基于人工智能对广州新型冠状病毒感染住院患者肺部炎症负荷的分层量化分析及临床匹配度研究  

Artificial intelligence-based stratified quantitative analysis of pulmonary inflammatory load and clinical matching of hospitalized patients with coronavirus disease 2019 in Guangzhou

作　　者：陈熙来 吴宏楷 甘清鑫[3] 曾庆思[1] 刘晋新[3] 张烈光[3] 邓宇[1] 李征途 李时悦 黎毅敏 钟南山 Chen Xilai;Wu Hongkai;Gan Qingxin;Zeng Qingsi;Liu Jinxin;Zhang Lieguang;Deng Yu;Li Zhengtu;Li Shiyue;Li Yimin;Zhong Nanshan(Department of Radiology,The First Hospital of Guangzhou Medical University,Guangzhou 510120,China;Guangzhou Institute of Respiratory Health,Guangzhou 510163,China;Department of Radiology,Guangzhou Eighth People’s Hospital,Guangzhou 510060,China)

机构地区：[1]广州医科大学附属第一医院放射科,广州510120 [2]广州呼吸健康研究院,广州510163 [3]广州市第八人民医院放射科,广州510060

出　　处：《中华生物医学工程杂志》2023年第5期495-502,共8页Chinese Journal of Biomedical Engineering

基　　金："十四五"国家重点研发计划(2021YFC2500703);国家重点研发计划(2023YFC3041700);广州国家实验室科研任务项目(SRPG23-001);呼吸疾病国家重点实验室开放课题(SKLRD-OP-201906)。

摘　　要：目的基于人工智能(AI)技术对奥密克戎变异株流行期间,广州新型冠状病毒感染(COVID-19)住院患者肺部炎症负荷进行不同分层的量化分析,并探索影像分型与临床分型间的匹配性。方法回顾性筛选广州医科大学附属第一医院2022年12月至2023年1月的COVID-19住院患者836例,收集符合纳入标准的348例临床资料及胸部CT影像学数据,同时使用AI胸部CT评估系统量化分析不同亚组的炎症负荷。纳入相同时段内广州医科大学附属第一医院及广州医科大学附属市八医院入院时临床分型为重型和危重型的COVID-19患者368例,基于视觉评价及AI进行临床分型及影像学分型的匹配度分析。结果基于双肺、单侧肺及肺叶等多维度对比分析,重症COVID-19患者总病灶,以及磨玻璃病灶、实变病灶的体积(cm^(3))及占比(%)(炎症负荷)均大于非重症患者(均P<0.001)。重症COVID-19患者的全肺及各肺叶(除右肺中叶外)肺炎评分高于非重症患者(均P<0.05),但全肺体积两组间差异无统计学意义(P>0.05)。危重型及重型2个亚组之间的炎症负荷指标差异无统计学意义(P>0.05)。重症监护室(ICU)患者的双肺病灶体积、磨玻璃影及实变影体积、双肺病灶占比、肺炎分级、全肺评分均大于普通病房患者,全肺体积小于后者(均P<0.05)。入院时临床诊断为重症COVID-19的368例患者,影像学诊断为重症-危重型病毒性肺炎(VP)为178例(48%),影像学轻型-普通型VP为145例(39%),影像学不确定COVID-19所致的呼吸重症为9例(3%),影像学非COVID-19所致的呼吸重症为36例(10%)。结论入院临床诊断重症患者有半数以上在影像学并非重症COVID-19,在临床救治中,应重视呼吸系统及其它系统的基础疾病。基于AI的肺部炎症负荷量化分析可为COVID-19患者的精准诊治提供影像学依据。Objective To quantitatively analyze the stratified pulmonary inflammation burden of hospitalized patients with coronavirus disease 2019(COVID-19)in Guangzhou during the epidemic of Omicron virus strain based on artificial intelligence(AI),and explore the correlation between imaging phenotype and clinical phenotype.Methods Retrospective screening of 836 patients hospitalized with COVID-19 from December 2022 to January 2023 at the First Hospital of Guangzhou Medical University.The clinical and chest CT imaging data of 348 cases meeting the inclusion criteria were collected.The CT images were analyzed using an AI chest CT assessment system.368 COVID-19 patients who were clinically classified as heavy and critical at the time of admission to the First Hospital of Guangzhou Medical University and the Eighth Hospital of Guangzhou Medical University were included within the same period.The matching analysis of clinical phenotype and imaging phenotype was performed based on visual evaluation and AI.Results The inflammation burden including the total lesions,volume(cm^(3))and percentage(%)of ground glass and solid lesions were greater in critically ill patients with COVID-19 than in non-critically ill patients(all P<0.001)in three dimensions including bilateral,unilateral and lobar lung.Pneumonia scores in the whole lung and in each lobe(except the middle lobe of the right lung)were higher in critically ill patients with COVID-19 than in non-critically ill patients(all P<0.05),but whole lung volumes did not differ between the two groups(P>0.05).There was no difference in inflammation burden between the two subgroups of critical and heavy phenotype(P>0.05).Bilateral lung lesion volume,ground glass and solid lesions volume,bilateral lung lesions percentage,pneumonia grading,and whole lung score were greater in ICU patients with COVID-19 than in general ward patients,and whole lung volume was less than the latter(all P<0.05).Of the 368 patients with a clinical diagnosis of severe COVID-19 at admission,the imaging diagnosis w

关 键 词：奥密克戎 新型冠状病毒感染 人工智能 体层摄影术 X线计算机 

分 类 号：R511[医药卫生—内科学]