CardiAWARE: A Novel ECG-based Deep Neural Network Algorithm for Early Detection of Cardiac Conditions

Electrocardiography (ECG) is a widely accessible and simple test that is used for initial screening of many cardiac disease states. Using machine learning methods for automated interpretation of ECG can not only lead to enhanced diagnostic accuracy but also provides a cost and time-effective alternative to physician interpretation. Thus, the goal is to create a neural network model which can process an ECG signal and detect the presence of multiple cardiac conditions. In this model, we aim to detect myocardial infarction, conduction disturbance, hypertrophy, or classify the ECG as normal. Our best model, the ResNet-50 network, has an accuracy of 97.41% over the training sample. By incorporating machine learning tools into the rapid and accurate detection of heart conditions, we increase access to quality and equitable health care and thereby reduce associated cardiac mortality and morbidity.