ESTRO 38 Abstract book

S496 ESTRO 38

A cohort of 20 patients was chosen from the institutional database to validate the two models, HN-RP-1 and HN-RP- 2, to assess and compare their plan quality. Results The HN-RP-2 model presented an improved mathematical process relative to HN-RP-1, showing increased R 2 parameters: 0.634/0.541 (parotids), 0.866/0.599 (oral cavity), 0.756/0.534 (larynx), 0.731/0.204 (spinal cord) for HN-RP-2/HN-RP-1, respectively. Also the regression plots presented general improvements. Regarding the plan quality, D 1% to the serial organs, spinal cord and brain stem, increased with HN-RP-2 model of 3-4%, not significant. All the parallel organs showed, on the contrary, an improvement with HN-RP-2 relative to HN- RP-1 model, in general around 1% (significant for some structures), with the larynx showing the highest mean dose reduction of 5% (p<0.01). The doses averaged over the 20 validation plans were: 26.3±0.8/26.5±0.8 Gy (parotid mean dose), 40.9±2.3/41.2±2.3 Gy (oral cavity mean), 27.5±2.1/29.0±2.0 Gy (larynx mean), 28.7±0.7/28.0±0.7 Gy (spinal cord D 1% ) for HN-RP-2/HN- RP-1. Conclusion A second model based on a previous model could improve the plan quality for the parallel organs at risk, and possibly the robustness of the model itself, considering the better mathematical parameters of HN-RP-2 relative to HN-RP-1. PO-0926 A novel approach to automatic planning: robust templates for lung VMAT SBRT L. Marrazzo 1 , C. Arilli 1 , S. Calusi 2 , M. Casati 1 , C. Talamonti 1,3 , P. Bonomo 4 , L. Livi 3,4 , S. Pallotta 1,3 1 Azienda Ospedaliera Universitaria Careggi, Medical Radiation Physics, Firenze, Italy ; 2 Istituto Fiorentino di Cura e Assistenza, Medical Physics and Radiation Oncology, Firenze, Italy ; 3 University of Florence, Department of Biomedical- Experimental and Clinical Sciences, Firenze, Italy ; 4 Azienda Ospedaliera Universitaria Careggi, Radiation Oncology, Firenze, Italy Purpose or Objective To develop and validate a planning class solution for VMAT SBRT of lung lesions, that achieved target and organs-at- risk (OAR) doses within established constraints using the multicriterial optimization (MCO) of Monaco treatment planning system (Elekta-CMS Software, MO, USA). Material and Methods The template, containing a list of planning objectives, was first established on a population of 10 lung SBRT patients planned for 55Gy/5fr (peripheral lesions, near to or partly overlapped with the thoracic wall) and refined with a stepwise process. In order to account for anatomical changes between patients, so to achieve personalized results, stage 1 (ideal fluence optimization) was conducted giving priority to OARs and using the MCO. MCO further pushes OARs dose, stopping just before compromising target coverage. To improve gradient and conformity a ring structure around the PTV was set in the list of objectives. Stage 2 (segmentation) was conducted giving priority to PTV coverage. The template was then applied (with no manual intervention) on 20 further patients and the resulting plan was compared with the manual clinical plan. Dose distributions were compared in terms of dosimetric plan parameters (dose to PTV, conformity and gradient index and dose to OARs). Dosimetric verification was performed and evaluated in terms of γ passing rate and point dose measurements, in order to assess that the planned dose distribution could be reliably delivered. Statistical significance of differences between automatic and manual plans was evaluated using paired two-sided Wilcoxon signed-rank test. Results No statistically significant differences in PTV coverage (p=0.6) and PTV maximum dose (p=0.2) were observed,

Table 1 lists average dose rates, monitor units (MU), and beam-on times for the clinical and FFF plans for all patient categories. On average, the increase in dose rate and MU are 64.8% and 41.8% respectively for the FFF plans in comparison to the clinical plans. The decrease in beam-on time is 39.2% on average for the FFF plans. Conclusion The standard AP-width based FFF TBI plans yield a clinically acceptable dose homogeneity that is similar to our current clinical plans while reducing both planning and beam-on time. This technique therefore offers the possibility of decreasing the overall treatment time, and improving patient experience. PO-0925 On the ability of a knowledge based planning process to improve itself A. Fogliata 1 , G. Reggiori 1 , C. Franzese 1 , D. Franceschini 1 , S. Tomatis 1 , M. Scorsetti 2 , L. Cozzi 2 1 Humanitas Research Hospital, Department of Radiation Oncology, Milan-Rozzano, Italy ; 2 Humanitas Research Hospital and Humanitas University, Department of Radiation Oncology and Faculty of Biomedical Science, Milan-Rozzano, Italy Purpose or Objective A knowledge based planning (KBP) engine generates a model from geometric and dosimetric data of a cohort of selected plans, in order to obtain mathematical parameters to estimate the possible DVHs of critical structures, according to the strategy adopted for the plans in the model. Those DVHs are used as optimization objectives for IMRT or VMAT planning for any new patient. As a machine learning process, the KBP is generally able to improve the plan quality. The aim of the present work is to determine if the KBP process RapidPlan (Varian, Palo Alto, CA) can improve itself, i.e. if plans generated by RapidPlan (improved in quality with respect to the clinical plans) can be the source to generate a new RapidPlan model to further improve the plan quality. Material and Methods Clinical VMAT plans from 83 patients presenting advanced head and neck cancer were selected to build a RapidPlan model. All the plans delivered prescription doses of 54.45 and 69.96 Gy to the elective and boost target volumes, respectively, in 33 fractions with simultaneous integrated boost. The optimization process was manual, using the PRO optimzer. The 83 clinical plans were used to generate a RapidPlan model, named HN-RP-1, using a defined and ad-hoc tuned list of objectives in the model. The HN-RP-1 model was then used to generate new plans for the same 83 patients as above, using the same plan geometry. Those new RapidPlan plans were the input of a new RapidPlan model, named HN-RP-2, having the same list of objectives as HN-RP-1.