Genetic and geometric optimization of three-dimensional radiation therapy treatment planning

The thesis of this report is that potentially useful treatment beams can be chosen based on geometric heuristics and that a genetic algorithm (GA) can be constructed to find an optimal combination of beams based on a formal objective function. The paper describes the basic principles of a GA and the particular implementation developed. The code represents each plan in the population as two paired lists comprised of beam identifiers and relative weights. Reproduction operators, which mimic sexual reproduction with crossover, mutation, cloning, spontaneous generation, and death, manipulate the lists to grow optimal plans. The necessary gene pool is created by software modules which generate beams, distribute calculation points, obtain clinical constraints, add wedges, and calculate doses. The code has been tested on a set of artificial patients and on four clinical cases: prostate, pancreas, esophagus, and glomus. All demonstrated consistent results, indicating that the code is a reliable optimizer. Additional experiments compared the results for a full set of open beams to the geometrically selected set and the GA code with simulated annealing. Geometric selection of beam directions did not significantly compromise optimization quality. Compared to simulated annealing, the genetic algorithm was equally able to optimize the objective function, and calculations suggest it may be the faster method when the number of beams to be considered exceeds approximately 70.