PÉEK: A cloud-based application for automatic electrocardiogram pre-diagnosis

Abstract

Electrocardiogram (ECG) visual analysis is a common task performed by a healthcare specialist as a cardiovascular-diseases pre-diagnostic technique. However, when an ECG specialist analyzes long-time duration records (such as a 24-h Holter), the task becomes tiresome, complicated, and a probable erroneous diagnostic. This article presents a cloud-based application called PÉEK that helps healthcare specialists automatically detect normal and abnormal heartbeats on ECG registers using Stationary Wavelet Transform (SWT) and Convolutional Neural Networks (CNN). In order to illustrate the functionality of PÉEK, we present the analysis of set ECG traces from the MIT-BIH Arrhythmia Database. This software can detect with a 99.09% accuracy normal heartbeats, premature ventricular contractions (PVC) beats, and others (Premature Atrial Contraction, Left Bundle Branch Block, and Right Bundle Branch Block).

Keywords

• Arrhythmia
• Electrocardiogram
• ECG
• Deep learning
• Convolutional neural networks