Abstract Book

S272

ESTRO 37

Material and Methods Datasets from 20 HNC patients, consisting of a p lanning CT and weekly control CTs (cCT), were analyzed. Intensity-modulated proton therapy (IMPT) plans with minimax robust optimization were calculated, accounting 3 mm and 3.5% for setup and range uncertainty, respectively. Prescribed doses to the low- and high-risk clinical target volume (CTV) were 57 and 70 Gy(RBE), respectively, delivered in 33 fractions. The organs at risk (OAR) spinal cord, brainstem, parotid glands, larynx, pharyngeal constrictor and esophageal inlet muscle were considered for plan optimization. Weekly cumulative doses, taking the anatomical variations of the cCTs into account, were compared with the nominal plan. When a reduction in target coverage and/or increased dose to OARs was detected, a plan adaptation was performed on the cCT where the dose degradation was observed. Furthermore, for these patients an additional robust plan was calculated, taking also anatomical changes from the first two cCTs into account in the robust optimization. It was evaluated if a subsequent plan adaptation would still be necessary. Results Nominal plans fulfilled the clinical specifications of D 98% ≥ 95% of the prescribed dose to the CTVs (range 96.58- 98.81% for low-risk CTV and 96.83-98.76% for high-risk CTV). During the treatment course, anatomical changes lead to reduced weekly cumulative D 98% values in five patients (25%; minimum 90.17% for low-risk CTV and 89.19% for high-risk CTV). Doses in OARs remained below the clinical constrains during the treatment course. One treatment adaptation was performed for each of these five patients, which allowed a target coverage improvement (range 97.68-99.72% for low-risk CTV and 95.89-98.46% for high-risk CTV). Total cumulative doses including adaptation were acceptable (range 96.67- 98.37% for low-risk CTV and 95.11-97.39% for high-risk CTV, see Figure 1a). The results for the more sophisticated robust plan, considering the first two cCTs, were diverse: whereas plan adaption became obsolete in one patient, the total cumulative dose would, without adaptation, still have been below clinical constraints in another (Figure 1b and 2).

Conclusion In a substantial number of patients, robust optimization only is not sufficient to account for anatomical changes occurring during the treatment course, resulting in severe target coverage degradation. Assessment of the cumulative weekly doses allowed detection of target coverage loss. The importance of frequent in-treatment imaging is underlined. OC-0518 Multi-criterial MLC segmentation with column generation, applied to robotic SBRT B.W.K. Schipaanboord 1 , S. Breedveld 1 , L. Rossi 1 , M. Keijzer 2 , B.J.M. Heijmen 1 1 Erasmus MC Cancer Institute, Radiotherapy, Rotterdam, The Netherlands 2 University of Technology, Department of Applied Mathematics, Delft, The Netherlands Purpose or Objective Fluence map optimization (FMO) for IMRT, based on convex cost functions, guarantees global optimality of the solution. On the other hand, plan quality loss may occur in the subsequent MLC segmentation, used to generate a deliverable plan. In this study, a novel multi- criterial MLC segmentation technique was developed, which minimizes plan quality loss with a focus on high priority clinical objectives. The proposed technique was evaluated with prostate SBRT dose distributions and investigations include dependencies on optimization grid size. Material and Methods In the developed segmentation technique, deliverable MLC segments are generated with column generation, which iteratively identifies promising apertures. If the segmented solution converges to a dose distribution that does not comply with one or more dose-volume criteria, as used in the clinical protocol, an iterative multi- criterial segmentation procedure is started; by putting emphasis on critical voxels, while considering the clinical priorities, an attempt is made to better comply with objectives. For 20 prostate SBRT patients, treatment plans with 25 non-coplanar beams were generated automatically using FMO with an in-house developed algorithm for multi- criterial optimization. FMO plans were generated with a commonly used 5x5 mm 2 beamlet resolution and with a finer 1x4 mm 2 resolution, corresponding to the MLC segmentation resolution. Segmentation was then performed, both with and without multi-criterial fine- tuning, using a 1x4 mm 2 MLC resolution. All generated plans were scaled to a PTV coverage of 95% and evaluated on 6 clinical dose-volume criteria (PTV V 38Gy >95%, Rectum D 1cc <32.3 Gy, Bladder D 1cc <38 Gy, Urethra D 5% <45 Gy, D 10% <42 Gy, D 50% <40 Gy). The success