Model-Based Clustering for Differentiating Lesion Tissue Types in Gadolinium fMRI Scans

• Activity Number: 405
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Roseline Bilina Falafala*+
• Companies: Cornell University
• Keywords: Model-Based clustering ; Spatial dependence ; Multiple sclerosis

Abstract

Multiple Sclerosis patients have brain lesions, believed to be caused by a breakdown in the blood-brain barrier (BBB). One way to examine the permeability of the BBB is via injection of Gadolinium (GA), which is visible on a functional Magnetic Resonance Imaging scan. Scans are taken, observing the diffusion of GA into the brain tissue in locations where the BBB is breached. Statistically, we want to identify, characterize, and cluster the lesions of a subject based on the GA diffusion over space and time. For example, GA concentration spikes quickly in young lesions, while older lesions and healthy tissue show very little absorption. We are also interested in the dynamics of GA absorption into different types of lesions. We develop statistical methods for these goals. First, we reduce the data associated with each voxel from a time series to a low-dimensional summary, using regularized functional principal components analysis. Then, we take a spatial model-based clustering approach to cluster the voxels into a small number of different tissue types, taking into account their spatial dependence. Computational methods from image analysis are applied and extended for this purpose