Artificial neural network for gait disorder classification

Abstract

Developments in motion analysis systems are distinctive in last decades. Those systemsbecame very important tools for diagnosis of various gait disorders. They evolved so much thatclinicians nowadays dare to use them in critical decisions. Thanks to advances in computer and motion capture technology, several biomechanical joint trajectories of human gait areavailable. Examining all parameters is wearisome and time consuming. Recent inclinations aretowards facilitation of neural networks in similar cases. An Artificial Neural Network could betrained and considered as a decision support system for gait analysis. In this study a neural network is trained for classification of four different gait patterns.Supervised learning method and Error Back-Propagation Algorithms are deployed for thetraining of the Multilayer Perceptron. Matlab programming language was exploited for writingthe code of the algorithm. Overall 150 subjects were used in this thesis. Their age range was between six and twelve years. Samples are collected for normal gait, Right Hemiplegia, LeftHemiplegia and Diplegia from Istanbul University Istanbul Medical Faculty MotionAnalysis Laboratory. Attained classification success for distinguishing normal and for three different abnormal gaits was on average 77%. Further increase in success was achieved after theimplementation of cross validation and early stopping methods, reaching at 85%. For the classification of normal and abnormal gaits into two groups a better classification success rate was achieved, up to 96%.There is still space to build upon the current research for further progress. This neuralclassifier could help clinician to support his/her decisions.|Keywords: Motion analysis, Neural network, Decision support.