Multi-Objective Genetic Optimization of a 10 Mev Small-Animal Irradiator for X-Ray Flash
Abstract
Purpose
Radiation therapy optimization problems frequently involve multiple competing objectives, requiring multi-objective optimization to navigate trade-offs. This work used a Non-dominated Sorting Genetic Algorithm (NSGA-II) optimizer to identify Pareto-optimal geometries for a simple 10 MeV X-ray FLASH small-animal beamline.
Methods
A 10 MeV electron source with an X-ray conversion target, electron filter and divergent collimator was simulated in TOPAS Monte Carlo to generate a 2×2 cm² field at source-to-surface distance (SSD). Dose was scored in a 6×6 cm² water phantom at SSD with 0.5 mm binning. Five geometric parameters were optimized: tungsten target thickness, SSD, copper electron filter thickness, collimator distance from the target, and collimator opening size. Multi-objective optimization was performed using TopasMOO, an open-source Python package extending TopasOpt, to implement NSGA-II. NSGA-II's population-based approach enables robust optimization with noisy Monte Carlo results, allowing lower-statistic simulations for faster evaluations. Three objectives were simultaneously evaluated: maximizing dose rate at dmax, minimizing the surface-to-dmax dose ratio to reduce electron contamination, and minimizing deviation from desired field size.
Results
The optimization was run for 50 generations with a population size of 20, for a total of 1,000 evaluations. The optimization process required 544 core-hours. The final Pareto front contained 25 solutions with a final hypervolume size of 6.98×103. Design dose rates ranged from 52.9 to 125 Gy/mAs. Surface dose percentages ranged from 9.1 to 30.7%. Field size deviation ranged from 0 to 19.4 mm. Solution design parameters spanned: target thickness [0.33–0.59 mm], filter thickness [3.75–4.98 mm], SSD [20.01–28.67 cm], aperture size [0.47–0.94 cm], and collimator position [5.01–6.55 cm].
Conclusion
Integrating NSGA-II with TOPAS via TopasMOO enabled systematic exploration of design trade-offs in a small-animal irradiator, yielding 25 Pareto-optimal solutions. Identifying a family of Pareto-optimal geometries enables informed decision-making, allowing the design to be tailored to address specific experimental and design constraints.