Pattern recognition of cardiac arrhythmias based on multivariate autoregressive modeling

Computer-assisted automatic diagnosis will play an important role in the diagnosis and treatment of critically ill patients. Multivariate autoregressive modeling (MAR) has been performed on two-lead ECG signals. MAR coefficients and K-L transformation of MAR coefficients have been used as ECG features for classification. Five types of ECG signals were obtained from the MIT-BIH database, namely normal sinus rhythm, atria premature contraction, premature ventricular contraction, ventricular tachycardia, and ventricular fibrillation. A quadratic discriminant function (QDF) based classification algorithm was employed. The results show MAR coefficients produced slightly better results than K-L transformation of MAR coefficients. The accuracy of classification based on MAR coefficients was 96.6% to 99.3%.