Dosimetry in brachytherapy : application of the Monte Carlo method to single source dosimetry and use of correlated sampling for accelerated dose calculations

• Three issues related to brachytherapy dosimetry are addressed in this thesis. (1) The prospect of using Compton scattering to measure energy spectra from a high dose-rate (HDR) 192Ir source is discussed and a Compton spectrometer evaluated. Promising results are demonstrated but further fine-tuning is needed to resolve problems with background subtraction. (2) Absorbed doses around an interstitial brachytherapy 125I-source (the Symmetra™ seed) are calculated in the TG-43 formalism using Monte Carlo (MC) techniques. A review of the literature on current measurements and MC simulations indicate systematic differences of 3-6% for the model 6702 seed and points to the need for renewed attention to the dosimetry of low-energy photons and careful estimations of uncertainty. The problems concern among other points: the energy response of LiF dosemeters including LET (Linear Energy Transfer) dependence and the influence of high atomic number doping materials, the atomic compositions of Solid Water and other phantom materials used in MC simulations, the proper MC simulation of the new NIST (National Institute of Standards and Technology) air-kerma strength calibration standard based on the wide-angle free air chamber, and the benchmarking of MC codes. (3) In order to promote development of MC based dose planning, correlated sampling as a means of speeding-up MC calculations is evaluated. In this pilot study, simplified physics is used. Only the photoelectric effect (disregarding the emission of characteristic x-rays) and Compton scattering (Klein-Nishina) are considered. Analogue (ANL) and expected value track-length (ETL) estimations are compared. Efficiency gains (relative to uncorrelated ETL estimations) are calculated for simplified geometries with a point isotropic source and a cylindrical heterogeneity of air, AI and W in a water medium. Efficiency gains of 103 - 104 were obtained for modest perturbations (heterogeneity correction factors HCF [0.8 <HCF < 1.2]). At large perturbations [HCF ≈ 0.4-0.5], in volume elements (voxels) behind the heterogeneity, correlated sampling can be even less efficient than uncorrelated sampling. With correlated ETL estimation, an overall gain in efficiency was, however, achieved and relative standard deviations less than 2% were obtained in 90% of the voxels for an 1251- source and 1-hour computing-time. Uncorrelated ETL estimation was 10-100 times more efficient than uncorrelated ANL estimation. Although promising, correlated sampling should be combined with some other variance reduction technique to reduce the variance everywhere in the volume. Analysis of the uncertainties of estimated efficiency gains shows that the use of the Fisher F distribution to derive their confidence intervals is suspect.

Nyckelord

MEDICINE

MEDICIN

Publikations- och innehållstyp

vet (ämneskategori)

dok (ämneskategori)