基于CT的影像组学在鉴别复杂前纵隔囊肿与低危组胸腺上皮肿瘤的应用价值  

Application value of CT-based radiomics in differentiating complex anterior mediastinal cyst from low-risk thymic epithelial tumor

作　　者：尹阳[1,2] 赵艳 李娜[1] 杨志远[1] 舒健[1] YIN Yang;ZHAO Yan;LI Na;YANG Zhiyuan;SHU Jian(Department of Radiology,Leshan People’s Hospital,Leshan,Sichuan Province 614000,China;Department of Radiology,the Affiliated Hosptital of Southwest Medical University,Luzhou,Sichuan Province 646000,China)

机构地区：[1]乐山市人民医院放射科,四川乐山614000 [2]西南医科大学附属医院放射科,四川泸州646000

出　　处：《实用放射学杂志》2022年第12期1945-1950,共6页Journal of Practical Radiology

摘　　要：目的探讨基于平扫及增强CT的影像组学方法对复杂前纵隔囊肿(AMC)与低危组胸腺上皮肿瘤(TET)鉴别诊断的应用价值。方法回顾性分析经手术病理证实的39例复杂AMC和26例低危组TET患者。使用3D Slicer软件在平扫及增强CT图像中勾画感兴趣体积(VOI),在Python软件中利用Pyradiomics包提取影像组学特征,采用独立样本t检验及最小绝对收缩和选择算子(LASSO)方法筛选影像组学特征,影像组学特征联合CT特征(分为组学组、CT组、联合组)分别利用Logistic回归(LR)、K邻近(KNN)、随机森林(RF)、支持向量机(SVM)等有监督学习分类器构建24个预测模型,通过受试者工作特征(ROC)曲线及曲线下面积(AUC)评估模型的预测效能,通过Delong检验比较不同模型间差异。结果从平扫及增强CT中共筛选出8个CT特征、8个平扫及6个增强影像组学特征。基于增强CT的各特征组在不同分类器中均具有高度预测效能(AUC 0.91~0.99),组学组及联合组在LR中表现最优(AUC=0.97,0.99)。基于平扫CT,除SVM中的组学组及CT组预测效能低(AUC=0.64,0.69),其余各模型均有中等预测效能(AUC 0.71~0.81),组学组在LR及联合组在RF中表现最优(AUC=0.80,0.81)。结论在鉴别复杂AMC及低危组TET方面,基于平扫及增强CT的影像组学方法具有中高度预测效能,在术前指导临床治疗方案选择及判断患者预后具有重要意义。Objective To explore the application value of radiomics based on plain and enhanced CT in differential diagnosis of complex anterior mediastinal cyst(AMC)and low-risk thymic epithelial tumor(TET).Methods Thirty-nine cases of complex AMC and 26 cases of low-risk TET patients confirmed by surgery and pathology were analyzed retrospectively.3D Slicer software was used to sketch volume of interest(VOI)on plain and enhanced CT images,and Pyradiomics package in Python software was used to extract radiomics features.The independent sample t-test and least absolute shrinkage and selection operator(LASSO)methods were used to screen the radiomics features.The screened radiomics and CT features were divided into radiomics group,CT group and combined group.4 supervised learning classifiers[Logistic regression(LR),K-nearest neighbor(KNN),random forest(RF)and support vector machone(SVM)]were used to construct 24 prediction models based on the three groups.The prediction efficiency of the models was evaluated by receiver operating characteristic(ROC)curve and area under the curve(AUC),and the differences among different models were compared by Delong test.Results Eight CT features and 14 radiomics features(8 plain and 6 enhanced)were screened out support vector machine.Each group based on enhanced CT had high prediction efficiency in different classifiers(AUC 0.91-0.99),and the combination of radiomics group+LR classifier and combined group+LR classifier had the best performance(AUC=0.97,0.99).All models had moderate predictive efficiency(AUC 0.71-0.81)based on plain CT,except for the combination of the radiomics group+SVM classifier and CT group+SVM classifier(AUC=0.64,0.69);the combination of radiomics group+LR classifier and combined group+RF classifier had the best performance(AUC=0.80,0.81).Conclusion In differentiating complex AMC from low-risk TET,radiomics model based on plain and enhanced CT have moderate-high predictive efficiency,which is of great significance in guiding the clinical treatment plan,making and judging

关 键 词：前纵隔囊肿 胸腺上皮肿瘤 影像组学 计算机体层成像 

分 类 号：R564[医药卫生—呼吸系统] R736.3[医药卫生—内科学] R445[医药卫生—临床医学] R814.42