Automatic Classification of Phonocardiogram:Processed by Linear Predictive Analysis

概要

論文の詳細を見る
This paper reports on a new algorithm for automatic classification of phonocardiogram and its experimental results. It is based on the frequency domain characteristics which are analysed by linear predictive method together with the time domain characteristics of phonocardiogram. It needs only a high frequency phonocardiogram of cardiac apex for data. Detection of the first and the second heart sounds and extraction of the spectral features for classification are executed by linear predictive analysis. Classification is performed by linear discriminant function method which is one of the statistical pattern classification methods.<BR>We have evaluated the performance of the system using 171 samples (744 beats) including normal and abnormal subjects. Recognition rates are calculated not only by the resubstitution method (R-method) but also by the leave-one-out method (L-method) which reflects practical performance more precisely than the R-method. For 2-class classification, more than 90% recognition rate is achieved even by L-method. Also for 6-class classification, high recognition rate is achieved by R-method and 2 classes are picked up from 6 classes by L-method with high reliability. Consequently, we conclude that one can use this system practically for high-performance screening.
一般社団法人日本生体医工学会の論文