首都医科大学学报 ›› 2020, Vol. 41 ›› Issue (3): 340-344.doi: 10.3969/j.issn.1006-7795.2020.03.004

• 新型冠状病毒肺炎的诊断、治疗及预后 • 上一篇    下一篇

人工智能在CT预测新型冠状病毒肺炎转归中的价值

吕志彬1, 关春爽1, 闫铄1, 陈七一1, 李晶晶1, 张羽君2, 陈步东1, 谢汝明1   

  1. 1. 首都医科大学附属北京地坛医院放射科, 北京 100015;
    2. 数坤(北京)网络科技有限公司, 北京 100102
  • 收稿日期:2020-03-20 出版日期:2020-06-21 发布日期:2020-06-17
  • 通讯作者: 谢汝明 E-mail:13911320739@163.com
  • 基金资助:
    北京市卫生健康科技成果与适宜技术推广专项(2020-TG-001)。

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

Lyu Zhibin1, Guan Chunshuang1, Yan Shuo1, Chen Qiyi1, Li Jingjing1, Zhang Yujun2, Chen Budong1, Xie Ruming1   

  1. 1. Department of Radiology, Beijing Ditan Hospital, Capital Medical University, Beijing 100015, China;
    2. Shukun(Beijing) Technology Co., Ltd., Beijing 100102, China
  • Received:2020-03-20 Online:2020-06-21 Published:2020-06-17
  • Supported by:
    This study was supported Beijing Health Science and Technology Achievements and Appropriate Technology Promotion Special(2020-TG-001).

摘要: 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。

关键词: 人工智能, 新型冠状病毒肺炎, 体层摄影术, X线计算机

Abstract: 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. 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.

Key words: artificial intelligence, COVID-19, tomography, X-ray computed

中图分类号: