CT影像组学与传统影像学预测甲状腺乳头状癌颈部淋巴结转移的对比研究

Comparative study of CT radiomics and conventional imaging for predicting cervical lymph node metastasis in papillary thyroid cancer

目的:比较CT影像组学与传统影像学在预测甲状腺乳头状癌颈部淋巴结转移中的价值。方法:收集2019年1月—2023年12月在重庆市巴南区中医院确诊的86例甲状腺乳头状癌患者,对颈部平扫132枚、增强(静脉期)157枚共289枚取得病理结果的淋巴结进行回顾性分析。将入组淋巴结按3:7的比例随机分为平扫组测试集(n=40)和训练集(n=92)、增强组测试集(n=48)和训练集(n=109)。通过影像组学软件提取CT平扫期和增强期(静脉期)的纹理特征,经最小绝对收缩与选择算子算法(LASSO)回归分析降维、筛选影像特征后,构建了线性支持向量机(SVM-Linear)分类器来预测淋巴结转移。传统影像学模型采用两名高年资主治医师对随机分配的淋巴结大小、形态、密度、纵横比、强化等特征进行盲法判读,以一致意见作为预测最终结果。通过受试者工作特征(ROC)曲线评价CT影像组学及传统影像学两种模型的预测性能。结果:提取CT平扫期和增强期(静脉期)图像纹理特征各130个后,筛选出平扫7个、增强8个共15个价值特征。平扫组CT影像组学模型ROC曲线下面积(AUC)为0.853,传统影像学模型AUC值0.660;增强组CT影像组学模型AUC值0.940,传统影像学模型AUC值0.765。结论:CT影像组学的预测性能明显优于传统影像学,可辅助临床术前准确预测甲状腺乳头状癌颈部淋巴结转移。

Objective Comparing the application value of CT radiomics and conventional imaging for predicting cervical lymph node metastasis in papillary thyroid cancer.Methods The clinical data of 86 patients diagnosed with papillary thyroid cancer in Chongqing Banan Hospital of Traditional Chinese Medicine from January 2019 to December 2023 were retrospectively analyzed,a total of 289 cervical lymph nodes were from 132 plain scan and 157 enhanced phase(venous phase)with pathological results.The enrolled lymph nodes were randomly divided into a plain scan group test set(n=40),training set(n=92)and a enhanced phase group test set(n=48),training set(n=109)in a ratio of 3:7.The texture features of plain scan and enhanced phase(venous phase)were extracted by radiomics software,after dimensionality reduction and feature filtering by least absolute shrinkage and selection operator(LASSO)regression analysis,a linear support vector machine(SVM-Linear)classifier was trained to build this model for predicting lymph node metastasis.In the conventional imaging model.Two senior attending physicians interpret blindly the characteristics of randomly assigned lymph nodes,including size,shape,density,aspect ratio,enhancement,etc,the final result was decided by consensus.The predictive performance of CT radiomics and conventional imaging models was assessed by receiver operating characteristic(ROC)curve.Results After extracting 130 texture features of every image from the plain scan and the enhanced phase(venous phase),we finally screened out 15 features,including 7 plain scan and 8 enhanced phase features.The area under ROC curve(AUC)of the CT radiomics model in the plain scan group was 0.853,and the AUC of the conventional imaging model was 0.660.The AUC of the enhanced CT radiomic model was 0.940,and the AUC of the conventional imaging model was 0.765.Conclusion The predictive performance of CT radiomics is obviously better than that of conventional imaging,and it can accurately predict cervical lymph node metastasis in papillary thyroid cancer before clinical operation.