基于智能手机内置加速度传感器的日常体力活动监测研究

Assessment of Daily Physical Activity Using Built in Accelerometer of Smart Phone

目的：探究智能手机内置加速度传感器测量人体活动原始信号的处理方法,并把其输出结果与ActiGraph进行比对,评估其准确性。方法：研究分为实验室测试（研究1）和日常生活测试（研究2）。在研究1中,50名受测者佩戴2部红米NOTE2、1部三星NOTE3和1个ActiGraph-GT3X完成4类10项活动,每项活动5 min,加上间歇时间全程约65 min,以其原始数据进行滤波、修正和合成,并在此基础上分析其与ActiGraph VM的一致性及划分体力活动强度的准确性。在研究2中,26名受试者连续携带ActiGraph和红米NOTE2手机共4天（2个工作日＋2个休息日）,分析两类设备测量每日各类强度活动时间的一致性。结果：研究1中,智能手机与ActiGraph VM高度相关（ρ=0.87~0.92）;红米NOTE2（荷包）、三星NOTE3（腰部）和红米NOTE2（腰部）之间的相关系数分别为0.942和0.989;以ActiGraph为校标,三星NOTE3（腰部）、红米NOTE2（荷包）和红米NOTE2（腰部）的RMSA分别为：470.2、700.4和506.9,红米（腰部）的测量误差最大;手机和ActiGraph在界定各活动强度类型的准确率上差异小。研究2显示,在每日VPA、MPA和LPA时间上,ActiGraph和手机的相关系数分别为0.819,0.762和0.669。Bland-Altman图显示,VPA的平均差为-2.8min/天,MPA为-5.2 min/天、LPA为22.5 min/天。结论：在VM以及每日LPA、MPA与VPA时间的测量上,手机与ActiGraph的结果一致性好,两类设备原始数据提供高度相似的信息,研究所采用的手机可以作为ActiGraph的补充应用于体力活动干预和运动健身实践中。

Objective：In order to make a foundation for future work,the purpose of this work was to explore the processing approach on original signal from built-in accelerometer of smart phone,compare its output to ActiGraph＇s,and make a assessment on its accuracy.Methods：A laboratory（study 1）and a free-living（study 2）protocol were conducted.In study 1,50 participants engaged in prescribed activities including 10 items of 4categories over a 65-minutes period wearing simultaneously wore two Redmi note2,one Sumsung note3 and ActiGraphGT3 X.Based on raw signal being filtered,modified and synthesized,this research make a analysis on consistency between each devices and their classification accuracy on intensity of physical activity.In Study 2,26 participants wored ActiGraph and Redmi NOTE2 during 4days（2days＋2rest days）,which its data is used to analyzed on consistency of result on time per day of each intensity level of physical activity.Result：In study 1,correlations between the ActiGraph and the 3phones were very strong（ρ=0.87~0.92）.The correlation coefficient between Redmi（pocket）,Sumsung（waist）,between Redmi（waist）were Respectively 0.942and0.989;Regarded output of ActiGraph as standard,Samsung（waist）,Redmi（pocket）and Redmi（waist）of RMSA were 470.2,700.4and 506.9,measurement error from Redmi（waist）was largest;The differences between 3phones and ActiGraph on defining accuracy of each type of activity intensity was small.Within the study 2,results suggested on daily time of VPA,MPA and LPA,the correlation coefficient were respectively 0.819.0.762 and 0.669.Results from Bland-Altman plots suggested close mean absolute estimates of VPA（mean difference=-2.8 min）,MPA（mean difference=-5.2 min）and VPA（mean difference=-22.5 min）.Conclusion：On VM and daily LPA,MPA and VPA time,consistency between phones and ActiGraph was good,which infer that two devices can provide similar information from raw signals,and that phones used in this research can provide an acceptable alternative to an ActiGraph for being applied to intervention for physical activity and research on Epidemiology.