Frequency Domain Features Based Atrial Fibrillation Detection Using Machine Learning And Deep Learning Approach

Abstract

Atrial Fibrillation (AF) is a serious heart disease which can be diagnosed using Electrocardiogram (ECG). Early detection of AF is very important so that the morbidity and mortality can be reduced and the patient can have quality life. This paper proposes frequency domain analysis using power spectrum estimation using Welch, Auto Regression (AR) and Lomb scargle peridiogram methods. The single lead ECG recordings obtained from Physionet Challenge 2017 dataset are used for the analysis. The deep learning methods such as Temporal Convolution Network (TCN) and Deep Convolutional Neural Network (DCNN) are compared with traditional machine learning methods such as Decision Tree (DT) and Support Vector Machine (SVM). The DCNN provided improved performance of 94.67% of accuracy which is superior compared to other methods. The proposed methodology can be used in practical hospital applications as adjunct/assisted tool for the physician.