Determining Multimodal Biomarkers for Neurodegenerative Diseases

• Activity Number: 65
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: SSC
• Author(s): DuBois Bowman*+ and Wenqiong Xue
• Companies: Emory University and Emory University
• Keywords: Imaging ; Bayesian model ; Spatial model ; Prediction ; Parkinson's disease

Abstract

Despite the considerable progress in understanding the biology of Parkinson's disease (PD), reliable biomarkers are still lacking for early stage detection of PD. Advances in biotechnology have led to studies that collect large-scale, multimodal imaging data sets. Leveraging data from magnetic resonance imaging (MRI), resting-state functional MRI, and diffusion tensor imaging (DTI), we present a statistical framework to identify multimodal PD biomarkers. Specifically, we consider a Bayesian hierarchical model to predict PD using imaging data reflecting both functional and structural properties of the brain. We consider a two-level brain parcellation and assume different spatial correlation structures between voxels within a subregion, between subregions (within a region), and between different regions. We perform Markov Chain Monte Carlo estimation via Gibbs sampler. Our model yields both whole-brain and voxel-level prediction, and we apply leave-one-out cross validation to assess prediction accuracy. We apply our model to our PD data and conduct simulation studies to evaluate its performance.