Bivariate markov model based analysis of ecg for accurate identification and classification of premature heartbeats and irregular beat-patterns

Abstract

. This paper describes a novel intelligent analysis technique based upon bivariate Markov model that integrates morphological and temporal features with a rule-based interval analysis of ECG signals to localize and accurately classify the premature beats to four major classes: (1) Premature Atrial Complex (PAC), (2) Blocked PAC (B-PAC), (3) Premature Ventricular Complex (PVC), and (4) Premature Junctional Complex (PJC). The paper also describes a beat-pattern classification algorithm to sub classify premature beat-patterns into bigeminy, trigeminy and quadrigeminy. The approach utilizes two phases: (1) a training phase that builds bivariate Markov model from standardized databases of ECG signals, and (2) a dynamic phase that detects embedded P and R waves in T-waves of premature beats using a combination of area subtraction and clinically significant rule-based analysis of R-R intervals. It detects and classifies premature beats using graph matching based upon the forward-backward algorithm and performs a look ahead pattern analysis for the sub-classification of beat-patterns. The algorithms have been presented. The software has been implemented that uses a combination of MATLAB and C++ libraries. Performance results show that processing time is realistic for real-time detection with 98%–99% sensitivity for the premature beat classification and 95%-98% sensitivity for the beat pattern identification.