beat-age

Beat-age

This is the official checkpoint release for the paper:

Beat-Level Electrocardiographic Biological Age and Its Variability as Digital Biomarkers for Cardiovascular Risk Stratification

Official GitHub repository: https://github.com/chiangfish/beat-age

Files

• v1_best.pth: Beat-age beat-level Net1D checkpoint trained on the UK Biobank Development Cohort.
• ckpt_manifest.json: checkpoint metadata, including file size, SHA-256 checksum, architecture, and intended use.

Model

Beat-age is a beat-level ECG biological age model. It predicts biological age from individual segmented 12-lead cardiac cycles and aggregates beat-level predictions at the ECG-recording level.

• Architecture: one-dimensional residual CNN (Net1D)
• Input: segmented 12-lead ECG beats
• Output: predicted biological age in years
• Age gap: predicted age minus chronological age

Usage

Download the checkpoint and place it under ckpts/ in the GitHub repository:

mkdir -p ckpts
hf download chiangfish/beat-age v1_best.pth --local-dir ckpts

Then run the inference scripts from the GitHub repository following its README.

Data

The model was developed using controlled-access UK Biobank ECG data. Downstream external validation used MIMIC-IV-ECG. These datasets are not redistributed in this model repository.

Citation

@article{beatage2026,
  title = {Beat-Level Electrocardiographic Biological Age and Its Variability as Digital Biomarkers for Cardiovascular Risk Stratification},
  author = {Zirui Jiang, Guangkun Nie, Qinghao Zhao, and Shenda Hong},
  year = {2026}
}