SU‐E‐T‐614: An Optimization Algorithm for Beam Angle, Beam Weight and Wedge Angle in Forward Treatment Planning of External‐Beam Radiotherapy Based on an Integer‐Representation Adaptive Mutation Probability Genetic Algorithm

Purpose: To present the development of an optimization algorithm for beam angle, beam weight and wedge angle in forward treatment planning of external‐beam radiotherapy using a genetic algorithm (GA). Methods: An adaptive mutation probability (AMP) integer‐representation GA was applied for this optimization process in the MATLAB programming environment. The code allows various user‐defined limits and starting points as well as comprehensive searches. We used an integer representation for all variables in the chromosomes pool for encoding the steps, because wedge angles often take discrete values (i.e. 0, 15, 30, 45 and 60 degrees). To improve performance, we designed a dynamic mutation probability assignment code in each generation so that the algorithm automatically adapts the mutation probability using the standard deviation of fitness values of the population at each generation. If the fitness diversity is great enough, a low mutation probability will be applied, and if the fitness values have low diversity, a high mutation probability will be applied. A dose calculation program using correction‐based techniques and the CTimages of the patient was also written within the same software. The GA code was tested using a standard test function both with AMP and with a constant mutation probability across all GA generations. Convergence of beam angle, beam weight and wedge angle was also investigated. Results: With the AMP technique, the GA maintained the population diversity in the chromosomes pool to avoid premature convergence into a local minimum. Test results showed that the algorithm with AMP (run time of 5 min for a simple standard test function) is more robust compared to the conventional method. Conclusions: This algorithm is a feasible and promising tool for optimization of treatment planning parameters with an acceptable computation time. Testing the algorithm against experienced treatment planners will be performed next.