摘要
目的探讨使用肝癌血清标志物构建的3种机器学习(ML)模型对原发性肝癌的诊断价值,选择最优诊断模型以提高肝癌标志物的临床应用价值。方法回顾性分析2021年9月至2022年6月于山西医科大学第一医院住院的273例肝病患者的临床资料,其中原发性肝癌患者83例,慢性肝病及肝硬化患者190例,另外纳入25例体检健康者。收集各研究对象的临床资料以构建人工神经网络(ANN)、决策树(DT)、Logistic回归(LR)模型。绘制ROC曲线并计算ROC曲线下面积(AUC^(ROC)),评估并比较3种模型的预测价值。结果ANN模型的敏感性为86.7%,显著高于LR模型(85.5%)、并联检测模型(80.7%)、DT模型(63.9%)和串联检测模型(39.8%);各诊断模型的AUC^(ROC)分别为:ANN(0.908)、LR(0.903)、DT(0.827)、串联检测(0.678)、并联检测(0.662)。结论机器学习模型能够提高血清标志物对肝癌的诊断价值,ANN模型和LR模型对肝癌的诊断价值相近,优于DT模型。
Objective To compare the diagnosis value of 3 types of machine learning(ML)models which were constructed based on serological markers for primary liver cancer,and screen the optimal model to improve the clinical application value of the serum markers of liver cancer.Methods The medical records of 273 patients with liver disease who were hospitalized in the First Hospital of Shanxi Medical University from September 2020 to June 2022 were retrospectively analyzed,including 83 patients with primary liver cancer(PLC),190 patients with chronic liver disease or hepatitis cirrhosis.The examination information of 25 healthy subjects were also included.The clinical data of the patients were collected and incluted into artificial neural network(ANN),decision tree(DT)and Logistic regression(LR)model.The receiver operating characteristic(ROC)curve of each model was drawn,and the diagnosis value of ML models for PLC were evaluated and compared among the three models by the area under the curve(AUC^(ROC))of ROC.Results The sensitivity of ANN was 86.7%,which was higher than that of LR(85.5%),parallel test(80.7%),DT(63.9%)and serial test(39.8%).The AUC^(ROC) of all the diagnosis models were 0.908(ANN),0.903(LR),0.827(DT),0.678(serial test)and 0.662(parallel test),respectively.Conclusion ML model could be able to improve the early diagnosis value of serum markers.The diagnostic value of ANN model was similar to LR model for the diagnosis of PLC,and both of them were superior to DT model.
作者
韩冰
遆亚楠
王蓉
仇丽霞
张缭云
HAN Bing;TI Yanan;WANG Rong;QIU Lixia;ZHANG Liaoyun(Department of Health Statistics,School of Public Health,Shanxi Medical University,Taiyuan 030001,Shanxi;Department of Infectious Diseases,The First Hospital of Shanxi Medical University,Taiyuan 030001,Shanxi,China)
出处
《临床检验杂志》
CAS
2023年第3期229-234,共6页
Chinese Journal of Clinical Laboratory Science
基金
山西省科技成果转化引导专项项目(201704D131025)。
