Bayesian Penalized Model for Classification and Selection of Functional Predictors Using Longitudinal MRI Data from ADNI

• Activity Number: 362 - SPEED: Food, Environment, Biomedical Imaging and Physical System Visualization/Learning, Part 2
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 11:35 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Asish Banik* and Taps Maiti and Andrew Bender
• Companies: Michigan State University and Michigan State University and Michigan State University
• Keywords: Bayesian Group Lasso; Spike-and-Slab prior; Gibbs Sampler; Alzheimer’s disease; Basis Spline; volumetric MRI

Abstract

The main goal was to employ longitudinal trajectories in a significant number of subregional brain volumetric MRI data as statistical predictors for Alzheimer’s disease (AD) classification. We used logistic regression in a Bayesian framework that included a large number of functional predictors. In high dimensional scenarios selections of predictors is paramount with the introduction of either Spike-Slab prior, Non-local priors or Horseshoe priors. We sought to avoid the complicated Metropolis-Hastings approach and to develop an easily implementable Gibbs Sampler. Bayesian estimation provides proper estimates of the model parameters, which are also useful for building inference. We combined approaches and also used a probability threshold for classifying individual patients. We used functional predictors consisting of volumetric estimates of brain subregions. Use of Spike-and-Slab priors ensures that a large number of redundant predictors are dropped from the model. The functional analysis of volumetric decline in AD classification, and the identification of specific brain sub-regions selected by our method offer value to future research on the detection of dementia progression.