Classification of Stabilograms in Healthy Subjects Using Neural Network

In this study we classified stabilograms to elucidate the characteristics of sway in the body center of the gravity using principal component analysis and neural network (NN) in 826 healthy subjects. Stabilography was performed with eyes open and closed with both feet close together for 60 seconds using a stabilometer. The area, length/time, length/area, power spectrum, position and velocity vectors and standard deviation and kurtosis of amplitude histogram were measured. Principal component analysis and classification using NN were conducted using the measurement values. Principal components were calculated by 80% of the cumulative proportion, and 5 components were obtained. From the eigenvectors of the 5 components, 8 sway types were found and these were used to classify the sways of the 826 subjects. Learning by NN was carried out with measurement values obtained from the sways represented by each pattern, and a weighted NN was obtained with an error of 0.005. Stabilograms of healthy subjects were classified using the weighted NN into 9 sway types; large sway type, low frequency type, high frequency type, forward-backward enlargement type, right-left enlargement type, centripetal type, forward-backward sway type, right-left sway type, and non-specific type. We considered that the NN classification of stabilograms was useful for the observation of daily fluctuation of equilibrium function in healthy subjects.