A Deep Brain Stimulation Optimization Strategy Based on Actor-Critic Network

Deep Brain Stimulation (DBS) is an effective approach for treating Parkinson's disease. In clinical practice, open-loop DBS parameters are typically selected based on the experience of the physician. However, this empirical approach restricts the effectiveness of DBS in different patients and various clinical contexts. Computational models provide valuable tools and methods for studying the optimization of DBS stimulation parameters. In this study, we introduced a computational model of the basal ganglia-thalamus-cortical circuit in Parkinson's disease. The Actor-Critic framework of reinforcement learning was employed to address the modulation of amplitude, frequency, and target selection in DBS optimization. By utilizing a simple network architecture, it achieved optimal decision-making for DBS within a continuous action space. The average energy consumption after control decreased by 29.39% compared to the constant stimulation. This model-independent control approach provides a novel solution for the treatment of brain disorders.