心梗特征提取与辅助诊断模型研究

发布时间：2018-03-25 13:32

本文选题：心肌梗死　切入点：心电信号　出处：《郑州大学》2017年硕士论文

【摘要】：随着人们生活水平和生活方式的变化,心血管病的发病率和死亡率逐年升高,同时伴随有年轻化的趋势。心肌梗死作为心血管疾病的最大死因之一,是严重影响人们生命健康的重大疾病。目前对心肌梗死的诊断主要有心肌标记物水平和心电图分析两种方式,而心肌标记物的提取和分析复杂多样,很难在短时间内完成,因此对心电图的自动分析成为目前专家学者研究的重点。本文基于心肌梗死数据的背景下,总结前人的研究经验,对心电信号的噪声预处理、心电特征波形检测以及心肌梗死辅助诊断模型的建立进行了详细的研究。(1)心电信号的噪声预处理:本文按照心电信号各种噪声的来源以及实时性分析的要求,设计了一种能一次性滤除三种干扰的滤波器组对信号进行滤波。这种方法不仅可以有效滤除信号的低频基线漂移、工频干扰及肌电干扰等不同频率的噪声,而且不影响原始信号的重要信息。(2)心电特征波形的检测:本文提出了一种改进小波算法检测心电特征波形,这种算法在连续小波多尺度变换的基础上通过自适应地改变阈值和增加窗口函数检测特征波形,极大的提高了特征提取的准确性。小波基函数的选取在很大程度上有效的抑制了在心电信号噪声预处理阶段残留的噪声干扰,而小波尺度的选择和自适应阈值的选取则是经过大量的实验分析决定的。用PTB诊断数据库中的部分数据对此算法进行验证,仿真结果表明,R波检测的准确率达到99%以上,能准确定位干扰严重和波形畸变较大的心电特征信号。(3)心肌梗死辅助诊断模型的建立:基于心电信号的各典型波幅值、间期以及电位偏移情况等特征参数,主要设计了Logistic回归、BP神经网络以及基于K-CV的SVM三种自动分类诊断模型。在建模分析的过程中,重点对SVM的参数c和g的最优值进行了研究,提出了一种8折K-CV的二次寻优方法,很大程度上提高了分类器的性能。最后对这三种模型的性能进行了验证,结果表明,基于K-CV的SVM模型准确率达到了99.19%,可以为心肌梗死疾病的辅助诊断提供重要的理论指导和临床意义。
[Abstract]:With the change of people's living standard and lifestyle, the morbidity and mortality of cardiovascular disease increase year by year, and accompanied by the trend of younger age. Myocardial infarction is one of the biggest causes of death of cardiovascular disease. At present, the diagnosis of myocardial infarction mainly includes two ways: myocardial marker level and electrocardiogram analysis. However, the extraction and analysis of myocardial markers are complex and diverse, so it is difficult to complete the diagnosis in a short time. Therefore, the automatic analysis of ECG has become the focus of experts and scholars. Based on the background of myocardial infarction data, this paper summarizes the previous research experience, the ECG signal noise preprocessing, The detection of ECG characteristic waveform and the establishment of auxiliary diagnosis model of myocardial infarction were studied in detail. The noise pretreatment of ECG signal was carried out. According to the source of various kinds of noise of ECG signal and the requirement of real-time analysis, A filter bank which can filter three kinds of interference at one time is designed to filter the signal. This method can not only effectively filter the noise of low frequency baseline drift, power frequency interference and myoelectric interference, etc. And does not affect the important information of the original signal. 2) ECG characteristic waveform detection: this paper proposes an improved wavelet algorithm to detect ECG characteristic waveform, On the basis of continuous wavelet multiscale transform, this algorithm adaptively changes the threshold value and increases the window function to detect the characteristic waveform. The accuracy of feature extraction is greatly improved. The selection of wavelet basis function can effectively suppress the residual noise interference in ECG signal preprocessing. The selection of wavelet scale and adaptive threshold are determined by a large number of experiments. The algorithm is verified by some data in PTB diagnostic database. The simulation results show that the accuracy of R wave detection is over 99%. The establishment of an auxiliary diagnostic model of myocardial infarction, which can accurately locate the ECG characteristic signal with serious interference and large waveform distortion: based on the characteristic parameters of ECG amplitude, interval and potential deviation, etc. This paper mainly designs Logistic regression BP neural network and SVM automatic classification and diagnosis model based on K-CV. In the process of modeling and analysis, the optimal values of parameters c and g of SVM are studied, and a quadratic optimization method of 80% K-CV is put forward. The performance of the classifier is improved to a great extent. Finally, the performance of the three models is verified, and the results show that, The accuracy of SVM model based on K-CV is 99.19, which can provide important theoretical guidance and clinical significance for the auxiliary diagnosis of myocardial infarction disease.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R542.22;TN911.7

【相似文献】