Classification of electrocardiogram using hidden Markov models

Abstract

The objective of this project is to develop models for the characterization of electrocardiogram (EGG). A fast and reliable QRS detection algorithm based on a one-pole filter has been developed. Automatic ECG classification using hidden Markov models (HMMs) is investigated. Models representing various types of beat are trained using the American Heart Association (AHA) ventricular arrhythmia ECG data. The types of beat being selected in the study are: normal (N), premature ventricular contraction (V), and fusion of ventricular and normal beats (F). Artificial ECG generated from the model shows that each model truly characterizes that particular type of beat. In the testing phase, ECG signals are classified using the trained models. The average classification accuracy is 93% for N beat, 65.55% for V beat, and 56.38% for F beat respectively.