Incorporation of uncertainties from setup errors and organ motion in 3-D dose calculations.

Abstract

External beam radiotherapy is one of several methods used in the treatment of intraheptatic cancer. Treatment plans are generated based on a static CT Scan of a patient, and treatment is delivered in multiple fractions. The dose distribution calculated using a single CT study does not include uncertainties from daily setup or organ motion. Monte Carlo-based methods have been proposed to address this concern, but these methods are cumbersome and inefficient. Methods based on a convolution of the static dose distribution with functions representing the setup uncertainties in three dimensions (for incorporating setup uncertainties) have been reported for prostate and nasopharynx treatments. In this thesis, we apply similar convolution methods to incorporate setup uncertainties and organ motion due to breathing for treatment of the liver. We develop a model that describes the motion of the liver due to breathing and we show that convolution-based techniques are sufficient to predict the dose distribution the patient receives for any number of fractions. For setup uncertainties, we show that convolution-based methods calculate the distribution of average dose values given an infinite number of fractions. However, because the patient receives a finite (not infinite) number of fractions, the actual dose distribution realized by the patient may differ from the distribution of average dose values. We extend the original model and derive an approach to calculate the distribution of standard deviation values at any point in the patient. The differences between a finite-fractioned treatment and the results of convolution-based calculations are characterized by the standard deviation at each point using the Central Limit Theorem. Further, we apply our convolution-based techniques to incorporate the effects of both setup uncertainties and organ motion due to breathing in a single distribution of dose values. Our results are confirmed via comparison to Monte-Carlo-based direct simulations of patient treatment and are generated on a time scale suitable for clinical applications. Finally, we address potential applications and future research associated with this thesis.