五导呼吸音采集分析研究

发布时间：2018-04-17 03:41

本文选题：呼吸音 + 特征提取　；参考：《西华大学》2015年硕士论文

【摘要】：呼吸疾病的机理信息在呼吸系统分布分散往往反应在整个胸部区域,呼吸音听诊作为一种主要的呼吸音监测手段,得到了越来越广泛的应用,但由于临床常见呼吸音疾病多为混合病变,传统的单导听诊只能依次反复听诊不同部位才能提供临床信息,而多导呼吸音采集有利于在同一时间对不同部位间的呼吸音及杂音的持续时间、强度等信息进行收集,提供丰富的临床信息,并对采集的呼吸信号进行进一步分析研究,本文拟在以下几个方面进行深入研究:(1)熟悉呼吸音的基本信息,掌握呼吸音的产生机理以及呼吸杂音与呼吸系统疾病的病例关系和内外的呼吸音的研究和方法。(2)分析与实验室有合作的山口大学自主研制的听诊头是否符合呼吸音采集,此听诊头主要由铁三角AT9904麦克风和3M公司生产的震动腔(Littmann,ClassicIISE)构成。(3)结合成都452医院医师指导,提出五导采集部位,基于听诊部位和处理方法已经申请一项实用新型专利目前已经采用此方案采集到20名正常呼吸音志愿者的50多例呼吸音信号和20名患者的21例异常呼吸音信号(其中有14列相同病理),完成了初步的呼吸音采集测试。(4)对采集到的呼吸音信号进行预处理分析,针对各种噪音设计相应的处理方法,例如:工频陷波器、小波阈值降噪、IIR滤波器设计,通过以上的方案尽可能的提高呼吸音信号的信噪比,通过对比实验,预处理效果比较明显。(5)包络提取与特征提取,针对呼吸音的特性,首先采用了频域分析为辅,然后主要采用时域的包络提取,采用常用包络提取方法的对比,基于希尔伯特变换的呼吸音包络提取、基于归一化平均香农能量的呼吸音包络提取、基于单自由度模型的呼吸音包络提取、基于Morlet小波呼吸音包络提取的对比性试验,选取Morlet包络提取算法,然后通过采用FCM算法对包络波进行阈值线划分,最后得到特征参数呼气相和吸气相的持续时间(T1,T2),呼气间隙时间和吸气间歇时间(D1,D2),呼气相和吸气相的峰值(P1,P2),以T1/T2,D1/D2作为区分呼吸音类型参数,进行初步的呼吸音二分类。(6)本实验室的最终目的在于能够下位采集,后端有服务器的支持诊断,于是如何利用较少的资源去储存和传输逐步增大的数据库和提升单次采集数据传输速度,也是本文需要研究的一个方向,实验以改写OMP算法和SAMP算法应用于呼吸音,通过对两种算法运算过程中所需要涉及的步骤和条件进行对比,得出需要在未知稀疏度(K)的前提下,改进算法,加快运算速度和运算精准度。综上所述,本文对于呼吸音的采集和分析进行了研究,并运用实际采集到的临床数据进行验证,实验证明:五导呼吸音采集相对于单导听诊采集的优势明显,能同时采集到数据后比较直观的对不同部位的数据进行对比和识别,基于FCM的聚类算法也能对呼吸音进行较好的特征提取分类,而今后的工作重点将放在采集更多异常呼吸音和异常呼吸音的种类和呼吸音算法的优化上。
[Abstract]:The mechanism of respiratory disease information in scattered respiratory system often in the chest area, respiratory auscultation as a main breathing monitoring means, has been more and more widely used, but because of clinical common breathing diseases are mixed lesions, single channel can only turn the traditional auscultation repeatedly in different parts in order to provide clinical auscultation and guide information, breathing is conducive to the acquisition at the same time the duration of breath sounds and murmurs among different parts of the strength, such as information collection, to provide clinical information rich, and the collection of respiratory signal for further analysis and research, this paper intends to conduct in-depth research in the following aspects: (1) familiar with the basic information of respiratory sounds, research and methods of respiratory sound master breath sound generating mechanism and respiratory murmur and respiratory diseases and abroad. (2) analysis The stethoscope head cooperate with laboratory developed by Yamaguchi University with respiratory sound collection, the stethoscope head is mainly composed of vibration cavity iron triangle AT9904 microphone and 3M company (Littmann, ClassicIISE). (3) Chengdu 452 hospital doctor guidance combined, put forward five guide collection parts, parts and processing method based on auscultation has been applied a utility model patent has been collected by using this method, more than 50 cases of 20 normal volunteers breathing respiratory sound signals and 20 patients with 21 cases of abnormal respiratory sounds (14 of the same pathological) completed respiratory sound acquisition, preliminary test. (4) pretreatment analysis of respiratory sounds the collected, processing method, the corresponding design for various noise such as frequency notch filter, wavelet threshold denoising, IIR filter is designed by the above scheme, as far as possible: the shouting signal sound The signal-to-noise ratio, through the contrast experiment, pretreatment effect is obvious. (5) envelope extraction and feature extraction, according to the characteristics of respiratory sounds, first used in the frequency domain analysis, and mainly uses the time domain envelope extraction, compared with the commonly used envelope extraction method, extraction of respiratory sound envelope based on Hilbert transform, extracting respiratory sound envelope the normalized average Shannon energy based on the extraction of respiratory sound envelope model of single degree of freedom based on the contrast test of extracting Morlet wavelet envelope based on respiration, select the Morlet envelope extraction algorithm, and then by using FCM algorithm to divide the threshold line envelope, and finally get the characteristic parameters of the expiratory phase and inspiratory phase duration (T1, T2), clearance time and expiratory inspiratory pause time (D1, D2), peak expiratory and inspiratory phase (P1, P2), T1/T2, D1/D2 as a division of respiratory sound type parameters, initial call Two sound classification. (6) the final purpose of this laboratory is able to support a collection, back-end diagnosis server, so how to use fewer resources to store and transfer to gradually increase the database and improve the single data transmission speed, also a direction of the study in this article, the experiment to rewrite the OMP algorithm and SAMP algorithm applied to the sound of breath through the steps and conditions involved need two kinds of algorithms in the process of comparison, found in unknown sparsity (K) algorithm under the premise, accelerate the operation speed and operation precision. To sum up, this paper does research on the collection and analysis of breath sounds, and verify the use of clinical data, the observed experimental results show: five guided breathing acquisition compared with single guide acquisition auscultation have obvious advantages, at the same time after collecting data more intuitive in different parts Comparing and identifying the data, the clustering algorithm based on FCM can also perform better feature extraction and classification of respiratory sounds, and the future work will focus on the collection of more abnormal breath sounds and the types of abnormal breathing sounds and the optimization of respiratory algorithm.

【学位授予单位】：西华大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R443

【参考文献】