Frequency-domain analysis of ventricular late potentials through time-variant autoregressive models

In this paper we propose an analysis of the high resolution ECG signal in the frequency domain which is focused on the ending part of the QRS complex and on the ST segment of the averaged modulus of the vectorcardiogram, in order to detect the presence of ventricular late potentials (VLP's). The influence of the QRS is excluded through a Tchebytcheff polynomial interpolation. Such a 'filtered' signal is then analyzed by means of a time variant autoregressive model that allows to obtain a power spectral density in correspondence of each sample and thus to track the dynamical changes in its frequency content. A joint time-frequency representation is obtained which may qualitatively discriminate between populations with or without VLP's. Finally, some parameters are proposed, such as the time duration, the associated power, the starting and ending points and the main frequency of the VLP's, through which it is possible to quantify the abnormal signal averaged ECG.