Matrix Decomposition for Modeling Multiple Sclerosis Lesion Development Processes

• Activity Number: 173 - Recent Advances on Neuroimaging Analysis
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Menghan Hu* and Ani Eloyan and Russell T. Shinohara and Ciprian Crainiceanu
• Companies: Brown University and Brown University and University of Pennsylvania Perelman School of Medicine and Johns Hopkins University
• Keywords: Functional Principal Component Analysis; Imaging Statistics; Longitudinal Data Analysis; Magnetic Resonance Imaging

Abstract

This project is motivated by a longitudinal MRI study of MS patients, where the objective was to study the progression of the MS disease by studying the intensity profiles of lesions. Understanding the dynamic behavior of the underlying trajectories is potentially useful for determining the stage of disease at the time of observation and predicting outcomes at future visits. We aim to develop a statistical model that characterizes the lesion development and find parameter estimates describing the intensity trajectories in MS. We model the longitudinal MRI data as discrete observations from a functional process over time and apply functional data analytic methods in our study. The model also accounts for the hierarchical structure of the data where the variability of the functional data can be decomposed into three levels. We group the MRI data by their spatial characteristics and align the trajectories by time. To further reduce the computational complexity, we use principal component bases for the functional processes. A three-way nested model is defined and an estimation procedure using functional PCA is developed and applied to MS imaging data with level-specific variability.