定量CT评估新型冠状病毒肺炎患者肺部改变的可行性分析  

Feasibility Analysis of Quantitative CT Assessment of Lung Changes in Corona Virus Disease 2019(COVID-19)

作　　者：陈静 金晨望[2] 郭佑民[2] 沈聪 于勇 段海峰 于楠 CHEN Jing;YU Yong;DUAN Hai-feng;JIN Chen-wang;SHEN Cong;YU Nan(Affiliated Hospital of Shaanxi University of Chinese Medicine,Radiology Department,Xianyang 712000,Shaanxi Province,China;The first Affiliated Hospital of Xi'An JiaoTong University,Radiology Department,Xi’an 71006,Shaanxi Province,China)

机构地区：[1]陕西中医药大学附属医院医学影像科,陕西咸阳712000 [2]西安交通大学第一附属医院医学影像科,陕西西安716000

出　　处：《中国CT和MRI杂志》2022年第10期28-30,共3页Chinese Journal of CT and MRI

基　　金：陕西省自然科学基础研究计划项目(2019JM-361);陕西中医药大学创新团队(2019-QN09);陕西中医药大学附属医院科研课题(2020QN012)。

摘　　要：目的探讨基于深度学习模型的定量CT方法对新型冠状病毒肺炎(COVID-19)肺部改变评估的可行性。方法回顾性分析2020年1月至3月确诊的22例COVID-19患者影像学资料,采用定量CT方法检测病变范围占全肺体积百分比(LOV%)和病变质量(Mass),并进一步评估CT定量检测的准确性;使用ROC曲线分析定量CT参数对不同临床分型COVID-19患者的区分力。结果22例确诊患者中,9例重型患者,13例普型患者。以人工判读作为参考标准,CT定量方法与参考标准对病变体积测量结果无显著统计学差异(P=0.934)。所获得的LOV%与Mass值在不同临床分型(普通型和重型)患者中存在差异。普通型患者LOV%在0.05%~5.67%之间,重型患者LOV%为5.00%-36.68%。普通型和重型患者的Mass值也有显著差异(56.8vs516.4)。识别重型患者的最佳LOV%阈值为10.37%,最佳Mass值阈值为160(曲线下面积1.000,敏感度100%,特异性100%)。结论采用基于深度学习的定量CT方法测量COVID-19肺部病变LOV%与Mass的变化,能够更加客观精准的对COVID-19患者进行临床分型,观测疾病进展,及早诊断危重型患者,对临床干预治疗具有重要提示意义。Objective To explore the feasibility of quantitative CT method based on deep learning model to evaluate the lung changes of Corona Virus Disease 2019(COVID-19).Methods The imaging data of 22 patients with COVID-19 diagnosed from January to February 2020 were retrospectively analyzed.Quantitative CT method was used to detect the percentage of lesions(LOV%)and the mass of lesions(Mass),and further evaluation accuracy of CT quantitative detection;using ROC curve to analyze the discriminatory power of quantitative CT parameters for patients with different clinical types of COVID-19.Results Among the 22 confirmed patients,9 were severe patients and 13 were ordinary patients.Taking manual interpretation as the reference standard,the CT quantitative method and the reference standard had no statistically significant difference in lesion volume measurement results(P=0.934).The obtained LOV%and Mass values are different in patients with different clinical types(normal and severe).The LOV%of normal patients is between 0.05%and 5.67%,and the LOV%of heavy patients is between 5.00%and 36.68%.There is also a significant difference in mass values between normal and severe patients(56.8 vs 516.4).The optimal LOV%threshold for identifying severe patients is 10.37%,and the optimal Mass threshold is 160(area under the curve 1.000,sensitivity 100%,specificity 100%).Conclusion Using the quantitative CT method based on deep learning to measure the changes of COVID-19 lung lesions LOV%and Mass can more objectively and accurately clinically classify COVID-19 patients,observe disease progression,and diagnose critically ill patients early.Clinical intervention therapy has important implications.

关 键 词：新型冠状病毒 肺炎 定量CT 计算机辅助诊断 

分 类 号：R445.3[医药卫生—影像医学与核医学]