The Harvard-Emory ECG Database

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• The Harvard-Emory ECG Database (HEEDB) is the largest open-access repository of 12-lead electrocardiogram (ECG) recordings.
• It comprises over 10 million ECGs from nearly 2 million patients across two major academic medical centers.

Purpose of the Study:

• To introduce the Harvard-Emory ECG Database (HEEDB) as a comprehensive resource for cardiovascular research.
• To facilitate the training and validation of machine learning algorithms for ECG analysis.

Main Methods:

• The database integrates 10,608,417 ECG recordings from Massachusetts General Hospital and 998,844 from Emory University Hospital.
• ECGs are 10-second, 12-lead recordings (250/500 Hz, WFDB format) linked with demographic and clinical data.
• Includes Marquette 12SL Software annotations, physician overreads, and ICD codes for comprehensive analysis.

Main Results:

• HEEDB provides a large-scale, multi-center dataset with extensive metadata and diagnostic annotations.
• The dataset enables robust evaluation of machine learning model performance and potential biases.

Conclusions:

• The HEEDB is a valuable, open-access resource for advancing AI in cardiology.
• It supports the development of more accurate and equitable diagnostic tools using electrocardiograms.