Assessment of gait nonlinear dynamics by inhomogeneous point-process models

Point-process linear models of stride intervals have been recently proven to provide a unique characterization of human gait dynamics through instantaneous time domain features. In this study we propose novel instantaneous measures characterizing nonlinear gait dynamics using a quadratic autoregressive inhomogeneous point-process model recently devised for the instantaneous assessment of physiological, natural, and physical discrete dynamical systems. Our mathematical framework accounts for long-term information given by the past events of non-stationary non-Gaussian time series, expressed by a Laguerre expansion of the Wiener-Volterra terms. Here, we present a study of gait variability from data gathered from physionet.org, including 15 recordings from young and elderly healthy volunteers, and patients with Parkinson's disease. Results show that our instantaneous polyspectral characterization provides an informative tracking of the inherent nonlinear dynamics of human gait, which is significantly affected by aging and locomotor disabilities.