基于小波变换与模糊神经网络的ECG自动分析

【作者】 张艳丽；

【导师】 陈振生；

【作者基本信息】 山东大学 ， 生物医学工程， 2005， 硕士

【摘要】 心血管疾病是当今危害人类健康的主要疾病之一,心电图(ECG)检查是临床诊断心血管疾病的重要方法。利用计算机对ECG信号进行自动准确的分析是国内外学者热衷的课题。ECG自动分析的任务主要包括对ECG信号进行滤波预处理、进行波形特征点的检测等,最终目的是对心电波形进行分类识别,根据其波形特点或者特征参数自动地做出疾病诊断。在各种心血管疾病中,室性早搏(PVC)是一种最常见的心律失常,它的实时、正确检测是ECG自动分析的一项重要技术,对提高心律失常检测、监护系统和动态心电图分析系统的性能起关键作用,对改善心脏病诊断有重要的实用价值。 利用小波变换进行ECG信号滤波处理与波形检测、用模糊推理技术及神经网络进行ECG分类识别等,在国内外已有不少报道,但算法还都不够完善。本课题就是要对小波变换和模糊神经网络在ECG自动分析中的应用,进行深入的研究和探讨,在总结前人工作的基础上,主要做了以下几方面的工作: 1.对基于小波变换的ECG信号滤波方法进行改进,即:将小波变换与自适应滤波技术相结合。这也是本论文的创新之处。经实验仿真验证,采用本文改进的滤波方法,可以在有效地去除噪声、干扰的同时,减少有用心电信息的损失,较好地保持心电信号的波形特征。 2.利用小波变换的信号奇异性检测理论,深入研究和探讨了ECG波形检测方法,即构造Marr小波滤波器,从等效滤波器的角度分析ECG信号的离散二进小波变换,通过对ECG信号小波变换模极大值列的探测,准确检测出R波、QRS波起止点、P波、T波等波形特征点。在R波检测中,采用了可变阈值、不应期、Lipschitz指数判据等策略,极大地提高了波形检测正确率。 3.进行室性早搏自动检测算法的研究:将模糊逻辑与神经网络进行有机结合,即利用具有模糊化输入、输出的前向神经网络的分类识别能力,实现ECG信号中正常心拍和室性早搏的自动识别,用MIT—BIH心电数据库中的样本数据对该模糊神经网络进行训练、检验,证明该网络具有较高的PVC自动识别率。更多还原

【Abstract】 Cardiovascular diseases are one of the main diseases that endanger human’s health nowadays. The electrocardiogram (ECG) examination is very important for clinical diagnosis of cardiovascular diseases. The application of computers at the accurate and automatic analysis of ECG has been a hotspot to researchers at home and overseas. The missions of the automatic analysis include the filter pretreatment of ECG signal and the examination of ECG characteristics etc. The final purpose is to classify ECG and diagnose diseases according to the characteristics of ECG. In various cardiovascular diseases, the Premature Ventricular Contract (PVC) is one of the most familiar diseases of the arrhythmia. Its real-time and accurate examination is an important technique of ECG automatic analysis. It operates a key function in increasing the performance of arrhythmia custody system and Holter ECG analysis system. It also has important practical value in improving the heart disease diagnosis.There have been had a lot of reports on the use of wavelet transform in carrying on the filter pretreatment of ECG signal and examining the ECG characteristics, and the use of fuzzy reasoning and artificial neutral network in classifying and identifying ECG. But these techniques aren’t perfect. In this paper, the applications of wavelet transform and fuzzy neural network at ECG automatic analysis were further researched and discussed. We mainly did following several works on the base of summarizing the past research work.Firstly, a modified method of ECG filter based on wavelet transform was put forward. That is to combine the wavelet transformand adaptive filter. It is an innovation of this paper. The experimental results demonstrated that the noises of ECG signal were successfully removed, the useful ECG information was perfectly preserved, and the modified denoising method was efficient and had better capability of filtering.Secondly, the ECG examination was further researched and discussed using the signal’s singularity examination theory of wavelet transform. Namely, Marr wavelet filters were designed, analyzed the discrete binary wavelet transform of ECG signal from the point of equivalent filter, the R wave、 jumping-off point and end point of QRS wave、 P wave and T wave were accurately examined by using Marr wavelet transform at ECG signal and examining the maximum line of wavelet transform. In the examination of R wave, several strategies such as volatile thresholds, disapprobatory interval, and Lipschitz exponent so on, were used for increasing the correctness of ECG characteristic examination.Thirdly, the automatic examination of PVC was researched. The automatic identification of normal ECG and PVC were realized using the classification and identification abilities of artificial neutral network with fuzzy input and output. The fuzzy neural network was proved having better automatic identification ability, after training and testing the neural network using the ECG datum of MIT-BIH database.更多还原