ESTRO 2023 - Abstract Book

S222

Saturday 13 May

ESTRO 2023

Conclusion The proton liver phantom is a particularly challenging phantom with a low pass rate (the lowest among all IROC phantoms). Range uncertainty, motion management, and underdosing are the main culprits of failures of the proton liver phantom. Clinically, careful attention to mutli-target liver proton therapy is needed to ensure optimal patient care. OC-0287 Fast and fully-automated robust multi-criterial IMPT planning via sparsity induced spot selection W. Kong 1 , M. Oud 1 , S. Habraken 2,1 , M. Huiskes 3 , E. Astreinidou 3 , C. Rasch 3,4 , B. Heijmen 1 , S. Breedveld 1 1 Erasmus MC Cancer Institute, University Medical Center Rotterdam, Department of Radiotherapy, Rotterdam, The Netherlands; 2 HollandPTC, Department of medical physics and informatics, Delft, The Netherlands; 3 Leiden University Medical Centre, Department of Radiation Oncology, Leiden, The Netherlands; 4 HollandPTC, Department of Radiation Oncology, Delft, The Netherlands Purpose or Objective In intensity modulated proton therapy (IMPT) planning, adequate spot selection is critical. A high-resolution spot distribution may enhance plan quality, but is associated with long computation time and has challenges in adhering to minimum monitor units (MU) per spot and the maximum number of beam energies used, the latter being proportional to delivery time. Therefore, we propose a novel, fast method for automated robust multi-criterial IMPT planning that utilises sparsity-induced spot selection (SISS) prior to automated multi-criteria optimisation (MCO). For various treatment sites, SISS was compared to iterative pencil-beam resampling (IPBR) (Van de Water, 2015). No manual fine-tuning of plans was performed. Materials and Methods The proposed algorithm is based on a wish-list that configures the goals and trade-offs in the MCO to enable high-quality automated multi-criterial treatment planning. It consists of three steps: 1) fast spot selection, 2) MCO of spot intensities, and 3) spot intensity refinement to adhere to physical delivery constraints. In step 1, 20.000 spot candidates are first distributed uniformly in the target volume. To enable fast spot selection, the wish-list is used to automatically define an unconstrained large-scale optimisation problem which is solved fast using a fast limited memory solver. Of the solution, the 20% spots with the highest MU are selected for step 2. In step 2, our in-house system for fully automated MCO treatment planning is used on the spots selected in step 1. In step 3, spots not complying with the minimum MU requirement are removed at once, and a re-optimisation is performed using the reference point method (RPM) to ensure adherence to clinical and physical constraints for the remaining spots. All generated SISS and IPBR plans were robustly optimised using 19 scenarios and constrained on the clinically imposed minimum and maximum MU. SISS and IPBR were compared for 40 head and neck (4 beam directions), 21 prostate (2 beam directions) and 19 cervical cancer (4 beam directions) patients. The clinically in use beam geometry was applied. Results For a proper perspective, SISS and IPBR were not only compared for computation time, but also for plan quality, number of applied spots and number of applied energy layers. For all three tumor sites, SISS produced similar CTV coverage compared to IPBR and slightly reduced OAR dose. Fig.1 shows average DVHs for the 40 head and neck cases. Table 1 shows that for all three tumor sites, the number of spots and the number of energy layers (and thereby delivery time) were similar for SISS and IPBR. Plan optimisation times are also presented in Table 1. On average SISS reduced the total optimisation by a factor of 10.1, i.e. from 362 min to 36 min.