ESTRO 2020 Abstract Book

S812 ESTRO 2020

select the clinically optimum balance of trade-offs for a selected calibration patient. Automated planning of new patients is then possible with the employment of algorithms designed to ensure trade-off balancing for a novel patients is consistent with that selected during calibration. The purpose of this work was to quantitatively assess whether or not a single patient’s calibration parameters are sufficient to generated automated plans for new patients of the same treatment site. Material and Methods 19 randomly chosen prostate seminal vesicles (PSV) patients previously treated at Velindre Cancer Centre were selected for this study. A previously calibrated EdgeVcc automated planning protocol, which included seven MCO parameters, was used as a base protocol. For each patient, an experienced operator used EdgeVcc’s MCO functionality to create a patient specific gold standard plan (GS). To simplify the MCO problem, four of the criteria were held constant across all plans (bowel dose volume, PTV homogeneity, and max dose objectives to bladder and rectum). The three criteria navigated (average dose to rectum, average dose to bladder, and PTV conformality) were considered to have the most clinically interesting trade-off relationships. For each patient, two additional plans were generated automatically with EdgeVcc using the following calibration methodologies: 1) calibration based on MCO navigation of a single randomly selected patient which was independent of the original set of 19 patient (SP), 2) calibration based on the average results of the GS MCO navigation using a leave-one-out validation methodology (Av). Results A summary of key dosimetric parameters for the three sets of plans can be found in Table 1. Compared to GS, SP yielded statistically significant increases in rectum V24.3 Gy and mean dose, and decreases in bladder V40.5Gy and mean dose. No differences found were deemed clinically significant.

[1] Physics and Imaging in Radiation Oncology, 10, 41-48. doi: 10.1016/j.phro.2019.04.005 PO-1505 Knowledge based treatment planning and validation of VMAT for Cervical Cancer. J. Swamidas 1 , S. Pradhan 1 , S. Panda 1 , S. Chopra 1 , A. Mangaj 1 , U. Mahantshetty 2 1 Advanced Centre for Treatment Research and Education in Cancer ACTREC- Tata Memoral Centre, Radiation Oncology, Mumbai, India ; 2 Tata Memorial Hospital- Tata Memorial Centre, Radiation Oncology, Mumbai, India Purpose or Objective To report our experience of knowledge based planning and its validation for VMAT for Cervical Cancer. Material and Methods 30 patients previously treated as part of Image guided intensity modulated E xternal beam radio-chemotherapy and M RI based adaptive BRA chytherapy in locally advanced CE rvical cancer (EMBRACE-II) protocol were used to build a model using knowledge-based planning module (RapidPlan v13.5.35, Eclipse v13.5,Varian Medical Systems). Plan geometry consists of two coplanar arcs of 360 ˚ , collimator angle of 5 ˚ or 355 ˚ , and field size 16x35cm 2 . Dose prescription to PTV pelvis was 45Gy/25fractions. 10 patients from the same clinical trial were randomly chosen to validate this model. A total of three plans were generated: Clinical plan (CP) made by an experienced planner manually, automatic Rapid Plan(RP), based on the model from single optimization without any manual tweaking, and Hybrid Plan(HP), combination of CP and AP. HP plans were generated from the RP, followed by minimal manual tweaking of organ priorities and dose constraints during optimization interactively by the planner. Dose volume parameters for PTV, ITV, and OARs were statistically analysed using paired t test and Wilcoxon signed test Results Out of the three plans HP was found to be superior in terms of organ sparing (bowel V 30Gy ,V 40Gy , and V 15Gy p = 0.001, bladder V 30Gy (%) and V 40Gy (%) p = 0.001) and conformality (V 43Gy /PTV vol and V 36Gy /PTV vol p=0.001) while maintaining target coverage. Table 1, lists the detailed DVH parameters for all the plans and various DVH parameters. However, no significant difference was observed between CP and RP. This confirms that the model is robust, efficient and equivalent to CP, however, a further improvement was observed when the model parameters were tweaked by the planner during optimization. This explains that the model is the average plan model, while, the manual tweaking takes into account the patient specific changes.

Conclusion This study shows evidence that a single patient is sufficient for calibrating an automated solution for PSV using MCO. Although there may be some statistical differences between GS and SP, those differences are considered clinically negligible. Whilst averaging calibrations over multiple patients (Av) may offer an improvement to SP, it is considered this improvement is not large enough to warrant the investment of obtaining more data.