基于双期增强CT影像组学模型对甲状腺乳头状癌被膜侵犯的预测价值

Pr ediction value of membrane invasion in papillary thyroid carcinoma based on biphasic enhanced CT imaging omics model

目的 探讨基于双期增强CT构建的支持向量机(SVM)及K-近邻(KNN)模型预测甲状腺乳头状癌被膜侵犯的可行性。方法回顾性分析2018年1月至2022年12月经手术病理确诊的157例甲状腺乳头状癌患者(160例病灶)的临床资料。根据术后病理结果分为被膜侵犯组(84例)和非被膜侵犯组(76例)。采用随机数表法按7∶3比例随机分为训练组(n=112)和验证组(n=48)。使用2种机器算法基于筛选后的影像组学特征构建模型,并对模型进行内部验证。采用受试者操作特征曲线及曲线下面积评价静脉期、动脉期及双期联合模型的预测效能。结果 在验证组中,基于静脉期CT建立的SVM模型、KNN模型AUC为0.752、0.698,基于动脉期CT建立的SVM模型、KNN模型AUC为0.880、0.716,基于双期联合CT建立的SVM联合模型、KNN联合模型AUC为0.936、0.764。联合模型预测甲状腺乳头状癌被膜侵犯的效能明显高于单时期模型。结论 本研究基于双期增强CT构建的SVM、KNN模型在一定程度上均能预测甲状腺乳头状癌被膜侵犯,其中双期联合的SVM模型表现出最佳的预测效能,在临床个体化诊治甲状腺乳头状癌中具有较高的应用价值。

Objective To investigate the feasibility of predicting the capsular invasion in papillary thyroid carcinoma based on the support vector machine model and K-nearest neighbor model by dual-phase enhanced CT.Methods In total 160 cases from 157 patients with PTC confirmed by pathology in our hospital from January 2018 to December 2022 were retrospectively analyzed.According to the postoperative pathological results,the cases were divided into the capsular invasion group(84 cases)and the non-capsular invasion group(76 cases).The enroll cases were randomly divided into the training group(n=112)and the validation group(n=48)in a 7:3 ratio.Two machine algorithms were used to construct the models based on selected radiomic features.The models were validated internally by the validation group.Finally,the operating characteristic curve and the area under the curve were used to assess the prediction effectiveness of different models.Results In the validation group,the AUCs of the SVM model and KNN model based on venous phase were 0.752 and 0.698,the AUCs of the SVM model and KNN model based on arterial phase were 0.880 and 0.716,the AUCs of the SVM model and KNN model based on dual-phase were 0.936 and 0.764.The dual-phase model was significantly more effective in predicting the capsular invasion in papillary thyroid carcinoma than the single-phase model.Conclusion In this study,both SVM and KNN models based on enhanced CT can predict capsular invasion in papillary thyroid carcinoma to some extent,and the SVM model based on dual-phase shows the best prediction effectiveness among them,which has high application value in individualized diagnosis and treatment of papillary thyroid cancer.