Comparative Performance of Various Wavelet Transformation for the Detection of Normal and Arrhythmia ECG Signal

Abstract

Cardiac Activity forms a signal of electrical potential waves in the heart that can be recorded using an Electrocardiogram (ECG). The results of the ECG signal can determine the conditions and abnormalities experienced by the heart, such as arrhythmias. Medical personnel diagnoses normal and arrhythmia heart conditions by looking at R peaks and R-R interval features. Normal conditions have regular R peaks and R-R intervals, whereas arrhythmias are irregular. The challenges in diagnosing ECG signals are that sometimes the signal has some noises that need reducing noise (denoising) are not required in the signal so it can be easier to detect abnormalities. This paper is a brief study of the comparison of the best performance in detecting ECG signals using various wavelet transforms and optimal threshold values based on empirical methods to obtain R peaks and R-R interval features. Wavelet transform describes the signals that can compress the ECG signal and reduce noise without losing important clinical information that can be achieved by medical personnel. The wavelet transform is suitable for approaching data with a discontinuity signal, so the frequency component will increase if noise or anomalies occur in the ECG signal. The various wavelet transforms used Daubechies (db4), Symlets (sym4), Coiflets (coif4), and Biorthogonal (bior3.7) with four types of Detail and Approximate levels; they are Level 1, 2, 3, and 4. The comparison result for the best performance of the various wavelet transforms is using Daubechies wavelet, and biorthogonal wavelet with an accuracy percentage of 100% at level 2 for diagnosing arrhythmia and 93.1% at level 1 for normal diagnosis from 31 data for arrhythmia and 18 for Normal sourced of the MIT-BIH Database. Hence, the total accuracy results obtained from all the data tested is 96.55%.