基于增强CT影像组学鉴别胸腺瘤组织学分型

Radiomics based on enhanced CT in differential diagnosis of histological classification of thymomas

目的观察基于增强CT影像组学鉴别胸腺瘤组织学分型的价值。方法回顾性分析226例经病理证实的胸腺瘤患者,按7∶3比例将其分为训练集(n=159)及测试集(n=67);利用最大相关最小冗余(mRMR)及最小绝对收缩和选择算子(LASSO)算法筛选最佳影像组学特征,构建鉴别胸腺瘤组织学分型的影像组学模型;以单因素及多因素logistic回归分析筛选鉴别胸腺瘤组织学分型相关的临床及CT表现,构建临床模型和联合影像组学特征及临床、CT特征的影像组学列线图。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评价3种模型鉴别胸腺瘤组织学分型的效能并比较其差异,评价影像组学列线图的临床价值。结果最终基于增强动脉期及静脉期CT筛选出19个最佳影像组学特征,以建立影像组学模型。临床模型由患者年龄、重症肌无力、病灶CT表现形态、侵犯邻近组织及动脉期CT值构成。训练集中,影像组学列线图及影像组学模型区分低危组胸腺瘤与高危组胸腺瘤的AUC(0.91、0.89)均高于临床模型(0.79,Z=3.62、2.49,P均<0.05),而影像组学列线图与影像组学模型AUC差异无统计学意义(Z=1.54,P=0.12);3种模型在测试集中的AUC差异均无统计学意义(P均>0.05)。阈值概率为0.1~1.0时,影像组学列线图的临床获益均大于临床模型及影像组学模型。结论基于增强CT影像组学模型和基于临床、CT表现及影像组学特征的影像组学列线图均有利于鉴别胸腺瘤组织学分型,后者临床获益更高。

Objective To explore the value of radiomics based on enhanced CT for differential diagnosis of histological classification of thymomas.Methods Clinical and abdominal CT data of 226 patients with pathologically confirmed thymomas were retrospectively analyzed and divided into training set(n=159)and test set(n=67)at a ratio of 7∶3.The maximum relevance minimum redundancy(mRMR),the least absolute shrinkage and selection operator(LASSO)were used to select best radiomics features for constructing radiomics model to identify the histologic classification of thymomas.Univariate and multivariate logistic regression analysis were performed to screen the clinical and CT features associated with histologic classification of thymomas for construction of the clinical model and radiomics nomogram based on clinical,CT features and radiomics features.Receiver operating characteristic(ROC)curve was drawn,and area under the curve(AUC)was calculated and compared among 3 models,and the clinical value of radiomics nomogram was analyzed.Results Based on enhanced arterial and venous phase CT,a total of 19 optimal radiomics features were selected to build radiomics model.The clinical model was composed of patients'age,myasthenia gravis,CT finding of lesions morphology,adjacent tissue invasion and CT value of arterial phase.In training set,AUC of radiomics nomogram and radiomics model(0.91,0.89)for identifying the histologic classification of thymomas were higher than that of clinical model(0.79,Z=3.62,2.49,both P<0.05),and there was no significant difference of AUC between radiomics nomogram and radiomic model(Z=1.54,P=0.12).In test set,there was no significant difference of AUC among 3 models(all P>0.05).Taken 0.1-1.0 as the threshold of probability,the clinical benefit of radiomics nomogram was greater than that of clinical model and radiomics model.Conclusion Both radiomics model based on enhanced CT and radiomics nomogram based on clinical,CT findings and radiomics features were help to identifying histological classification of thymomas,and the latter might bring higher clinical benefit.