Adaptive neuro-fuzzy as a closed-loop model for deep brain stimulation in Parkinson’s disease
AbstractDeep brain stimulation (DBS) is a widely used therapy to treat motor disorders such as Parkinson's disease (PD). To achieve a symptom suppression, an implantable pulse generator constantly delivers current pulses, at fixed parameters, in a deep brain structure such as subthalamic nuclei (STN). In recent years, advances have been published to incorporate a feedback loop in DBS devices to operate in a closed-loop manner in order to reduce on-stimulation time and side effects, optimize battery consumption and improve the patient welfare. In this work, the authors propose a control model for a DBS device based on an adaptive neuro-fuzzy model (ANFIS) whose feedback variables are the beta band power calculated from local field potentials and the magnitude of the acceleration obtained from inertial registers acquired from a smart watch. The proposed model achieved an important decrease in the total electrical energy delivered (TEED) by the stimulator of 111,81 ± 47,36 µW and a theoretical charge density of ( ) of 6,78 ± 1,40 µC/cm2. When comparing the ANFIS model regarding to the non-adaptive fuzzy inference models, the proposed model achieved the lowest TEED and , being 68% lower than conventional DBS and 20% lower than non-adaptive models. In addition, the ANFIS model used 95% less rules in its fuzzy rule base in regards to the non-adaptive models, optimizing resources and processing. Thus, the ANFIS model was the most adequate to control a closed-loop DBS in patients with PD under the experimental conditions of the study.