Chapter 26 A system identification approach to balance testing

Abstract

By modelling the posture control system as a linear transfer function cascaded with a saturation-like nonlinearity, the problem of quantifying relative stability was reduced to a linear system identification problem. The relative stability of different individuals was quantified in terms of the nominal ‘saturation amplitude’ (SA), the transient perturbation amplitude at which the resulting COP displacement would saturate (or equal the length of) the BOS. To ensure subject safety and to minimize anticipatory adaptations, the transient response was determined indirectly. A small-amplitude pseudorandom perturbation was applied and the results used to identify the linear component of the cascade model. The linear transfer function was then used to predict the large-amplitude transient response, thereby allowing the saturation amplitude to be estimated. Three system identification methods were used: cross spectral, least squares and maximum likelihood. In order to approximate the kinematics and sensory input of typical falling situations, the test perturbation was selected to be an anterior-posterior acceleration of the platform on which the subject stood, with a visual field that moved with the platform. The perturbation waveform, power spectrum, bandwidth, duration and amplitude were selected to meet requirements for persistent excitation, accurate identification, stationarity and subject safety and tolerance. Pilot experiments were performed to aid in the selection of the perturbation parameters. Testing of 64 young and elderly normal adults has shown the balance test to be a sensitive measure of the deterioration in postural control known to occur in normal ageing. Depending on the system identification method used, the balance test has been able to identify up to three of five patients with peripheral vestibular lesions and five of five elderly subjects with a history of falling, at a false positive rate of 25%. Comparison of the model predictions with direct measurements of transient response has indicated that the model may overestimate the response to large transient perturbations; however, the transient test results may have been confounded by adaptive effects. The balance test results have shown no significant correlation with measures of spontaneous postural sway, except in normal young adults.