On classification of simulated sensory deficiency

A feedforward, backpropagation neural network is developed for the purpose of classifying balance disorders using simulated sensory deficiency. Balance of the human multi-segmental system requires the integration of sensors with unique internal and external frames of reference. Thus, spatial information is provided by visual, vestibular and somatosensory inputs (Buchanan, J.J. and Horak, F.B., 1999; Szturm, T. and Fallang, B., 1998), loss or deterioration of which can lead to balance disorders. Using the center-of-foot pressure (COP) as a measure of postural control, three groups were classified: no sensory deficiencies; simulated somatosensory deficiency; simulated visual and somatosensory deficiencies. Features were extracted from the second derivative of the COP, i.e., acceleration. The two selected features, with which the network was trained, were the distances from the maximum absolute value of the acceleration to the average RMS value and average energy value, respectively. The results of this study show that only 1 out of 48 cases was misclassified, supporting neural network modeling as a promising method for classifying sensory deficiency.