【作者】 赵治栋；

【导师】 陈裕泉；

【作者基本信息】 浙江大学 ， 生物医学工程， 2004， 博士

【摘要】 冠心病是世界上严重危害人们身体健康的最重要的疾病之一。近10年来，随着生活水平的提高和平均寿命的延长，我国冠心病发病率正在逐年上升，而且还有年轻化的趋势。冠心病的危害具有严重性、广泛性、和复杂性，这给冠心病的诊断和治疗提出更高的要求，关键在于及时诊断、及时治疗。因此，积极广泛开展冠心病无损检测的研究并为大众提供方便可行的检测手段是十分急需的。心音信号是人体重要的生理信号之一，蕴含了人体心脏和血管中丰富的原始生理、病理信息，临床上的应用潜力很大。传统心音学与现代信号处理技术相结合，可能会给冠心病无损诊断技术带来突破性的进展。本文正是针对此展开的。 本文主要研究了基于舒张期心音分析及特征提取的冠心病无损诊断技术。在分析了冠状动脉堵塞与舒张期心音关系的基础上，首先研究了心音信号的预处理(消噪和分段)，对舒张期心音信号进行了定位；其次利用Hilbert Huang Transform(简称HHT)对舒张期心音信号进行分析研究，提取了冠心病病理特征；最后通过适用于小样本机器学习的支持向量机(Support Vector Machine，简称SVM)实现了冠心病的智能无损诊断，获得了较高的诊断率。 本文主要研究内容有： 研究了生物医学信号的去噪问题。建立了小波阈值消噪的广义框架：提出了广义阈值函数以及推导了广义阈值函数的阈值确定公式。推导出小波阈值消噪的偏差、方差以及风险与小波系数和阈值的关系式，从理论上研究了阈值函数及阈值的选择对消噪性能的影响。通过对不同特性的模拟信号的处理，得到了最优的小波阈值消噪方案。用此消噪方案对心音信号及心电信号进行了处理，取得了较好的效果。 在分析了目前存在的心音分段算法优缺点的基础上，提出了一种新的心音自动分段算法，给出了分段算法步骤及策略。考虑到心音信号的多样性及复杂性，用大量临床病例对此算法进行了验证，达到了很高的分段正确率。该算法无需额外信号做参考，正确率高，鲁棒性强。 对HHT进行了系统的研究。回顾了传统的信号分析方法；深入研究了多分浙江大学博士学位论文 量信号的瞬时频率，从瞬时频率存在的条件出发引申出了内蕴模式函数的概念。 阐述了HHT的原理:IMF、EMD、Hi lbert谱和边界谱等。研究了影响HHT性 能的主要因素，并提出了相应的改善措施，采取的这些措施对HHT方法在舒张 期心音分析中的实际应用具有重要意义。最后用各种不同性质的信号表征了 HHT分析方法的优越时频分析性能。 将HHT引入心音信号的研究，建立了基于HHT的冠心病舒张期心音信号分析及特征提取方法。设计了自适应增强器，提高舒张期心音信号信噪比。探讨了舒张期心音信号的平稳性。利用HHT方法分析了冠心病病人以及非冠心病人的舒张期心音的瞬时频率以及谱分布特性。提出了加权瞬时频率的概念，得到了舒张期心音的瞬时频率变化规律。提出了基于舒张期心音的冠心病特征提取的新方法，所提取的特征有效的揭示了冠状动脉堵塞的信息，刻画了冠状动脉阻塞时舒张期心音的变化。表明HHT方法能有效的提取舒张期心音信号中隐含的关于冠状动脉堵塞相关的病理信息，这对于冠心病的诊断有着重要的意义。 提出了基于支持向量机的冠心病诊断方法。介绍了统计学习理论以及SVM的原理，构建了SVM分类器，将其应用于冠心病临床诊断，在有限的病例样本情况下获得了很高的诊断率。同时与传统神经网络方法的诊断性能做了比较，结果表明，基于支持向量机的冠心病诊断方法具有更高的诊断率和较好的推广性。 经过临床实践证明，本文所提出的基于舒张期心音F旧T分析、病理特征提取和SVM模式识别的冠心病无损诊断方法取得了较高的诊断率，为冠心病的及时治疗奠定了良好的基础。更多还原

【Abstract】 Coronary Artery Disease (CAD) is one of the leading causes of death in the world. In recent 10 years, with the raising of life quantity and the extending of the average longevity, Morbidity for CAD is gradually increasing with the trend of patients being young. CAD is harmful with the characteristic of Severity popularity. and complexity. Seasonable diagnosis and cure are important. Developing the study of the noninvasive detection of CAD is essential. Detection of CAD is a most important medical research area. Heart sound is one of the most physiological signals of body. It provides a lot of valuable diagnostic and prognostic information concerning the heart and hemodynamics. It had important potentiality in clinical practice. Combination the traditional phonocrdiogrm (PCG) with the modern signal processing may bring the breakthrough in the noninvasive detection of CAD.This paper mainly concerns the diagnosis of CAD automatically based on the analysis and features extraction of diastolic murmurs. On the basis of analyzing the relation between the coronary artery block and the diastolic murmurs, firstly this paper studied the preprocessing of PCG (denoising and segmentation), the diastolic periods were positioned; secondly the diastolic murmurs was analyzed by use of HHT, The features of CAD were acquired; In the end, the CAD was diagnosed automatically by use of Support Vector Machine (SVM), which was suitable in the situation of small samples and better diagnosis ratio was acquired.The main contributions of this thesis are as follows:The denoising of biomedical signal was studied. The generalized frame of wavelet shrinkage denoising was build. The generalized threshold function was proposed and the threshold value was also studied for the generalized threshold function. Efficient formulas for computing bias, variance and risk of generalized threshold function were derived. On the basis of this, the relation of bias, variance and risk of generalized threshold function with threshold value and wavelet coefficient were compared. These comparisons gave the performances of wavelet shrinkage denoising in finite sample situations. Kinds of signals with different features were considered to illustrate the performances of the different threshold functions and different threshold values. The optimal denoising scheme was achieved. . In end, the method was used to denoise the electrocardiogram signal and phonocardiogram signal, and the better performance was acquired.On the basis of reviewing the virtues and defects of traditional PCG segmentation algorithms, a novel segmentation algorithm was proposed. The steps and strategies of segmentation were given. The algorithm was tested using a lot of clinic normal and abnormal heart sound data. Whereas the heart sound signals were intricate, the algorithm achieved a high detection rate. The correct detection rate wasvery high by the segmentation algorithm and the algorithm was robust without other signal as reference.HHT method was studied. Traditional signal processing methods were reviewed; the instantaneous frequency of multicomponent signal was further researched. The concept of intrinsic mode function was induced by the existing conditions of instantaneous frequency for multicomponent signal. The principles of HHT were introduced, including IMF, EMD Hilbert spectrum and marginal spectrum. Some factors related the performances of HHT were studied, and the methods were proposed to improve the performance. The methods were very important for the applications of HHT. In the end, some kinds of signals were analyzed by HHT; the result indicated the HHT method had better qualities.The diastolic murmurs was analyzed by HHT. The methods of analyzing diastolic murmurs and the features obtained by HHT were presented. The adaptive linear enhancer was designed to improve the signal noise ration of diastolic murmurs. The stationarity of diastolic murmurs was analyzed. The instantaneous frequency and Hilbert spectrum and marginal spectrum of diastolic murmurs for CAD更多还原