摘要
本文提出了一种基于心电数据相邻数据点斜率的心电图异常检测方法。该方法首先对数据进行高斯滤波,然后计算相邻数据点的斜率,接着根据时序将各个斜率进行拼接得到一维数据串,最后将数据串通过宽度为50的窗口以滑动的方式读取到RNN神经网络进行异常判定。本文通过实验分析后得到神经网络节点个数、遗忘率等各项参数与心电异常诊断准确率之间的联系,从而确定了神经网络三层结构以及各参数值。本文使用国际权威心电数据库四类共计1600条数据对神经网络进行训练,使用课题组采集的四类共计1491条心电数据进行算法验证。实验结果表明,本文提出的方法准确率可以达到97.0%的准确率,并且该方法具有良好的可靠性与适用性。
This paper presents an ECG anomaly detection method based on the slope of adjacent data points of ECG data.The method first performs Gaussian filtering on the data,then calculates the slope of the adjacent data points,then splices the respective slopes according to the timing to obtain a one-dimensional data string,and finally reads the data string to the RNN through the window of width 50in a sliding manner.The neural network makes an abnormality determination.Through experimental analysis,the relationship between the number of neural network nodes and the forgetting rate and the diagnostic,accuracy of ECG abnormality are obtained,and the three-layer structure of the neural network and the values of each parameter are determined.In this paper,a total of 1600data from four types of international authoritative ECG database were used to train the neural network,and four types of fourteen ECG data collected by the research group were used for algorithm verification.The experimental results show that the accuracy of the proposed method can reach 97.0%,and the method has good reliability and applicability.
作者
李锋
王泽南
LI Feng;WANG Ze-nan(College of Computer Science and Technology,Donghua University,Shanghai 201620)
出处
《智慧健康》
2018年第31期5-8,共4页
Smart Healthcare
关键词
斜率
循环神经网络
心电信号
异常诊断
Slope
Recurrent neural network
ECG signal
Abnormal diagnosis