Estimating of balance status in hemiparetics: An artificial neural network implementation

Abstract

Purpose: The aim of the study is to estimate balance status of hemiparetics using an Artificial Neural Network(ANN) implementation. Methods: Fourty-five hemiparetics(Female:14,Male:31) were included. Balance Evaluation Systems Test(BESTest) was used to evaluate balance status of the participants. We also used Functional Reach Test(FRT), One Leg Standing Test(OLST) (left and right side), 6 Meter Timed Walk(6MTW), and Timed Up&Go Test(TUG) to predict the BESTest total score. IBM SPSS Statistics 21 was used to analyze descriptive statistics. Matlab Neural Network Toolbox was used to build ANN and Multiple Linear Regression (MLR) models. Results: Mean age of the sample was 54.24±8.19 years. Mean of the BESTest total score, FRT, OLST-left side, OLST-right side, 6MTW and TUG were 26.99±7.12, 12.23±13.23 cm, 11.44±10.71 sn, 12.86±10.44 sn, 18.77±14.01 sn and 71.11±16.91 sn, respectively. The designed ANN included five inputs and one output. The number of hidden layers of this ANN with a single hidden layer has been tested from one to ten. Randomly, the data set was used for learning(70%), evaluation(15%), and testing purposes(15%), respectively. The results obtained from this study showed that, it was observed that ANN estimations were in agreement with the measured BEST test total score as well as with the MLR estimations(Root mean squared error-RMSE- for ANN:3.86, forMLR:6.97). Conclusions: We found modeling through ANN as a promising and useful tool to estimate balance status of hemiparetics. Predicting balance status of hemiparetics makes contribution to plan the most suitable rehabilitation program providing useful and better suggestions to establish home modification and relevant exercise program.