摘要
目的 评价基于人工智能(artificial intelligence, AI)的自动肺部超声评分(lung ultrasound score, LUS)对急性呼吸窘迫综合征(ARDS)患者血管外肺水(extravascular hung water, EVLW)评估的价值。方法 选择2019年1月至2022年6月上海长征医院急诊重症监护病房(EICU)符合ARDS诊断标准的28例患者作为研究对象,采用两种肺部超声评分方法:(1)基于两阶段级联深度学习模型评估ARDS的自动LUS(自动组LUS);(2)临床医生评估的LUS(人工组LUS),并采用脉波指示剂连续心排血量监测技术(PiCCO)监测血管外肺水指数(EVLWI),年龄18~80岁,性别不限,在行PiCCO检测前0.5 h内行肺部超声检查,计算LUS,并行动脉血气分析,记录氧合指数(PaO2/FiO2),比较两种LUS方法在不同严重程度ARDS患者诊断中的作用。结果 自动组LUS和人工组LUS与EVLWI高度相关(R^(2)=0.924 vs.R^(2)=0.910),在评估EVLW方面均显示出较高的准确性,LUS的受试者工作特征(ROC)曲线分析以PiCCO计算所得:(1)以EVLWI>7 mL/kg为界,对自动组和人工组LUS方法绘制ROC曲线,两组ROC曲线下面积(AUC)分别为0.956和0.947,两组方法的敏感度分别为90.8%和87.0%,特异度分别为94.3%和92.5%。(2)以EVLWI≥10 mL/kg为界,对自动组和人工组LUS方法绘制ROC曲线,两组ROC曲线AUC分别为0.979和0.978,两组方法的敏感度分别为92.2%和89.1%,特异度分别为97.9%和96.7%。(3)以EVLWI≥15 mL/kg为界,对自动组和人工组LUS方法绘制ROC曲线,两组ROC曲线AUC分别为0.997和0.996,两组方法的敏感度分别为94.5%和93.0%,特异度分别为98.8%和97.8%。结论 人工智能自动LUS可用于临床ARDS的诊断及严重程度的评估。
Objective To evaluate the value of automated lung ultrasound score(LUS)based on artificial intelligence in the assessment of extravascular lung water(EVLW) in acute respiratory distress syndrome(ARDS) patients. Methods Twenty-eight patients who met the diagnostic criteria of ARDS in the Emergency Intensive Care Unit(EICU) of Shanghai Changzheng Hospital from January 2019 to June 2022 were selected as the research objects. Two pulmonary ultrasound scoring methods were used:(1)Automated lung ultrasound score of ARDS based on two-stage cascade deep learning model(automatic group LUS);(2) clinician assessed lung ultrasound score(manual group LUS), and pulse indicator continuous cardiac output monitoring(PiCCO) was used to monitor extravascular lung water index(EVLWI). The patients′ ages were 18-80 years old regardless of gender. Lung ultrasound examination was performed within half an hour before PiCCO test, LUS was calculated, arterial blood gas analysis was performed, oxygenation index was recorded, and the role of two LUS methods in the diagnosis of ARDS with different severity was compared. Results LUS in the automatic group and LUS in the manual group were highly correlated with EVLWI(R^(2)=0.924 vs. R^(2)=0.910), and there was high accuracy in assessing EVLW. ROC analysis of LUS was bounded by PiCCO calculation(1)EVLWI>7 mL/kg. The ROC curves were drawn for the automatic and manual LUS methods. The areas under the ROC curve of the two groups were 0.956 and 0.947, respectively. The sensitivity of the two groups were 90.8% and 87.0%, and specificity was 94.3% and 92.5%, respectively.(2) With EVLWI≥10 mL/kg as the bound, ROC curves were drawn for the automatic and manual LUS methods. The areas under the ROC curve of the two groups were 0.979 and 0.978, respectively. The sensitivity of the two groups were 92.2% and 89.1%, and specificity was 97.9% and 96.7%, respectively.(3)EVLWI≥15 mL/kg was considered as the bound. The ROC curves were drawn for the automatic and manual LUS methods. The areas under the ROC curve of the two groups were 0.997 and 0.996, respectively. The sensitivity of the two groups were 94.5% and 93.0%, and specificity was 98.8% and 97.8%, respectively. Conclusions Automated LUS can be used to diagnose and evaluate the severity of clinical ARDS.
作者
范浩浩
姜倩倩
邢文宇
陈建刚
何超
李文放
Fan Hao-hao;Jiang Qian-qian;Xing Wen-yu;Chen Jian-gang;He Chao;Li Wen-fang(Department of Emergency Intensive Care Unit,the Second Affiliated Hospital of Naval Medical University(Shanghai Changzheng Hospital),Shanghai 200003,China)
出处
《中国急救医学》
CAS
CSCD
2023年第1期24-29,共6页
Chinese Journal of Critical Care Medicine
基金
上海市“科技创新行动计划”医学创新研究专项项目(21Y11902500)。