Atrial Fibrillation Pattern Recognition using Features of Second Order Dynamic System

According to World Health Organization (WHO), an estimated 17.9 million people died from cardiovascular diseases (CVD) in 2019, representing 32 % of all global deaths. Of these deaths, 85 % were due to heart attack and stroke. The occurrence and prevalence of atrial fibrillation (AF) is growing worldwide. Limited tools are available to evaluate clinical outcomes and response to thrombolysis in stroke patients with AF. Therefore, this study analysed the ECG features of AF and the normal sinus rhythm (NSR) signals for AF recognition. The first objective is to extract AF features using second-order dynamic system (SODS) algorithm. The following objective is to investigate the effect of windowing length towards AF classification. Next, to compare the two-pattern recognition machine learning support vector machine (SVM) and artificial neural network (ANN) on the accuracy, specificity, and sensitivity of AF classification. In this study, the Physiobank database, included MITBIH Atrial Fibrillation Dataset and MITBIH Normal Sinus Rhythm Dataset were used. For signal pre-processing, butterworth filter are used to diminish the muscle noise and the features signals were extracted. Multiple episodes of the windowing size of 2s, 4s, 6s, 8s and 10s were included in this design to evaluate the appropriate windowing size for AF recognition. The pattern recognition machine learning of SVM algorithm has higher accuracy compared to ANN, which are 100 % with 4s windowing size. In conclusion, the 4s windowing size having the highest detection rate in AF classification system.