Parkinson's Disease Classification, Monitoring and Optimization of Deep Brain Stimulation

• Main Author: Shah, Syed Aamir Ali
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2021
• Subjects: Algorithms; Artificial intelligence; Brain research; Computer science; Deep brain stimulation; Deep learning; Dopamine; Dynamical systems; Electrodes; Electroencephalography; Mathematics; Neural networks; Neurons; Neurosciences; Optimization; Parameter estimation; Parkinsons disease; Patients; Time series
• ISBN: 9798352982464

Parkinson’s disease (PD) alters the brain’s anatomy when dopamine releasing neurons die in basal ganglia neural pathways. The resulting alterations force synchronized neuronal activity in β frequency components in subthalamic nucleus (STN). This synchronization affects the chaotic behavior of the brain activities, which induce motor related impairments in patient’s limbs. This can be seen in the electroencephalogram (EEG) signals of PD patients. Deep brain stimulation (DBS) is the therapy of choice in later stage of disease when the side-effects of medication supersede its benefits. In this research work, we utilize the relationship between chaotic systems and brain for the analysis of EEG signals of PD patients. The overall objective is to develop a framework that can monitor, analyze the chaotic behavior, and optimize the stimulation of various regions of the brain to restore normal state. The first component of the overall objective is to monitor the activities of brain regions. In this component, we use embedding reconstruction, a tool from chaos theory, to highlight PD-related alterations in dynamical properties of EEG and presentit as a potentially reliable bio-marker for PD related classification. Additionally, there is a strong synchronization between amplitude of higher frequency components and phase of β components for PD patients. This information is used to improve the performance of this classification. We apply embedding reconstruction to design a new architecture of a deep neural network called Dynamical system Generated Hybrid Network. We report that this network outperforms the state of the art in terms of classification accuracy of 99.2. (+0:52%) with approximately 24% of the computational resources. The second component is a step towards analysis of chaotic behavior of brain in which we investigate the use of recurrent neural networks to adapt the chaotic characteristics of chaotic time series. We define three unique topologies of long shortterm memory (LSTM) network and analyze those for chaotic parameter estimation in seven different test cases. We show that the deep LSTM networks are capable of modeling the chaotic behavior of a wide range of parameters and that the network performs the best when the architecture is driven by chaotic attributes of the time series data. To mimic the behavior of brain regions, the third component of the overall objective, we utilize a neural reservoir of spatially distributed spiking neurons to our advantage. This reservoir is inspired by the anatomy and operations of biological brain. It provides a testing platform for final component of the overall objective. The final component is to restore the brain activities using DBS. In this component, we propose a simulation framework to define an optimized DBS stimulus using EEG signals. The objective of this framework is to provide simulation environment inspired by realistic brain. This framework utilizes spiking neurons, similar to realistic neurons, distributed in a reservoir which is inspired by real brain anatomy. This reservoir is trained via spike-time-dependent-plasticity algorithm which is inspired by biologically realistic neural learning process. This reservoir is initially set to OFFmedication state. Then, the synaptic changes are optimized in such a way that the reservoir is forced to drift from OFF-medication state to ON-medication state. The stimulus signal is generated by accumulating the variations in synaptic weights in neural reservoir in target brain region. For validation we analyze this signal and show that the application of this signal as stimulus results in local field potential (LFP) of STN region neurons with increased chaotic level compared to LFP without stimulation using SIM4LIFE simulation software.