Wavelet analysis techniques and Hidden Markov Model for classification of cardiac activity

Abstract

Non-invasive diagnostic procedures in medical applications stimulated a remarkable increase in advanced signal processing techniques and sophisticated algorithms. This study explores time-frequency methods for pattern recognition in the context of detecting and classifying high-risk patients prone to ventricular tachyarrhythmia. Cardiologists have focused their interest in determining what actually provokes the onset of ventricular fibrillation, which is almost always preceded by an episode of ventricular tachycardia. In order to retain both temporal and spectral information, a Continuous Wavelet Transform (CWT) algorithm was considered as a potential tool for discriminating between normal sinus rhythm, ventricular tachycardia and fibrillation. The results obtained from these tests were highly accurate in discriminating between Ventricular Tachycardia and Ventricular Fibrillation with eight-seconds segments having an average accuracy of 95%. The Hidden Markov Model proved to be ideal in accounting for the probabilistic nature of the observed dataset. By adjusting the scale used for the wavelet transform coefficients also increased the reliability of the detector increased. This allowed for better modelling of the actual ECG characteristics. This study sought to determine the early detection of such a phenomena, in order to eventually help in reducing the number of patients who are predisposed to sudden cardiac death.