Multi-Dynamics Analysis of QRS Complex for Atrial Fibrillation Diagnosis

Abstract

This paper presents an effective atrial fibrillation (AF) diagnosis algorithm based on multi-dynamics analysis of QRS complex. The idea behind this approach is to produce a variety of heartbeat time series features employing several linear and nonlinear functions via different dynamics of the QRS complex signal. These extracted features from these dynamics will be connected through machine learning based algorithms such as Support Vector Machine (SVM) and Multiple Kernel Learning (MKL), to detect AF episode occurrences. The reported performances of these methods were evaluated on the Long-Term AF Database which includes 84 of 24-hour ECG recording. Thereafter, each record was divided into consecutive intervals of one-minute segments to feed the classifier models. The obtained sensitivity, specificity and positive classification using SVM were 96.54%, 99.69%, and 99.62%, respectively, and for MKL they reached 95.47%, 99.89%, and 99.87%, respectively. Therefore, these medical-oriented detectors can be clinically valuable to healthcare professional for screening AF pathology.