基于机器学习算法构建血常规指标肺结核诊断模型  

Pulmonary tuberculosis diagnosis model for blood routine test based on machine learning algorithms

作　　者：黄莺[1] 周颖[1] 宋云霄[2] 茅俊杰[1] 管超[1] 赵金燕 倪佩青 HUANG Ying;ZHOU Ying;SONG Yunxiao;MAO Junjie;GUAN Chao;ZHAO Jinyan;NI Peiqing(Department of General Practice,Shanghai Xuhui Central Hospital,Shanghai 200237,China;Department of Clinical Laboratory,Shanghai Xuhui Central Hospital,Shanghai 200237,China)

机构地区：[1]上海市徐汇区中心医院全科医学科,上海200237 [2]上海市徐汇区中心医院检验科,上海200237

出　　处：《检验医学》2024年第7期668-672,共5页Laboratory Medicine

基　　金：江苏大学医教协同创新基金(JDYY2023091)。

摘　　要：目的基于机器学习算法,利用血常规检验数据构建肺结核诊断模型,并分析其临床应用价值。方法选取2019年1月—2022年12月上海市徐汇区中心医院469例初诊肺结核患者(肺结核组),以年龄、性别相匹配的506名体检健康者作为正常对照组。收集所有研究对象22项血常规检验数据和人口学参数。采用LASSO回归分析评估共线性。将数据集随机分为训练集(75%,用于机器学习模型构建)和测试集(25%,用于模型性能评估)。采用分布式随机森林(DRF)、深度学习、梯度提升机和广义线性模型这4种机器学习算法进行测试,采用5倍交叉法进行验证。采用受试者工作特征(ROC)曲线评估模型的诊断效能。结果基于Logistic回归分析和LASSO回归分析结果进行模型特征重要性排序,共筛选出10个非共线性指标,结果显示,DRF是构建肺结核诊断的最佳机器学习算法。在训练集和测试集中,DRF模型的曲线下面积分别为0.9921和0.8474,敏感性分别为99.16%和92.04%,特异性分别为80.91%和55.22%,准确度分别为89.84%和72.06%。结论基于机器学习算法构建的血常规检验数据肺结核诊断模型是一个有效的诊断工具,但其临床应用价值需要进一步验证。Objective To construct a pulmonary tuberculosis diagnosis model based on machine learning algorithms for blood routine test,and to analyze its clinical application value.Methods Totally,469 newly diagnosed patients with pulmonary tuberculosis(pulmonary tuberculosis group)from Shanghai Xuhui Central Hospital from January 2019 to December 2022 were enrolled,and 506 healthy subjects matched by age and sex were enrolled as healthy control group.The data of 22 blood routine test items and demographic parameters of all the subjects were collected.The collinearity was analyzed by LASSO regression analysis.The datum set was randomly divided into 2 parts:75%was used as the training set for the construction of the machine learning model;25%was used as the test set for the performance evaluation of the model.Four machine learning algorithms,distributed random forest(DRF),deep learning,gradient elevator and generalized linear model,were used to test the model,and the diagnostic efficiency of the model was verified by 5-fold crossover method.The diagnostic performance of the model was evaluated by receiver operating characteristic(ROC)curve.Results Based on Logistic regression analysis and LASSO regression analysis,10 non-collinear indicators were selected.DRF was the opitmal machine learning algorithm for the construction of pulmonary tuberculosis diagnosis.In the training set and test set,the areas under curves of the DRF model were 0.9921 and 0.8474,the sensitivities were 99.16%and 92.04%,the specificities were 80.91%and 55.22%,and the accuracies were 89.84%and 72.06%,respectively.Conclusions The pulmonary tuberculosis diagnosis model based on machine learning algorithm is an effective diagnostic tool,but its clinical application value needs to be further verified.

关 键 词：机器学习 诊断模型 肺结核 血常规 

分 类 号：R446[医药卫生—诊断学]