A large-scale 12-lead electrocardiogram dataset for acute coronary syndrome prediction containing 19,955 ECGs.

Acute coronary syndrome(ACS) is a common cardiovascular disease and a severe type of coronary heart disease. Electrocardiograms(ECGs) are the initial and indispensable examination for patients suspected of ACS in emergency treatment. Particularly in patients with occlusion myocardial infarction(OMI), prompt medical intervention is imperative. Although ECG-based artificial intelligence(AI) models have achieved high diagnostic accuracy in conditions such as atrial fibrillation and heart failure, their application to ACS prediction has been limited by the scarcity of large, openly available ECG datasets. In this study, we present a comprehensive 12-lead ECG dataset for ACS prediction, comprising 19,955 ten-second recordings sampled at 500 Hz from 18,909 patients-all of whom underwent digital subtraction angiography. The dataset includes labels for ST-elevation myocardial infarction, non-ST-elevation myocardial infarction, unstable angina, OMI, and infarction locations. Two physicians also assessed ACS and OMI identification, providing clinical diagnostic benchmarks. Furthermore, we developed and open-sourced a baseline deep learning model for OMI detection. This database and baseline model offer valuable resources for advancing AI research in ACS.