Optimization Of Beam Angles In Imrt Using Ant Colony Optimization Algorithm
L. Yongjie,D. Yao,J. Yao
DOI: https://doi.org/10.1016/j.ijrobp.2005.07.838
2005-01-01
Abstract:Purpose/Objective: Intensity-modulated radiotherapy (IMRT) is being increasingly used for treatment of malignant cancer. Beam angle optimization (BAO) is an important problem in IMRT, and it has been demonstrated that beam configuration may have significant influence on the quality of an IMRT treatment. Clinically, however, the selection of beam angles is generally based on the experience of the treatment planner, which is generally difficult due to the non-intuitive dose contribution of each beam to the target and critical structures. The tremendous computation hampers those BAO techniques to be widely used in IMRT practice. The purpose of this work is to introduce a novel emerging meta-heuristic algorithm named ant colony optimization (ACO) to improve the efficiency of BAO in IMRT planning. Materials/Methods: In essence, ACO is a class of evolutionary approach where several generations of artificial ants search for good solutions. The idea behind ACO is to use artificial ants to keep track of promising areas of the search space by laying trails of pheromone on the visited path. This pheromone information is used to guide the search, as a heuristic for choosing values to be assigned to variables. In the context of BAO, every ant of a generation builds up a beam configuration step by step going through several probabilistic decisions. To map the BAO problem to ACO, we designed a wheel-shaped graph, which center resembles the isocenter of the plan, and spokes resemble the discretized beam directions to be selected. Starting from the center, each ant selects directions step by step according to a probabilistic state transition rule, which gives the probability of an angle to be selected. After selecting a direction, the ant returns to the center and prepare to select next one, until a beam configuration is constructed. After selecting a path by an ant, the corresponding beam intensity maps are optimized using a conjugate gradient (CG) method, and the ant deposits some pheromone on the spokes (i.e. angles) in the path, the amount of which is a function of the objective function value. When an ant chooses paths at the isocenter, the spokes with a larger amount of pheromone have a greater probability to be chosen. One of the particular important issues is the selection of heuristic function, which is problem-dependant and important for evolution process and used to calculate the heuristic value in the probabilistic state transition rule. In this study, we introduce the beam′s-eye-view dosimetrics (BEVD) score proposed by Pugachev and Xing (2001) as our heuristic function. The heuristic values for each beam are previously calculated before the optimization. Results: The proposed algorithm was applied to a clinical prostate tumor case, in which the target (prostate) is surrounded by four critical structures: bladder, rectum, left and right femur head. Seven coplanar photon beams are used for the treatment. The ACO-based algorithm ran ten times, and an averaged time of about 25 min was taken to find the optimal beams with angles 10, 60, 110, 155, 200, 250 and 300 degrees. The plan was also optimized ten times using a genetic algorithm (GA), and the same optimal beam angles were found. However, the averaged computation time was about 35 min, about 40% longer than that of ACO. Furthermore, compared with the optimization results with manually selected equispaced beams, the plan with optimal beams improved the dose distribution meaningfully. Conclusions: The ACO-based algorithm presented here results in optimal beams that improve dose distribution within a reasonable computation time. Also, the presented algorithm appears more efficient than GA-based approach according to the limited preliminary results.