基于熵和卷积神经网络的抗按压伪迹干扰心电自动识别模块研究

Research on ECG automatic recognition module against compression artifact interference based on entropy and convolutional neural network

目的胸外按压及电除颤是救治心脏骤停患者的两种常用手段,对患者进行电除颤前需分析其心电图(electrocardiogram,ECG),而胸外按压会干扰患者的心电分析。本研究设计了一种用于心肺复苏的心电分析算法以及配套心电采集模块,以解决因胸外按压伪迹导致的心电误识别为室颤的问题。方法该心电采集模块采用若干模拟数字电路搭建而成。自动分析程序采用经近似熵算法改进的卷积神经网络(convolutional neural network,CNN)模型进行心电分类。采用MIT-BIH心律失常数据库进行验证,并在实机测试中通过在心电采集模块中耦合同频段干扰进一步验证。结果相比无近似熵改进版本的分类程序,使用该CNN模型分析程序可将含同频段伪迹的正常心电识别准确率由CNN的86.3%提升至97.78%;在存在同频段伪迹的实机测试中,使用该分析程序可将心电识别准确率由20%提升至96%。测试结果表明,当存在非同频段伪迹干扰时,该设计模块分类准确;当存在同频段伪迹的情况下,对正常心电信号的心电分类准确率依然高达90%以上。结论基于熵和卷积神经网络模块的抗按压伪迹干扰心电自动识别性高、抗干扰能力强,后续可推荐其用于自动胸外按压电除颤一体机等高同频段伪迹场合的心电采集与自动诊断,以减少误判。

Objective External chest compression and electric defibrillation are two common methods to treat patients with cardiac arrest.Before electric defibrillation,the electrocardiograms(ECG)of patients need to be analyzed,and external chest compression may interfere with the ECG analysis of patients.In this study,an ECG analysis algorithm for cardiopulmonary resuscitation and a supporting ECG acquisition module are designed to solve the problem of ECG misidentification as ventricular fibrillation caused by chest compression artifacts.Methods The ECG acquisition module is built with several analog and digital circuits.The automatic analysis program uses the convolutional neural network(CNN)model improved by the approximate entropy algorithm to classify ECG.MIT-BIH arrhythmia database is used for verification,and in the real machine test,it is further verified by coupling the contract frequency band interference in the ECG acquisition module.Results Compared with the improved version of classification program without approximate entropy,this CNN model analysis program can improve the recognition accuracy of normal ECG containing artifacts in the same frequency band from 86.3%of CNN to 97.78%.The accuracy of ECG recognition can be improved from 20%to 96%by using this analysis program in real-world tests with artifacts in the same frequency band.The test results show that when there is artifact interference in different frequency bands,the classification of the design module is accurate.When there are artifacts in the same frequency band,the accuracy of ECG classification of normal ECG signals is still more than 90%.Conclusions The anti compression artifact interference ECG automatic recognition based on entropy and convolution neural network module has high anti-interference ability.It can be recommended for ECG acquisition and automatic diagnosis in the same high frequency band artifact occasions such as automatic external chest compression defibrillation machine,so as to reduce misjudgment.