design choices and training hyperparameters; the axioms are the assumptions about data, labels, and statistical inference that the comparison depends on.

free parameters (5)
  • R-peak tolerance window = 20 ms
    Chosen post-hoc because the literature-standard 150 ms window yielded near-perfect F1 scores for transformer and xLSTM models; affects detection task rankings.
  • BenchECG task weights = equal mean (1/10 per task)
    The composite score is an unweighted mean of normalized task metrics; the ranking depends on this equal-weighting choice.
  • xECG temporal patch size = 250 ms (25 samples at 100 Hz)
    Hand-chosen tokenization granularity; determines temporal resolution of representations and cost.
  • SimDINOv2 view crop fractions = global 80%, local 40%
    Hand-chosen multi-crop sizes for SSL pretraining; influence what invariances the model learns.
  • Per-model downstream hyperparameters = see Supplementary Tables 13-22
    Learning rates, weight decay, drop path, batch sizes, and patch representations are set per model per task; central comparison assumes each model was tuned fairly.
axioms (5)
  • domain assumption Public dataset labels and clinical annotations are correct
    All benchmark tasks rely on labels from PTB-XL, CPSC2018, MIT-BIH, Apnea-ECG, DeepBeat, MIMIC-IV, and CODE-15% without independent label verification.
  • domain assumption ECG signals contain sufficient information for atypical tasks (age, blood tests, mortality)
    The benchmark assumes these predictions are learnable from ECG alone; prior work supports some but not all markers.
  • domain assumption Resampling to 100 Hz preserves clinically relevant information
    xECG pretraining and evaluation resample all signals to 100 Hz, citing prior work that lower sampling rates do not hurt downstream performance.
  • standard math Welch's t-tests across five runs give valid significance estimates
    Pairwise comparisons use two-sided Welch's t-tests across independent runs; assumes approximate normality and representative seed variation.
  • domain assumption Pretrained weights and preprocessing from ST-MEM, ECG-JEPA, and ECGFounder are used correctly
    The benchmark relies on official pretrained weights and preprocessing from the original papers; any mismatch would bias comparisons.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of BenchECG and xECG: a benchmark and baseline for ECG foundation models." pith.science (2026). https://pith.science/paper/S2RAEY4J

@misc{pith2026250910151,
  author       = {Pith},
  title        = {Pith review of: BenchECG and xECG: a benchmark and baseline for ECG foundation models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S2RAEY4J}},
  note         = {Machine review of arXiv:2509.10151}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Electrocardiograms (ECGs) are inexpensive, widely used, and well-suited to deep learning. Recently, interest has grown in developing foundation models for ECGs - models that generalise across diverse downstream tasks. However, consistent evaluation has been lacking: prior work often uses narrow task selections and inconsistent datasets, hindering fair comparison. Here, we introduce BenchECG, a standardised benchmark comprising a comprehensive suite of publicly available ECG datasets and versatile tasks. We also propose xECG, an xLSTM-based recurrent model trained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score compared to publicly available state-of-the-art models. In particular, xECG is the only publicly available model to perform strongly on all datasets and tasks. By standardising evaluation, BenchECG enables rigorous comparison and aims to accelerate progress in ECG representation learning. xECG achieves superior performance over earlier approaches, defining a new baseline for future ECG foundation models.

discussion (0)

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

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Pretraining Strategies and Scaling for ECG Foundation Models: A Systematic Study

    eess.SP 2026-05 unverdicted novelty 7.0

    Contrastive predictive coding pretraining combined with structured state space models yields the strongest ECG foundation models, with continued gains from scaling data to 11 million samples.

  2. Do ECG Foundation Models Transfer to Rare Cardiac Diseases? Evidence from Brugada Syndrome Detection

    cs.LG 2026-07 conditional novelty 6.5

    For Brugada syndrome detection, ECG foundation-model pre-training mainly stabilizes optimization rather than encoding transferable clinical knowledge, and fails to improve zero-shot cross-site generalization.

  3. Cross-Modal Contrastive Learning of ECG and Angiography Representations for Severe Stenosis Classification

    cs.LG 2026-05 unverdicted novelty 6.0

    StenCE uses cross-modal contrastive learning on paired ECG-angiography data to learn ECG features that classify severe coronary stenosis, reporting the first high performance on this task.

  4. How Do Electrocardiogram Models Scale?

    cs.LG 2026-05 conditional novelty 6.0

    Empirical scaling study of ECG models finds SSL scales robustly while ResNets show 1.3-2.5x better parameter efficiency and SSL up to 16x better data efficiency than supervised baselines on out-of-distribution tasks.

  5. DeepArrhythmia: Segment-Contextualized ECG Arrhythmia Classification via Selective Evidence Acquisition

    cs.LG 2026-05 unverdicted novelty 5.0

    DeepArrhythmia introduces a segment-contextualized multimodal framework for beat-level ECG arrhythmia classification that uses tool-grounded evidence extraction and selective acquisition routed by segment-level confidence.

  6. Echo2ECG: Enhancing ECG Representations with Cardiac Morphology from Multi-View Echos

    cs.LG 2026-03 conditional novelty 5.0

    Multi-view Echo contrastive pretraining yields a compact ECG encoder that outperforms larger SOTA models on structural cardiac phenotype tasks and morphology-aware Echo retrieval.

Reference graph

Works this paper leans on

71 extracted references · 1 canonical work pages · cited by 6 Pith papers

  1. [1]

    Lindstrom, M.et al.Global Burden of Cardiovascular Diseases and Risks Collaboration, 1990-2021.J. Am. Coll. Cardiol.80, 2372–2425 (2022)

    1990

  2. [2]

    Neural Inf

    Beck, M.et al.xLSTM: Extended long short-term memory.Adv. Neural Inf. Process. Syst.37, 107547–107603 (2024)

    2024

  3. [3]

    Wu, Z.et al.Simplifying DINO via Coding Rate Regularization.arXiv preprint arXiv:2502.10385(2025)

    Pith/arXiv arXiv 2025

  4. [4]

    H.et al.Automatic diagnosis of the 12-lead ECG using a deep neural network

    Ribeiro, A. H.et al.Automatic diagnosis of the 12-lead ECG using a deep neural network. Nat Commun11, 1760 (2020)

    2020

  5. [5]

    NEJM AI2, AIoa2401033 (2025)

    Li, J.et al.An Electrocardiogram Foundation Model Built on over 10 Million Recordings. NEJM AI2, AIoa2401033 (2025)

    2025

  6. [6]

    Moor, M.et al.Foundation models for generalist medical artificial intelligence.Nature 616, 259–265 (2023)

    2023

  7. [7]

    & Ding, C

    Han, Y., Liu, X., Zhang, X. & Ding, C. Foundation Models in Electrocardiogram: A Review.arXiv preprint arXiv:2410.19877(2024)

    arXiv 2024

  8. [8]

    & Joo, S

    Na, Y., Park, M., Tae, Y. & Joo, S. Guiding Masked Representation Learning to Capture Spatio-Temporal Relationship of Electrocardiogram.arXiv preprint arXiv:2402.09450 (2024)

    Pith/arXiv arXiv 2024

  9. [9]

    Learning General Representation of 12-Lead Electrocardiogram with a Joint- Embedding Predictive Architecture.arXiv preprint arXiv:2410.08559(2024)

    Kim, S. Learning General Representation of 12-Lead Electrocardiogram with a Joint- Embedding Predictive Architecture.arXiv preprint arXiv:2410.08559(2024)

    Pith/arXiv arXiv 2024

  10. [10]

    AAAI Conf

    Henderson, P.et al.Deep Reinforcement Learning That Matters.Proc. AAAI Conf. Artif. Intell.32(2018)

    2018

  11. [11]

    Pineau, J.et al.Improving Reproducibility in Machine Learning Research(A Report from the NeurIPS 2019 Reproducibility Program).J. Mach. Learn. Res.22, 1–20 (2021)

    2019

  12. [12]

    Roberts, M.et al.Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans.Nat Mach Intell3, 199–217 (2021)

    2021

  13. [13]

    & Narayanan, A

    Kapoor, S. & Narayanan, A. Leakage and the reproducibility crisis in machine-learning- based science.Patterns4, 100804 (2023)

    2023

  14. [14]

    Is AI leading to a reproducibility crisis in science?Nature624, 22–25 (2023)

    Ball, P. Is AI leading to a reproducibility crisis in science?Nature624, 22–25 (2023)

    2023

  15. [15]

    L., Montanari, A

    Bernardini, A., Brunello, A., Gigli, G. L., Montanari, A. & Saccomanno, N. AIOSA: An approach to the automatic identification of obstructive sleep apnea events based on deep learning.Artif. Intell. Med.118, 102133 (2021)

    2021

  16. [16]

    Deng, J.et al.ImageNet: A large-scale hierarchical image database.CVPR248–255 (2009)

    2009

  17. [17]

    & Tuytelaars, T

    Lin, T.-Y.et al.Fleet, D., Pajdla, T., Schiele, B. & Tuytelaars, T. (eds)Microsoft COCO: Common Objects in Context. (eds Fleet, D., Pajdla, T., Schiele, B. & Tuytelaars, T.) Comput. Vis. – ECCV 2014, 740–755

    2014

  18. [18]

    Wang, Y.et al.MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark.NeurIPS(2024). 18

    2024

  19. [19]

    S.et al.The DRAGON benchmark for clinical NLP.npj Digit

    Bosma, J. S.et al.The DRAGON benchmark for clinical NLP.npj Digit. Med.8, 289 (2025)

    2025

  20. [20]

    Ektefaie, Y.et al.Evaluating generalizability of artificial intelligence models for molecular datasets.Nat Mach Intell6, 1512–1524

  21. [21]

    Karargyris, A.et al.Federated benchmarking of medical artificial intelligence with MedPerf.Nat Mach Intell5, 799–810

  22. [22]

    Auer, A.et al.TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning.arXiv preprint arXiv:2505.23719(2025)

    arXiv 2025

  23. [23]

    Kong, Y.et al.Unlocking the Power of LSTM for Long Term Time Series Forecasting. Proc. AAAI Conf. Artif. Intell.39, 11968–11976 (2025)

    2025

  24. [24]

    Bommasani, R.et al.On the Opportunities and Risks of Foundation Models.arXiv preprint arXiv:2108.07258(2021)

    Pith/arXiv arXiv 2021

  25. [25]

    Data7, 1–15 (2020)

    Wagner, P.et al.PTB-XL, a large publicly available electrocardiography dataset.Sci. Data7, 1–15 (2020)

    2020

  26. [26]

    Liu, F.et al.An Open Access Database for Evaluating the Algorithms of Electrocardio- gram Rhythm and Morphology Abnormality Detection.J. Med. Imaging Health Inform. 8, 1368–1373 (2018)

    2018

  27. [27]

    Gow, B.et al.MIMIC-IV-ECG: Diagnostic Electrocardiogram Matched Subset.PhysioNet (2023)

    2023

  28. [28]

    L.et al.PhysioBank, PhysioToolkit, and PhysioNet.Circulation101, e215–e220 (2000)

    Goldberger, A. L.et al.PhysioBank, PhysioToolkit, and PhysioNet.Circulation101, e215–e220 (2000)

    2000

  29. [29]

    Johnson, A.et al.MIMIC-IV.PhysioNet(2023)

    2023

  30. [30]

    Johnson, A. E. W.et al.MIMIC-IV, a freely accessible electronic health record dataset. Sci Data10, 1 (2023)

    2023

  31. [31]

    H.et al.CODE-15%: A large scale annotated dataset of 12-lead ECGs (2021)

    Ribeiro, A. H.et al.CODE-15%: A large scale annotated dataset of 12-lead ECGs (2021). URL https://doi.org/10.5281/zenodo.4916206

    doi:10.5281/zenodo.4916206 2021

  32. [32]

    Moody, G. B. & Mark, R. G. The impact of the MIT-BIH Arrhythmia Database.IEEE Eng. Med. Biol. Mag.20, 45–50 (2001)

    2001

  33. [33]

    & Lin, C.-C

    Chang, H.-Y., Yeh, C.-Y., Lee, C.-T. & Lin, C.-C. A Sleep Apnea Detection System Based on a One-Dimensional Deep Convolution Neural Network Model Using Single-Lead Electrocardiogram.Sensors20, 4157 (2020)

    2020

  34. [34]

    & Atienza, D

    De Giovanni, E., Teijeiro, T., Meier, D., Millet, G. & Atienza, D. ECG in high intensity exercise dataset (2021). URL https://doi.org/10.5281/zenodo.5727800

    work page doi:10.5281/zenodo.5727800 2021

  35. [35]

    & Ashley, E

    Torres-Soto, J. & Ashley, E. A. Multi-task deep learning for cardiac rhythm detection in wearable devices.npj Digit. Med.3, 116 (2020)

    2020

  36. [36]

    Ribeiro, A. L. P.et al.Tele-electrocardiography and bigdata: The CODE (Clinical Outcomes in Digital Electrocardiography) study.J. Electrocardiol.57, S75–S78 (2019)

    2019

  37. [37]

    Neural Inf

    Vaswani, A.et al.Attention is all you need.Adv. Neural Inf. Process. Syst.30(2017)

    2017

  38. [38]

    Schmidinger, N.et al.Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences.arXiv preprint arXiv:2411.04165(2025). 19

    Pith/arXiv arXiv 2025

  39. [39]

    & Karatzas, D

    Kang, L., Fu, X., Vazquez-Corral, J., Valveny, E. & Karatzas, D. xLSTM-ECG: Multi-label ECG Classification via Feature Fusion with xLSTM.arXiv preprint arXiv:2504.16101 (2025)

    Pith/arXiv arXiv 2025

  40. [40]

    & Brandstetter, J

    Alkin, B., Beck, M., P¨ oppel, K., Hochreiter, S. & Brandstetter, J. Vision-LSTM: xLSTM as generic vision backbone.ICLR(2025)

    2025

  41. [41]

    & Hinton, G

    Chen, T., Kornblith, S., Norouzi, M. & Hinton, G. A Simple Framework for Contrastive Learning of Visual Representations.ICML(2020)

    2020

  42. [42]

    He, K.et al.Masked Autoencoders Are Scalable Vision Learners.CVPR16000–16009 (2022)

    2022

  43. [43]

    Assran, M.et al.Self-Supervised Learning From Images With a Joint-Embedding Predictive Architecture.CVPR15619–15629 (2023)

    2023

  44. [44]

    & Hospedales, T

    Ericsson, L., Gouk, H. & Hospedales, T. M. How Well Do Self-Supervised Models Transfer? CVPR(2021)

    2021

  45. [45]

    Caron, M.et al.Emerging Properties in Self-Supervised Vision Transformers.ICCV 9650–9660 (2021)

    2021

  46. [46]

    Oquab, M.et al.DINOv2: Learning Robust Visual Features without Supervision.TMLR (2024)

    2024

  47. [47]

    Med.6, 74 (2023)

    Huang, S.-C.et al.Self-supervised learning for medical image classification: A systematic review and implementation guidelines.npj Digit. Med.6, 74 (2023)

    2023

  48. [48]

    & Zhang, X

    Liu, Z., Alavi, A., Li, M. & Zhang, X. Self-Supervised Contrastive Learning for Medical Time Series: A Systematic Review.Sensors23, 4221 (2023)

    2023

  49. [49]

    & Wang, K

    Xia, Y., Xiong, Y. & Wang, K. A transformer model blended with CNN and denoising autoencoder for inter-patient ECG arrhythmia classification.Biomed. Signal Process. Control86, 105271 (2023)

    2023

  50. [50]

    & Hwang, H

    Sellami, A. & Hwang, H. A robust deep convolutional neural network with batch-weighted loss for heartbeat classification.Expert Syst. Appl.122, 75–84 (2019)

    2019

  51. [51]

    Li, Y., Qian, R. & Li, K. Inter-patient arrhythmia classification with improved deep residual convolutional neural network.Comput. Methods Programs Biomed.214, 106582 (2022)

    2022

  52. [52]

    B.et al.A novel electrocardiogram feature extraction approach for cardiac arrhythmia classification.Future Gener

    Marinho, L. B.et al.A novel electrocardiogram feature extraction approach for cardiac arrhythmia classification.Future Gener. Comput. Syst.97, 564–577 (2019)

    2019

  53. [53]

    & Liu, Z

    Li, F., Xu, Y., Chen, Z. & Liu, Z. Automated Heartbeat Classification Using 3-D Inputs Based on Convolutional Neural Network With Multi-Fields of View.IEEE Access7, 76295–76304 (2019)

    2019

  54. [54]

    Ali, S. T. A., Kim, S. & Kim, Y.-J. Towards Reliable ECG Analysis: Addressing Validation Gaps in the Electrocardiographic R-Peak Detection.Appl. Sci.14, 10078

  55. [55]

    Koscova, Z.et al.The Harvard-Emory ECG database.medRxiv(2025)

    2025

  56. [56]

    Kaplan, J.et al.Scaling laws for neural language models.arXiv preprint arXiv:2001.08361 (2020)

    Pith/arXiv arXiv 2001

  57. [57]

    & Beyer, L

    Zhai, X., Kolesnikov, A., Houlsby, N. & Beyer, L. Scaling Vision Transformers.CVPR 12104–12113 (2022). 20

    2022

  58. [58]

    & Reilly, R

    De Chazal, P., O’Dwyer, M. & Reilly, R. B. Automatic classification of heartbeats using ECG morphology and heartbeat interval features.IEEE Trans. Biomed. Eng.51, 1196– 1206 (2004)

    2004

  59. [59]

    B., Mark, R

    Penzel, T., Moody, G. B., Mark, R. G., Goldberger, A. L. & Peter, J. H. The apnea-ecg database.Computers in Cardiology 2000. Vol. 27 (Cat. 00CH37163)255–258 (2000)

    2000

  60. [60]

    A.et al.Deep learning cardiac motion analysis for human survival prediction

    Bello, G. A.et al.Deep learning cardiac motion analysis for human survival prediction. Nat Mach Intell1, 95–104 (2019)

    2019

  61. [61]

    & Strodthoff, N

    Miguel Lopez Alcaraz, J. & Strodthoff, N. CardioLab: Laboratory Values Estimation from Electrocardiogram Features - An Exploratory Study.2024 Computing in Cardiology Conference(2024)

    2024

  62. [62]

    Strodthoff, N., Lopez Alcaraz, J. M. & Haverkamp, W. MIMIC-IV-ECG-Ext-ICD: Diagnostic labels for MIMIC-IV-ECG (version 1.0.1).PhysioNet(2024)

    2024

  63. [63]

    Dosovitskiy, A.et al.An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale.ICLR(2021)

    2021

  64. [64]

    Bolya, D.et al.Perception Encoder: The best visual embeddings are not at the output of the network.arXiv preprint arXiv:2504.13181(2025)

    Pith/arXiv arXiv 2025

  65. [65]

    & Choi, E

    Oh, J., Chung, H., Kwon, J.-m., Hong, D.-g. & Choi, E. Lead-agnostic Self-supervised Learning for Local and Global Representations of Electrocardiogram.Proc. Conf. Health Inference Learn.338–353 (2022)

    2022

  66. [66]

    Zheng, J.et al.A 12-lead electrocardiogram database for arrhythmia research covering more than 10,000 patients.Sci Data7, 48 (2020)

    2020

  67. [67]

    Zheng, J.et al.Optimal Multi-Stage Arrhythmia Classification Approach.Sci Rep10, 2898 (2020)

    2020

  68. [68]

    & Strodthoff, N

    Mehari, T. & Strodthoff, N. Advancing the State-of-the-Art for ECG Analysis through Structured State Space Models.arXiv preprint arXiv:2211.07579(2022)

    Pith/arXiv arXiv 2022

  69. [69]

    & Dandapat, S

    Jyotishi, D. & Dandapat, S. An Attentive Spatio-Temporal Learning-Based Network for Cardiovascular Disease Diagnosis.IEEE Trans. Syst. Man Cybern. Syst.53, 4661–4671 (2023)

    2023

  70. [70]

    Murugesan, B.et al.ECGNet: Deep Network for Arrhythmia Classification.IEEE MeMeA 1–6 (2018)

    2018

  71. [71]

    & Liu, G

    Shen, Q., Qin, H., Wei, K. & Liu, G. Multiscale Deep Neural Network for Obstructive Sleep Apnea Detection Using RR Interval From Single-Lead ECG Signal.IEEE Trans. Instrum. Meas.70, 1–13 (2021). 21 Supplementary Information Individual BenchECG Results In this section, we report the individual results of each model for each task in BenchECG. Each model is ...

    2021

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.