Recurrent Neural Network Modeling and Low-Dimensional Analysis of Parkinson's Dynamics

This paper constructs a cortex-basal ganglia-thalamus loop model based on a recurrent neural network (RNN) to explore the intrinsic mechanism of the rhythmic oscillatory activity of Parkinson's state, and characterizes the internal transfer characteristics by observing its manifold dynamics. Starting from the rhythmic oscillatory phenomenon in the Parkinson's state, a RNN model is proposed to act as a research tool to probe the Parkinson's kinetic properties. In addition, variational autoencoders are used to extract low-dimensional manifolds from the high-dimensional network activities of RNN model, and the dynamic characteristics behind different rhythmic oscillatory activities and stimuli are captured. It is shown that the RNN model can effectively simulate the progress and evolution of the dynamic system of the cerebral cortex-basal ganglia-thalamus circuit model in different states. And the low-dimensional manifold can effectively observe the suppression effect of external stimuli on the abnormal rhythm through resonance.