对截取波段处理的室性早搏分类方法研究

Classification method of ventricular premature beat processed with intercepted wavebands

以往检测室性早搏多是直接对整个心电周期进行处理,噪声干扰大,分类准确率难以提高,分类效率较低,文中直接处理截取波段从而提高检测准确率。用小波变换对心电信号进行预处理,标记R波,计算RR间期,将心电信号周期短的信号(不含P波)筛选出来,根据医生的建议及医学统计规律自动截取包含R波和T波的候选波段,用卷积神经网络对候选波段进行训练和分类。将候选波段输入卷积神经网络进行训练,识别率达到了预期效果。使用MIT-BIH心电数据库中的数据验证,其自动检测识别率达到97.3%,能够对医生的诊断提供有效帮助。

In the past,the detection of ventricular premature beat(VPB)was mostly performed directly on the whole ECG cycle,in which the noise interference was serious,the classification efficiency was low and the classification accuracy was difficult to be improved. In this paper,the intercepted waveband is directly processed to improve the detection accuracy. The wavelet transform is used to preprocess ECG signals,mark R wave,calculate RR intervals and screen out ECG signals(excluding P wave) with short cycles. The candidate wavebands including R wave and T wave are automatically intercepted according to doctor′s advice and medical statistical laws,and then are trained and classified by convolutional neural network(CNN). After training,the recognition rate of detection reaches the expected effect. The data in MIT-BIH ECG database verifies that the recognition rate of the automatic detection reaches 97.3%,which can provide effective help for doctors′ diagnosis.