新型冠状病毒肺炎不同临床分型的定量CT影像参数表现

目的探讨新型冠状病毒肺炎(COVID-19)患者不同临床分型的CT影像定量表现。方法多中心回顾性分析2020年1月至2020年3月在浙江省五家医院确诊的COVID-19患者81例,搜集患者的一般资料、临床表现、临床分型、胸部CT影像和实验室检查资料。基于深度学习技术和患者的胸部CT影像定量参数:受累肺叶数、不同肺叶肺段的感染体积占比(%)与密度分布占比(%),采用多组别独立样本的Kmskal-Wallis秩和检验,分析比较非正态分布量化参数。结果COVID-19轻型29例,COVID-19普通型44例,COVID-19重型8例,三组患者的年龄差异有统计学意义(P=0.018)。COVID-19轻型出现发热(>38℃)的比例为55.2%,普通型为72.7%,重型为100%。实验室检查,COVID-19轻型和普通型均表现为中粒细胞计数正常或降低,重型患者的中粒细胞计数显著升高(P=0.042)。三组的受累肺叶个数分别为0(2)、4(2)、4.5(1),组间差异有统计学意义(P<0.001)。三组的肺炎病灶体积占比分别为0(0.18%)、1.81%(3.69%)、5.14%(18.03%),轻型显著小于普通型和重型,且各肺叶肺段均具有统计学意义(P<0.001);感染病灶密度分布呈现同样趋势,且在CT值区间[-570,-60]HU轻型显著小于普通型和重型(P<0.001),而普通型与重型的肺炎病灶体积占比、感染病灶密度分布均无明显统计学差异(P>0.05)。结论基于人工校正的深度学习分割模型可实现对COVID-19不同临床分型的受累肺叶个数、CT病灶分布、病灶密度分布的快速自动评估。胸部CT不仅可以早期诊断COVID-19,还能对其临床病程及严重程度进行评估。

Objective To investigate the quantitative CT images of different clinical types in COVID-19 patients.Methods A multi-center retrospective analysis was performed on 81 patients diagnosed with COVID-19 in five hospitals in Zhejiang province from January 2020 to March 2020.General information,clinical manifestations,clinical typing,chest CT images and laboratory examination data were collected.Based on deep learning techniques and quantitative parameters of chest CT images of patients:number of affected lobes,proportion of infected volume(%)and proportion of density distribution(%),kruskal-Wallis rank sum test of multiple groups of independent samples was used to analyze and compare quantitative parameters of non-normal distribution.Results There were 29 cases of niild COVID-19,44 cases of ordinary COVID-19,and 8 cases of severe COVID-19.The age difference among the three groups was statistically significant(P=0.018).COVID-19 fever(>38 XI)occurred in 55.2% of mild cases,72.7% of normal cases,and 100% of severe cases.Laboratory tests showed normal or reduced granulocyte counts in both mild and normal forms of COVID-19,and significandy increased granulocyte counts in severe patients(P=0.042).The number of affected pulmonary lobes in the three groups was 0(2),4(2)and 4.5(1),respectively,and the difference between the groups was statistically significant(P<0.001).The proportion of the volume of pneumonia lesions in the three groups was 0(0.18%),1.81%(3.69%)and 5.14%(18.03%),respectively.The proportion of the volume of pneumonia lesions in the three groups was significantly smaller than that in the normal type and the severe type,and the lung segments of each lung lobe had statistical significance(P<0.001).The density distribution of infected lesions showed the same trend,and in the CT value range[-570,-60]HU mild type was significantly smaller than that of normal type and severe type(P<0.001),while there were no significant differences in the proportion of pneumonia lesions volume and density distribution of infected lesions between normal type and severe type(P>0.05).Conclusion The deep learning segmentation model based on manual correction can realize the rapid and automatic evaluation of the number of affected lung lobes,CT lesion distribution and lesion density distribution of different clinical types of COVID-19.Chest CT can not only diagnose COVID-19 early,but also assess its clinical course and severity.