增强CT影像组学结合深度学习算法预测甲状腺乳头状癌颈部淋巴结转移  

Enhanced CT radiomics combined with deep learning algorithm for predicting cervical lymph node metastasis of papillary thyroid carcinoma

作　　者：叶媛媛 贺克武 刘奇峰 洪文敏 YE Yuanyuan;HE Kewu;LIU Qifeng;HONG Wenmin(Department of Medical Imaging,the Third Affiliated Hospital of Anhui Medical University,Hefei 230000,China)

机构地区：[1]安徽医科大学第三附属医院(合肥市第一人民医院)影像中心,安徽合肥230000

出　　处：《中国介入影像与治疗学》2025年第3期196-200,共5页Chinese Journal of Interventional Imaging and Therapy

摘　　要：目的观察增强CT影像组学结合深度学习(DL)算法预测甲状腺乳头状癌(PTC)颈部淋巴结转移(CLNM)的价值。方法回顾性分析100例单发PTC患者,按7:3比例将其分为训练集(n=70)与测试集(n=30);基于颈部动脉期CT提取并筛选病灶最优影像组学特征和最优DL特征,分别计算影像组学评分(Radscore)及DL评分(Deepscore)并据以构建影像组学模型及DL模型。将临床资料、常规CT表现、Radscore及Deepscore纳入多因素logistic回归分析,筛选PTC CLNM独立预测因素并构建联合模型。绘制受试者工作特征曲线,计算曲线下面积(AUC),评估各模型预测PTC CLNM的效能。结果共筛选出13个最优影像组学特征和12个最优DL特征。Radscore(OR=1.698,P=0.002)及Deepscore(OR=1.872,P=0.021)均为PTC CLNM的独立预测因素。影像组学模型、DL模型及联合模型预测训练集PTC CLNM的AUC分别为0.775、0.876及0.880,在测试集分别为0.739、0.776及0.789;联合模型在训练集的预测效能高于影像组学模型(Z=2.551,P=0.011)。结论结合DL算法可有效提高增强CT影像组学预测PTC CLNM的效能。Objective To observe the value of enhanced CT radiomics combined with deep learning(DL)algorithm for predicting cervical lymph node metastasis(CLNM)of papillary thyroid carcinoma(PTC).Methods Totally 100 patients with single PTC were retrospectively enrolled and divided into training set(n=70)and test set(n=30)at the ratio of 7∶3.The optimal radiomics features and DL features of lesions were extracted and screened based on arterial phase cervical CT,and the radiomics score(Radscore)and DL score(Deepscore)were calculated to construct radiomics model and DL model,respectively.Clinical data,routine CT findings,Radscore and Deepscore were enrolled in multivariate logistic regression analysis to screen the independent predictors of PTC CLNM,and a combined model was then constructed.The receiver operating characteristic curve was plotted,and the area under the curve(AUC)was calculated to evaluate the efficacy of each model for predicting PTC CLNM.Results Thirteen optimal radiomics features and 12 DL features were selected.Radscore(OR=1.698,P=0.002)and Deepscore(OR=1.872,P=0.021)were both independent predictors of PTC CLNM.The AUC of radiomics model,DL model and combined model for predicting PTC CLNM was 0.775,0.876 and 0.880 in training set,which in test set was 0.739,0.776 and 0.789,respectively.In training set,the prediction efficacy of combined model was better than that of radiomics model(Z=2.551,P=0.011).Conclusion Combined with DL algorithm could effectively increase the efficacy of enhanced CT radiomics for predict PTC CLNM.

关 键 词：甲状腺肿瘤 淋巴转移 体层摄影术 X线计算机 影像组学 深度学习 

分 类 号：R736.1[医药卫生—肿瘤] R814.42[医药卫生—临床医学]