心电信号的分析与识别方法研究

发布时间：2018-02-21 02:38

本文关键词： 心电图决策表心律失常分类高阶统计量粗糙集理论　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着现在人们生活水平的提高,心血管类的疾病的发病率也变得更高了,甚至被称为全球“头号杀手”。而对传统的人工对心电信号进行识别,容易由于医务人员疲劳等原因易产生误判或漏判。对心电信号的自动诊断能够将医务人员从烦琐的心电图形识别中解脱出来,提高诊断心血管疾病的效率。对此,本文提出了一个新的心律失常分类的方法,利用高阶统计量的方法和决策表分类器的方法对心电信号进行分析与识别。本文按照心电信号的自动诊断的顺序,从预处理到最后的心电波形识别的具体操作一步步作了介绍。通过心电仪器从病人身上得到的心电信号,需要经过一些预处理操作,去除工频干扰,基线漂移等噪音干扰。由于本文采用的是MIT-BIH心律失常标准数据库,因而预处理的除噪过程和波形检测的过程本文并没有详细介绍。本文所研究的主要内容有如下几点:1.由于特征提取的2个基本原则,一是让同类间的数据特征差异尽量小,二是让不同类间的数据特征差异尽量大。而本文经过对一些原始心电波形的形态进行观察和分析,发现基于原始心电信号的波形个体差异变得很大。本文采用了高阶统计量的方法,将原始波形变换到三个累积量(二阶累积量、三阶累积量和四阶累积量)上,通过在matlab中绘图并观测可知,同类型的心电信号的波形经过变换之后个体差异明显变小了。2.经过提取到高阶统计量的特征,结合心电信号波形的RR间期特征,得到一个含有18个特征的特征向量。本文利用一个基于粗糙集理论的决策表分类器进行分类。利用在特征提取中得到的18个特征,将其作为决策表的条件属性,而根据AAMI标准将所有的心电信号波形所分成5种类型便作为决策表的决策属性。利用MIT-BIH数据库中的心电信号记录对决策表进行训练,并约简,得到一个简化的决策表。本文便是利用这个简化的决策表对心电信号波形进行识别的。3.本文将MIT-BIH数据库中第101条记录中前1000个正常搏动类型心电波形训练决策表分类器,然后再对后续500条正常搏动类型的心电波形进行测验。可以得到该决策表的分类精度达到了90.2%。当所训练的样本足够大时,决策表分类器的分类效果很可观。
[Abstract]:With the improvement of people's living standards, the incidence of cardiovascular diseases has become higher, even known as the "number one killer" in the world. It is easy to misjudge or miss judgment because of medical staff fatigue and other reasons. Automatic diagnosis of ECG signals can free medical workers from tedious electrocardiogram recognition and improve the efficiency of diagnosing cardiovascular diseases. In this paper, a new method of arrhythmia classification is proposed, which uses the method of high order statistics and the method of decision table classifier to analyze and recognize ECG signal. The specific operation of ECG waveform recognition from pretreatment to final ECG waveform recognition is introduced step by step. The ECG signals obtained from patients by ECG instruments need some pretreatment operations to remove power frequency interference. Noise interference such as baseline drift. Because the MIT-BIH arrhythmia standard database is used in this paper, Therefore, the process of pre-processing noise removal and waveform detection is not introduced in detail in this paper. The main contents of this paper are as follows: 1.Because of the two basic principles of feature extraction, one is to make the difference of data features between the same kind as small as possible. The second is to make the data characteristics of different classes as different as possible. And this paper has observed and analyzed the shape of some original ECG waveforms. It is found that the individual difference of the waveform based on the original ECG signal becomes very large. In this paper, the original waveform is transformed into three cumulants (second-order cumulant, third-order cumulant and fourth-order cumulant) using the method of high-order statistics. By drawing and observing in matlab, we can see that the individual difference of the same type of ECG signal is obviously reduced after the transformation. 2. After extracting the feature of high order statistics, combining with the RR interval feature of ECG waveform, A feature vector containing 18 features is obtained. In this paper, a decision table classifier based on rough set theory is used to classify. 18 features obtained in feature extraction are used as conditional attributes of decision table. According to the AAMI standard, all ECG waveforms are divided into five types as decision attributes of decision tables. The ECG records in MIT-BIH database are used to train and reduce the decision tables. A simplified decision table is obtained. This paper uses this simplified decision table to identify ECG waveform. In this paper, the first 1000 normal pulse type ECG waveform training decision table classifiers in MIT-BIH database are used. Then, 500 ECG waveforms of normal pulsatile type are tested. The classification accuracy of the decision table is 90.2. When the trained sample is large enough, the classification effect of the decision table classifier is remarkable.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R540.4;TN911.7

【相似文献】