A Comparison of Metaheuristic Techniques for Optimizing Gamma Knife Inverse Treatment Planning
J Langas*, John Jay Science and Engineering AcademySU-E-T-664 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To determine if the Firefly Algorithm (Yang) or a genetic algorithm would provide viable alternatives to the simulated annealing technique used in Leksell GammaPlan 10 to create potential tumor treatment plans for use with Leksell Gamma Knife.
Methods: To determine the viability of each algorithm, separate programs were created to generate treatment plans. The first utilized an annealing algorithm based on Leksell GammaPlan, and was intended to provide an accurate baseline for comparing the performance of the other algorithms. The second utilized an algorithm (Yang) that mimicked the behavior of fireflies in order to search all possible treatment plans to find an acceptable solution. The third mimicked natural selection and breeding among a population of potential solutions to generate acceptable solutions. Each of the programs were then run through a variety of tests to determine its effectiveness, after which each generated treatment was evaluated with the fitness function detailed by Elekta.
Results: Of the compared algorithms, the Firefly Algorithm consistently generated the highest quality solutions when all three were run for a set number of generations. The Firefly Algorithm required the least amount of time to generate an acceptable solution for cubic tumors, but the genetic algorithm required the least amount of time for spherical tumors. The genetic algorithm produced the highest quality solutions in tests with shorter maximum run times, while the Firefly Algorithm produced the highest quality solutions in tests with longer maximum run times.
Conclusion: Both the Firefly algorithm and genetic algorithm outperformed annealing in almost all test cases. This supports their potential use as alternatives to the existing method used in Leksell GammaPlan. Future studies with a wider variety of more extensive tests (i.e. more tumor varieties, longer tests, etc.) could provide additional support for their potential.