Learning-Based Control Design for Deep Brain Stimulation
By employing low-voltage electrical stimulation of the basal ganglia (BG) regions of the brain, deep brain stimulation (DBS) devices are used to alleviate the symptoms of several neurological disorders, including Parkinson's disease (PD). Recently, we have developed a Basal Ganglia Model (BGM) that can be utilized for design and evaluation of DBS devices. In this work, we focus on the use of a hardware (FPGA) implementation of the BGM platform to facilitate development of new control policies. Specifically, we introduce a design-time framework that allows for development of suitable control policies, in the form of electrical pulses with variable temporal patterns, while supporting tradeoffs between energy efficiency and efficacy (i.e., Quality-of-Control) of the therapy. The developed framework exploits machine learning and optimization based methods for design-space exploration where predictive behavior for any control configuration (i.e., temporal pattern) is obtained using the BGM platform that simulates physiological response to the considered control in real-time. To illustrate the use of the developed framework, in our demonstration we present how the BGM can be utilized for physiologically relevant BG modeling and design-state exploration for DBS controllers, as well as show the effectiveness of obtained controllers that significantly outperform conventional DBS controllers.
