人工智能在新型冠状病毒肺炎临床分型中的应用  

Application of artificial intelligence in clinical classification of COVID-19

作　　者：云永兴 王立非[1] 安超 张谍 杨根东[1] 冯凯 周昀 黄华[1] 刘文浩 YUN Yongxing;WANG Lifei;AN Chao;ZHANG Die;YANG Gendong;FENG Kai;ZHOU Yun;HUANG Hua;LIU Wenhao(Department of Radiology,the Third People's Hospital of Shenzhen,Shenzhen,Guangdong Province 518112,China)

机构地区：[1]深圳市第三人民医院放射科,广东深圳518112

出　　处：《实用放射学杂志》2021年第2期207-210,共4页Journal of Practical Radiology

基　　金：深圳市医疗卫生匚名匸程项目(SZSM201612053)。

摘　　要：目的探讨人工智能在诊断新型冠状病毒肺炎(COVID-19)临床分型中的应用价值.方法对158例最终经深圳疾控中心核酸检测阳性确诊COVID-19患者的胸部CT影像资料进行回顾性分析.把临床分型为普通型、重型、危重型的患者分为普通型和重型/危重型2组,应用推想科技影像人工智能软件(InferReadTM CT Pneumonia)对2组的肺炎体积及肺炎体积占全肺体积百分比(肺炎占比)进行自动识别和半定量计算.采用两独立样本t检验(Levene检验)比较2组资料的肺炎体积及肺炎占比间的差异;采用Spearman相关评价临床分型与肺炎体积及肺炎占比的相关性;采用受试者工作特征(ROC)曲线评估肺炎体积和肺炎占比诊断重型/危重型患者的诊断效能.结果158例确诊患者中,普通型125例,重型/危重型33例,其中男女比例89︰69,年龄14~86岁,平均(48.9±17.7)岁.普通型组的肺炎体积[(112.62±11.87)cm^(3)]与重型/危重型组[(307.91±39.67)cm^(3)]的差异有统计学意义(F=18.49,P<0.01);普通型组的肺炎占比[(2.62±0.36)%]与重型/危重型组[(9.69±1.55)%]的差异有统计学意义(F=26.59,P<0.01);肺炎体积与肺炎占比与临床分型呈轻度正相关(r=0.421,P<0.01;r=0.487,P<0.01);在临床分型鉴别中,肺炎体积与肺炎占比都有较高的准确性,曲线下面积(AUC)分别为0.799、0.846,肺炎体积以阈值56.725 cm3诊断重型/危重型的敏感度和特异度分别为0.939、0.504,肺炎占比以阈值2.13%诊断重型/危重型的敏感度和特异度分别为0.879、0.704.结论应用影像人工智能可以半定量计算COVID-19的肺炎体积及肺炎占比,能较准确地区分COVID-19的普通型和重型/危重型患者.Objective To explore the clinical value of artificial intelligence in the diagnosis of COVID-19 classification.Methods Retrospective analysis was performed on chest CT images of 158 patients who were confirmed positive nucleic acid test for COVID-19 by Shenzhen Center for Disease Control and Prevention.Patients with clin ical classifications of common,severe,and critically severe were divided into two groups:common group and severe/critically severe group.Artificial intelligence software(InferRead^(TM) CT Pneumonia)was used to automatically identify and semi-quantitative calculate pneumonia volume and the percentage of pneumonia volume in the total lung volume(pneumonia proportion)for the two groups.Two independent sample t test(Levene test)was used to compare the differences between pneumonia volume and pneumonia proportion in the two groups,and Spearman correlation was used to evaluate the correlation between clinical classification and pneumonia volume and pneumonia proportion,and receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficacy of pneumonia volume and pneumonia proportion in the diagnosis of severe/critically severe patients.Results 158 confirmed patients were collected in the study,125 common and 33 severe/critically severe.Male/female ratio:89/69.The age range:14—86 years,and the mean age was(48.9±17.7)years.The difference between the pneumonia volume in common group[(112.62±11.87)cm3]and the severe/critically severe group[(307.91±39.67)cm^(3)] were statistically significant(F=18.49,P<0.01),and the difference between the pneumonia proportion in the common group[(2.62±0.36)%]and the severe/critically severe group[(9.69±1.55)were statistically significant(F=26.59,P<0.01).And there was a slight positive correlation between the pneumonia volume and the pneumonia proportion with clinical classification(r=0.421,P<0.01;r=0.487,P<0.01).In the differential diagnosis of clinical classification,the pneumonia volume and the pneumonia proportion both had high accuracy,and t

关 键 词：人工智能 新型冠状病毒肺炎 计算机体层成像 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R563.1[自动化与计算机技术—控制科学与工程]