基于增强CT特征的logistic模型预测胸腺上皮肿瘤组织学分型  

Logistic regression model based on contrast-enhanced CT for predicting histopathologic classification of thymic epithelial tumors

作　　者：何晓东[1] 舒震宇[1] 李杰[1] 徐玉芸[1] 林毅[1] HE Xiaodong;SHU Zhenyu;LI Jie;XU Yuyun;LIN Yi(Department of Radiology,Zhejiang Provincial People's Hospital(People's Hospital of Hangzhou Medical College),Hangzhou 310014,China)

机构地区：[1]浙江省人民医院(杭州医学院附属人民医院)放射科,杭州310014

出　　处：《浙江医学》2021年第2期167-171,179,共6页Zhejiang Medical Journal

摘　　要：目的探讨基于增强CT影像学特征构建并验证logistic模型在预测胸腺上皮肿瘤组织学分型中的应用价值。方法回顾性分析2015年12月至2020年6月在浙江省人民医院行增强CT检查并经病理证实的178例胸腺上皮肿瘤患者资料,根据组织学分型分为低危组91例(A、AB、B1)、高危组(B2、B3)43例和胸腺癌44例。将所有患者依据检查时间先后按7:3比例诊断时间分为训练组124例和测试组54例。由2位放射科医师采用盲法对肿瘤CT特征进行评估,分析不同影像特征与组织学分型的相关性,建立并验证logistic预测模型,使用ROC曲线评估模型的诊断效能,使用决策曲线分析(DCA)评估模型效益。结果3组患者年龄、性别、钙化、强化程度的差异均无统计学意义(均P>0.05),肌无力、大小、形状、边界、坏死、强化方式、心包侵犯、大血管侵犯、胸膜侵犯、心包积液、胸腔积液、肿大淋巴结比较差异均有统计学意义(均P<0.05)。使用多因素logistic回归构建包含了形态、边界、心包侵犯的预测模型,该模型预测高危胸腺瘤及胸腺癌的AUC在训练组和测试组中分别为0.900和0.889。结论基于增强CT特征构建的logistic预测模型是一种无创性预测胸腺上皮肿瘤组织学分型的工具,有助于识别高危胸腺瘤及胸腺癌。Objective To develop and verify a logistic regression model for predicting histopathologic classification of thymic epithelial tumors(TETs)based on contrast-enhanced CT.Methods The CT findings of thymic epithelial tumors in 178 patients,who underwent surgery at Zhejiang Provincial People's Hospital from December 2015 to June 2020,were retrospectively analyzed by two radiologists without knowing histopathologic classification of the cases.Among 178 patients there were 91 cases of low-risk(A,AB,B1),43 cases of high-risk(B2,B3)and 44 cases of thymic cancer.There were 124 cases in the training group and 54 cases in the test group with a ratio of 7:3.The correlation between CT findings and histopathological classification was analyzed.And the predictive model of TETs was constructed by multivariate logistic regression combined with clinical related factors and CT features.The diagnostic accuracy of the model was evaluated by receiver operating characteristic(ROC)in the training set,and reliability of the model was verified with test set.Decision curve analysis(DCA)was used to examine the clinical practicability of the model.Results There was significant difference in tumor size,shape,boundary,necrosis,enhancement pattern,vascular invasion,pleural invasion,pericardial invasion,pleural effusion,pericardial effusion,lymphadenectasis and myasthenia gravis among low risk group,high risk group and thymic carcinoma group.While there was no significant difference in sex,age,calcification and enhancement degree(P>0.05).The predictive model was constructed by multivariate logistic regression based on tumor shape,boundary and pericardial invasion.ROC analysis showed that the AUC was 0.900 and 0.889 in the training and test sets,respectively.The DCA curve confirmed the clinical practicability of the predictive model.Conclusion The logistic regression model based on contrast-enhanced CT is a non-invasive tool for predicting histopathologic classification of TETs,which is helpful for identifying high-risk thymomas and thymic carcin

关 键 词：胸腺肿瘤 体层摄影术 组织病理学 logistic回归 

分 类 号：R730.44[医药卫生—肿瘤] R736.3[医药卫生—临床医学]