ECG Noise Removal and Efficient Arrhythmia Identification Based on Effective Signal-Piloted Processing and Machine Learning

Abstract

For a viable classification of electrocardiogram (ECG) signals, a signal-piloted adaptive rate processing approach is suggested for the efficient reduction of noise and extraction of features. By using an adaptive rate wavelet decomposition scheme, recognizable features are derived from the preconditioned signal. These attributes are then analyzed for arrhythmia recognition. By using a known arrhythmia, MIT-BIH, database, the output of the framework is studied. It is demonstrated that the system is able to adapt its parameters by analyzing the incoming signal variations. It permits the processing of a lower dimension dataset, for arrhythmia recognition, by the computationally efficient adaptive-rate denoising and subbands decomposition stages. This results in a major decrease in the system's computational costs. The amount of information, required to be sent to the health server is also drastically diminished. This aptitude shows a measurable decrease in the activity of data transmission and processing load of the post classifier. Moreover, the classification performance of the devised method is tested. Findings demonstrated a good performance by achieving 99.3 percent accuracy.