多种机器学习模型预测甲状腺微小乳头状癌患者的中央区淋巴结转移风险  

Multiple machine learning models predict the risk of central lymph node metastasis in patients with papillary thyroid microcarcinoma

作　　者：刘明昊 王子昂 符瑜 LIU Minghao;WANG Ziang;FU Yu(Medical School of Chinese PLA,Beijing 100853,CHN;Medical Department of General Surgery,the First Medical Centerof Chinese PLA General Hospital,Beijing 100853,CHN;Department of General Surgery,Daping Hospital of Army Medical University,Chongqing 400042,CHN)

机构地区：[1]解放军医学院,北京1100853 [2]中国人民解放军总医院第一医学中心普通外科医学部,北京100853 [3]陆军军医大学大坪医院普通外科,重庆400042

出　　处：《河南大学学报(医学版)》2025年第1期47-56,共10页Journal of Henan University:Medical Science

摘　　要：目的:甲状腺微小乳头状癌(papillary thyroid microcarcinoma, PTMC)的发病率逐年升高,部分患者会发生中央区淋巴结转移(central lymph node metastasis, CLNM),本研究旨在提供可靠的机器学习(machine learning, ML)模型来预测PTMC患者发生CLNM的概率。方法:我们从2010-2017年的监测、流行病学和结果(surveillance, epidemiology, and end results, SEER)数据库中提取27 251例PTMC患者的数据。并构建极端梯度提升(eXtreme gradient boosting, XGB)、随机森林(random forest, RF)、决策树(decision tree, DT)、逻辑回归(logistic regression, LR)、多层感知器(multilayer perceptron, MLP)与朴素贝叶斯(bernoulli naive bayes, BNB)等6种机器学习模型来预测患者的淋巴结转移风险。此外,我们通过各种指标对模型的性能进行评估来选择最佳模型,并对影响患者出现CLNM的变量重要性进行排序。结果:在6种机器学习模型中,XGB模型的表现最佳,在训练集的曲线下面积(area under curve, AUC)为0.87,准确率为0.89,精确率为0.89,在测试集的AUC为0.77,准确率为0.88,精确率为0.84。在该模型中对患者发生CLNM影响最大的因素为甲状腺腺外侵犯(extra-thyroidal extension, ETE)。结论:在本研究中,我们开发了多种机器学习模型来预测PTMC患者的CLNM风险,其中XGB模型具有最佳预测效能,通过该模型更有助于临床医生进行决策。Objective: The incidence of papillary thyroid microcarcinoma(PTMC) has been increasing annually, with some patients experiencing central lymph node metastasis(CLNM). This study aims to provide a reliable machine learning(ML) model to predict the probability of CLNM in patients with PTMC. Methods: We extracted data from 27 251 patients with papillary thyroid microcarcinoma(PTMC) from the Surveillance, Epidemiology, and End Results(SEER) database spanning the years 2010 to 2017. We developed six machine learning models—eXtreme Gradient Boosting(XGB), Random Forest(RF), Decision Tree(DT), Logistic Regression(LR), Multilayer Perceptron(MLP), and Bernoulli Naive Bayes(BNB)—to predict the risk of lymph node metastasis in these patients. Furthermore, we evaluated the performance of these models using various metrics to identify the optimal model and ranked the importance of variables influencing the occurrence of central lymph node metastasis(CLNM). Results: Among the six machine learning models, the XGB model demonstrated the best performance, achieving an area under the curve(AUC) of 0.87 with an accuracy of 0.89 and a precision of 0.89 on the training set. In the test set, the AUC was 0.77, with an accuracy of 0.88 and a precision of 0.84. The most significant factor influencing the occurrence of CLNM in this model was extra-thyroidal extension(ETE). Conclusion: In this study, we developed various machine learning models to predict the risk of CLNM in patients with PTMC, among which the XGB model exhibited the best predictive performance. This model can significantly aid clinicians in making decisions.

关 键 词：甲状腺微小乳头状癌 中央区淋巴结转移 机器学习模型 SEER数据库 

分 类 号：R736.1[医药卫生—肿瘤]