ニューラルネットの学習による健常例，迷路・中枢障害例の重心動揺の識別

Abstract

The present paper investigated methods of discriminating stabilograms of healthy subjects from those of patients with labyrinthine and central disorders using a neural network (NN).Stabilometry was performed with eyes open and closed with both feet close together for 60 seconds using a stabilometer. From the stabilograms, area, locus length, deviations of the sway, Romberg ratio, power spectrum, position and velocity vectors, and amplitude probability density distribution were measured. Stabilograms were evaluated using a NN program produced by Anima corporation. Data file for learning was composed of healthy subjects and patients with various labyrinthine and central disorders. Data file for evaluation was composed of patients diagnosed with labyrinthine disorders.As a result of learning, stabilograms of healthy subjects and patients of labyrinthine and central disorders were clearly discriminated. The square error was 0.005. In evaluations using a NN weighted by learnig, the square error was 0.326.In medical examinations of patients with vertigo and equilibrium disturbances, discrimination of stabilograms is important for diagnosis of sites of lesion and classification of types and stages of disease. Learning and evaluation using a neural network are expected as a useful method for discriminating stabilograms. However, errors in evaluation (i.e. generalization error) remain high value. In the future, reducing generalization errors is a key issue.