Deep learning based myocardial ischemia detection in ECG signals

Electrocardiograms (ECG) are used for monitoring and diagnosing the cardiac electrophysiology of a patient and have various medical uses. The recent advancements offered by deep learning-based data-driven approaches for clinical decision support have led to the development of competitive solutions, typically focusing on diagnosis and treatment planning. Herein, we introduce and evaluate different deep learning-based approaches for myocardial ischemia detection on ECG signals. A publicly available large dataset containing both ECG traces and annotations was employed. Three models were evaluated: convolutional neural network (performing binary classification), an autoencoder (performing outlier detection to identify pathological heart beats), and a hybrid model. The classification accuracies of the three models were 85.7%, 75.0%, and 88.6% respectively.