心电信号质量评估与去噪方法的研究与实现

发布时间：2018-06-13 20:04

  本文选题：心电信号 + 质量评估　； 参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：心血管疾病是对人类健康造成威胁的最严重的疾病之一,且具有高发病率、高患病率、高致残率、高死亡率等特点。心电图作为检测和诊断心血管疾病的重要方法之一,在采集过程中,经常受到各种噪声的干扰。被噪声干扰的心电信号,不但增加了医师的工作量,降低了诊断效率,而且增加了心脏监视器的错误报警率。为了减少噪声带来的影响,对心电信号的质量进行评估和信号去噪具有重要的意义。本文主要研究心电信号的质量评估方法和去噪方法,根据心电采集设备和使用对象的不同,分为十二导联心电信号和单导联心电信号。在对十二导联心电信号质量评估中,本文提出了一种基于多特征融合的心电信号质量评估方法,在该方法中提出了两种特征融合方式:一种是基于规则的方法,另一种是基于统计特性和机器学习的方法。并用两种方法在Physio Net/Computing in Cardiology Challenge 2011提供的数据库中的训练集和测试集上进行了测试,第一种方法获得的分类准确率是92.8%和90.4%,时间性能是0.78秒,第二种方法获得的分类准确率是94.0%和91.6%,时间性能是2.03秒。在对单导联心电信号质量评估中,本文从MIT-BIT心律不齐数据库中选取了干净的心电信号,向其中加入了三种来自MIT-BIH噪声压力测试数据库中的真实噪声,根据噪声水平的不同,制作了具有5个质量等级的单导联心电信号数据集。本文从单导联心电信号中总共提取了10个信号质量指数,并用支持向量机分类器在数据集上进行训练和测试,经过交叉验证,获得的分类准确率是79.94%,单类重叠准确率是98.75%。根据单导联心电信号中噪声的类型不同,本文针对每种类型的噪声使用了不同的去噪方法,包括数字滤波、自适应滤波和小波滤波等方法,并用每种方法对不同质量等级的心电信号进行了去噪处理,然后从视觉效果和信噪比两个方面进行评价和比较,最后为不同噪声类型和不同质量等级的心电信号选择了适当的去噪方法。基于前面的理论研究和分析,本文设计并实现了一个测试及应用平台,对平台的各个功能进行了测试。测试结果验证了平台各个功能的正确性和有效性,并对本文提出的质量评估方法和采用的去噪方法提供了较好的支持。
[Abstract]:Cardiovascular disease is one of the most serious diseases that threaten human health, and has the characteristics of high incidence, high morbidity, high rate of disability, high mortality, etc. electrocardiogram is one of the most important methods to detect and diagnose cardiovascular diseases. In the process of collecting, it is often disturbed by various noises. But it increases the workload of doctors, reduces the efficiency of diagnosis, and increases the false alarm rate of the heart monitor. In order to reduce the influence of noise, it is of great significance to evaluate the quality of the ECG signal and to denoise the signal. This paper mainly studies the quality evaluation method and denoising method of ECG signal, and set up the ECG acquisition according to the ECG acquisition. Different objects are divided into twelve lead ECG signals and single lead ECG signals. In the quality evaluation of twelve lead ECG signals, this paper presents a quality evaluation method based on multi feature fusion. In this method, two feature fusion methods are proposed: one is a rule based method, the other is the other. Based on statistical characteristics and machine learning methods, and using two methods to test the training set and test set in the database provided by Physio Net/Computing in Cardiology Challenge 2011, the classification accuracy of the first method is 92.8% and 90.4%, the time performance is 0.78 seconds, and the classification accuracy of the second methods is 9. 4% and 91.6%, the time performance is 2.03 seconds. In the quality evaluation of single lead ECG signal, this paper selects clean ECG signals from the MIT-BIT arrhythmia database, and adds three kinds of real noise from the MIT-BIH noise pressure test database. According to the difference of noise level, there are 5 quality grades. In this paper, a total of 10 signal quality indexes are extracted from the single lead ECG signal, and the support vector machine classifier is trained and tested on the data set. After cross validation, the accuracy of the classification is 79.94%, and the single class overlap accuracy is 98.75%. based on the type of noise in the single lead ECG signal. In the same way, this paper uses different denoising methods for each type of noise, including digital filtering, adaptive filtering and wavelet filtering, and uses each method to denoise the ECG signals of different quality grades, and then evaluates and compares the two sides of the visual effect and the signal to noise ratio, and finally the different noise types. Based on the previous theoretical research and analysis, a test and application platform is designed and implemented, and the functions of the platform are tested. The test results verify the correctness and effectiveness of the various functions of the platform, and the quality evaluation method proposed in this paper. And the method of denoising is well supported.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R540.4;TN911.7
，

本文编号：2015246