摘要
目的探讨人工智能(AI)辅助定量测量评估新型冠状病毒肺炎(COVID-19)胸部CT动态变化的价值。方法回顾性分析2020年1月15日至3月10日在华中科技大学同济医学院附属武汉市中心医院接受治疗的99例确诊COVID-19患者的临床和胸部CT动态影像资料。依据最终诊断,99例患者分为普通型(36例)、重型(33例)和危重型(30例),分析3型间CT影像学表现,包括CT基本征象、肺炎病灶总体积及肺炎病灶总体积占全肺体积的百分比(体积比)。使用AI软件定量评价胸部CT影像的动态变化,定量指标有CT病灶峰值时间、病灶峰值总体积、病灶峰值体积比、总体积最大增长率、体积比最大增长率。采用Kruskal-Wallis秩和检验比较3型间定量指标的差异,以χ²检验或Fisher确切概率法比较3型间定性指标的差异。采用序列测量及散点图显示3型COVID-19病灶体积比的演变趋势,采用ROC曲线分析肺炎病灶体积比及其最大增长率预判普通型肺炎转为重型或危重型的价值。结果普通型、重型、危重型COVID-19患者年龄及性别分布差异有统计学意义(P<0.05),其中重型与危重型患者年龄显著高于普通型(P<0.01)。与普通型[2.5(1.0,5.0)d]和危重型[2.5(1.0,4.0)d]相比,重型发病至首次胸部CT扫描的时间延长[5.0(2.5,8.0)d,P<0.01]。普通型、重型、危重型COVID-19患者首诊肺部多叶受累的差异有统计学意义(分别为20例、29例、25例,χ²=10.403,P=0.006),其中重型和危重型患者多肺叶受累发生率显著高于普通型(P=0.002)。普通型、重型、危重型COVID-19患者首诊体积比差异有统计学意义[分别为1.0%(0.2%,4.7%)、9.30%(1.63%,26.83%)、2.10%(0.64%,8.61%),Z=14.236,P=0.001],其中重型患者体积比显著高于普通型(P<0.001),普通型与危重型差异无统计学意义(P=0.062)。随访CT显示肺炎病灶呈进展及恢复的动态转变,可见多期相病灶共存。3型COVID-19患者病灶体积比散点图中趋势线显示进展期病灶均呈由少到多的发展。普通型、重型及危重型病灶峰值体积比分别为9.75%(4.83%,13.18%)、29.80%(23.99%,42.36%)和61.81%(43.73%,72.82%),差异有统计学意义(Z=74.147,P<0.001),病灶体积比的最大增长率分别为1.27%(0.50%,1.81%)/d、4.39%(3.16%,5.54%)/d和6.02%(4.77%,9.96%)/d,差异有统计学意义(Z=52.453,P<0.001),病灶峰值时间分别为12.0(9.0,15.0)d、13.0(10.0,16.0)d和16.5(12.0,25.0)d,差异有统计学意义(Z=9.524,P=0.009)。以肺炎病灶峰值体积比22.60%、体积比的最大增长率1.875%/d为界值,诊断普通型转重型或危重型的灵敏度分别为92.10%、96.83%,特异度分别为100%、80.56%,曲线下面积分别为0.987、0.925。结论COVID-19肺炎病灶在CT上呈类抛物线样变化,采用AI技术对CT肺炎病灶体积比进行动态精确测量有助于评价疾病的严重程度和预判疾病的发展趋势,病灶体积比快速增长的患者更容易转变为重型或危重型。
Objective To investigate the value of artificial intelligence(AI)-assisted quantitative measurement in evaluation of the dynamic changes of CT for COVID-19 pneumonia.Methods The clinical and chest CT dynamic imaging data of 99 patients with confirmed COVID-19 pneumonia who were hospitalized in Wuhan Central Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology from January 15,2020 to March 10,2020 were retrospectively analyzed.According to the definitive diagnosis,the 99 patients were classified into common(n=36),severe(n=33)and critical(n=30)type,the CT imaging findings of each type were analyzed,including CT basic signs,total volume of pneumonia lesions and percentage of pneumonia lesions of the total lung volume(volume ratio).AI software was used to quantitatively evaluate the dynamic changes of chest CT images.The quantitative indicators included CT peak time of lesions,total volume of lesions peak,volume ratio of lesions peak,maximum growth rate of total volume and maximum growth rate of volume ratio.Kruskal-Wallis rank sum test was used to compare the difference of quantitative indexes between the 3 types,andχ2 test or Fisher exact probability test was used to compare the difference of qualitative indexes between the 3 types.Sequence measurement and scatter plots were used to show the evolution trend of the volume ratio of the three types of COVID-19 pneumonia lesions.The ROC curve was used to analyze the value of the volume ratio of pneumonia lesions and its maximum growth rate in predicting the conversion of common pneumonia to severe or critical pneumonia.Results There were statistically significant differences in age and gender distribution among patients with common,severe and critical COVID-19(P<0.05),the age of severe and critical types were significantly higher than that of common type(P<0.01).Compared with common[2.5(1.0,5.0)d]and critical type[2.5(1.0,4.0)d],the time from onset to the first chest CT scan of severe type was prolonged[5.0(2.5,8.0)d,P<0.01].There were statistically significant differences in involvement of multiple lung lobes(20 cases,29 cases,25 cases,χ²=10.403,P=0.006)in patients with common,severe and critical COVID-19 at the first scan,the incidence of the involvement of multiple lung lobes in severe and critical types was significantly higher than that of common type(P=0.002).The volume ratios of patients with common,severe and critical COVID-19 at the first scan were statistically significant[1.0%(0.2%,4.7%),9.30%(1.63%,26.83%),2.10%(0.64%,8.61%),Z=14.236,P=0.001],and the volume ratio of severe type was significantly higher than that of common type(P<0.001),there was no statistically significant difference between common type and critical type(P=0.062).Follow-up CT showed that the pneumonia lesions showed a dynamic transformation of progress and recovery,and it was seen that the coexistence of multiphase lesions.The trend line in the scatter plot of the three types of COVID-19 pneumonia lesions showed that the lesions in the advanced stage developed from less to more.The lesion peak volume ratios of the common,severe and critical types were 9.75%(4.83%,13.18%),29.80%(23.99%,42.36%)and 61.81%(43.73%,72.82%),respectively,the difference was statistically significant(Z=74.147,P<0.001).The maximum growth rates of lesion volume ratio were 1.27%(0.50%,1.81%)/d,4.39%(3.16%,5.54%)/d and 6.02%(4.77%,9.96%)/d,respectively,the difference was statistically significant(Z=52.453,P<0.001).The peak times of lesions were 12.0(9.0,15.0)d,13.0(10.0,16.0)d and 16.5(12.0,25.0)d,respectively,the difference was statistically significant(Z=9.524,P=0.009).Taking the volume ratio of pneumonia lesion 22.60%and the maximum growth rate of the volume ratio 1.875%/d as the boundary value,the sensitivity of diagnosing common type to severe or critical type was 92.10%and 96.83%,and the specificity was 100%and 80.56%,respectively.The area under the curve was 0.987 and 0.925,respectively.Conclusions The lesions of COVID-19 pneumonia show a similar parabolic change on CT imaging.The use of AI technology to dynamitcally and accurately measure the CT pneumonia lesion volume ratio is helpful to evaluate the severity of the disease and predict the development trend of the disease.Patients with a rapid growth of volume ratio are more likely to become severe or critical type.
作者
杜丹
谢元亮
李惠
赵胜超
丁义
杨培
刘彬
孙建清
王翔
Du Dan;Xie Yuanliang;Li Hui;Zhao Shengchao;Ding Yi;Yang Pei;Liu Bin;Sun Jianqing;Wang Xiang(Department of Radiology,Central Hospital of Wuhan,Tongji Medical College,Huazhong University of Science and Technology,Wuhan 430014,China;Philips(China)Investment Co.,LTD.,Shanghai 200072,China)
出处
《中华放射学杂志》
CAS
CSCD
北大核心
2021年第3期250-256,共7页
Chinese Journal of Radiology