Using hidden Markov toolkit for arrhythmia recognition

This paper describes a recognition system based on diverse features combination for the automatic heartbeat recognition purpose. The method consists of three stages: at the first stage, we extract a set of features including the morphological ones, high order statistics and pitch synchronous decomposition from ECG data using QT database; at the second stage, we use the hidden Markov tree classifier, then the third stage is added as a tool on which we have implemented the hidden Markov tree. The classification accuracy of the proposed system is measured by sensitivity and specificity measures. These measures for average sensitivity and average specificity are 95,79%, 98,93% in case of separated features and 97,46%, 99,22% in case of combined features.