Ultra-Local Model Control of Parkinson's Patients Based on Machine Learning

Parkinson’s disease (PD) is one of the most privileged neurodegenerative, which has had an upward trend in recent decades. The most important complications of PD are tremor, rigidity, and slow movement. A surgery method namely Deep brain stimulation (DBS) plays a vital role in the treatment of advanced Parkinson’s patients. In the past decades, stimulating one nucleus of basal ganglia including Globus pallidus internal (GPi) or Subthalamic nucleus (STN) without any feedback (open-loop manner) has had a common strategy, which leads to several different side-effects like muscle tonic and forgetfulness. In the present paper, two nuclei of BG are stimulated in a closed-loop structure (feedback signal) to reduce the entrance electric field intensity to the brain, and in addition to shrinking hand tremor in Parkinson’s patients. For this purpose, an ultra-local model (ULM) control based on a deep deterministic policy gradient (DDPG) is designed to stimulate the STN and a conventional feedback controller is considered for stimulating GPi. In this method, the coefficients of the ULM are adaptively assumed as the control objective parameters, which are designed by the critic and actor neural networks (NNs) of DDPG. To demonstrate the effectiveness and suitability of the suggested approach is compared to state-of-the-art strategies such as ULM, SMC, and PI controllers.