【作者】 姚成；

【导师】 司玉娟；

【作者基本信息】 吉林大学 ， 通信与信息系统， 2012， 博士

【摘要】 伴随着人们生活水平的逐步提高，人们的健康意识不断增强；而现今心脏病的发病率也在逐年升高，已严重地危害了人类的生存和健康。但由于心血管疾病的发病时期不规律，且发病现象较隐蔽；因此，人们试图通过各种方式来提前预防和治疗心血管疾病比较困难。而院外监护、初步诊断、积极预防与及时治疗是行之有效的心脏疾病处理手段，这也对研究心电信号分析、诊断技术及心电监护产品提出了更高的要求。本文以珠海市高新技术领域科技攻关及高新技术产业化项目“院外多生理参数监护诊断系统”（2010B020102021）为背景。依据项目的研究内容，针对目前心电信号处理、分析和智能诊断算法中存在的不足，就心电信号的预处理（去噪）、波形检测、波形特征向量的选择与提取以及波形自动分类等关键技术进行研究。针对算法面向心电监护仪等硬件应用平台移植的关键技术也做了相应的探讨。旨在提高计算机智能分析的准确率和实用性，提高异常心电波形自动分类的精度和性能，这对于加快国内心电监护医疗器械的研制，取得具有自主知识产权的心电信号智能分析核心技术，提高心电智能监护的质量，普及心电智能监护的应用都具有非常重要的现实意义和很好的经济效益。本文对心电信号智能分析关键技术的研究取得了一定的成果，主要工作集中在：1、心电信号预处理（去噪）算法的研究充分研究了心电信号中噪声的特性。依据小波去噪原理，构造了一种基于软、硬阈值之间的新阈值函数；构造了一种加权阈值收缩函数，并提出了基于两种新阈值函数的心电信号去噪算法。利用MIT-BIH数据库对两种去噪方法进行了实验，结果表明，两种方法都比以往小波阈值去噪法在心电信号去噪的效果上有所改善，并且提出的基于加权阈值收缩去噪法，可以更好地保留心电信号P、T波形的细节特征，满足心电信号波形特征识别的需要。2、心电信号波形特征识别算法的研究提出了基于连续小波变换的心电信号QRS波识别算法。该算法采用高斯函数一阶导数作为小波基函数，利用考察小波变换相应层次中的模极大值对位置作为搜索QRS波中R波顶点的范围；根据R波顶点位置，结合平均心电周期，提出了一种P、T波搜索窗口宽度自适应方法，在此窗口中利用原信号的差分值，确定P、T波形的关键点。该算法对QRS波和P、T波各关键点的检出准确率较以往算法有所提高。3、面向硬件实现心电信号处理和分析快速算法的VLSI实现研究提出了基于DB4小波提升的心电信号处理和识别快速算法。该算法利用小波提升速度快的特性，使算法整体执行速度大大提高。对于算法向硬件平台移植的关键难点DB4小波提升的VLSI实现问题进行了研究，提出了利用FPGA实现DB4小波提升分解和重构的方案，通过实验验证了以上算法的有效性和方案的可行性。4、异常心电信号分类算法的研究提出了一种对平均心电周期长度具有自适应性的心电波形向量提取算法，提出了一种利用逻辑判断提取正常心电波形的判断依据，并提出了一种将逻辑判断（Logic）、聚类（Clustering）和模糊聚类（FCM）三者结合对异常心电实现准确聚类的算法（LCFCM）。算法对存在个体差异的心电信号具有很好的适应性，依据提取的心电向量波形进行聚类和模糊聚类分析，保证了算法对象信息的完整性，使算法整体具有很好的准确性。最后利用MIT-BIH数据库作为样本进行实验，LCFCM算法对异常心率分类的准确率达到了93%。更多还原

【Abstract】 With the gradual improvement of people’s living standard, people have becomemore conscious about their health. The incidence of heart disease also increases year byyear, which seriously endangers human health and survival. The incidence ofcardiovascular disease does not occur regularly. And it is not easy to notice theincidence of the phenomenon. As a result, it is very difficult for people to make theearly prevention and treatment of cardiovascular disease. And the effective heartdisease treatments such as monitoring outside hospital, primary diagnosis, activeprevention and the timely treatment have higher demand toward the study on ECGsignal analysis, the diagnostic techniques and ECG monitoring products.Multiple Physiological Parameter Monitoring and Diagnostic Systems Outside theHospital （2010B020102021）is one of the science research projects and high-techindustrialization projects in Zhu Hai high-tech field of science and technology. In thecontext, based on the research content of this project, as far as the weakness of ECGsignal processing, analysis and the intelligent diagnosis algorithm is concerned, thisthesis makes the study on the key technology of ECG signal preprocessing (denoising),waveform detection, the selection and extraction of the waveform feature vector andthe waveform automatically classification. This thesis also makes the relevant study onthe key technology of the transplantation for the hardware platform of thealgorithm-oriented ECG application. It aims at improving the accuracy andpracticability of computer intelligence analysis. It also aims at improving the accuracyand performance of abnormal ECG waveform automatic classification. This can speedup the development of ECG medical devices in China and make intelligent analysis ofECG signals with independent intellectual property core technology. In addition, thiscan improve the quality of ECG intelligent guardianship and make the application ofECG intelligent guardianship become universal. The study in this thesis has veryimportant significance and can bring enormous economic benefits.The research on key technology of ECG signal intelligent analysis in this thesishas made some achievements. The achievements are focused on the following aspects. 1. The Study on the Algorithm of ECG Signal Preprocessing (Denoising)After making research on the characteristics of noise in ECG signal, based onwavelet denoising principle, a new threshold function based between the soft and hardthreshold is created. And a weighted threshold shrinkage function is created. Inaddition, this thesis puts forward an ECG signal denoising algorithm based on two newthreshold functions. The two denoising methods are used to make experiments on thetypical data in MIT-BIH Data Base. The experimental results indicate that, comparedwith the previous wavelet threshold denoising methods, the two methods improve a lotin the effect of denoising. The proposed based on weighted threshold shrinkagedenoising method can preserve much more details of the waveform of P wave and Twave in ECG signal, which is much easier to satisfy the need of recognizing thefeatures of ECG signal waveform.2. The Study on the Identification Algorithm of Features of ECG SignalWaveformThe QRS wave identification algorithm based on the continuous wavelettransform is proposed. This algorithm takes the first derivative of Gaussian function asthe wavelet basis function. The relative position of the modulus maximum is used todefine the range of searching the R wave vertex in QRS complex, by examining thewavelet transform in the corresponding levels. According to the position of R wavevertex and the average ECG cycle, a window width adaptive method of searching for Rand P wave is proposed. In this window, the differential values of the original signalsare used to identify the key points of the waveform of P and T wave. The detection rateof the key points in QRS complex, P wave and T wave improves a lot than the previousalgorithms.3. The Study on the Hardware Oriented Implementation of ECG SignalProcessing, Analysis of the Fast Algorithm and VLSI ImplementationThe fast algorithm of ECG signal processing and identification based on DB4wavelet lifting is created. The algorithm uses the wavelet lifting’s characteristic ofpossessing fast speed, and greatly improves the algorithm’s overall execution speed.After making the study on DB4wavelet lifting VLSI implementation issues, as far asthe key technical points of the algorithm’s transplantation toward the hardwareplatform are concerned, the program of the lifting decomposition and reconstruction ofDB4wavelet by using FPGA is proposed. The effectiveness and the above algorithm and the feasibility of realizing FPGA program are proved by the experiments.4. The Study on the Sorting Algorithm of the Abnormal ECG SignalThe ECG waveform vector extraction algorithm which is self adaptive to theaverage length of ECG cycle is created. The judgment of using the logical judgment toexact the normal ECG waveform is given. Accordingly, LCFCM algorithm whichcombines logical judgment, cluster analysis and fuzzy clustering together to realize theaccurate clustering of the abnormal ECG heart rate is proposed. The algorithm has agood adaptability to the ECG signal with individual differences. And the clusteranalysis and fuzzy clustering analysis based on the extracted ECG vector waveform canguarantee the integrity of algorithm’s object information, which makes the wholealgorithm become very accurate. Finally, the experiment by using MIT-BIH databaseas the sample is made. The accuracy rate of the classification of the abnormal heart rateby LCFCM algorithm reaches93%.更多还原