ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

Summary

ECGFounder (Harvard Medical School / Emory University) is built on the Harvard-Emory ECG Database (HEEDB), currently the largest open-access ECG dataset with 10,771,552 expert-annotated ECGs from 1,818,247 unique subjects. This dwarfs prior ECG FM datasets (ECG-FM: 1.66M, HuBERT-ECG: 9.1M). The model uses real-world cardiologist annotations with 150 diagnostic label categories — a supervised pretraining paradigm rather than self-supervised, distinguishing it from most other FMs.

Reported benchmarks