Hidden Markov Models in Wavelet Analysis

The paper deals with a mathematical model using a structure designed for an application of pattern recognition in ECG signals. The overall process of recognition is consist of generation of a code book, vector quantization, Markov model learning, and recognition. A Hidden Markov Model (HMM) structure with vector–valued observation sequences can be used for the characterization of cardiac arrythmias and other irregularities in multiple–lead ECG recordings. 1. INTRODUCTION Hidden Markov models are widely used in speech recognition [1,2] and biosignal analysis [3,4] and pattern recognition [5]. The fundamental hypothesis used in the project follows the idea of wave-shape changes during pathological states of the heart [6]. Recorded data are processed by wavelet transform to obtain time-frequency patterns. The patterns are used to learn hidden Markov model. Then, the model serves as a recognition engine. The states of the HMM are defined by waveform events of the ECG. Thus for a normal ECG the states S 1 to S N of the HMM are respectively the isoelectric, P, PR, R, S, ST, T waves and finally the isoelectric line again [7].