基于胸部CT平扫的放射组学特征在胸腺瘤与其他前纵隔病变鉴别诊断中的价值

The value of radiomics based on chest CT plain scan in the differential diagnosis of thymoma and other anterior mediastinal lesions

目的探讨基于CT平扫的放射组学特征在胸腺瘤与其他前纵隔病变鉴别诊断中的价值。方法回顾性分析2018年1月至2021年1月江苏大学附属医院收治的128例前纵隔病变患者资料。以病理结果为金标准,将患者分为胸腺瘤组(67例)、非胸腺瘤组(61例)。采用基于MATLAB平台的放射组学分析模块对CT平扫图像进行分析,提取整个病变的组学特征,依次采用组间差异分析、Boruta算法以及共线性检测进行特征筛选。根据最终被选择特征,绘制其单独及联合诊断胸腺瘤的受试者工作特征(ROC)曲线,并计算曲线下面积(AUC),分析所选特征的诊断效能。结果共提取851个病变的组学特征。经过多步骤降维后,最终选择4个差异有统计学意义的组学特征,分别为鲁棒平均绝对偏差、灰度不均匀性、游程方差(小波-LLH)以及依赖不均匀性(小波-HLL)。ROC曲线分析显示,上述4个组学特征单独诊断时的AUC分别为0.712、0.634、0.660、0.699,特异度分别70.2%、61.2%、61.2%、61.2%,灵敏度分别为60.7%、60.6%、68.8%、70.5%;四者联合检测的AUC为0.881,灵敏度和特异度分别为75.4%、89.6%,诊断效能明显提升。结论基于CT平扫的放射组学特征对于胸腺瘤及其他前纵隔病变的鉴别诊断有一定的价值及应用潜力。

Objective To explore the diagnostic value of radiomics features based on chest CT plain scan in differentiating thymoma from other anterior mediastinal lesions.Methods The data of 128 patients with anterior mediastinal lesions from January 2018 to January 2021 in the Affiliated Hospital of Jiangsu University were retrospectively analyzed.According to the pathological criteria,all patients were divided into thymoma group(67 cases)and non-thymoma group(61 cases).The radiomics analysis module based on MATLAB platform was used to analyze images of CT plain scan,and then radiomics features of the whole lesions were extracted.The radiomics features were screened by using group difference analysis,Boruta algorithm and collinearity detection stepwise.The receiver operating characteristic(ROC)curves for the single diagnosis and the combined diagnosis of thymoma were plotted with the selected features,and the area under the curve(AUC)was calculated to analyze the diagnostic performance of the selected features.Results A total of 851 radiomics features were extracted,and 4 radiomics features with statistically significant differences were finally selected after multi-step dimensionality reduction,including robust mean absolute deviation,gray level non-uniformity,wavelet-LLH run variance and wavelet-HLL dependence non-uniformity.ROC curves analysis showed that the AUC of 4 radiomics features was 0.712,0.634,0.660 and 0.699,respectively in the single diagnosis;the specificity was 70.2%,61.2%,and 61.2%,respectively;and the sensitivity was 60.7%,60.6%,68.8%and 70.5%,respectively.AUC value of the four combined detection was 0.881,the sensitivity and specificity was 75.4%and 89.6%,respectively;and the diagnostic efficiency was significantly improved.Conclusion The radiomics features based on CT plain sans have a certain value and application potential in the differential diagnosis of thymoma and other anterior mediastinal lesions.