Adaptive multi symptoms control of Parkinson's disease by deep reinforcement learning. (February 2023)

• Record Type: Journal Article
• Authors: Faraji, Behnam; Rouhollahi, Korosh; Mollahoseini Paghaleh, Saeed; Gheisarnejad, Meysam; Khooban, Mohammad-Hassan
• Is Part Of: Biomedical signal processing and control. Volume 80:Part 2(2023)
• Publication Date: 2023-02
• Subjects: Parkinson's Disease (PD) -- Deep Brain Stimulation (DBS) -- Rigidity -- Hand tremor -- Deep Deterministic Policy Gradient (DDPG) -- Sliding Mode (SM); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104410
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: In response to the aforementioned issues, an intelligent control strategy has been established for reducing hand tremors and rigidity in Parkinson's patients, respectively. An ULM controller was developed in a model-free framework to mitigate simultaneously tremor and rigidity by stimulating the basal ganglia system in the current study. In the ULM controller, an SMO is used to estimate the ULM's poorly understood dynamics. By adopting the input and output of the BG system, the ULM controller was designed without the need for the model identification of the basal ganglia system. The learning capability of DDPG is used to reduce tremor and rigidity by interacting the agent with the BG model. The Actor and Critic of the learning control are trained in a model-free way by introducing a reward function as the optimizing goal. The hardware-in-the-loop (HiL) simulations under various scenarios of the dynamic system were accomplished for real-time analysis of the designed controllers. Abstract: Parkinson's disease (PD) is one of the really frequent disorders, with hand and head tremors and rigidity being the most common sequelae. Deep brain stimulation (DBS) is a common treatment used to alleviate the symptoms of this disease. This work investigates an ultra-local model (ULM) based on a sliding mode observer (SMO) to simultaneously reduce hand tremor and rigidity. specifically, a deep deterministic policy gradient (DDPG) controller is adaptively designed in the current study to reduce observer estimation error and improve the nonlinear dynamic features of a central neural network (CNN). The DDPG is designed with an actor that produces policy demands and a critic that measures the effectiveness of the actor's policy orders. The offered methodology employs a DDPG-based mechanism to compensate for the shortcomings of the ULM-based SMO. In the present mechanism, training of the weight values of both networks (actor and critic) is by the gradient descent way that relies on the tremor fault's reward return. Finally, the following methodology is analyses by computer simulation in a variety of contexts (robustness and controller performance) and compared to current practices to prove the benefits and adaptability of the procedure with varied models and patients. Additionally, the controllers are implemented in the hardware-in-the-loop (HiL) simulations testbed to validate the performance of the developed scheme's profitability from a realistic perspective.