Deep Learning assisted tool for Atrial Fibrillation detection using RR Intervals

Abstract

Atrial Fibrillation (AF) is a life-threatening heart rhythm disorder. AF diagnosis is very essential and important aspect for healthcare professionals. Early detection of AF using Electrocardiogram (ECG) plays an important role in the clinical practice. Manual interpretation of ECG signals to detect AF is time-consuming and needs higher expertise, and it is subject to variability among experts. Detecting AF in a timely and effective manner still remains a difficult challenge. In this paper, we propose a Deep Learning (DL) based AF detection method using Physionet challenge 2017 dataset. VGG16 architecture is used for the classification purpose. With the help of Discrete Wavelet Transform (DWT) the ECG signals are denoised. The RR intervals are computed and are subjected to VGG16 for classification. The class specific accuracies of normal, AF, and other rhythms are calculated. The proposed method achieves overall accuracy of 97.60%. The proposed method can be used as an assisted tool by the physician in their clinical practice.