Using Postural Control System Measures to Detect Hypovigilance

Posturography is a method to assess the postural control system quantitatively providing the possibility for testing vigilance. In this paper we present pilot experiments and analysis investigating the discriminatory abilities of posturography. A total of 10 young adults participated in postural assessment within a study with extended time awake. Acquired measurements were assigned to two classes “vigilant” and “hypovigilant” according to subject’s continuous time-since-sleep (TSS). Two features sets were extracted from posturographical recordings. In time domain 7 kinds of features from other authors were extracted, including measures of sway velocity and sway area. In spectral domain features were extracted by estimating power spectral densities and subsequent averaging in equidistant spectral bands. In addition to static feature extraction this paper introduces analysis of temporal dynamics within the features of both domains. The ability to discriminate between both classes “vigilant” and “hypovigilant” was evaluated in terms of mean test set errors estimated using 25-fold delete-d cross validation. Different algorithms of computational intelligence including artificial neural networks (ANN) and SupportVector Machines (SVM) were applied. SVM using Gaussian kernel functions performed best with achieved mean test set error rates of 9.0 ± 4.2 %. Author Keywords