ECG
ECG Datasets: Signals, Labels and Sourcing
By GetDATA Team
An overview of 12-lead ECG datasets, common arrhythmia labels, and signal formats used in machine-learning research.
Electrocardiogram (ECG) datasets
The electrocardiogram records the heart's electrical activity over time, sampled in millivolts across multiple leads. A standard resting study captures 12 leads; wearable and Holter studies often record a single lead such as lead II for rhythm analysis.
What a good ECG dataset includes
- Raw waveforms in WFDB or EDF, with sampling rate and lead configuration documented.
- Expert rhythm and arrhythmia labels (e.g. atrial fibrillation, sinus rhythm, ST-segment changes).
- Demographics and acquisition metadata, fully de-identified.
Researchers use GetDATA to request balanced ECG cohorts by rhythm class, while hospitals contribute de-identified recordings that match the request.