人工智能辅助诊断技术在超声心动图诊断房间隔缺损中的应用研究  

Application of artificial intelligence assisted diagnostic technology in echocardiography diagnosis of atrial septal defect

作　　者：李文秀[1] 罗涛[2] 张文静[2] 武兴坤 刘爱军[1] 司锐 苏俊武[1] Li Wenxiu;Luo Tao;Zhang Wenjing;Wu Xingkun;Liu Aijun;Si Rui;Su Junwu(Pediatric Cardiovascular Center,Beijing Anzhen hospitul,Capitul Medicul University,Beijing lnstitute of HeartLung and Blood lesselDiseases,Beijing 100029,China;Beijing Laboratory of Adranced Information Networks,Beijing University of Posts and Telecommunications,Beijing 100876,China;Beijing Key Laboratory of Network System Architecture and Conrergence,Beijing University of Posts and Telecommuniceations,Beijing 100876,China)

机构地区：[1]首都医科大学附属北京安贞医院小儿心脏中心,北京市心肺血管疾病研究所,北京100029 [2]北京邮电大学北京先进信息网络实验室,北京100876 [3]北京邮电大学网络体系构建与融合北京市重点实验室,北京100876

出　　处：《中国医药》2022年第2期274-277,共4页China Medicine

基　　金：北京市自然科学基金(7202040)。

摘　　要：目的探讨人工智能辅助诊断技术在超声心动图诊断房间隔缺损(ASD)中的应用价值。方法回顾性分析2020年6—9月首都医科大学附属北京安贞医院小儿心脏中心确诊的80例ASD患者的临床资料。另选取同期因心脏杂音行超声心动图检查并排除ASD的体检者39例。收集所有受试者的二维及彩色多普勒超声心动图静态图像,按60%和40%的比例划分为训练集和测试集,训练集分别在RESNET50(残差神经网络)、NSENET121(密集连接卷积网络)以及人工智能辅助诊断模型中进行训练。比较3种模型在测试集中对ASD的诊断准确率、假阴性率及假阳性率。结果训练集由44例ASD患者与27例体检者组成,测试集由36例ASD患者与12例体检者组成。RESNET50、DENSENET121及人工智能辅助诊断模型对ASD的诊断准确率分别为83.3%(40/48)、85.4%(41/48)、97.9%(47/48),假阴性率分别为2.8%(1/48)、16.7%(8/48)、2.8%(1/48),假阳性率分别为58.3%(28/48)、13.9%(6/48)、0(0/48)。3个模型对ASD的诊断准确率、假阴性率、假阳性率比较差异均有统计学意义(χ^(2)=6.05,P=0.04;χ^(2)=10.53,P=0.01;χ^(2)=29.67,P<0.01)。结论人工智能辅助诊断技术可提高超声心动图诊断ASD的准确率,有推广应用价值。Objective To explore the application value of artificial intelligence(AI) assisted diagnostic technology in echocardiography diagnosis of atrial septal defect(ASD). Methods From June to September 2020, clinical data of 80 patients with ASD diagnosis in Pediatric cardiovascular center, Beijing Anzhen hospital, Capital Medical University were retrospectively analyzed. Another 39 cases of physical examinees who underwent echocardiography for heart murmur and excluded ASD diagnosis were enrolled at the same period. Static images of two-dimensional and color Doppler echocardiography of the all subjects were collected. The subjects were divided into training set and test set according to the proportion of 60% and 40%. The training set were trained in RESNET50(residual neural network), NSENET121(dense convolutional network) and AI assisted diagnostic model respectively. The diagnostic accuracy, false negative rate and false positive rate for ASD in test set were compared among the three groups. Results The training set consisted of 44 patients with ASD and 27 physical examinees, and the test set consisted of 36 patients with ASD and 12 physical examinees. The diagnostic accuracy for ASD in RESNET50, DENSENET121 and AI assisted diagnostic model was 83.3%(40/48), 85.4%(41/48) and 97.9%(47/48), false negative rate was 2.8%(1/48), 16.7%(8/48) and 2.8%(1/48), and false positive rate was 58.3%(28/48), 13.9%(6/48) and 0(0/48) respectively. There were significant differences in diagnostic accuracy, false negative rate and false positive rate for ASD among the three groups(χ^(2)=6.05, P=0.04;χ^(2)=10.53, P=0.01;χ^(2)=29.67, P<0.01). Conclusion AI assisted diagosis can improve the accuracy in echocardiography diagnosis of ASD and has application value.

关 键 词：房间隔缺损 人工智能 超声心动图 辅助诊断 

分 类 号：R541.1[医药卫生—心血管疾病]