基于机器学习构建肺腺癌骨转移自动化模型  

Constructing an automated model for lung adenocarcinoma bone metastasis detection based on machine learning

作　　者：李晓[1] 李侠 葛静 刘亚锋 张鑫[1] 陈英[3] LI Xiao;LI Xia;GE Jing;LIU Yafeng;ZHANG Xin;CHEN Ying(Department of Blood Transfusion,Rizhao People's Hospital,Shandong Province,Rizhao 276800,China;Department of Laboratory Medicine,Cancer Hospital Affiliated to Huainan Oriental Hospital Group,Anhui Province,Huainan 232001,China;Department of Laboratory Medicine,Rizhao People's Hospital,Shandong Province,Rizhao 276800,China)

机构地区：[1]山东省日照市人民医院输血科,山东日照276800 [2]安徽省淮南东方医院集团附属肿瘤医院检验科,安徽淮南232001 [3]山东省日照市人民医院检验科,山东日照276800

出　　处：《中国当代医药》2024年第23期114-119,共6页China Modern Medicine

摘　　要：目的采用机器学习算法对关键变量进行识别,并对肺腺癌(LUAD)患者骨转移风险进行预测。方法回顾性分析2019年1月至2022年6月淮南东方医院集团附属肿瘤医院收治的132例确诊非小细胞肺癌(NSCLC)患者的临床资料,包括是否发生骨转移、年龄、性别、病理类型、吸烟状况、T分期、N分期、骨转移前是否有其他部位的转移,以及癌胚抗原(CEA)、碱性磷酸酶(ALP)、鳞状细胞癌抗原(SCCA)、糖类抗原125(CA125)、细胞角蛋白19片段抗原21-1(CYFRA21-1)、神经元特异性烯醇化酶(NSE)、钙(CA)水平。采用LASSO回归分析方法来筛选与骨转移相关的关键特征,并将其用于构建6种机器学习模型,另收集63例NSCLC患者的临床数据用于模型的外部验证。不同模型的预测性能通过受试者工作特征曲线(ROC曲线)来评估。校准曲线和DCA曲线用于验证所建模型的准确性和获益能力。使用SHAP包对logistic模型进行解释。结果LASSO回归分析最终筛选了4个重要变量,包括性别、N分期、CEA水平和糖类抗原CA125水平。在6种机器学习模型中,logistic模型在训练集(AUC=0.710)、测试集(AUC=0.705)和外部验证集(AUC=0.655)均具有最佳的预测效能和稳定性。SHAP图显示在logistic模型中4个变量的权重从高到低依次为CEA、性别、T分期和CA125。成功构建了LUAD骨转移的机器学习模型和网页计算器。结论logistic预测模型可以识别LUAD骨转移高风险患者,这有助于临床医生指导高危患者做出适当预防措施。Objective To identify key variables and predict the risk of bone metastasis in patients with lung adenocarcinoma(LUAD)using machine learning algorithms.Methods The clinical data from 132 patients diagnosed with non-small cell lung cancer(NSCLC)who were admitted to Cancer Hospital Affiliated to Huainan Oriental Hospital Group from January 2019 to June 2022 were retrospectively analyzed,encompassing whether bone metastasis occurred,age,gender,pathological type,smoking status,T stage,N stage,the presence of metastasis in other areas before bone metastasis,and the levels of carcinoembryonic antigen(CEA),alkaline phosphatase(ALP),squamous cell carcinoma antigen(SCCA),carbohydrate antigen 125(CA125),cyto-keratin 19 fragment antigen 21-1(CYFRA21-1),neuron specific enolase(NSE),and calcium(CA).The LASSO regression analysis method was used to select key features associated with bone metastasis and to construct six machine learning models,with an additional collection of clinical data from 63 NSCLC patients for external validation of the models.The predictive performance of different models was assessed by the receiver operating characteristic(ROC)curve.Calibration curves and DCA curves were used to verify the accuracy and benefit capability of the constructed models.The SHAP package was used to interpret the logistic model.Results The LASSO regression analysis ultimately selected four significant variables,including gender,N stage,CEA levels,and carbohydrate antigen CA125 levels.Among the six machine learning models,the logistic model had the best predictive efficacy and stability in the training set(AUC=0.710),test set(AUC=0.705),and external validation set(AUC=0.655).The SHAP plot showed that in the logistic model,the weights of the four variables from highest to lowest were CEA,Sex,T stage,and CA125.The machine learning model and a web calculator for LUAD bone metastasis were successfully constructed.Conclusion Logistic model can identify the LUAD,high-risk patients with bone metastases,it can assist clinical physician

关 键 词：非小细胞肺癌 骨转移 预测模型 机器学习 

分 类 号：R734.2[医药卫生—肿瘤]