Preliminary Results of Neuromorphic Controller Design and a Parkinson's Disease Dataset Building for Closed-Loop Deep Brain Stimulation

TL;DR

This paper introduces a neuromorphic controller for closed-loop deep brain stimulation in Parkinson's Disease, reducing power consumption and increasing suppression efficiency, and also presents a new dataset of neural activity biomarkers.

Contribution

It proposes a novel neuromorphic LIF-based controller for CL-DBS, improving energy efficiency and effectiveness, and creates a Parkinson's Disease neural activity dataset.

Findings

Power consumption reduced by up to 56%.

Suppression efficiency increased by up to 6.77%.

New Parkinson's Disease neural activity dataset created.

Abstract

Parkinson's Disease afflicts millions of individuals globally. Emerging as a promising brain rehabilitation therapy for Parkinson's Disease, Closed-loop Deep Brain Stimulation (CL-DBS) aims to alleviate motor symptoms. The CL-DBS system comprises an implanted battery-powered medical device in the chest that sends stimulation signals to the brains of patients. These electrical stimulation signals are delivered to targeted brain regions via electrodes, with the magnitude of stimuli adjustable. However, current CL-DBS systems utilize energy-inefficient approaches, including reinforcement learning, fuzzy interface, and field-programmable gate array (FPGA), among others. These approaches make the traditional CL-DBS system impractical for implanted and wearable medical devices. This research proposes a novel neuromorphic approach that builds upon Leaky Integrate and Fire neuron (LIF) controllers to adjust the magnitude of DBS electric signals according to the various severities of PD patients. Our neuromorphic controllers, on-off LIF controller, and dual LIF controller, successfully reduced the power consumption of CL-DBS systems by 19% and 56%, respectively. Meanwhile, the suppression efficiency increased by 4.7% and 6.77%. Additionally, to address the data scarcity of Parkinson's Disease symptoms, we built Parkinson's Disease datasets that include the raw neural activities from the subthalamic nucleus at beta oscillations, which are typical physiological biomarkers for Parkinson's Disease.