基于CT平扫的影像组学列线图模型对高密度胸腺囊肿和低危胸腺瘤的鉴别诊断价值

Value of radiomic nomogram based on nonenhanced CT for the differential diagnosis of high-density thymic cysts and low-risk thymoma

目的:探讨基于CT平扫的影像组学列线图模型鉴别高密度胸腺囊肿和低危胸腺瘤的价值。方法:回顾性分析经手术病理证实的88例胸腺肿物患者,其中高密度胸腺囊肿40例,低危胸腺瘤48例;将88例随机分为训练集70例和验证集18例。由2名放射科医师独立提取CT平扫图像中肿瘤所有层面的影像组学特征,采用组内相关系数(ICC)评价2名医师提取组学特征的一致性;依次利用ANOVA+KW检验、单因素逻辑回归分析、相关性分析(去冗余r>0.9)、LASSO回归分析对影像组学特征降维;通过绘制ROC曲线评价模型诊断效能,构建影像组学特征列线图模型。结果:2名医师提取的影像组学特征的一致性良好。利用LASSO模型从CT图像中筛选出3个影像组学特征(Sphericity、SizezoneVariability、ShortRunEmphasis_AllDirection_offset1_SD),并建立Logistic回归模型。该模型在训练集中具有良好的校准和较高的鉴别能力,AUC为0.916(95%CI 0.825~0.969),敏感度为100.0%,特异度为71.1%;在验证集中,AUC为0.963(95%CI 0.753~1.000),敏感度为100.0%,特异度为80.0%。建立基于影像组学评分的可视化微分列线图模型,决策曲线具有良好的一致性。结论:高密度胸腺囊肿和低危胸腺瘤的CT平扫影像组学特征有重要的鉴别诊断价值,基于CT平扫的影像组学特征的可视化列线图模型可能具有良好的临床应用前景。

Objective:To establish a radiomic nomogram model based on CT plain scans to differentiate high-density thymic cysts from low-risk thymoma.Methods:A retrospective analysis was performed on thymic masses confirmed by operation and pathology including 40 thymic cysts and 48 low-risk thymomas.88 patients were randomly divided into 70 cases in the training set and 18 cases in the testing set.Two radiologists independently extracted the image features of all levels of the tumor in the CT plain scan,evaluated the consistency of the image features by using the intragroup correlation coefficient(ICC),used ANOVA+KW test,single factor logistic regression,correlation analysis(redundancy removal r>0.9),and LASSO regression to reduce the dimension of the image feature parameters,and drew the ROC curve to evaluate the diagnostic efficiency of the model and construct a radiomics nomogram model of image characteristics.Results:The two radiologists extracted the imaging features of patients with good consistency.Three features(Sphericity,SizezoneVariability,ShortrunEmphasis_AllDirection_offset1_SD)were selected from the nonenhanced CT images using the LASSO model,following which the logistic regression model was constructed.The model had good calibration and high discrimination ability in the training set,the AUC was 0.916(95%CI 0.825~0.969),the sensitivity was 100.0%,and the specificity was 71.1%.In the validation set,the AUC was 0.963(95%CI 0.753~1.000),the sensitivity was 100.0%,and the specificity was 80.0%.A visual differential nomogram based on the image group score was established.The decision curve had a good consistency.Conclusions:The CT features of high-density thymic cysts and low-risk thymomas are of great value in differential diagnosis.The visual nomogram based on the CT features may have good clinical application prospects.