基于人工神经网络及小波分析的心音诊断方法

由于传统的心音听诊就是凭医生的经验用听觉分析心音信号, 不能满足医学上所要求
的高精确度性能而且听诊技能要花多年时间才能掌握,针对这些弊端本文提出了一种新的心音诊断方法。它对电子听诊器录制的心音数据，经过去噪预处理后用小波变换进行分析并提取特征值，再将选取的特征值输入到前馈型神经网络进行训练和识别。实验中我们用节点数分别为9，5，5 的BP 神经网络能成功识别出主动脉关闭不全，主动脉狭窄，二尖瓣关闭不全，二尖瓣狭窄，和正常心音五类心音, 能为相应心脏疾病的诊断提供有力的依据，为临床应用提供有效的分析手段。
关键词：心音 小波变换 特征值选取 神经网络

The method of heart sound diagnosis based on neural networks and wavelet transform
Cao-Li¹ ,Zhao De-an¹ ,Sun Yueping¹ ,Liu Jian-yue² 1. School of Electrical and Information Engineering , Jiangsu University, Zhenjiang 212013 2. Hospital of Jiangsu University, Zhenjiang 212013 ABSTRACT: Heart auscultation with the bare ear and the stethoscope can not satisfy the high precision in medicine and forming a diagnosis based on heart sounds is a skill that can take years to acquire. So this paper presents a diagnosis method on heart auscultation. A library of heart sound files, recorded via an electronic stethoscope are used, features from these samples are extracted using discrete wavelet transform and the classification is trained and carried out by using a feed forward neural network. We collected sound samples of five categories, they are aortic regurgitation, aortic stenosis, mitral regurgitation, mitral stenosis and normal heard sound.The BP can regonise them succesfully. It can make up the shortage of heart auscultation and save the time to grasp the diagnosis skill for doctors.
Key words:heard sound; wavelet transform; eigenvalue; neural networks