Room: Stars at Night Ballroom 1
Purpose: We propose a novel optimization approach, for robust optimization in proton therapy, which reduces the active number of optimization scenarios to significantly reduce the optimization time.
Methods: Instead of evaluating all uncertainty scenarios, a dynamic pool of candidate-worst scenarios is set up and used in the optimization process. This pool contains a reduced number of scenarios compared to conventional worst-case robust optimization. At each iteration, only the scenarios present in the pool are evaluated by computing their objective functions. The pool is then dynamically updated, selecting only the scenarios that are frequently a worst case (i.e. maximum objective function value) and discarding those that are not addressed by the optimizer (replacing them with yet unconsidered scenarios). Discarded scenarios may be introduced back to the pool and might become active again depending on the convergence towards a particular solution. Treatment plans for dynamic and conventional worst-case robust optimization were calculated for a lung cancer patient using the open-source robust optimizer MIROpt, with 60 Gy as prescription dose. Conventional optimization (63 scenarios: combinations of +/-5 mm setup errors in the three directions, +/-3% range error and maximum inhale/exhale breathing phases) was used as reference, with the MidP phase the nominal scenario. Plan robustness was evaluated with the MCsquare dose engine by recalculating the dose distribution on a set of 100 randomly sampled error scenarios.
Results: The proposed method used only 10 active optimization scenarios, reducing the plan optimization time by 84%, whilst achieving a similar plan quality w.r.t. the conventional worst-case optimization (using 63 scenarios): average target coverage difference (D95) of 0.4 Gy and average lung V20 difference of 0.4% between both methods.
Conclusion: The proposed dynamic scenario-selection optimization method achieved a very promising 84% optimization time reduction for the patient considered in this study, without compromising plan quality or robustness.