Adaptive Predictive Control-Based Noise Cancellation With Deep Learning for Arrhythmia Classification from ECG Signals

Abstract

Arrhythmia is a disorder in the heart produced due to irregular electrical activities of the heart, and an electrocardiogram (ECG) represents a modality used by clinicians to discover arrhythmias. The occurrence of noise in ECG and irregular heartbeat is the complexity faced while diagnosing arrhythmias. Thus, there is a requirement for a technique that can attain high accuracy in arrhythmia classification. This paper utilizes the Deep Maxout network (DMN) for classifying arrhythmia using ECG signals. Here, the ECG signal is considered as an input to adaptive prediction noise control wherein the noise cancellation is performed using adaptive Least Mean Square (LMS) to discard noise and obtain a clean signal. From a clean signal, the features, like Empirical Mode Decomposition (EMD), wave component detection, wave features, such as RR interval, QT interval, PR interval, PP interval, and R peak whereas statistical features, such as kurtosis, eccentricity are mined for improved analysis. At last, the classification of arrhythmia is done using DMN. The greatest accuracy of 92.1%, the sensitivity of 92.6%, and the specificity of 91.9% are measured using DMN.