STC ECG Classification

ECG is the routine method in cardiac health examination and the need for intelligent diagnostic approaches for ECG is becoming more urgent at this stage. Wearable ECG device can continuously monitor the health status of patients for 24h, and instantly alerting cardiologists to arrhythmia and requesting intervention, which can greatly improve patient survival and prognosis. ST segment change (STC) on the ECG can be classified as ST segment elevation (STE) or ST segment depression (STD), indicating the possibility of myocardial infarction and myocardial ischaemia respectively. The real-time recognition and detection of STC on the ECG is therefore of great importance.STE and STE are not mutually exclusive and may occur simultaneously in different leads of the same ECG. All ECGs are 12-lead, digital sampled at 500 Hz and the time length of each signal is 15s. The data were saved as “. mat”, and the label information present in each file and finally summarised in an Excel table in the folder. The data set consisted of 7000 ECGs, with a training set consisting of 4000 ECGs for training (containing 180 STEs, 180 STDs, 180 simultaneous STEs and STDs, and 3460 other samples, which are open-sourced), and two hidden test sets with1000 and 2000 ECGs respectively (for evaluation of the algorithm only, which not be public).The ratio of positives to total samples in this dataset is approximate 1:11, which is a typical unbalanced multi-label classification task. The aim of this task is to mine the information and knowledge contained in signals, and to improve the sensitivity of positives while minimising false alarms. The use of external ECG dataset is therefore prohibited, and researchers are prohibited from inviting cardiologists to correct labels. The recommended evaluation indicators is Area Under the Precision-Recall Curve (AUPRC).