基于可视化图方法的体征时间序列数据分类分析研究  

Classification of Biological Signals Time Series by Extracting the Network Features Based on Visibility Graph

作　　者：焦晓宇[1] 周雪忠[1] 胡镜清[2] 谢琪[3] 周洪伟[4] 

机构地区：[1]北京交通大学计算机与信息技术学院交通数据分析与挖掘北京市重点实验室,北京100044 [2]中国中医科学院中医基础理论研究所,北京100700 [3]中国中医科学院,北京100700 [4]中国中医科学院中医中医药数据中心,北京100700

出　　处：《世界科学技术-中医药现代化》2016年第4期664-670,共7页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：科学技术部国家科技支撑计划项目(2013BAH06F03):服务老年公寓的健康服务应用系统研发与应用;负责人:谢琪;科学技术部国家中医药行业科研专项(201307003):基于中医特色的老年社区的健康监测与干预关键技术研究;负责人:胡镜清

摘　　要：目的:本研究采用复杂网络理论研究体征时序数据,使用网络特征刻画体征时序数据的动态特征,分析网络特征与人体生理系统健康状态的关系。方法:本文采用可视化图方法将标准心率时序数据和穿戴设备采集的老年人心率等体征时序数据网络化,提取网络特征,采用决策树分类方法分析网络特征与心脏疾病和年龄因素关系。结果:决策树模型对心脏疾病和年龄因素有较好的分类结果,标准心率时序数据的分形特性使网络度分布为幂律分布,网络图密度特征是与心脏疾病和年龄因素相关的主要因素。结论:网络拓扑结构继承体征时序数据的动态特性并将之体现在网络特征上。体征时序数据的动态特性和网络特征的对应关系还待进一步研究阐明。This study aimed to transform the time series to network features using complex network approaches, and investigate associations between physiological network features and human health state. In this study, networks of standard heart rate time series and physiologic time series of the elderly that collected by wearable devices were built using visibility graph method. Then network features were extracted from these networks, and decision tree model was applied to analyze the main factors of network features contributing to heart disease and age. It was found that the fractal characteristic of heart rate time series brought out powerful law distribution for the degree distribution, and the network density became one of the major factors which were relevant to heart disease and age. In conclusion, it was indicated that topological features of networks underlay the dynamic characteristics of human physiologic time series. However, the correspondence between them still remained to be clarified.

关 键 词：时间序列 复杂网络 网络特征 生理体征 老年健康 

分 类 号：R19[医药卫生—卫生事业管理]