Towards a "gold-standard" approach to address the presence of long-range auto-correlation in physiological time series
Towards a "gold-standard" approach to address the presence of long-range auto-correlation in physiological time series
复制标题
DOI：
10.1016/j.jneumeth.2010.07.017
复制
发表时间：
2010-09-30
影响因子：
3
通讯作者：
Lejeune, T. M.
中科院分区：
文献类型：
--
作者：
Crevecoeur, F.;Bollens, B.;Lejeune, T. M.
由循环运动产生的一系列电机输出通常很复杂，这表明时间序列的相关函数跨越大量连续样本。著名的例子包括步幅间隔、心跳变异性、自发神经放电模式或与外部起搏的运动同步。长程相关性对于基础研究可能很重要，因为产生这些相关性的神经和生物力学机制仍然未知，并且对于临床应用来说，考虑到长程相关性的丧失可能是疾病的标志。然而，尚未使用系统方法或稳健的分析方法来支持生理序列中相关函数的研究。本研究研究了四种选定的方法（赫斯特指数、功率谱密度分析、矩收敛率和多尺度熵方法）。我们展示了对已知自相关函数的人工计算机生成的序列进行的每次分析的结果，然后对从步态和上肢节律运动中提取的时间序列进行分析。我们的结果表明，使用 Hurst 指数和功率谱密度的组合分析适合相当短的序列（512 点）。矩收敛率直接支持功率谱密度分析，多尺度熵进一步证实了长程相关性的存在，尽管这种方法似乎更适合较长的序列。所提出的方法增加了对生理序列是长记忆过程这一假设的信心，这对于未来的基础和临床研究至关重要。 (C) 2010 Elsevier B.V. 保留所有权利。
Series of motor outputs generated by cyclic movements are typically complex, suggesting that the correlation function of the time series spans over a large number of consecutive samples. Famous examples include inter-stride intervals, heartbeat variability, spontaneous neural firing patterns or motor synchronization with external pacing. Long-range correlations are potentially important for fundamental research, as the neural and biomechanical mechanisms generating these correlations remain unknown, and for clinical applications, given that the loss of long-range correlation may be a marker of disease. However, no systematic approach or robust analysis methods have yet been used to support the study of correlation functions in physiological series. This study investigates four selected methods (the Hurst exponent, the power spectral density analysis, the rate of moment convergence and the multiscale entropy methods). We present the result of each analysis performed on artificial computer-generated series in which the auto-correlation function is known, and then on time series extracted from gait and upper limb rhythmic movements. Our results suggest that combined analysis using the Hurst exponent and the power spectral density is suitable for rather short series (512 points). The rate of moment convergence directly supports the power spectral density analysis, and the multiscale entropy further confirms the presence of long-range correlation, although this method seems more appropriate for longer series. The proposed methodology increases the level of confidence in the hypothesis that physiological series are long-memory processes, which is of prime importance for future fundamental and clinical research. (C) 2010 Elsevier B.V. All rights reserved.