解剖M型超声人工智能模型测量儿童左室收缩功能的研究  

Artificial Intelligence Model of Anatomic M-mode Echocardiography for Measuring Left Ventricular Systolic Function in Children

作　　者：张丽 徐楠 邢佳怡 张婷婷 齐晓玲 王蓉[2] 逄坤静 Zhang Li;Xu Nan;Xing Jiayi;Zhang Tingting;Qi Xiaoling;Wang Rong;Pang Kunjing(Department of Echocardiography,Fuwai Hospital,National Center for Cardiovascular Diseases,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100037,China;Chongqing Medical University,Chongqing 400016,China)

机构地区：[1]中国医学科学院阜外医院超声影像中心,北京市100037 [2]重庆医科大学,重庆市400016

出　　处：《中国超声医学杂志》2023年第5期531-535,共5页Chinese Journal of Ultrasound in Medicine

基　　金：首都特色诊疗技术研究及示范应用(No.Z191100006619003);首都卫生发展科研专项(No.2022-1-4032)。

摘　　要：目的研究智能测量左室收缩功能的解剖M型超声心动图人工智能(AI)模型。方法从现有儿童心脏病超声心动图库中选择400例不同年龄段及不同心血管疾病的儿童超声心动图解剖M型图像进行人工标测,建立AI模型。另外纳入100例年龄和病种与训练集相匹配的儿童病例的解剖M型图像作为测试集,对模型进行测试。分析AI模型与有经验超声医师相比测量左室收缩功能相关参数的准确度和一致性。结果AI模型与超声医师测量的左室射血分数(LVEF)绝对值差7.09%±5.86%(62.21%±12.38%vs 65.77%±13.02%)。超声医师测量LVEF1的观察者内差异绝对值是9.30%±7.97%,观察者间是8.10%(2.0%,8.75%)。AI模型测量的LVEF与超声医师测量的LVEF1高度相关(r=0.780,P<0.001)。AI模型与超声医师测量的其余左室相关参数同样高度一致,包括左室舒张末容积(LVEDV,r=0.978)、左室收缩末容积(LVESV,r=0.977)、左室舒张末内径(LVEDD,r=0.983)、左室收缩末内径(LVESD,r=0.963)、左室短轴缩短率(FS,r=0.740),P<0.001。以LVEF1<50%定义为左室收缩功能减低,AI模型测量的LVEF预测左室收缩功能减低的ROC曲线下面积(AUC)是0.872,P<0.05。结论超声心动图解剖M型AI模型能够准确测量儿童的左室收缩功能相关参数,可用于识别儿童左室收缩功能减低。Objective To construct and assess the artificial intelligence(AI)model that can accurately measure left ventricular systolic function using anatomical M-mode images.Methods The 400 anatomical M-mode images of echocardiography in different-aged children with various type of cardiovascular diseases were chosen from our pediatric echocardiographic database to be measured manually to construct AI model.To assess the performance of the model,another loo anatomical M-mode images of children matched in disease and age with the training set were selected as the test set.The accuracy and consistency of the parameters of left ventricular systolic function were compared between the AI model and experienced echocardiographers.Results The left ventricular ejection fraction(LVEF)measured by the AI model was 7.09%±5.86%lower on average than LVEF1 measured by echocardiographers(62.21%±12.38%vs 65.77%±13.02%).The absolute value of inter-observer variation was 9.30%±7.97%and of intra-observer variation was 8.10%(2.0%,8.75%).The value of LVEF was highly correlated with LVEF1(r=0.780,P<0.001).Additionally,the rest left ventricular parameters measured by the AI model were also highly consistent with manual measurements,including left ventricular end-diastolic volume(LVEDV)(r=0.978),left ventricular end-systolic volume(LVESV)(r=0.977),left ventricular end-diastolic diameter(LVEDD)(r=0.983),left ventricular end systolic diameter(LVESD)(r=0.963)and left ventricular short axis fractional shortening(FS)(r=0.740),and their respective p values were all equal to<0.001.The area under the ROC curve(AUC)of LVEF measured by the AI model predicting decreased left ventricular function,defined as LVEF1 less than 50%,was 0.872(P<o.05).Conclusions AI model of anatomic M-mode echocardiography can accurately measure the parameters in children concerning left ventricular systolic function and is capable in detection decreased left ventricular systolic function.

关 键 词：人工智能 模型 左室收缩功能 解剖M型 超声心动图 儿童 

分 类 号：R445.1[医药卫生—影像医学与核医学] R725.4[医药卫生—诊断学]