Analysis of Multiple Sclerosis Lesions via a Bivariate Spatial GLM with Spatially Varying Coefficients

• Activity Number: 107
• Type: Invited
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Timothy Duane Johnson*+
• Companies: University of Michigan
• Keywords: Bayesian Analysis ; Imaging ; Prediction ; Spatially Varying Coefficients ; Multiple Sclerosis ; High-Dimensional Data

Abstract

Multiple Sclerosis (MS) is a disease in which the myelin sheaths surrounding the axons of the brain and spinal cord are damaged. This leads to dymelination and scarring of the white matter tracks in the brain. There are four main subtypes of MS: 1) relapsing/remitting, 2) secondary progressive, 3) progressive relapsing and 4) primary progressive. Treatment options are subtype dependent and therefore clinicians are interested in using MRI lesion location to predict MS subtype in patients. We use a model based approach to prediction, which can lead to substantial increases in prediction accuracy versus non-model based approaches. The data are MS lesions from T1-weighted and T2-weighted MRI images. A bivariate spatial GLM is employed to model the probability of lesion location as a function of subject specific covariates, including disease subtype, with spatially varying coefficients. Including an apriori probability of MS subtype, we can invert the model to make valid predictions about a specific person's subtype given their imaging data and covariates.