A comparison of mixed integer programming and fast simulated annealing for optimizing beam weights in radiation therapy.

Two competing methods for assigning intensities to radiation treatment beams were tested. One method was derived from mixed integer programming and the other was based on simulated annealing. The methods faced a common objective and identical constraints. The goal was to maximize the minimum tumor dose while keeping the dose in required fractions of normal organ volumes below a threshold for damage. The minimum tumor doses of the two methods were compared when all the dose-volume constraints were satisfied. A mixed integer linear program gave a minimum tumor dose that was at least 1.8 Gy higher than that given by simulated annealing in 7 of 19 trials. The difference was > or = 5.4 Gy in 4 of 19 trials. In no case was the mixed integer solution one fraction size (1.8 Gy) worse than that of simulated annealing. The better solution provided by the mixed integer program allows tumor dose to increase without violating the dose-volume limits of normal tissues.