Auto-detection of Atrial Fibrillation with Improved Classification and Noise Removal Algorithm along with Dimensionality Reduction Methods

Abstract

Atrial fibrillation (AF) is an irregular manner of the heart rhythm commonly called arrhythmia. Most of the cases this type will associated with significant mortality. It is important to diagnosis at early stage to minimize this consequence. This type of timely diagnosis of arrhythmia is difficult since patients may be asymptomatic. In this study, we describe a robust algorithm for the automatic detection of AF with effective noise removal technique. Proximal splitting-based noise removal method was used to evade noise from the signal. Next, 19 features were extracted from the denoised signal, which included features like RR interval, R peak, P wave morphology, power and spectrum. Application of raw extracted feature directly to the classifier reduces its efficiency. The classifier, quadratic Renyi entropy feature selection method and dimensionality reduction algorithm using principle component analysis (PCA) used to improve the performance. Then the reduced feature set is applied in the Support Vector Machine (SVM) classifier where the samples were classified into normal signals and AF signals. Analysis of the performance of the classifier indicated an accuracy of 97.62%in detecting the AF signals.