Incorporating prior knowledge into beam orientaton optimization in IMRT

Purpose: Selection of beam configuration in currently available intensity-modulated radiotherapy (IMRT) treatment planning systems is still based on trial-and-error search. Computer beam orientation optimization has the potential to improve the situation, but its practical implementation is hindered by the excessive computing time associated with the calculation. The purpose of this work is to provide an effective means to speed up the beam orientation optimization by incorporating a priori geometric and dosimetric knowledge of the system and to demonstrate the utility of the new algorithm for beam placement in IMRT. Methods and Materials: Beam orientation optimization was performed in two steps. First, the quality of each possible beam orientation was evaluated using beam's-eye-view dosimetrics (BEVD) developed in our previous study. A simulated annealing algorithm was then employed to search for the optimal set of beam orientations, taking into account the BEVD scores of different incident beam directions. During the calculation, sampling of gantry angles was weighted according to the BEVD score computed before the optimization. A beam direction with a higher BEVD score had a higher probability of being included in the trial configuration, and vice versa. The inclusion of the BEVD weighting in the stochastic beam angle sampling process made it possible to avoid spending valuable computing time unnecessarily at "bad" beam angles. An iterative inverse treatment planning algorithm was used for beam intensity profile optimization during the optimization process. The BEVD-guided beam orientation optimization was applied to an IMRT treatment of paraspinal tumor. The advantage of the new optimization algorithm was demonstrated by comparing the calculation with the conventional scheme without the BEVD weighting in the beam sampling. Results: The BEVD tool provided useful guidance for the selection of the potentially good directions for the beams to incident and was used to guide the search for the optimal beam configuration. The BEVD-guided sampling improved both optimization speed and convergence of the calculation. A comparison of several five-field IMRT treatment plans obtained with and without BEVD guidance indicated that the computational efficiency was increased by a factor of ∼ 10. Conclusion: Incorporation of BEVD information allows for development of a more robust tool for beam orientation optimization in IMRT planning. It enables us to more effectively use the angular degree of freedom in IMRT without paying the excessive computing overhead and brings us one step closer to the goal of automated selection of beam orientations in a clinical environment.