基于小波倒谱系数和概率神经网络的先心病心音信号识别

Recognition of Heart Sound Based on Wavelet Cepstrum Coefficient and Probabilistic Neural Network

摘要: 目前先心病的初诊主要依靠心脏听诊，对心音信号进行分析研究有助于先心病初诊阶段的辅助诊断。本文首先对从临床采集的正常心音信号和先心病心音信号进行去噪预处理，之后提取心音信号的小波倒谱系数(WCC)作为特征，采用概率神经网络(PNN)作为分类器，用154例正常心音和105例先心病心音对分类器进行训练，用66例正常心音和45例先心病心音进行了测试。实验结果为：对正常心音的正确识别率为91%，对异常心音的正确识别率为86.7%，平均识别率89.2%。

Abstract: At present, the initial diagnosis of congenital heart disease mainly relies on cardiac auscultation. The analysis and study of the heart sound signal is helpful to the auxiliary diagnosis of congenital heart disease (CHD) at the initial stage. Firstly, the normal heart sounds and CHD heart sounds were denoised and pretreated. Then the wavelet cepstrum coefficients (WCC) of heart sounds were extracted as features. The probability neural network (PNN) was used as classifier. The classifier was trained with 154 cases of normal heart sound and 105 cases of CHD heart sound. It was tested by using 66 cases of normal heart sound signal and 45 cases of CHD heart sound. The experiment results show that the correct recognition rate of normal and abnormal heart sounds is 91% and 86.7% respectively, and average identification rate is 89.2%.