Abstract
In this paper, two basic methods of signal processings to make the systematic diagnosis of subtle heart malfunctions possible are proposed, by utilizing the time-frequency information of heart sounds and noises detected only on the chest surface. That is, a simple method of determining the average time length for nonstationary power spectral analysis, based on the evaluation of the local 1 st and 2 nd order cumulant variation of the observed data, along with the heart dynamics, and an advanced method of discriminant analysis, that can extract the time-frequency characteristic parameters, properly fitted to the analysed heart sounds and noises, by the additional use of multiple regression analysis. The principles of these methods and their fundamental features as well as the effectiveness of sophisticated processings for computer-aided automatic diagnosis of heart malfunctions are shown, along with fundamental results of experiments.
