Estimation of Parameter Covariance for a Penalized Likelihood Approach to Estimation of Kinetic Models in PET (Positron Emission Tomography)

• Activity Number: 535
• Type: Contributed
• Date/Time: Thursday, August 10, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Michelle Byrtek*+ and Finbarr O'Sullivan
• Companies: Western Washington University and University College Cork
• Keywords: penalized likelihood ; components of variance ; mean square error ; Bayes ; kinetic parameters

Abstract

Extraction of kinetic parameters from PET image data involves the use of non-linear weighted least squares regression. Current estimation techniques often have poor mean square error properties. Penalized likelihood techniques have been shown to improve parameter estimation accuracy in this context. The effectiveness of penalized likelihood, however, depends on the appropriate specification of a penalty term. A Bayesian formulation for the penalty, based on a variance component model using a rudimentarily estimated subject-to-subject parameter covariance matrix, was shown to improve parameter loss by as much as 19%. In this study, well-established methods from mixed effects modeling are used to obtain a more robust estimate of subject-to-subject variability. The error in estimation of this covariance matrix, and corresponding error in parameter estimation is also evaluated.