ECGFounder: An Electrocardiogram Foundation Model Built on over 10 Million Recordings with External Evaluation across Multiple Domains

This is the official implementation of our paper "An Electrocardiogram Foundation Model Built on over 10 Million Recordings with External Evaluation across Multiple Domains".

Authors: Jun Li, Aaron Aguirre, Junior Moura, Jiarui Jin, Che Liu, Lanhai Zhong, Chenxi Sun, Gari Clifford, Brandon Westover, Shenda Hong.

🚀 Getting Started

🚩 News (Aug 2025): The out-of-the-box feature — the 150-class classification validation function of ECGFounder — is now online.

(Mar 2025): The pre-training checkpoint is now available on 🤗 Hugging Face!

⚠️ Important Notice
If you intend to use the pretrained model weights for validation or fine-tuning, you must strictly follow the preprocessing steps in dataset.py — including filtering, z-score normalization, and any other specified procedures.
Failure to do so will make it difficult to reproduce the results reported in the paper!

Installation

To clone this repository:

git clone https://github.com/PKUDigitalHealth/ECGFounder.git

Environment Set Up

Install required packages:

conda create -n ECGFounder python=3.10
conda activate ECGFounder
pip install -r requirements.txt

150-class classification validation

• PTB-XL: Please download the PTB-XL dataset from physionet.

Next, please download the model's checkpoint from the 🤗 Hugging Face. And place the model weights in path ./checkpoint

You can run the ptbxl_eval.py to do the 150-class classification validation on PTB-XL dataset.

Fine-tune on Downstream Tasks

In our paper, downstream datasets we used are as follows:

• MIMIC-ECG: Please download the MIMIC-ECG dataset from physionet.

You can run the jupyter notebook to finetune the model by the example dataset.

References

If you found our work useful in your research, please consider citing our works at:

@article{li2025electrocardiogram,
  title={An Electrocardiogram Foundation Model Built on over 10 Million Recordings},
  author={Li, Jun and Aguirre, Aaron D and Junior, Valdery Moura and Jin, Jiarui and Liu, Che and Zhong, Lanhai and Sun, Chenxi and Clifford, Gari and Brandon Westover, M and Hong, Shenda},
  journal={NEJM AI},
  volume={2},
  number={7},
  pages={AIoa2401033},
  year={2025},
  publisher={Massachusetts Medical Society}
}