Three Essays on Radiotherapy Treatment Planning Optimization

Abstract

Radiation therapy is one of the most common and effective methods of treating cancer. There are two main types: external and internal. External to the patient, a linear accelerator aims beams of radiation toward the patient; internal to the patient, radioactive sources are placed temporarily or permanently at the treatment site to deposit dose locally. Both methods of treatment can deliver complex dose distributions to a patient. The radiation damages both tumorous tissue and nearby healthy organs; treatment planning optimization determines how to deliver a dose distribution that maximizes tumor kill while sparing nearby healthy organs as much as possible. This thesis studies three treatment planning problems: the first two are in the context of external radiation therapy and the third is in the context of internal radiation therapy.

Conventional planning is based on only the physical geometry of the patient anatomy. In chapter II, we propose two models that incorporate (additional) liver function information for planning liver cancer treatment to preserve as much post-treatment liver function as possible and compare this to a conventional approach that ignores liver function information.

Conventional plans assume the patient geometry does not change between the time of patient imaging and later treatments. Although the patient geometry can be updated at treatment for plan adaptation, current practice may lead to plans that result in significantly worse quality than originally intended due to its myopic nature. In chapter III, we propose a model that produces a plan that caters to each potential patient geometry while considering both day-of and cumulative impact.

In high-dose rate brachytherapy, the patient undergoes anesthesia due to the need to implant catheters for radiation source placement before planning. Consideration of multiple conflicting criteria in treatment planning results in challenging optimization problems. Current commercial systems require iterative guess-and-checking of optimization input parameters to make trade-offs among criteria, but a plan must be finalized quickly to minimize anesthesia administration. In chapter IV, we develop a practical optimization engine that generates a trade-off surface and feeds into a graphical user interface that provides the clinician more control to make trade-offs without trial-and-error optimizations.