Bayesian classification of Alzheimer's disease stages from longitudinal volumetric MRI data
Asish Banik, Department of Statistics and Probability, Michigan State University
The primary objective of this article is to build a classification method using longitudinal volumetric magnetic imaging (MRI) data from five regions of interest (ROIs) (hippocampus (H), entorhinal cortex (EC), middle temporal cortex (MTC), fusiform gyrus (FG) and whole brain (WB)). A functional data analysis method is used to handle the longitudinal measurement of ROIs and later the functional coefficients are used in the classification models. We propose a Polya-gamma augmentation method to classify normal controls and diseased patients based on the functional MRI measurements. We get a fast posterior sampling by avoiding slow and complicated Metropolis-Hastings algorithm. Our main motivation is to determine the important ROIs which have the highest separating power for classifying our dichotomous response. We compared the sensitivity, specificity and accuracy of classification based on single ROIs and also with various combinations of them. We obtained sensitivity over 85\% and specificity around 90\% for most of the combinations. Addition of few baseline Mental state exam scores in the model improve our results. The combination in which all five important ROIs and baseline patients' scores are included provides the best result among all other combinations.