基于小波包和神经网络的心电信号分类方法研究

发布时间：2018-03-29 01:09

本文选题：心电信号　切入点：特征提取　出处：《天津工业大学》2017年硕士论文

【摘要】：心电图是人体心脏电活动最直接的反映,是医生进行心脏病诊断治疗的重要依据之一。随着科学和技术的发展,基于心电图的自动分析诊断技术已被广泛地用于心脏病检测和诊断的研究。基于心电图的自动分析诊断技术不仅大大降低了医生的工作量,而且可以显著地提高心电图分类的效率和准确率,对于心脏病及时的诊断和治疗具有重要的实际应用价值。因此,本文主要针对心电自动分析诊断技术中的心电信号分类识别方法进行了深入的研究,主要研究内容包括心电信号的特征提取和特征分类。提取稳定有效的心电信号特征是心电自动分析诊断技术中的重要环节,本文提出了一种基于小波包分解与统计分析相结合的心电信号特征提取算法。该算法首先采用小波包分解方法对心电信号进行四尺度分解,然后结合统计分析方法计算小波包分解后第四尺度上的16个小波包系数的奇异值、标准差和最大值,将求得的48维小波包系数统计特征组成心电信号特征空间。为了尽可能地提高心电信号分类识别的效率和准确率,本文提出了一种基于遗传算法优化神经网络的心电信号特征选择和分类算法。通过遗传算法对心电信号特征空间进行降维得到25维心电信号特征,同时采用遗传算法对误差反向传播神经网络分类器的权值和阈值进行优化,将降维得到的心电信号特征输入到分类器中进行训练和预测,从而实现对MIT-BIH心律失常数据库中六类心电信号:正常心跳、左束支传导阻滞、右束支传导阻滞、起搏心跳、室性早搏和房性早搏的分类,测试集的识别准确率为97.78%,平均灵敏度、平均特异度和平均阳性预测值分别为97.86%、99.54%和97.81%。最后,本文通过基于MPS450多参数模拟仪组成的心电信号采集实验系统对六类心电信号进行采集,并对其进行特征提取和分类算法验证,识别准确率达到了 99.33%,平均灵敏度、平均特异度和平均阳性预测值分别为99.33%、99.87%和 99.36%。实验结果表明本文提出的特征提取算法和分类算法能够有效地提取稳定的心电信号特征,并通过遗传算法优化的神经网络分类器实现了对六类心电信号的高精度分类。因此,本文提出的心电信号分类方法可以有效地用于心律失常识别,对于心脏病的预防、诊断和治疗具有重要的意义。
[Abstract]:Electrocardiogram (ECG) is the most direct reflection of human heart electrical activity and one of the important bases for doctors to diagnose and treat heart disease. With the development of science and technology, Automatic analysis and diagnosis technology based on electrocardiogram has been widely used in heart disease detection and diagnosis. The automatic analysis and diagnosis technology based on electrocardiogram not only greatly reduces the workload of doctors, Moreover, it can improve the efficiency and accuracy of ECG classification, and has important practical value for the timely diagnosis and treatment of heart disease. In this paper, the classification and recognition method of ECG signal in ECG automatic analysis and diagnosis technology is studied deeply. The main research contents include the feature extraction and feature classification of ECG signals. The extraction of stable and effective ECG features is an important part of ECG automatic analysis and diagnosis technology. In this paper, a new ECG feature extraction algorithm based on wavelet packet decomposition and statistical analysis is proposed. Then the singular value, standard deviation and maximum value of 16 wavelet packet coefficients on the fourth scale after wavelet packet decomposition are calculated by using statistical analysis method. In order to improve the efficiency and accuracy of ECG classification and recognition, the statistical features of 48 dimensional wavelet packet coefficients are used to form the ECG feature space. In this paper, an algorithm of ECG feature selection and classification based on genetic algorithm optimization neural network is proposed. By using genetic algorithm to reduce the dimension of ECG feature space, 25 dimensional ECG feature can be obtained. At the same time, genetic algorithm is used to optimize the weights and thresholds of the neural network classifier with error back propagation, and the reduced dimension ECG features are input into the classifier for training and prediction. Thus, the classification of six kinds of ECG signals in MIT-BIH arrhythmia database: normal heartbeat, left bundle branch block, right bundle branch block, pacing heartbeat, ventricular premature beat and atrial premature beat was realized. The accuracy of recognition of the test set was 97.78 and the average sensitivity was 97.78. The average specificity and average positive predictive value are 97.86% and 97.81%, respectively. Finally, six kinds of ECG signals are collected by an experimental system of ECG signal acquisition based on MPS450 multiparameter analog instrument, and their feature extraction and classification algorithm are verified. The recognition accuracy is 99.33, the average sensitivity, average specificity and average positive predictive value are 99.33% and 99.36%, respectively. The experimental results show that the proposed feature extraction algorithm and classification algorithm can effectively extract stable ECG features. The neural network classifier optimized by genetic algorithm is used to realize the high accuracy classification of six kinds of ECG signals. Therefore, the ECG classification method proposed in this paper can be used to identify arrhythmia effectively and prevent heart disease. Diagnosis and treatment are of great significance.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R540.4;TP18

【参考文献】