Deep Learning Based Atrial Fibrillation Detection Using Effective Denoising Methods and Dimensionality Reduction Techniques

Abstract

Atrial Fibrillation (AF) is one of the most common heart rhythm disorder observed by the physician on a daily basis. Automatic detection of AF is one of the major challenges in the field of heart arrhythmia. In this paper we propose an algorithm to classify Electrocardiogram (ECG) signal into three classes namely Normal, AF and other rhythms. Three different methods namely Discrete Wavelet Transform (DWT), Butterworth filter and Savitzky-Golay filter are used separately to remove high frequency noise and baseline wander. Two dimensionality reduction techniques namely Principal Component Analysis (PCA) and Independent Component Analysis (ICA) are used for feature extraction. Finally class specific accuracy of three classes viz: normal, AF and other rhythms are calculated using Decision Tree (DT) and Deep Convolutional Neural Network (DCNN) classifier separately. DWT method combined with ICA and DCNN classifier provided improved performance of 91.71% as overall accuracy which is higher compared to other methods. The proposed method can be used for mass screening in hospitals for detecting cardiac abnormalities