基于复杂度的心脏瓣膜病心音信号的分析及识别研究

发布时间：2018-06-21 20:19

本文选题：心音信号 + 心脏瓣膜病　；参考：《重庆大学》2012年硕士论文

【摘要】：心脏瓣膜病日益成为危害人类健康的最重要的心脏病之一，伴随着老年化的加剧，其发病率日渐增加。心音是人体重要的生理信号之一，包含了心脏瓣膜大量的生理病理信息，其杂音特征对于心脏瓣膜病的检查和诊断至关重要。一直以来，人们对非线性、复杂的心脏系统采用抽象简化，通过建立理想线性模型，并用时域、频域及时频域联合分析。虽然取得了很好的效果，，但人们同时发现，采用线性方法分析并不足以研究本质为非线性、非平稳的生命系统。为研究心脏瓣膜病的心音信号，本课题从非线性复杂度的角度对其分析，以期利用非线性动力学方法提取其特征，实现基于心音信号的心脏瓣膜病计算机辅助诊断。论文阐述了正常心音和心脏瓣膜病心音信号的产生机理，并介绍了心音在心脏瓣膜病诊断中的应用。在分析了心脏瓣膜活动和心音关系的基础上，采用了平均经验模式分解(Ensemble Empirical Mode Decomposition, EEMD)的自适应滤波方法进行了心音消噪研究，并对消噪的心音进行Lemple-Ziv复杂度分段；其次采用非线性复杂度方法对瓣膜性心音信号进行分析，并将EEMD和近似熵复杂度相结合来提取心音中的生理病理特征；最后运用二叉树向量机技术对常见的几种心脏瓣膜病心音信号进行分类识别，为心脏瓣膜病的早期诊断和治疗提供了一种有效的手段。本文研究分析了经验模式分解的算法原理和不足，针对模态混叠问题提出了基于EEMD阈值的自适应消噪方法，分析了临床采集到的心音信号，通过对比分析消噪前后的心音信号，可以有效的去除噪声并解决了模态混叠的问题，并进一步提出了基于Lemple-Ziv复杂度的心脏瓣膜病心音的分段算法，结果表明其分段准确率高，且不需要额外的生理信号来辅助。为了提高心音信号的识别精度和分类准确性，在分析心脏瓣膜病心音信号特点和复杂性的基础上，选择基于EEMD和近似熵复杂度的心音非线性动力学算法来进行特征提取，首先对其进行EEMD分解得到反映心音本体特征的固有模态函数（Intrinsic Mode Function，IMF）；然后计算每阶IMF的非线性复杂度特征，通过互相关系数准则对IMF进行筛选，组成心音的特征向量，结果表明该方法能有效揭示了心脏瓣膜病变引起的瓣膜缺损和病变信息。本文选取二叉树支持向量机（Binary Tree SVM，BT-SVM）来实现心脏瓣膜病心音信号的识别。首先分析了SVN对多子类分类问题研究和核函数的选择，提出了基于BT-SVM的心脏瓣膜病心音信号的分类和识别算法，对5种临床常见的心脏瓣膜病心音信号和正常心音信号进行识别，其中心脏瓣膜病心音信号包括二尖瓣关闭不全心杂音、二尖瓣狭窄心杂音、主动脉瓣关闭不全心杂音、主动脉狭窄心杂音、室间隔缺损心杂音共计90例，实验结果验证了所提出的常见瓣膜性心音信号分类方法的可行性，正确率可以高达93.23%，从而为心脏瓣膜病的诊断和临床应用提供有力的根据。
[Abstract]:Heart sound is one of the most important heart diseases which is harmful to human health . It is one of the most important physiological signals of human body . Heart sound is one of the most important physiological signals of human body . It is very important for the examination and diagnosis of valvular heart disease .

Based on the analysis of the relationship between heart valve activity and heart sound , an adaptive filtering method based on empirical mode decomposition ( EEMD ) is used to study the heart sound and noise , and the Lemple - Ziv complexity segmentation is carried out on the noise elimination heart sound .
Secondly , nonlinear complexity method is used to analyze the valvular heart sound signal , and the physiological and pathological characteristics of heart sounds are extracted by combining EEMD and approximate entropy complexity .
Finally , using binary tree vector machine technique to classify the common heart sound signals of valvular heart disease , provides an effective means for early diagnosis and treatment of valvular heart disease .

In this paper , the principle and disadvantage of the empirical mode decomposition are analyzed . Based on the EEMD threshold - based adaptive noise elimination method , this paper analyzes the sound signals collected in clinic , analyzes the sound signals before and after noise elimination , effectively removes noise and solves the problem of modal aliasing , and further proposes a segmentation algorithm based on Lemple - Ziv complexity heart sounds . The results show that the segmentation accuracy is high , and no extra physiological signal is needed to assist .

In order to improve the recognition accuracy and classification accuracy of the heart sound signal , based on the analysis of the characteristics and complexity of the heart sound signal of the valvular heart disease , a nonlinear dynamic algorithm based on EEMD and approximate entropy complexity is selected to extract feature extraction , which is firstly decomposed by EEMD to obtain the intrinsic mode function ( IMF ) which reflects the characteristic of the heart sound body ;
Then the nonlinear complexity of the IMF is calculated , and the IMF is screened by the correlation coefficient criterion to form the characteristic vector of the heart sound . The results show that the method can effectively reveal the valve defect and the lesion information caused by the heart valve disease .

In this paper , we choose binary tree SVM ( BT - SVM ) to realize the recognition of heart sound signal of valvular heart disease . Firstly , we analyze the classification and recognition algorithm of heart sound signal of valvular heart disease based on BT - SVM .
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R542.5;R318.0

【参考文献】