Date of Award
2024
Document Type
Campus Access Master's Thesis
Degree Name
Master of Science in Biomedical Engineering (MS)
Administrative Home Department
Department of Biomedical Engineering
Advisor 1
Chunxiu Yu
Advisor 2
Lan Zhang
Committee Member 1
Bruce Lee
Abstract
Parkinson’s Disease (PD) is a neurodegenerative disorder which affects movement and has no cure. Electrical stimulation to the brain, known as deep brain stimulation (DBS), is known to alleviate the symptoms of PD. Adaptive Deep Brain Stimulation (aDBS) uses an algorithm to generate electrical stimulation patterns with respect to the state of the patient to treat the motor symptoms of PD. A novel aDBS algorithm was created by training a reinforcement learning model in a biophysical model of parkinsonism. The reinforcement learning algorithm only takes in one biomarker from the model and determines an appropriate stimulation pattern from the collected data. The algorithms trained with an average frequency of 30 and 35 Hz were able to effectively reduce neural oscillations indicative of parkinsonism. The model trained with an average frequency of 35 Hz was able to be quantized, reducing the size of the model for eventual use in an implantable pulse generator (IPG). Quantization of the algorithm did not reduce the performance of the model. Applying a conventional stimulation pattern during training was found to increase the rate for the model to create an algorithm but did not increase the performance. These findings show that reinforcement learning can be used to create an adaptable and personalized stimulation pattern; furthermore, the algorithm can be quantized, reducing the size for eventual use in an IPG.
Recommended Citation
Mehregan, Jessica E., "APPLICATION OF MACHINE LEARNING TECHNIQUES IN THE DESIGN OF ADAPTIVE DEEP BRAIN STIMULATION FOR PARKINSONISM", Campus Access Master's Thesis, Michigan Technological University, 2024.