Automatic detection of PVCs using autoregressive models

Abstract

An algorithm is presented for the identification of ventricular ectopic beats (PVC) by performing a beat by beat analysis of the electrocardiogram (ECG). The discrete cosine transform (DCT) of a windowed ECG cycle is decomposed into spectra of system and excitatory functions representing action potential and excitation pattern of the heart muscle during the cardiac cycle. The autoregressive (AR) modeling of the system function provides necessary information for identification of PVCs. The partial energy spectrum derived from the DCT coefficients characterises the decay rate of DCT of the system function and is related to bandwidths of resonances in the AR spectrum. The algorithm was able to successfully identify PVCs from the recordings of MIT-BIH database. Under noisy conditions, the algorithm clearly distinguishes PVC patterns from those of normal beats up to a signal to noise ratio (SNR) of 10 dB.