Machine Learning Based Decision Support System for Atrial Fibrillation Detection using Electrocardiogram

Abstract

Atrial Fibrillation (AF) is a common sustained arrhythmia encountered in regular clinical practice. In order to diagnose AF, Electrocardiogram (ECG) is used in correlation with clinical symptoms. ECG is noninvasive and cost effective modality in order to diagnose cardiac abnormalities using AF. The complexity of ECG and its interrelationship with other physiological parameters make the AF detection a challenging task in the clinical practice. The traditional practice of diagnosing AF manually by the physician can cause intra physician variability leading to a need for automated algorithm based assisting system to detect AF. In the present methodology, the QRS complex is detected and each beat in the entire signal is segmented, the median beat is calculated for a given signal, the dimensionality is reduced using Principal Component Analysis (PCA) and the resultant components along with energy values are used for classification using decision tree. The methodology provided an improved average accuracy of 85.1 percent which is reasonably high. The system developed can be used in many practical applications and can provide acceptable results in clinical implementations. The developed methodology can be used as an adjunct tool by the physician in his clinical practice.