Automatic Berg Balance Scale assessment system based on accelerometric signals

This paper presents the automatic system for computer-aided balance assessment reflecting the Berg Balance Scale. The system employs inertial sensors for the data acquisition purposes. A set of features is extracted from multiple signal representations during the examination. A multilevel Fisher's linear discriminant is used to select most suitable features for each of the BBS tasks. The feature space dimensionality reduction and the multilayer perceptron classifier training both involve expert scoring on the observed examinations. The system is verified using data acquired during the BBS scoring of 64 elderly patients. Both assessment modes: the entire examination as well as separate BBS items, are evaluated and discussed using introduced assessment metrics.