Parkinson's disease (PD) is the second most common neurodegenerative disease, affecting 1 in 100 people over the age of 60, and more than 5 million people worldwide. Deep brain stimulation (DBS) has emerged as an effective treatment for the motor signs of PD. However, DBS is often limited by partial efficacy, side effects, and long and difficult clinical visits to identify and program therapeutic stimulation settings. Additionally, PD symptoms fluctuate from moment-to-moment depending on a multitude of factors, and therefore the demand for stimulation (or medication) changes throughout the day. There is a clear need for automated, individualized DBS programming strategies that can both reduce programming time and adapt to deliver optimal stimuli as each patient's symptoms fluctuate. The ultimate goal of this proposal is to improve the quality of life for PD patients by providing them with an intelligent, adaptive, closed-loop DBS algorithm.
In Aim 1, we propose to create a closed-loop DBS algorithm in silico that can learn from and adapt to electro-physiological signatures of parkinsonian states in order to deliver optimal, patient- specific stimulation. The algorithm will be built upon the framework of reinforcement learning, which has several desirable properties that make it ideally suited for use in a closed-loop DBS algorithm. Given a biomarker for parkinsonian severity and a set of possible actions, the algorithm will autonomously learn how to deliver stimulus to reduce the biomarker and therefore PD symptoms. Preliminary results shown in this proposal suggest that the algorithm is able to learn complex actions without a priori knowledge, and can help to control a selected biomarker.
In Aim 2, we propose to test out the algorithm in a large animal model of PD with high-density DBS arrays that have been shown to enable more advanced current steering and higher resolution local field potential recordings. We will then apply the algorithm and compare its performance to continuous DBS in terms of effects on individual parkinsonian motor signs as well as the electrophysiological signal targeted by the algorithm. This project will not only provide a means to critically evaluate putative biomarkers of DBS therapy for alleviating parkinsonian motor signs, but also provide a much more versatile and powerful closed-loop algorithm for use in titrating DBS therapy for individuals with Parkinson's disease and other neurological disorders amenable to deep brain stimulation therapy. As a result of this project, we will create a closed-loop DBS algorithm that can help improve the quality of life for Parkinson's disease patients.
Parkinson's disease (PD) is a neurodegenerative disorder that affects 1 in 100 people over the age of 60, and while dopamine replacement medication is well established for alleviating many of the motor symptoms of PD, but deep brain stimulation (DBS) is often indicated, in conjunction with medication as the disease progresses. One of the significant challenges with DBS is patient-specific dosimetry of stimulation settings to efficiently achieve a strong and consistent therapeutic effect. This project will develop an innovative reinforcement learning algorithm that autonomously learns which stimulation settings are most effective at reducing parkinsonian motor signs based on closed-loop feedback of oscillatory activity within the subthalamic nucleus and globus pallidus using high-density electrode arrays.
