人工智能在CT预测新型冠状病毒肺炎转归中的价值

Value of CT findings in predicting transformation of clinical types of COVID-19

目的探讨人工智能(artificial intelligence,AI)在CT预测新型冠状病毒肺炎(COVID-19)转归中的价值。方法分析2020年1月25日至2020年2月17日首都医科大学附属北京地坛医院确诊的普通型COVID-19肺炎首诊胸部CT、首次复查胸部CT的影像资料62例,其中,男性31例,女性31例。依据是否随病程发展为重型分为普通型组(51例)和转重型组(11例),应用AI技术定量分析组间差异。结果首次胸部CT转重型组患者肺炎病灶占整肺体积百分比3.3%,高于普通组1.3%,以病灶占整肺体积百分比2.0%为界值,诊断普通型转重型患者的敏感度(72.7%)、特异度(66.7%)最高,受试者工作特征(receiver operating characteristic,ROC)曲线下面积为0.744。复查CT两组患者病灶体积变化存在明显差异,与普通型组相比,转重型组病灶体积占比10.0%,明显增大,以增加的病灶占整肺体积百分比2.65%为界值,诊断普通型转重型的敏感度(90.9%)、特异度(78.4%)最高,ROC曲线下面积为0.896。两次CT比较两组患者病灶内部磨玻璃密度、实性密度成分比例变化差异有统计学意义(P<0.05)。结论人工智能在CT预测COVID-19转归中具有重要意义,可在早期对重症及危重症的发生进行预警,病灶占整肺体积百分比、病变短期内迅速进展可能对普通型COVID-19肺炎的预后产生重要影响。

Objective To explore the value of artificial intelligence(AI)in computed tomography(CT)prediction for outcome of 2019 Coronavirus disease(COVID-19).Methods The initial and first follow-up CT imaging of 62 cases with common COVID-19 pneumonia diagnosed in Beijing Ditan Hospital,Capital Medical University from January 25 to February 17,2020 were retrospectively analyzed.Ratio of male:female was 31∶31.Based on the fact whether patients conditions had deteriorated into severe type,all the cases were stratified into common type group(51 cases)and deteriorated type group(11cases).Differences of quantitative CT findings by AI in two groups of patients were analyzed.Results Based on initial CT,the percentage of volume of inflammatory lesions to volume of lungs in deteriorated group was 3.3%(1.6%,7.2%),higher than that of common group[1.3%(0.4%,2.4%)].Sensitivity and specificity of diagnose for deteriorated cases reached the best diagnostic value by setting cut-off value as 2.0%.The sensitivity was 72.7%,specificity was 66.7%,and area under a receiver operating characteristic(ROC)curve was 0.744.On first follow-up CT,significant difference was observed on the change of lesion volume.Compared to common type group,percentage of lesion volume in lungs increased significantly to 10.0%(8.9%,18.1%).The sensitivity(90.9%)and specificity(78.4%)of diagnose for deteriorated cases reached the best diagnostic value by setting cut-off value of increased percentage of lesions in lungs as 2.65%.Area under ROC was 0.896.Changes on percentage of ground glass opacity and consolidation in lesions were significantly different between the groups(P<0.05).Conclusions AI has a profound value in CT prediction for outcome of COVID-2019.It can help in early warning for severe and critical type of patients.The percentage of volume of lesions in lungs and rapidly increase may predict outcome of COVID-19.