Phonocardiogram Sound Event Detection-Delineation Based on Gaussian Smoothing Filtering

This paper presents a new method for detection, delineation and unsupervised classification of PCG sounds. First, the PCG signal was preprocessed and then for detection and delinea-tion of the PCG genuine sounds incidence locations, two windows were slid on the pre pro-cessed PCG signal. In each slid of the analysis window, the value of mid-sample of the first window was replaced by a frequency based feature of the signal excerpted by the window and the value of mid-sample of the second window was replaced by an amplitude based fea-ture of the signal excerpted by the window. Then a final feature was extracted by the combi-nation of these two features. By these calculations peaks representing location of the sounds incidence appears and can be delineated by various methods such as application of a mini-mum slope forward-backward local search starting from the sound location. The operating characteristics of the proposed method is the average accuracy Acc=98.91% and 93.50% for sound detection and unsupervised classification, respectively

Keywords:

Valvular Vibration , Phonocardiography , heart sound , Auscultation