摘要
目的构建基于卷积神经网络(convolutional neural network,CNN)的流出道室性期前收缩(outflow tract premature ventricular complex,OTPVC)心电信号的人工智能定位诊断模型。方法研究纳入了2015年10月至2019年11月就诊于广东省人民医院,经心电生理检查确诊为OTPVC的421例患者,其中右心室流出道(RVOT)330例,左心室流出道(LVOT)91例,采集每例患者各1份包含室性期前收缩(premature ventricular complex,PVC)的体表12导联心电信号及心电图。心电信号经过预处理,输入VGGNet、ResNet和MobileNet等3种CNN进行训练,构建3种人工智能OTPVC左右侧二分类定位诊断模型,比较3种模型的诊断效能,并与两名心电生理专科医师的诊断结果进行对比。结果基于VGGNet、ResNet和MobileNet架构的3种CNN模型的曲线下面积均为0.88,对OTPVC二分类定位诊断的准确度分别为87.65%、88.84%、89.31%,敏感度为92.12%、95.45%、95.45%,特异度为71.43%、64.84%、67.03%。两名心电生理医师对OTPVC的准确度分别为77.78%、79.01%,敏感度为93.70%、97.64%,特异度为20.00%、11.43%。3种CNN模型的诊断效能相近,其敏感度与两名医师相近,但准确度和特异度高于两名医师。结论本研究成功构建了OTPVC心电信号的3种CNN人工智能左右侧定位诊断模型,构建的人工智能模型均具有良好的诊断效能,有望在体表心电信号定位诊断中发挥重要作用。
Objectives on convolutional neural network(CNN)for outflow tract premature ventricular complex(OTPVC).Methods of 421 patients with OTPVC were enrolled in this study in Guangdong Provincial People′s Hospital from October 2015to November 2019. Cardiac electrophysiology study revealed that premature ventricular complex(PVC)originated from right ventricular outflow tract(RVOT)in 330 patients,and originated from left ventricular outflow tract(LVOT)in 91patients. 12-lead body surface ECG signal and ECG picture contained at least one PVC arrythmia were collected from each patient. After preprocessing,ECG signals were fed to VGGNet,ResNet and MobileNet to construct 3 CNN models to discriminate right or left origination OTPVC. Diagnostic efficiencies of these 3 CNN models were compared,and comparisons of CNN models with two cardiac electrophysiologists were also performed.Results curve(AUC)of VGGNet,ResNet and MobileNet models were all 0.88,and the accuracies were 87.65%,88.84%,89.31%,respectively. The sensitivities of VGGNet,ResNet and MobileNet models were 92.12%,95.45% and 95.45%,and the specificities were 71.43%,64.84% and 67.03%,respectively. For two cardiac electrophysiologists,the accuracies were 77.78% and 79.01%,the sensitivities were 93.70% and 97.64%,and the specificities were 20.00% and11.43%,respectively. The diagnostic efficiencies of these 3 CNN models were similar to each other. The accuracies and specificities of VGGNet,ResNet and MobileNet models were higher than those of the two cardiac electrophysiologists.Conclusions Three artificial intelligence models for OTPVC localization based on CNN were successfully constructed.This study highlights the feasibility of ECG signal artificial intelligence localization for OTPVC. Artificial intelligence ECG signal localization models might play prominent roles in clinical practice.
作者
文龙
胡鑫荣
梁东坡
林敏茵
刘甜
王树水
WEN Long;HU Xin-rong;LIANG Dong-po;LIN Min-yin;LIU Tian;WANG Shu-shui(Department of Pediatric Cardiology,Guangdong Cardiovascular Institute,Guangdong Provincial People′s Hospital,Guangdong Academy of Medical Sciences,Guangzhou 510080,China;Department of Computer Science and Engineering,College of Engineering,University of Notre Dame,United States)
出处
《岭南心血管病杂志》
CAS
2022年第2期144-150,共7页
South China Journal of Cardiovascular Diseases
基金
医疗流程辅助智能软硬件系统与应用(项目编号:2020AAA0109605)
广东省医学科学技术研究基金项目(项目编号:A2021072)。
关键词
室性期前收缩
人工智能
卷积神经网络
心电图
premature ventricular complex
artificial intelligence
convolutional neural network
electrocardiogram
