a gray-box neural network model of parkinson’s disease using gait signal

in this study, we focused on the gait of parkinson’s disease (pd) and presented a gray box model for it. we tried to present a model for basal ganglia structure in order to generate stride time interval signal in model output for healthy and pd states. because of feedback role of dopamine neurotransmitter in basal ganglia, this part is modelled by “elman network”, which is a neural network structure based on a feedback relation between each layer. remaining parts of the basal ganglia are modelled with feed-forward neural networks. we first trained the model with a healthy person and a pd patient separately. then, in order to extend the model generality, we tried to generate the behaviour of all subjects of our database in the model. hence, we extracted some features of stride signal including mean, variance, fractal dimension and five coefficients from spectral domain. with adding 10% tolerance to above mentioned neural network weights and using genetic algorithm, we found proper parameters to model every person in the used database. the following points may be regarded as clues for the acceptability of our model in simulating the stride signal: the high power of the network for simulating normal and patient states, high ability of the model in producing the behaviour of different persons in normal and patient cases, and the similarities between the model and physiological structure of basal ganglia.