Date of Award
Master of Public Health (MPH)
INTRODUCTION: Alzheimer’s Disease is a neurodegenerative disorder that affects millions of individuals worldwide and the association of brain regions to diagnosis is not presently known. Current methods for diagnosis are not sufficient, with the only true method for knowing if an individual has Alzheimer’s Disease being a post mortem analysis of brain tissue. Due to the high dimension of data, a classic principal component analysis to determine which variables to include in a model would not suffice. Sparse Principal Component Analysis deals with the limitations of Classic PCA and can produce which variables are highly correlated to include.
AIM: Compare the results of logistic regression, classic principal component analysis, and sparse principal component analysis to determine the variables to include in a model to differentiate between Mild Cognitive Impairment and Alzheimer’s Diagnosis.
METHODS: We analyzed brain scans from the Alzheimer’s Disease Neuroimaging Initiative. Variables were predefined by the dataset by individual. We used these variables to run a regular logistic regression on all the variables, ran classic PCA on every stepwise increase in components included in the model, and finally ran the Sparse PCA model, comparing error rate to differentiate between the models and select the variables to include.
RESULTS: We identified the error rate for every model, with SPCA with 8 components and a tuning parameters of 6 having the lowest, and then the variables included in that model were selected as the variables for prediction.
DISCUSSION: By applying this method to high dimensional brain scan data, we identified 59 variables to include in the model. Majority of these 59 variables agreed with the current literature for association with Alzheimer’s Disease.
Vashi, Tejal, "Prediction of Disease Status Based on MRI Brain Scans Using Sparse Principal Component Analysis." Thesis, Georgia State University, 2017.