心音信号的聚类分析

Clustering Analysis of Phonocardiagrams

通过对ＡＲＴ２神经网络进行改进，构造了适合于对信号的时变谱进行分类的二维ＡＲＴ神经网络，而后根据心音信号的特点，我们选择了心音的小波变换谱作为网络的输入特征。调整网络的参数，使之能够将输入的心音信号特征聚类成为两类，实验结果表明我们所设计的分类系统能够从连续３２个正常人的心音信号中区分出两个正常分裂的心音。这个结果在基础生理研究和临床诊断上都有一定的应用价值。

Through the development of the ART 2 neural network,we had constructed the 2 dimensional ART neural network appropriate for clustering the time varying spectrums.After that ,according to the character of phonocardiagram(PCG),we chose the scalogram of PCG to be the input feature of the network.By adjusting parameters the network was enabled to cluster the input PCGs to two clusters.The result demonstrates that the system we designed can pick out,from among 32 continuous PCGs,two PCGs which have the tendency toward abnormality.This result may be applicable to basic physiological studies and clinical diagnosis.