Doctoral Dissertations
Date of Award
5-2002
Degree Type
Dissertation
Degree Name
Doctor of Philosophy
Major
Nuclear Engineering
Major Professor
Peter G. Groer
Committee Members
Frank M. Guess, Laurence F. Miller, Lawrence W. Townsend
Abstract
Bayesian statistical methods are applied to several problems in internal and external radiation dosimetry. In each case, Bayesian methods produce results that are clear, coherent (i.e.-following the rules of probability calculus), and free of "ad-hockery" often associated with alternative frequentist statistical methods.
Research in the area of internal radiation dosimetry focuses on compartmental modeling of ⁴⁵Ca biokinetics. Several approaches are taken to predict compartmental activity at fixed times. One approach is numerical calculation of the expected value and variance of compartmental activity using quasi-Monte Carlo integration. This approach considers the uncertainty of parameter estimates in the calculation, an improvement on the usual practice of estimating compartmental activity with point estimates of transfer coefficients. The second approach focuses on the calculation of posterior probability distributions for the compartmental activities. Predictive densities for observations of compartmental activity and posterior distributions for expected values of compartmental activity are calculated for a 2-compartment caternary model with simulated data and for the ⁴⁵Ca model. The predictive density for an observation is shown to be computable in all cases, but less desirable than the posterior distribution of the expected value due to greater uncertainty. The more desirable posterior distribution for the expected value of compartmental activity cannot be calculated except for extremely simple cases. Instead, these posterior distributions for the expected value can be approximated using point estimates of the expectation and variance and an assumption of normality. The final topic considered in the area of internal radiation dosimetry is compartmental moderling with missing observations. The general approach of Box et al. (1970) for multiresponse nonlinear fitting with missing observations is made specific to compartmental modeling. The method is demonstrated using the ⁴⁵Ca biokinetic data by omitting one and two observations. In both cases, posterior distributions are derived for transfer coefficients as well as for the missing observations.
Research in the area of external radiation dosimetry focuses on dose estimation with chromosome aberrations. The topics considered include: chromosome dosimetry for a single radiation type, for doses received during a criticality accident, with uncertain calibration doses, and with overdispersed calibration data. The work of Groer et al. (1987) for the chromosome dosimetry following exposure to a single radiation type to extended to several new calibration data sets. The results reinforce the practicality and usefulness of the calibrative density as a tool for chromosome dosimetry. A new method is developed and demonstrated for chromosome dosimetry following criticality accidents that permits estimation of the total dose as well as the individual dose contributions from fission-spectrum neutrons and gamma rays with an imprecise neutron to gamma dose ratio. This new approach is an improvement over the frequentist methods currently used for estimating the dose received from a criticality excursion, which rely on an assumption of precisely-known neutron to gamma dose ratios. Another novel method is developed for chromosome dosimetry with uncertain calibration doses. It is clear that all doses in calibration data sets are estimates, and as such they are subject to uncertainty. This method produces dose estimates that consider the inherent uncertainty of the calibration doses. The last topic considered is model selection/identification. The concerns are the assumption of Poisson statistics for high linear energy transfer radiations, and the assumption of a linear dose response for low doses of neutrons. Bayes Factors and Partial Bayes Factors are used to compare the probabilities of the competing models. The results show that the Polya model is preferred over the Poisson model for chromosome dosimetry with high linear energy transfer radiations, and that an assumption of linear dose response is appropriate for the induction of dicentric chromosome aberrations by low doses of neutrons.
Recommended Citation
Brame, Ryan Scott, "Bayesian methods for internal and external radiation dosimetry. " PhD diss., University of Tennessee, 2002.
https://trace.tennessee.edu/utk_graddiss/7066