Wavelet Packet Descriptors for Physiological Signal Classification

Signal processing plays an indispensable role in the diagnosis of cardiac and pulmonary disorders. This study identifies a new set of features to facilitate in the classification of physiological signals examined for cardiac and pulmonary disorders. Accordingly, electrocardiogram (ECG), phonocardiogram (PCG) and respiratory signals are considered for this study. The feasibility of utilizing auto regression coefficients, Shannon entropy and multifractal estimates derived from maximal overlap discrete wavelet packet transform (MODWPT) coefficients of the signal as features for signal classification is explored in this study. These descriptors, by the virtue of time-invariance property of MODWPT due to the absence decimation in decomposition, characterize the raw signal very prominently and demonstrated a phenomenal performance with a classification accuracy of 97.96%, 88.86% and 89.62% respectively for ECG, PCG and respiratory signals.