基于XGBoost算法的冠心病阵列式脉图参数特征分析和辅助预测模型研究

Research on Parameter Feature Analysis and Auxiliary Prediction Model of Coronary Heart Disease Array Pulse Graph Based on XGBoost Algorithm

目的研究冠心病患者的阵列式脉图参数特征,探索基于极限梯度提升算法(extreme gradient boosting,XGBoost)建立冠心病辅助预测模型。方法纳入社区老年体检中心的正常人群106例(正常对照组)和冠心病患者300例(冠心病组),使用24点阵列式脉诊仪采集受试者的脉象数据,分析冠心病组与正常对照组的阵列式脉图容积(array pulse volume,APV)差异,并基于XGBoost算法建立冠心病辅助预测模型。结果3种脉图通道数据分析方法的对比差异结果趋同,均有冠心病组h 1、h 4、h 5、h 1/t 1、As显著低于正常对照组(P<0.05),是3种分析方法中共有的冠心病脉图诊断特征。冠心病组APV h1、APV h2、APV h3、APV h4、APV h5显著低于正常对照组(P<0.05)。最大幅值通道均值法的模型综合性能最好,指标包括t 1、t 3、t 4、w 1、w 2、w 1/t、w 2/t。结论阵列式脉图特征参数一定程度上可以反映冠心病患者的心血管功能状态,APV指标可以提升模型的冠心病辅助预测性能。

Objective To study the characteristics of array pulse diagram parameters of patients with coronary heart disease(CHD),and explore the establishment of auxiliary prediction model of CHD based on extreme gradient boosting(XGBoost).Methods 106 normal people(normal control group)and 300 patients with CHD(CHD group)were included in the community geriatric physical examination center.The pulse data of the subjects were collected by 24-point array pulse detector.The differences of array pulse pattern parameters between the CHD group and the normal control group were analyzed,and the auxiliary prediction model of CHD was established based on XGBoost algorithm.Results The comparative difference results of the three pulse map channel data analysis methods were similar,and h 1,h 4,h 5,h 1/t 1 and As in the CHD group were significantly lower than those in the normal control group(P<0.05),which was a common diagnostic feature of CHD pulse map among the three analysis methods.APV h1,APV h2,APV h3,APV h4 and APV h5 in CHD group were significantly lower than those in normal control group(P<0.05).The maximum value channel mean method has the best comprehensive performance,including t 1,t 3,t 4,w 1,w 2,w 1/t and w 2/t.Conclusion The characteristic parameters of the array pulse map can reflect the cardiovascular function status of patients with CHD to a certain extent,and the APV index can improve the auxiliary prediction performance of the model.