基于机器学习的单相抑郁与双相抑郁鉴别的脉图参数分类模型  

Classification model of pulse parameters of unipolar depression and bipolar depression based on machine learning

作　　者：刘鑫子 李自艳 郑思思 朱虹 尹冬青[1,2] 贾竑晓 Liu Xinzi;Li Ziyan;Zheng Sisi;Zhu Hong;Yin Dongqing;Jia Hongxiao(The National Clinical Research Center for Mental Disorders & Beijing Key Laboratory of Mental Disorders, Beijing Anding Hospital, Capital Medical University, Beijing 100088, China;Advanced Innovation Center for Human Brain Protection, Capital Medical University, Beijing 100069, China)

机构地区：[1]首都医科大学附属北京安定医院,国家精神心理疾病临床医学研究中心,精神疾病诊断与治疗北京市重点实验室,北京100088 [2]人脑保护高精尖创新中心首都医科大学,北京100069

出　　处：《首都医科大学学报》2022年第3期407-414,共8页Journal of Capital Medical University

基　　金：首都卫生发展科研专项(2020-4-2123);北京市医院管理中心临床医学发展专项经费资助(ZYLX202129);北京医院管理中心登峰人才(DFL20191901)。

摘　　要：目的基于机器学习方法研究针对单相抑郁与双相抑郁脉图参数的特征差异。方法采用道生中医四诊仪DS01-A信息采集系统,对31名单相抑郁患者与57名双相抑郁患者分别采集脉图信息,应用SPSS 26进行脉图参数比较,应用R 4.0.5建立向量机算法(support vector machines,SVM)与随机森林算法(random forest,RF)模型,评估模型性能并获得模型特征重要性排序。结果单相抑郁与双相抑郁脉图参数在脉力、脉率、AD、T1/T4、H1、H3/H1、(H3-H1)/H1共7个变量差异具有统计学意义(P<0.05);SVM模型与RF的鉴别模型都具有较好的鉴别准确率和稳定性。SVM模型鉴别准确率为80.56%,曲线下面积(area under curve,AUC)值为83.04%;RF模型鉴别准确率为80.56%,AUC值为84.62%。模型特征重要性前5位分别为H4、H2、AD、AGE、(T4-T1)/T与H3、H4、AD、AGE、H2。结论单相抑郁与双相抑郁在脉图参数上具有明显差异特征,可辅助临床医生进行单双相抑郁鉴别。Objective To study the characteristic difference of pulse parameters between unipolar depression and bipolar depression based on machine learning method.Methods Thirty-one patients with unipolar depression and 57 patients with bipolar depression were collected with Daosheng DS01-A TCM four diagnostic instruments,and the pulse parameters were compared with SPSS 26.The models of support vector machines(SVM)and random forest(RF)were established with R 4.0.5.We evaluated the model performance and obtained the importance ranking of model features.Results There were significant differences in pulse force,pulse rate,AD,T1/T4,H1,H3/H1,and(H3-H1)/H1 between unipolar depression and bipolar depression(P<0.05).Both SVM model and RF model have good accuracy and stability.The accuracy of SVM model is 80.56%,area under curve(AUC)value is 83.04%.The accuracy of RF model is 80.56%,AUC value is 84.62%.The top 5 importance of model features are H4,H2,AD,AGE,(T4-T1)/T and H3,H4,AD,AGE,H2 respectively.Conclusion Unipolar depression and bipolar depression have obvious differences in pulse parameters,which can assist clinicians in the differentiation of unipolar and bipolar depression.

关 键 词：单相抑郁 双相抑郁 脉图参数 机器学习 

分 类 号：R749[医药卫生—神经病学与精神病学]