高原肺水肿肺超声筛查深度学习模型构建与优化  

Construction and optimization of the deep learning model for pulmonary ultrasound screening with plateau pulmonary edema

作　　者：张铭杰 李梦娜 程明瑶 王宇亮[2] 粘永健 赵瑞臣[2] 杨宇 刘慕源 廖元 唐超 Zhang Mingjie;Li Mengna;Chen Mingyao;Wang Yuliang;Nian Yongjian;Zhao Ruicheng;Yang Yu;Liu Muyuan;Liao Yuan;Tang Chao(Medical School of Medicine,Xizang University,Lhasa 850032;Department of Critical Care Medicine,General Hospital of the Tibet Military Region,Lhasa 850032;School of Basic Medicine,Army Military Medical University,Chongqing 400000)

机构地区：[1]西藏大学医学院,拉萨850032 [2]西藏军区总医院重症医学科,拉萨850032 [3]陆军军医大学基础医学院,重庆400038

出　　处：《中华肺部疾病杂志(电子版)》2025年第1期68-73,共6页Chinese Journal of Lung Diseases(Electronic Edition)

基　　金：中国博士后科学基金(2017M623356)。

摘　　要：目的利用西藏军区总医院收集的肺超声图像数据,基于深度学习技术,建立一种适用于高原肺水肿(high altitude pulmonary edema,HAPE)的肺超声筛查自动技术,用于自动筛选HAPE病症,提升诊断的准确性。方法收集2021年1月至2023年12月期间,我院收治的高原肺水肿确诊病例共174例,按照分层的方式划分为训练集121例、验证集18例和测试集35例。采集患者肺超声图像,通过卷积神经网络(Convolutional Neural Network,CNN)模型进行图像识别和分析,系统进行多次训练和验证。模型被整合并优化以满足实时性和用户友好性需求,比较自动筛查技术系统与传统人工筛查方法诊断准确性。结果模型性能评估中,AI模型的敏感性为95.00%,特异性为96.00%和总体准确率为95.50%(包含训练集115张,验证集17张,测试集33张),高于医师组的敏感性84.33%,特异性87.67%和总体准确率85.50%(包含训练集106张,验证集16张,测试集31张)。统计学分析表明,AI系统与人工筛查方法在诊断敏感性、特异性及准确率上的差异具有统计学意义(P<0.05)。结论与传统人工筛查方法相比,AI模型在诊断敏感性、特异性和准确率表现优异,可提高临床诊断。Objective To establish an automatic lung ultrasound screening technology for high altitude pulmonary edema based on deep learning technology based on the ultrasonic lung image data collected by the General Hospital of Xizang Military Region,which can automatically screen the disease of HAPE and improve the diagnostic accuracy.Methods This study investigated the application effect of Convolutional Neural Network(CNN)based artificial intelligence(AI)model in the diagnosis of high altitude pulmonary edema.The study included 174 confirmed cases of high altitude pulmonary edema admitted to our hospital from January 2021 to December 2023.The cases were divided into 121 training sets,18 verification sets and 35 test sets according to random stratification.The research methods include the collection of patients'lung ultrasound image data,the automatic recognition and analysis of the image using CNN model,and the training and verification of the model for several times to improve the diagnostic performance.In the model performance evaluation,the diagnostic accuracy,recall(sensitivity)and specificity of the AI system wTo establish an automatic lung ultrasound screening technology for high altitude pulmonary edema based on deep learning based on the data of lung ultrasound images collected by the General Hospital of Xizang Military Region,and to automatically screen HAPE and improve the diagnostic accuracy.Methods:This study investigated the application effect of Convolutional Neural Network(CNN)based artificial intelligence(AI)model in the diagnosis of high altitude pulmonary edema.The study included 174 confirmed cases of high altitude pulmonary edema admitted to our hospital from January 2020 to December 2023.The cases were divided into 121 training sets,18 verification sets and 35 test sets according to random stratification.The research methods include the collection of patients′lung ultrasound image data,the automatic recognition and analysis of the image using CNN model,and the training and verification of the mode

关 键 词：高原肺水肿 肺部超声 自动筛选技术 模型 人工智能 

分 类 号：R563[医药卫生—呼吸系统]