探测数据缺失情况下心率变异性信号的复杂性

Detecting complexity changes for heart rate variability in the case of missing data

在人口普查,环境监测及医学科学纵向研究等大型资料分析中,经常会遇到数据缺失或不能利用的问题,它给数据分析与应用带来很多困难.因此衡量一种数据分析方法的优劣性时,其在数据缺失情况下依然能够保持分析结果的稳定性就显得尤为重要.为了模拟最普遍的数据缺失情况,本文设计了一种构造对比数列的方法,以便对完整信号做高斯分布和指数分布的随机数据去除,使其存在缺失值,并运用两种熵测度—基本尺度熵和近似熵对存在数据缺失的信号进行复杂性对比分析.结果显示,去除数据的比例以及缺失数据片段的长度均值这两个关键性参数都会引起序列复杂性的改变,而缺失数据片段的长度服从一个怎样的分布,对分析结果影响不大.而且,近似熵方法对数据缺失有较强的敏感性,不适用于分析存在缺失的心率变异性信号,而运用基本尺度熵方法计算出的结果有较好的稳定性,分析实际信号有其独特的优越性.

Data are frequently missing in the analysis of large databases, such as those for census, environmental, and medical longitudinal studies, causing significant difficulties for data analysis and application. And it is very important to maintain stability of analysis results in the case of missing data when determining the quality of a data processing method. To simulate the case of general data loss in this paper, we developed an approach to generate synthetic signals, which randomly wiped data from the original dataset according to a Gaussian distribution and an exponential distribution. We then used two methods of entropy measure-base-scale entropy and approximate entropy-to perform a complexity comparative analysis. Research has shown that two key parameters-the percentage and average length of missing data segments-can cause changes to sequence complexity. However, no matter what kind of distribution is used for the length of the missing data segments, it has little effect on the results of the analysis. Moreover, approximate entropy has strong sensitivity to missing data,so it was unsuitable for the analysis of signals with missing data. However, the calculation results had preferable stability for base-scale entropy, whose superiority suits it for analysis of actual signals.