Date Approved


Embargo Period


Document Type


Degree Name

M.S. in Engineering


Electrical & Computer Engineering


Henry M. Rowan College of Engineering

First Advisor

Polikar, Robi


Electroencephalography; Nervous system--Diseases--Diagnosis


Electrical and Computer Engineering


The prevalence of Alzheimer's disease (AD), Parkinson's disease (PD), and mild cognitive impairment (MCI) are rising at an alarming rate as the average age of the population increases, especially in developing nations. The efficacy of the new medical treatments critically depends on the ability to diagnose these diseases at the earliest stages. To facilitate the availability of early diagnosis in community hospitals, an accurate, inexpensive, and noninvasive diagnostic tool must be made available. As biomarkers, the event related potentials (ERP) of the electroencephalogram (EEG) - which has previously shown promise in automated diagnosis - in addition to volumetric magnetic resonance imaging (MRI), are relatively low cost and readily available tools that can be used as an automated diagnosis tool.

16-electrode EEG data were collected from 175 subjects afflicted with Alzheimer's disease, Parkinson's disease, mild cognitive impairment, as well as non-disease (normal control) subjects. T2 weighted MRI volumetric data were also collected from 161 of these subjects. Feature extraction methods were used to separate diagnostic information from the raw data. The EEG signals were decomposed using the discrete wavelet transform in order to isolate informative frequency bands. The MR images were processed through segmentation software to provide volumetric data of various brain regions in order to quantize potential brain tissue atrophy. Both of these data sources were utilized in a pattern recognition based classification algorithm to serve as a diagnostic tool for Alzheimer's and Parkinson's disease. Support vector machine and multilayer perceptron classifiers were used to create a classification algorithm trained with the EEG and MRI data. Extracted features were used to train individual classifiers, each learning a particular subset of the training data, whose decisions were combined using decision level fusion. Additionally, a severity analysis was performed to diagnose between various stages of AD as well as a cognitively normal state.

The study found that EEG and MRI data hold complimentary information for the diagnosis of AD as well as PD. The use of both data types with a decision level fusion improves diagnostic accuracy over the diagnostic accuracy of each individual data source. In the case of AD only diagnosis, ERP data only provided a 78% diagnostic performance, MRI alone was 89% and ERP and MRI combined was 94%. For PD only diagnosis, ERP only performance was 67%, MRI only was 70%, and combined performance was 78%. MCI only diagnosis exhibited a similar effect with a 71% ERP performance, 82% MRI performance, and 85% combined performance. Diagnosis among three subject groups showed the same trend. For PD, AD, and normal diagnosis ERP only performance was 43%, MRI only was 66%, and combined performance was 71%. The severity analysis for mild AD, severe AD, and normal subjects showed the same combined effect.