基于高斯混合模型的标准心电波形筛选

Complete ECG waveform selection based on Gaussian Mixture Model

针对基于心电信号身份识别中,由于随机选取的心电波形不完整(如T波缺失、R波缺失等)、变形等而影响身份识别准确率的问题,提出基于高斯混合模型的心电波形筛选方法.该方法将信号发生器产生的标准的心电信号切割成单心动周期,提取每个心动周期幅值、斜率、弧长及面积等46个特征,建立单心动周期标准心电波形的高斯混合模型,通过计算马氏距离判断某心电波形是否符合标准心电波形的高斯模型分布,实现具有明显PQRST特征的心电波形筛选.该方法在52人实际采集心电数据和ECG-ID数据库上分别进行了三组实验,不完整波形的平均拒绝率分别为97.87%、98.69%,能够有效挑选出完整的心电波形.

Randomly selected waveform may be incomplete or deformed,which result in low accuracy for electrocardiograph（ECG）identification. A method of ECG waveform selection based on Gaussian Mixture Model（GMM）is proposed. The standard ECG signal generated by the signal generator is cut into a single cardiac cycle. The amplitude,slope,arc length and area of each cardiac cycle are extracted,which formed the vector parameter feature of 46 Dimension. The GMM of single cardiac cycle standard ECG waveform was established with these 46 characteristics. By calculating Mahalanobis distance,we can determine whether a ECG waveform conforms to the GMM distribution of the standard ECG waveform. ECG waveform selecting with obvious PQRST characteristics is realized. The performance of the presented method is tested on the 30 individuals of ECG-ID data set and 52 individuals of the hand ECG data collected in real scenery.The average incomplete waveform rejection rate is 97.87% and 98.69% respectively. The results show that the complete ECG waveform can be selected.