ESTRO 2024 - Abstract Book

S4582

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTR0 2024

55

Digital Poster

Fast many-scenario IMPT planning through constrained dose-mimicking

Franziska Knuth 1 , Michelle Oud 1 , Wens Kong 1 , Hazem Nomer 1 , Joep van Genderingen 1 , Linda Rossi 1 , Steven Habraken 2,3 , Ben Heijmen 1 , Sebastiaan Breedveld 1 1 Erasmus MC Cancer Institute, University Medical Center Rotterdam, Department of Radiotherapy, Rotterdam, Netherlands. 2 HollandPTC, Medical Physics & Informatics, Delft, Netherlands. 3 Leiden University Medical Center, Radiation Oncolocy, Leiden, Netherlands

Purpose/Objective:

Intensity modulated proton therapy (IMPT) is susceptible to patient setup variations and proton range uncertainties. To account for this, treatment planning is performed using a set of scenarios, i.e. patient shifts and range variations, to ensure adequate CTV coverage and adherence to maximum doses for critical OARs. A disadvantage is that the optimisation problem increases with each included scenario, resulting in long optimisation times. In clinical practice, 21 scenarios are often used for planning, with more scenarios desirable for more accurate modeling of the impact of variations and uncertainties. For fast robust optimisation with a large set of scenarios, we propose a two-step approach: 1) automated multi-criteria treatment planning using a small set of scenarios, 2) fast constrained dose mimicking to make the plan robust for a much larger set of scenarios.

Material/Methods:

For 15 oropharynx patients, Erasmus-iCycle [1] was used to automatically generate two robust IMPT plans, one for 13 scenarios (iCycle13) and the other for 53 scenarios (iCycle53). A third plan (Mimic53) was generated using the proposed dose mimicking approach with iCycle13 as input, aiming at robustness for the 53 scenarios. Setup- and range robustness settings of 3 mm/3% were used. The scenarios were distributed on a unit sphere following Korevaar et. al. [2], where the 13 scenarios were a subset of the 53. Doses of 70 GyE and 54.25 GyE were prescribed to the primary and elective CTV, respectively. All three plan generations were constrained for adherence to minimum and maximum MU constraints and maximum number of energy layers, following the requirements in our clinic.

In dose mimicking, a quadratic penalty function was minimised aiming at close reproduction of all voxel doses of iCycle13 in all 13 scenarios, however, using clinical constraints for CTV voxel-wise minimum dose for all 53 scenarios.

For validation of dose mimicking, Mimic53 plans were compared to ground truth iCycle53 plans and to the input iCycle13 plans. The plans were compared using 53 rotated scenarios, i.e. not overlapping with the 53 scenarios as used for planning (with the exception of the nominal scenario).

All plan generations were performed on a 2xIntel Xeon E2690. Optimisation times for iCycle13, iCycle53 and Mimic53 were recorded, excluding time for dose computations, which was the same for all three.

Results: