|
Activity Number:
|
132
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #304238 |
|
Title:
|
Bayesian Normal Mixture Modeling: A Case Study in CT Scanning
|
|
Author(s):
|
Clair L. Alston*+ and Kerrie Mengersen
|
|
Companies:
|
Queensland University of Technology and Queensland University of Technology
|
|
Address:
|
GPO Box 2434, Brisbane, International, 4001, Australia
|
|
Keywords:
|
Bayesian mixture models ; prior distributions ; hyperparameters ; computing
|
|
Abstract:
|
The implementation of Bayesian mixture models in data analysis is not always straightforward. In this paper we use a Bayesian mixture model to estimate the area of each of three tissue types of interest, fat, muscle and bone present in individual CT scan slices of a live sheep. We then use the Cavalieri principle to estimate the proportion of the carcass attributable to each tissue. The approach is validated by analysis of experimental sheep carcasses. Choice of values for the hyper-parameters of the prior distribution on component parameters can be of issue and consequences of different schemes are examined, including the usual approach of setting hyperparameters for component means to the overall data mean for all components. We also illustrate a technique which allows the MCMC chains to begin with starting values that lead to successful computation when data sets are large.
|
