Detecting a Local Change in Brain Structures by the Matrix Normal Model

• Activity Number: 655
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Michelle Liou*+ and Wei-Chen Cheng and Aleksandr A Simak and Philip E. Cheng
• Companies: Academia Sinica and Academia Sinica and National Taiwan University and Academia Sinica
• Keywords: brain structures ; Alzheimer's disease ; matrix normal distribution ; factor analysis ; dependent data ; LONI database

Abstract

Deformation- or tensor-based morphometry has been widely applied to measuring local distortion in MR brain images. Distortion parameters over time such as the curvature or jacobian determinant can be summarized in different anatomical regions using morphometric techniques. Statistical analyses of longitudinal and group comparisons can be carried out based on the distortion parameters. In this study, we will show that factor analsis based on the matrix normal model by assuming dependence within subjects and independence between subjects can partition longitudinal changes in cortical structures into important dimensions which successfully identify normal subjects from patients with either mild cognitive impairment or Alzheimer's disease. The parameters in the factor analysis model are estimated by a two-stage algorithm with reasonable prior assumptions on the unknown parameters. The use of the factor analysis model will be demonstrated by analyzing the MR brain images supported by the LONI database.