神经网络智能诊断系统在心室流出道室性期前收缩定位诊断中的开发和应用  

Development and application of artificial neural network based intelligent system in the localization and diagnosis of premature ventricular contractions originating in the ventricular outflow tract

作　　者：李芳[1] 罗玉寅[1] 王萍[1] 诸帆[1] LI Fang;LUO Yuyin;WANG Ping;ZHU Fan(Cardiovascular Interventional Treatment Center,Huzhou First People's Hospital,Huzhou 313000,China)

机构地区：[1]湖州市第一人民医院心血管介入诊疗中心,313000

出　　处：《心电与循环》2022年第3期275-278,共4页Journal of Electrocardiology and Circulation

基　　金：浙江省医药卫生科技计划项目(2021PY075);湖州市科学技术局公益性应用研究项目(2020GY30)。

摘　　要：目的探讨神经网络智能诊断系统在心室流出道室性期前收缩定位诊断中的开发和应用。方法选取湖州市第一人民医院2018年1月至2020年12月就诊的4398例患者为研究对象,利用北京谷山丰心电网络工作站采集实时传输标准心室流出道室性期前收缩常规12导联心电图数据4398份并开发神经网络智能诊断系统,评价其F1值、灵敏度、特异度、阴性预测值和阳性预测值。选取本院2017年1月至2021年1月经射频导管消融证实为心室流出道室性期前收缩300例进行临床应用测试,比较神经网络智能诊断系统与人工诊断对心室流出道室性期前收缩定位诊断符合率。结果神经网络智能诊断系统在心室流出道室性期前收缩定位诊断中,总的F1值为0.809,多标签均值阴性预测值为0.780,多标签均值阳性预测值为0.820;标签阈值以最优F1值作为调整标准,整体更偏向于阴性预测值,可有效降低漏诊率。神经网络智能诊断系统与人工诊断对心室流出道室性期前收缩定位诊断符合率比较,差异均无统计学意义(均P>0.05)。结论基于深度学习开发的神经网络智能诊断系统在心室流出道室性期前收缩定位诊断中具有重要应用价值,有助于室性期前收缩的快速定位分型。Objective To explore the development and application of artificial neural network based intelligent system in the localization and diagnosis of premature ventricular contractions(PVCs)originatin in the ventricular outflow tract.Methods Four thousand three hundred and ninety-eight patients were selected from Huzhou First People’s Hospital from January 2018 to December 2020.Real-time transmission of 4398 routine 12-lead ECG data with PVCs from standard ventricular outflow tract were collected to develop the artificial neural network based intelligent system.The F1 value,sensitivity,specificity,negative predictive value and positive predictive value were evaluated.A total of 300cases of PVCs originating in ventricular outflow tract were confirmed by radiofrequency catheter ablation in our hospital from January 2017 to January 2021.These data were selected for clinical application testing,and the consistency rate of localization diagnosis of ventricular outflow tract premature contraction confirmed by artificial neural network based intelligent system and manual diagnosis was compared.Results The total F1 value of the artificial neural network based intelligent system was 0.809,the negative predictive value of the multi-label mean was 0.780,and the positive predictive value of the multi-label mean was 0.820.The optimal F1 value was taken as the adjustment standard,and the label threshold was more inclined to negative predictive value on the whole,which could effectively reduce the rate of missed diagnosis.There was no significant difference in the coincidence rate between manual diagnosis and the artificial neural network based intelligent system(all P>0.05)in the diagnosis of PVCs originating in ventricular outflow tract.Conclusion The artificial neural network based intelligent system based on deep learning has important application value in the localization and diagnosis of ventricular outflow tract premature contraction,which is helpful for rapid localization and classification of PVCs.

关 键 词：心电图 智能诊断系统 心室流出道 室性期前收缩 定位 深度神经网络 

分 类 号：R541.7[医药卫生—心血管疾病]