学术文献库

Background: Human gait exhibits complex fractal fluctuations among consecutive strides. The time series of gait parameters are long-range correlated (statistical persistence). In contrast, when gait is synchronized with external rhythmic cues, the fluctuation regime is modified to stochastic oscillations around the target frequency (statistical anti-persistence). To highlight these two fluctuation modes, the prevalent methodology is the detrended fluctuation analysis (DFA). The DFA outcome is the scaling exponent, which lies between 0.5 and 1 if the time series exhibit long-range correlations, and below 0.5 if the time series is anti-correlated. A fundamental assumption for applying DFA is that the analyzed time series results from a time-invariant generating process. However, a gait time series may be constituted by an ensemble of sub-segments with distinct fluctuation regimes (e.g., correlated and anti-correlated). Methods: Several proportions of correlated and anti-correlated time series were mixed together and then analyzed through DFA. The original (before mixing) time series were generated via autoregressive fractionally integrated moving average (ARFIMA) modelling or actual gait data. Results: Results evidenced a nonlinear sensitivity of DFA to the mix of correlated and anti-correlated series. Notably, adding a small proportion of correlated segments into an anti-correlated time series had stronger effects than the reverse. Significance: In case of changes in gait control during a walking trial, the resulting time series may be a patchy ensemble of several fluctuation regimes. When applying DFA, the scaling exponent may be misinterpreted. Cued walking studies may be most at risk of suffering this issue in cases of sporadic synchronization with external cues.