Feature extraction of high-frequency patterns with the a priori unknown parameters in noised electrograms using spectral entropy

Abstract

Statement of the problem: the article looks into a class of problems which require identification of hidden regularities in adjustment of the bioelectrical activity of living organisms registered against various stimuli, through search and temporal localization of patterns in noised electrograms containing useful information. One of the approaches to solving such problems is based on an analysis of the Shannon entropy calculated based on the components of the power spectrum and called the spectral entropy function. It is found that under the conditions providing that the patterns in question pertain to high-frequency rhythms, and the boundaries of their energy spectra are a priori unknown, the criterial functions of spectral entropy are of low sensitivity. Purpose of the research: to develop cost functions of entropy analysis which are sufficiently sensitive for searching for high-frequency patterns with a priori unknown parameters in noised electrograms. Results: development of a cost function that makes it possible to find the frequency sub-band where the spectral components of the patterns in question maximally contribute to the total spectrum power. The subsequent computation of the spectral entropy in the identified frequency sub-band provides a solution to the problem of search for the response patterns in noised electrograms under the above conditions. Practical significance: the results confirm the effectiveness of the developed functions whose use is limited by the requirement that the electrogram should be recorded on more than one lead.