基于随机森林算法构建冠心病人群左心室舒张功能不全的诊断模型  

Random Forest Algorithm-based Diagnostic Model for Left Ventricular Diastolic Dysfunction in Patients With Coronary Heart Disease

作　　者：刘梦怡 马红 林静茹 朱振辉 陆敏杰 吴伟春 王浩[1] LIU Mengyi;MA Hong;LIN Jingru;ZHU Zhenhui;LU Minjie;WU Weichun;WANG Hao(Department of Echocardiography,National Center for Cardiovascular Diseases and Fuwai Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100037,China;Department of Cardiology,The First Affiliated Hospital of Nanjing Medical University,Nanjing 211103,China;The Key Laboratory of Cardiovascular Imaging,National Center for Cardiovascular Diseases and Fuwai Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100037,China)

机构地区：[1]中国医学科学院、北京协和医学院国家心血管病中心、阜外医院超声影像中心,北京100037 [2]南京医科大学附属第一医院心内科,南京211103 [3]中国医学科学院、北京协和医学院国家心血管病中心、阜外医院、心血管影像重点实验室,北京100037

出　　处：《中国循环杂志》2023年第7期730-733,共4页Chinese Circulation Journal

基　　金：中国医学科学院心血管影像重点实验室(培育)建设项目(2019PT310025)。

摘　　要：目的:应用随机森林(random forest,RF)算法构建冠心病人群左心室舒张功能不全的诊断模型。方法:纳入2013年10月至2014年7月就诊于中国医学科学院阜外医院的84例冠心病患者及30例正常志愿者,以5:5的比例随机分配至训练组(n=57)和测试组(n=57)。整合15项临床特征及13项超声心动图特征搭建模型,对左心室舒张功能进行分类预测。结果:RF模型预测冠心病人群左心室舒张功能不全的AUC为0.95,平均准确率达92.9%,F1分数为95.5%。灵敏度、特异度、阳性预测值及阴性预测值分别为89.5%、95.6%、93.3%、93.0%。RF模型与左心导管检查之间的Kappa值为0.85,2016年美国超声心动图学会发布的《关于评估左心室舒张功能不全的建议》(简称2016舒张功能指南)Kappa值为0.35。结论:本研究证明了人工智能模型预测冠心病人群左心室舒张功能不全的可行性。与2016年舒张功能指南相比,RF模型诊断准确性及一致性较好。Objectives:To establish a diagnostic model for left ventricular diastolic dysfunction in patients with coronary heart disease by using random forest(RF)algorithm.Methods:84 patients with coronary heart disease and 30 healthy volunteers in Fuwai Hospital from October 2013 to July 2014 were randomly divided into training group(n=57)and test group(n=57)at the ratio of 5:5.15 clinical features and 13 echocardiographic features were integrated to build a model to predict left ventricular diastolic function.Results:The AUC of RF model for predicting left ventricular diastolic dysfunction in patients with coronary heart disease was 0.95,with an average accuracy of 92.9%and F1 score of 95.5%.The sensitivity,specificity,positive predictive value and negative predictive value were 89.5%,95.6%,93.3%and 93.0%,respectively.The Kappa value was 0.85,while the Kappa value of 2016 diastolic function guide was 0.35.Conclusions:This study proves the feasibility of using artificial intelligence model in predicting left ventricular diastolic dysfunction in patients with coronary heart disease.RF model shows satisfactory diagnostic accuracy and consistency as compared with the 2016 recommendations for the evaluation of left ventricular diastolic function by echocardiography.

关 键 词：超声心动图 左心室舒张功能不全 随机森林模型 人工智能 

分 类 号：R54[医药卫生—心血管疾病]