慢性阻塞性肺疾病肺功能的CT人工智能评估  

CT artificial intelligence assessment of pulmonary function in chronic obstructive pulmonary diseases

作　　者：符浩男 张珊珊 张敏鸽[1] 刘子姗 杨海[1] FU Haonan;ZHANG Shanshan;ZHANG Minge;LIU Zishan;YANG Hai(Department of Radiology,Taizhou Hospital of Zhejiang Province,Enze Hospital of Taizhou Enze Medical Center(Group),Taizhou 317000,Zhejiang,China;Department of Radiology,Taizhou Central Hospital,Taizhou University Hospital,Taizhou 318000,Zhejiang,China)

机构地区：[1]浙江省台州医院台州恩泽医疗中心(集团)恩泽医院放射科,浙江台州317000 [2]台州市中心医院台州学院附属医院放射科,浙江台州318000

出　　处：《中国现代医生》2025年第6期1-5,78,共6页China Modern Doctor

基　　金：浙江省医药卫生科技计划面上项目(2021KY1224)。

摘　　要：目的 探讨慢性阻塞性肺疾病(chronic obstructive pulmonary disease,COPD)患者行胸部计算机断层扫描(computed tomography,CT)时,基于人工智能(artificial intelligence,AI)模型算法的肺气肿全自动量化与肺功能的相关性。方法回顾性分析2020年12月至2021年5月于台州恩泽医疗中心(集团)恩泽医院接受胸部CT平扫检查的COPD住院患者的临床及影像资料。根据患者通气功能下降程度分为5个等级。利用AI模型计算COPD患者的肺气肿病变范围,识别低于–950HU的低衰减区域,并计算低衰减区域百分比(low attenuation area percentage,LAA%)。结合AI模型输出结果,根据变量是否满足正态分布的特征,分别计算COPD不同分级患者的1秒末用力呼气容积实测值占预计值百分比(percentage of measured forced expiratory volume at the end of 1 second to estimated value,FEV_(1)%)与各肺叶LAA%之间的Pearson相关系数及FEV_(1)占用力肺活量百分比(FEV_(1)as a percentage of forced vital capacity,FEV_(1)/FVC)与各肺叶LAA%之间的Spearman相关系数。结果 中度COPD患者全肺LAA%与FEV_(1)/FVC存在负相关关系(r=–0.632,P=0.001);极重度COPD患者全肺LAA%与FEV_(1)/FVC和FEV_(1)%均呈负相关(r=–0.562,P=0.045和r=–0.701,P=0.004)。肺段分析结果表明极重度COPD患者左肺上叶LAA%与肺功能指标相关性更强(r=–0.650,P=0.016和r=–0.731,P=0.002);中度COPD患者左肺下叶LAA%与FEV_(1)/FVC相关性更显著(r=–0.712,P=0.000)。吸烟患者中,右肺下叶LAA%与FEV_(1)%呈中度相关(r=–0.534,P=0.006),左肺下叶LAA%与FEV_(1)/FVC亦呈中度相关(r=–0.564,P=0.003)。结论 基于AI的肺气肿量化结果与FEV_(1)/FVC和FEV_(1)%具有良好的相关性,可为基于CT平扫图像的COPD诊断和分级提供有力支持。Objective To analyse correlation between automatic quantification of emphysema and lung function based on artificial intelligence(AI)model algorithm by chest computed tomography(CT)in patients with chronic obstructive pulmonary disease(COPD).Methods The clinical and imaging data of hospitalized COPD patients who received chest CT plain scan in Taizhou Hospital of Zhejiang Province,Enze Hospital of Taizhou Enze Medical Center(Group)from December 2020 to May 2021 were retrospectively collected,patients were classified into five levels of ventilator-function decline.By using the AI model,the extent of emphysema lesions in COPD patients were calculated,low-attenuation areas below–950HU were identified and their low attenuation area percentage(LAA%)were calculated.Combined with the output results of AI model and whether each variable met the characteristics of normal distribution,Pearson correlation coefficient between percentage of measured forced expiratory volume at the end of 1 second to estimated value(FEV_(1)%)and LAA%of each lung lobe,and the Spearman correlation coefficient between FEV_(1) as a percentage of forced vital capacity(FEV_(1)/FVC)and LAA%of each lung lobe in patients with different COPD grades were calculated respectively.Results There was a negative correlation between total lung LAA%and FEV_(1)/FVC in moderate COPD(r=–0.632,P=0.001).Total lung LAA%in very severe COPD was negatively correlated with both FEV_(1)/FVC and FEV_(1)%(r=–0.562,P=0.045 and r=–0.701,P=0.004).The results of lung segment analysis showed that LAA%of the left upper lung lobe was more strongly correlated with pulmonary function indicators in extremely severe COPD(r=–0.650,P=0.016 and r=–0.731,P=0.002).The correlation between left inferior lobe LAA%and FEV_(1)/FVC was stronger correlation in patients with moderate COPD(r=–0.712,P=0.000).In smoking patients,LAA%was moderate correlated with FEV_(1)(r=–0.534,P=0.006),and LAA%was moderate correlated with FEV_(1)/FVC(r=–0.564,P=0.003).Conclusion AI-based emphysema q

关 键 词：慢性阻塞性肺疾病 人工智能 肺功能 定量CT 

分 类 号：R563[医药卫生—呼吸系统] R816[医药卫生—内科学]