Work

A novel fuzzy reinforcement learning approach for personalization deep brain stimulation with communication delay

Year: 2025

Type: article

Abstract: • A novel adaptive closed-loop model is proposed for the regulation of hand tremors. • The proposed model incorporates a nonlinear differential backward controller (NLDBC). • A fuzzy reinforcement learning algorithm is developed to fine-tune the parameters of NLDBC. • Simulation findings illustrate the capability of the proposed model to mitigate hand tremors. • The technique may help patients with Parkinson’s disease effectively manage their hand tremors. This study presents a groundbreaking adaptive closed-loop methodology for the regulation of hand tremors associated with Parkinson’s disease (PD) via a deep brain stimulation (DBS) framework. The proposed methodology incorporates a nonlinear differential backward controller (NLDBC) meticulously engineered to stimulate the subthalamic nucleus (STN) and globus pallidus internus (GPi). Given the incessant character of tremors, a fuzzy reinforcement learning algorithm is developed to fine-tune the parameters of the NLDBC controller, thereby facilitating real-time adaptations. This investigation tackles two principal challenges: i) alleviating the detrimental impacts of wireless communication technologies; while contemporary stimulation systems employ wireless sensors and programming devices for the remote modulation of hand tremors, these advancements engender complications, such as latency and temporal delays that adversely affect neurostimulation procedures; ii) stimulating two targets (the STN and GPi) in contrast to either target alone. A comparative evaluation is performed in silico to measure the effectiveness of dual-target stimulation relative to single-target stimulation. A series of tests assess the system’s performance under diverse conditions, revealing that the dual-target stimulation approach demonstrates approximately 83 % improvement in tremor amplitude and 76 % in produced electrical field intensity (EFi) compared to a single target, and the improvement of the developed system is almost 80% compared to alternative systems.

Source: Biomedical Signal Processing and Control

DOI: 10.1016/j.bspc.2025.107736