Nonlinear Mixed Effect Model for Wavelet-transformed Longitudinal MRI Data in Non-demented and Demented Older Adults
Date of Award
Master of Arts (AM/MA)
Alzheimer's disease (AD) is the most common form of dementia. How AD affects brain structure and tissues has aroused a lot of attention. Due to the high dimensionality of Magnetic Resonance Imaging data, it is difficult to conduct data analysis on them. My thesis uses data from Open Access Series of Imaging Studies, a longitudinal collection of 373 MRI sessions from 150 subjects aged 60 to 96. Instead of processing images directly, I use wavelet transformation to change image data into the wavelet domain for more efficient data reduction. A linear mixed effect model is then fitted for every dominating wavelet coefficient that contains the major information of the images. I discover that the association between the clinical covariates and the features of the brain images that are represented by the dominating wavelet coefficients is monotone but nonlinear over the index of the coefficients. To capture such a nonlinear trend and integrate all features in one model, I try to fit a nonlinear mixed effect model. The estimates from nonlinear least square models are used as the initial values for the parameters. For the better interpretation and visualization of the estimates, I predict the wavelet coefficients from the model and reconstruct the predicted images. After reconstruction, the effects of Clinical Dementia Rating and baseline age can be easily observed while gender has a significant effect.
Chair and Committee
Gu, Tianhui, "Nonlinear Mixed Effect Model for Wavelet-transformed Longitudinal MRI Data in Non-demented and Demented Older Adults" (2014). Arts & Sciences Electronic Theses and Dissertations. 330.