342例肺系疾病患者的语音信号采集和特征分析

来源:用户上传      作者:陈春凤 王忆勤 徐�Q 颜建军

　　摘 要 目的：运用现代声学技术，采集和分析肺系疾病患者的声音信号，为肺系疾病的中医辨证提供一定的声诊客观数据。方法：运用“中医闻诊采集系统”采集肺系疾病患者声音样本342例，其中原发性支气管癌79例、慢性支气管炎119例、支气管哮喘144例，另选上海中医药大学在校健康师生102名为正常对照组。运用小波包分析技术对信号分析处理，提取各组样本信号的小波包能量和Shannon熵值特征，通过独立变量非参数检验方法对各组样本的小波包特征参数进行检验，并运用这两种特征进行中医病证的分类识别研究。结果：多个时域频段的小波包能量特征值及Shannon熵值特征差异均有统计学意义（P<0.05），总熵值比较，正常对照组总熵值均低于患病各M。运用本研究得到的小波包能量和Shannon熵值两种特征，分别采用支持向量机和BP（back propagation）神经网络两种方法，对肺系不同证型样本进行分类识别，分类识别准确率分别为83.67%和71.95%。结论：肺系不同病证患者的语音信号特征存在差异，小波包能量和Shannon熵值特征能初步区分肺系常见病证的语音特征，辅助肺系常见病证的临床诊断和辨证；运用支持向量机和小波包能量、Shannon熵值特征进行肺系病证的分类识别，有较好的效果，为中医声诊的分类识别提供新的思路。
　　关键词 肺系疾病；中医声诊；客观化；语音信号
　　中图分类号：R273 文献标志码：A 文章编号：1006-1533（2022）14-0021-05
　　引用本文 陈春凤， 王忆勤， 徐Q， 等. 342例肺系疾病患者的语音信号采集和特征分析[J]. 上海医药， 2022， 43（14）： 21-25.
　　Voice signal acquisition and characteristic analysis of 342 patients with pulmonary diseases
　　CHEN Chunfeng1， WANG Yiqin2， XU Jin2， YAN Jianjun3（1. Lingyun Community Health Service Center of Xuhui District， Shanghai 200237，China； 2.Basic Medicine School of Shanghai University of Traditional Chinese Medicine， Shanghai 201203， China； 3.School of Mechanical and Power Engineering of East China University of Science and Technology， Shanghai 200237， China）
　　ABSTRACT Objective： Using modern acoustic technology， to collect and analyze the sound signals of patients with pulmonary diseases， so as to provide certain objective data for acoustic diagnosis of traditional Chinese medicine（TCM） syndrome differentiation of pulmonary diseases. Methods： Using the “TCM auscultation and diagnosis collection system” sound samples of 342 patients with lung diseases were collected， among them， there were 79 cases of primary bronchogenic carcinoma， 119 cases of chronic bronchitis and 144 cases of bronchial asthma. Another 102 healthy teachers and students in the University of Traditional Chinese Medicine were selected as the normal control group. Wavelet packet analysis technology was used to analyze and process the signal， and the wavelet packet energy and Shannon entropy characteristics of each group of sample signals were extracted. The wavelet packet characteristic parameters of each group of samples were tested by independent variable nonparametric test method， and the two characteristics were used for the classification and recognition of TCM disease and syndromes. Results： The differences of wavelet packet energy eigenvalues and Shannon entropy characteristics in multiple time-domain bands were statistically significant（P<0.05）. Compared with the total entropy， the total entropy in the normal group was lower than that in the diseased groups. Using the two characteristics of wavelet packet energy and Shannon entropy obtained in this study， support vector machine and back propagation（BP） neural network were used to classify and recognize the samples of different syndrome types of lung system， and the accuracy of classification and recognition was 83.67% and 71.95%， respectively. Conclusion： The speech signal characteristics of patients with different lung diseases and syndromes are different， the wavelet packet energy and Shannon entropy characteristics can preliminarily distinguish the speech characteristics of common lung diseases and syndromes， and assist in the clinical diagnosis and syndrome differentiation of common lung diseases and syndromes； using support vector machine， wavelet packet energy and Shannon entropy to classify and recognize lung diseases and syndromes has a good effect， which provides a new idea for the classification and recognition of TCM acoustic diagnosis.

nlc202208181454

转载注明来源:https://www.xzbu.com/6/view-15437914.htm