P-QRS-T localization in ECG using deep learning

Abstract

This paper describes a work using the capabilities of deep neural networks to predict key wave locations in a cardiac complex on an electrocardiogram (ECG) as part of a challenge introduced by Physionet, a provider of ECG collections, on detecting critical waveforms that contain essential information in cardiology. The key waves include P-wave, QRS-wave, and T-wave. Recent attempts to extract hierarchical features of cardiac complexes have been reported in literature, but finding the accurate position of critical cardiac waves has been a challenge in the ECG signal processing research. This study investigates multiple architectures and learning rates of the deep neural networks and adopts a four-step procedure to find the best one that can predict the wave locations. A remarkable rate of 96.2% of accuracy in the localization task has been achieved. This study consists of four parts to produce output predictions; obtaining the cardiac complexes from QT Databse (QTDB); introduce multiple architectures, including fully-connected networks, LeNet-style ConvNet with dropout, LeNet-style ConvNet without dropout and train these networks; use an unseen test set to calculate the accuracy of the system with different tolerance in each wave interval; compare all these architectures together to analyze the most suitable architecture for this task.