Heart disease recognition based on extended ECG sequence database and deep learning techniques

Mortality caused by cardiovascular diseases (CVDs) has been steadily increasing over the years. For this reason, numerous studies have addressed this issue, introducing innovative techniques for automatic detection of heart disease using ECG/PCG signals and convolutional neural networks (CNNs). The present paper proposes a system for automatic diagnosis of heart disease (five pathology classes) using electrocardiogram (ECG) signals and CNNs. Specifically, ECG signals are passed directly to an appropriately trained CNN network. The database comprises a combination of two public datasets: MIT-BIH Arrhythmia and MIT-BIH Atrial Fibrillation database. The results obtained from testing the network show average classification accuracy of about 93% when a 2second ECG signal is fed to the network; conversely, applying a post-processing filter results in about 100% accuracy after around 38 seconds.