Atrial fibrillation detection through heart rate variability using a machine learning approach and poincare plot features

Abstract

Atrial Fibrillation (AF) is a common cardiac arrhythmia, and it has a high rate of morbidity and mortality. In this paper, an algorithm for automatic AF episodes detection based on novel low computational cost features is proposed. The features are based on Poincare plots calculated from heart rate variability signal. A supervised classification technique, Support Vector Machines, optimized with Particle Swarm Optimization, was implemented. The data was obtained from MIT-BIH Atrial Fibrillation and Normal Sinus Rhythm Databases. This method shows an accuracy of 92.9% to detect AF spontaneous episodes in signals from AF patients, and 97.8% to classify between AF episodes from AF patients and episodes from subjects with normal sinus rhythm. The proposed method can be employed in real time applications due to its performance as well for its low computation time around 8.8 ms per episode.