Date Approved

12-31-2006

Embargo Period

4-7-2016

Document Type

Thesis

Degree Name

M.S. in Electrical Engineering

Department

Electrical & Computer Engineering

College

Henry M. Rowan College of Engineering

First Advisor

Polikar, Robi

Subject(s)

Alzheimer's disease--Diagnosis; Evoked potentials (Electrophysiology)

Disciplines

Electrical and Computer Engineering

Abstract

The recent advances and knowledge in medicine and nutrition have greatly improved our average life expectancy. An unfortunate consequence of this longer life span, however, is a dramatic increase in the number of individuals suffering from dementia, and more specifically, from Alzheimer's disease (AD). Furthermore, AD remains under-diagnosed and under-treated until its more severe stages due to lack of standard diagnostic tools available to community clinics. A search for biomarkers that will allow early diagnosis of the disease is therefore necessary to develop effective medical treatments. Such a biomarker should be non-invasive, simple to obtain, safe, inexpensive, accurate, and most importantly, must be made available to local health clinics for maximum effectiveness. Event related potentials (ERPs) of the electroencephalogram have the potential to become such a diagnostic biomarker for AD.

This work investigates the use of ERP signals for the early detection of AD. The analysis of the ERP signals is accomplished through multiresolution wavelet decomposition, producing time-frequency features in successive spectral bands. In previous studies, these feature sets were concatenated and used as inputs to a neural network classifier. This contribution investigates training an ensemble of classifiers on each feature set separately, and combining the ensemble decisions in a data fusion setting. Comparisons of intra-signal and inter-signal ensemble combinations are presented in along with the benefits of using an ensemble of classifiers in data fusion.

Share

COinS