摘要
目的:以常规临床检验大数据为基础,使用机器学习算法构建乳腺癌诊断模型。以探讨临床检验组学在乳腺癌诊断中的应用价值。方法:采用回顾性研究方法,收集6089例乳腺癌患者及6830例乳腺良性疾病患者临床信息和检验数据。分别运用极限梯度提升、神经网络、支持向量机、随机森林、最近邻算法、逻辑回归、线性判别分析算法、朴素贝叶斯、梯度提升机算法和C5.0决策树等机器学习算法建立乳腺癌诊断模型。采用10折交叉验证进行模型训练,应用准确度、AUC、平均准确度、特异度、灵敏度、阳性预测值、阴性预测值及Kappa值评估各模型性能。结果:从28项常规临床检验指标中筛选出GLU、DBIL、RDW-CV、MONO、TG、ALB、RBC、LYMPH、UREA等9项指标再加上age用于模型构建。通过10种机器学习算法进行模型评估,发现梯度提升机算法相较其它算法具有最优的诊断性能。梯度提升机算法诊断乳腺癌准确度为0.80、AUC为0.80、平均准确度为0.80、特异度为0.77、灵敏度为0.82、阳性预测值为0.78、阴性预测值为0.81、Kappa值为0.59。结论:以常规临床检验数据为基础,使用机器学习算法构建乳腺癌诊断模型,可以为乳腺癌的辅助诊断提供决策支持。
Objective:To construct a breast cancer diagnostic model using machine learning algorithms based on big data from routine clinical laboratories,in order to explore the application value of clinlabomics in the diagnosis of breast cancer.Methods:A retrospective study was conducted,collecting clinical and laboratory test data from 6089 breast cancer patients and 6830 patients with benign breast diseases.Various machine learning algorithms including extreme gradient boosting(XGBoost),neural network(NN),support vector machine(SVM),random forest(RF),K-nearest neighbors(KNN),Logistic regression(LR),linear discriminant analysis(LDA),naive bayes,gradient boosting machine(GBM)algorithm,and C5.0 decision tree were utilized for establishing the breast cancer diagnostic model.Ten-fold cross validation was used to train the model,and the accuracy,AUC,average accuracy,specificity,sensitivity,positive predictive value,negative predictive value,and Kappa value were used to evaluate the performance of each model.Results:Nine indicators were selected from 28 conventional clinical test indicators,including GLU,DBIL,RDW-CV,MONO,TG,ALB,RBC,LYMPH and UREA,and then age was added for model construction.The model was evaluated by using 10 machine learning algorithms,and the results show that the gradient elevator algorithm,has the best diagnostic performance compared with other models.The accuracy,AUC,average accuracy,specificity,sensitivity,positive predictive value,negative predictive value,and Kappa value of gradient elevator algorithm in diagnosing breast cancer were 0.80,0.80,0.80,0.77,0.82,0.78,0.81 and 0.59,respectively.Conclusion:Based on the routine clinical laboratory data,the machine learning algorithm is used to construct the breast cancer diagnosis model,which can provide decision support for the auxiliary diagnosis of breast cancer.
作者
卢峰
张开炯
吴立春
蒋叙川
冀承杰
刘靳波
LU Feng;ZHANG Kaijiong;WU Lichun;JIANG Xuchuan;JI Chengjie;LIU Jinbo(Medical Laboratory Department of the Affiliated Hospital of Southwest Medical University,Sichuan Luzhou 646000,China;Department of Experimental Medicine,Jianyang City People's Hospital,Sichuan Jianyang 641400,China;Sichuan Cancer Hospital Institute/Sichuan Cancer Prevention Center/Medical Laboratory Department of the Affiliated Tumor Hospital of University of Electronic Science and Technology,Sichuan Chengdu 610000,China)
出处
《现代肿瘤医学》
CAS
2024年第7期1264-1272,共9页
Journal of Modern Oncology
基金
四川省科技厅重点研发项目第二版(编号:2022YFS0335)
四川省成都市医学科研课题(编号:2022417)
四川省简阳市人民医院科研课题(编号:JY202234)。
关键词
乳腺癌
机器学习
临床检验组学
诊断模型
人工智能
breast cancer
machine learning
clinlabomics
diagnostic model
artificial intelligence
