不同机器学习模型预测甲状腺乳头状癌颈部淋巴结转移  

Different machine learning models for predicting cervical lymph node metastasis of papillary thyroid carcinoma

作　　者：胡蓓蓓 张英霞[1] 邓伟[1] HU Beibei;ZHANG Yingxia;DENG Wei(Department of Ultrasound Diagnosis,the Affiliated Hospital of Inner Mongolia Medical University,Hohhot 010050,China)

机构地区：[1]内蒙古医科大学附属医院超声诊断科,内蒙古呼和浩特010050

出　　处：《中国介入影像与治疗学》2025年第3期164-168,共5页Chinese Journal of Interventional Imaging and Therapy

基　　金：内蒙古自治区教育厅硕士研究生学科创新项目(S20231192Z)。

摘　　要：目的对比观察不同机器学习(ML)模型预测甲状腺乳头状癌(PTC)颈部淋巴结转移(CLNM)的价值。方法纳入207例经病理诊断的PTC患者,根据术后淋巴结病理结果将其分为转移组(n=103)与非转移组(n=104),并以7:3比例分为训练集(n=144)与验证集(n=63)。结合临床信息和淋巴结超声特征构建随机森林(RF)、决策树(DT)、K近邻法(KNN)、逻辑回归(LR)及支持向量机(SVM)模型,以预测PTC发生CLNM。绘制受试者工作特征曲线,计算曲线下面积(AUC),评价各模型预测PTC CLNM的效能。结果患者年龄,超声所见淋巴结横径/纵径比值≥0.5、囊性变、微钙化、淋巴门消失、呈偏高回声团及边界不清均为PTC CLNM的影响因素,尤以微钙化对模型的贡献度最高。所获DT、RF模型预测训练集PTC CLNM的准确率达93.06%,AUC均为0.987。DT模型及RF模型预测验证集PTC CLNM的AUC分别为0.817及0.895,高于其他模型;RF模型的准确率、特异度及阳性预测值分别为84.13%、93.10%及92.86%,为5种ML模型中的最佳者。结论ML模型中,RF模型用于预测PTC发生CLNM较其他模型更具优势。Objective To comparatively observe the value of different machine learning(ML)models for predicting cervical lymph node metastasis(CLNM)of papillary thyroid carcinoma(PTC).Methods Totally 207 patients with pathologically diagnosed PTC were enrolled and divided into metastasis group(n=103)and non-metastasis group(n=104)according to lymph nodes pathology findings after surgical resection,also divided into training set(n=144)and validation set(n=63)with a ratio of 7∶3.Random forest(RF),decision tree(DT),K-nearest neighbor(KNN),logistic regression(LR)and support vector machine(SVM)models were constructed through combining clinical information and ultrasonic manifestations of lymph nodes.Receiver operating characteristic curves were drawn,and the areas under the curve(AUC)were calculated to evaluate the efficacy of these ML models for predicting PTC CLNM.Results Patients’age,transverse diameter/longitudinal diameter ratio of lymph node≥0.5,lymph nodes with cystic change,microcalcification,disappearance of lymphatic gate,hypoechoic mass and unclear border were all impact factors of PTC CLNM,among which microcalcification had the highest contribution to the models.In training set,AUC of DT and RF models were both 0.987,with accuracy reached 93.06%.In validation set,AUC of DT and RF models was 0.817 and 0.895,respectively,all higher than those of other models.The accuracy,specificity and positive predictive value of RF model in validation set was 84.13%,93.10%and 92.86%,respectively,and RF model was the best one among all 5 ML models.Conclusion Among different ML models,RF model was the best one for predicting PTC CLNM.

关 键 词：甲状腺肿瘤 癌 乳头状 肿瘤转移 机器学习 超声检查 

分 类 号：R736.1[医药卫生—肿瘤] R445.1[医药卫生—临床医学]