A1641
Title: Functional nonlinear mixed-effects modeling of convolved data when direct observations are sparse
Authors: Baoyi Shi - Columbia University (United States)
Granville Matheson - Columbia University (United States)
Todd Ogden - Columbia University (United States) [presenting]
Abstract: In a common PET imaging framework, functional data observed for each of n subjects and k regions are modeled as a convolution between two unknown functions, one of which is common across all regions for each subject. When direct observations from each of these subject-specific common functions are also available, these functions may each be estimated and plugged in prior to model fitting across subjects. However, if such direct data are very sparsely observed, it becomes necessary to model both types of functions simultaneously across subjects and regions. Nonparametric techniques are discussed for model fitting in this situation based on shape constraints that are imposed on the subject-specific common functions. These modeling strategies are illustrated through application to PET neuroreceptor mapping data.
