Pattern recognition of cardiac arrhythmias using scalar autoregressive modeling

Abstract

Arrhythmia classification is introduced for automatic diagnosis and treatment of cardiac diseases. Scalar autoregressive (AR) modeling was performed on two-lead electrocardiogram (ECG) signals to extract features. AR coefficients were estimated from each channel and concatenated together to form the ECG features. Five types of ECG signals were obtained from MIT-BIH database including normal sinus rhythm, atria premature contraction, premature ventricular contraction, ventricular tachycardia and ventricular fibrillation. A stage-by-stage quadratic discriminant function (QDF) based classification algorithm was employed. The results show two ECG lead based classification can obtain better results than that of single ECG lead. The accuracy of classification based on two ECG leads is over 98.3%.