Paperdoi:10.64898/2026.07.15.26358182
Scaling ECG Foundation Models and Identifying a Threshold for Effective Representation Learning
medRxiv preprint v1, posted 2026-07-17; CC BY 4.0
TL;DR
Data-scale sweep with everything else fixed. A 1-D ViT masked autoencoder (60% random masking, 50-sample patches) is pretrained on MIMIC-IV-ECG subsets of 100k, 200k, 400k, 600k and 800k ECGs. Input is 8 leads (I, II, V1–V6) × 10 s at 500 Hz. Each model is fine-tuned and compared with the same architecture trained from random init. Downstream: PTB-XL rhythm, form and diagnostic labels; EchoNext Mini structural heart disease.
- Form factor: clinical
Abstract
We conducted a scaling evaluation of unlabeled pretraining for electrocardiogram foundation model performance. One-dimensional vision transformer (1D-ViT) masked autoencoders were pretrained across increasing ECG volumes and fine-tuned for rhythm, morphology, diagnostic, and structural heart disease tasks. Models pretrained at ≤400,000 ECGs failed to consistently exceed controls without self-supervised pre-training, whereas 600,000–800,000 ECGs improved AUROC across tasks, suggesting a minimum threshold for effective ECG representation learning.
Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: http://creativecommons.org/licenses/by/4.0/
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Signals
Keywords
Datasets
MIMIC-IV-ECGDiagnostic Electrocardiogram Matched Subset (v1.0)160,000 subjects · ECG
PTB-XL21,799 clinical 12-lead ECGs (PhysioNet)18,869 subjects · ECG
EchoNext-Mini100,000 clinical 12-lead ECGs (250 Hz) with echocardiogram-confirmed structural heart disease labels (Columbia, PhysioNet restricted)36,286 subjects · ECG
How it works
Cite this paper
APA
Sriram, R., Nenadic, I., Shahrabani, E., Goonewardena, S., Yao, S., Farrell, B., Loring, Z., & Murthy, V. L. (2026). Scaling ECG Foundation Models and Identifying a Threshold for Effective Representation Learning. medRxiv preprint v1, posted 2026-07-17. https://doi.org/10.64898/2026.07.15.26358182
BibTeX
@misc{sriram2026scaling,
title = {{Scaling ECG Foundation Models and Identifying a Threshold for Effective Representation Learning}},
author = {Sriram, R. and Nenadic, I. and Shahrabani, E. and Goonewardena, S. and Yao, S. and Farrell, B. and Loring, Z. and Murthy, V. L.},
year = {2026},
howpublished = {medRxiv preprint v1, posted 2026-07-17},
doi = {10.64898/2026.07.15.26358182}
}Built from the hub record. Check it against the publisher before submitting.
Added to the ChatPPG research hub
