Date of Award

8-1-2016

Document Type

Masters Thesis

Degree Name

M.S.

Organizational Unit

Daniel Felix Ritchie School of Engineering and Computer Science, Electrical and Computer Engineering

First Advisor

Jason Jun Zhang, Ph.D.

Second Advisor

Kimon Valavanis

Keywords

Parkinson’s disease, Local field potentials, Subthalamic nucleus, Deep Brain Stimulation

Abstract

This thesis aims to develop of methods for behavior onset detection of patients with Parkinson's disease (PD), as well as to investigate the models for classification of different behavioral tasks performed by PD patient. The detection is based on recorded Local Field Potentials (LFP) of the Subthalamic nucleus (STN), captured through Deep Brain Stimulation (DBS) process.

One main part of this work is dedicated to the research of various properties and features of the STN LFP signals of several patients' behavior conditions. Features based on temporal and time-frequency analysis of the signals are developed and implemented. Evaluation and comparison of the features is conducted on several patients' data during a classification process, using onset windows of preprocessed signals.

Another part of this research is concentrated on automated onset detection of behavioral tasks for patients with PD using the LFP signals collected during DBS implantation surgeries. Using time-frequency signal processing methods, features are extracted and clustered in the feature space for onset detection. Then, a supervised model is employed which used Discrete Hidden Markov Models (DHMM) to specify the onset location of the behavior in the LFP signal.

Finally, a method for simultaneous onset detection and task classification for patients with PD is presented, which classifies the tasks into motor, language, and combination of motor and language behaviors, using LFP signals collected during DBS implantation surgeries. Again, time-frequency signal processing methods are applied, and features are extracted and clustered in the feature space. The features extracted from automated detected onset are used to classify the behavior task into predefined categories. DHMM is merged with SVM in a two-layer classifier to boost up the behavior classification rate into 84%, and the presented methodology is justified using the experimental results.

Publication Statement

Copyright is held by the author. User is responsible for all copyright compliance.

Rights Holder

Nazanin Zaker Habibabadi

Provenance

Received from ProQuest

File Format

application/pdf

Language

en

File Size

68 p.

Discipline

Electrical engineering, Nanoscience



Share

COinS