ESTRO 2021 Abstract Book

S1581

ESTRO 2021

calculated and compared to plans made by experts. Results

Overall, plan quality generated by DL improved when manual delineations of tissue/applicator are replaced with automatic DL delineations. The time for the DL image countering are fully meet the clinical needs and are significantly reduced compared to manual work by experts (P<0.003). The delineation accuracy is improved by 2% compared to the plan made by physicists. Mean D2cc OARs dose in the plans using DL was equal to or smaller than the plans were delineated manually. Mean bladder, rectum, sigmoid, and small bowel D2cc OARs were reduced by 0.8%, 1.0%, 0.5%, 0.2% (respectively), and CTV dose remained the same compared to the plan made by experts. Conclusion DL is a promising technique for target delineations and applicator reconstructions. The next step is to include a fast DL dose calculations model against Monte Carlo simulations for treatment planning automation. PO-1855 Protocol compliance of two Knowledge Based models in two Geo-Ethnic populations for cervical cancer J. Swamidas 1 , M. Assenholt 2 , M. Serban 3 , J. Jain 4 , S. Chopra 4 , V. Hande 4 , S. Pradhan 5 , K. Tanderup 2 1 ACTREC, Tata Memorial Centre, Radiation Oncoclogy, Mumbai, India; 2 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark; 3 McGill University Health Centre, Montreal, , Department of Medical Physics, Montreal, Canada; 4 ACTREC, Tata Memorial Centre, Department of Radiation Oncology, Mumbai, India; 5 All India Institute of Medical Sciences, Department of Medical Physics, Delhi, India Purpose or Objective To compare and validate the performance of knowledge-based planning (KBP) models developed in two geo-ethnic patient populations for image guided VMAT in locally advanced cervical cancer. Materials and Methods Two KBP models were created, at the Tata Memorial Centre – India (Rapid plan v13.5.23, Varian Medical Systems) and Aarhus University Hospital – Denmark (Rapidplan, v16.1)using patients treated at their respective departments and was validated through comparison with the manually optimised clinical plans. Clinical plans were created by expert dose planners who had undergone accreditation according to EMBRACE II quality criteria. The KBP model was developed using training (n=125) based on clinical treatment plans based on dose constraints of the EMBRACE-II protocol. For this analysis, the KBP models were shared between the two centres and prospective planning was performed. Validation was carried out in a total of 20 patient data sets: 10 node positive (N+) and negative (N-) at each institution. Three plans for each data set, clinical plan, Tata KBP, and Aarhus KBP, were created, thereby generating a total of 120 plans. Plans based on KBP were generated in a single run optimization without any manual intervention. Quantitative comparison of dosimetric indices was performed between clinical vs. Tata KBP, Clinical vs. Aarhus KBP and Tata KBP vs. Aarhus KBP, using Wilcoxon sign rank sum test. p value <0.05 was considered statistically significant. Results Both models have excellent compliance to the protocol specific dose constraints. Between the two KBP, Aarhus model resulted in high conformity for N+ Tata patients (CI43Gy 1.0 vs 1.09 p=0.0051, CI36Gy 1.6 vs 1.4 p=0.0051), improved OAR sparing, while under-dosing targets marginally, whereas Tata models resulted in comparable plans for Aarhus patients, both in terms of OAR sparing and conformality, however, marginal overdose of target coverage was observed. The details are as follows: For Tata N- patients, Aarhus model resulted in significant bowel (V40Gy p= 0.009, V30Gy = 0.005, V15Gy =0.005), bladder (V40Gy = 0.007, V30Gy = 0.006) and rectum sparing (V40Gy = 0.006), while under dosing the target marginally i.e. PTV45 (V42.75Gy 96.8±1.1 vs 94±1 p=0.005, CTV-N D98 55.4±0.2 vs 54.5±0.5 p=0.0051) as compared to Tata model (Table 1). This under-dose in target coverage may overcome with plan normalization. For Aarhus N- patients, Tata model resulted in comparable plans both for target coverage (PTV45V42.75Gy 96.1±0.7 vs. 97.2±1.2 p=0.96, CTV-N D98 55.9±1.5 vs 55.8±0.1 p=0.07) and OAR sparing (bowel V40Gy , V30Gy, V15Gy, bladder V40Gy , V30Gy and rectum V40Gy, V30Gy p>0.36) as compared to Aarhus clinical plans.. For N+ patients, Tata model resulted in comparable target coverage, with marginally less OAR sparing (bowel V40Gy p= 0.01, V30Gy = 0.007, V15Gy =0.44, bladder V40Gy and V30Gy = 0.028 and rectum V40Gy = 0.025, V30Gy = 0.48). Tata model resulted in good conformality in N- patients (CI V43 0.99 vs 1.00 p = 0.2, V36 1.46 vs 1.42 p=0.02), however for N+ patients, it was marginally less (CI V43 1.04 vs 1.075 p = 0.0051, V36 1.53 vs 1.6 p=0.0051) as compared to clinical plans. Conclusion Protocol compliance was good in both clinical and KBP plans, across the institutions. Geo-ethnic validation of these KBP models suggests that with appropriate training, KBP models result in plans with similar quality as compared to clinical plans carried out by expert dose planners. Given that unexperienced dose planners may produce plans of less quality than experienced planners, KBP has potential to improve quality and trial specific protocol compliance. Purpose or Objective Treatment planning for a brain tumor can be challenging due to the close proximity of numerous critical structures. Plan optimization is usually based on generic dose constraints and manual adjustments. While this may yield an acceptable plan, the full potential is not easily exploited. We investigated the benefit of a semi-automatic process, in which the optimization is driven by patient-specific objectives from a DVH estimation model. Materials and Methods A brain DVH estimation model was configured using the RapidPlan tool within the Eclipse v15.6 environment (Varian, Palo Alto, USA). Dose distribution data was extracted from 100 treatment plans in the department’s database. The resulting model is applicable to the majority of curative brain cases. When applied to a new case, it can generate line objectives for 14 critical structures. The study concentrated on a separate set of 20 patients previously irradiated for glioblastoma multiforme (GBM): 59.4 or 60 Gy in 33 or 30 fractions. The treatment plans employed 6 MV photons in a dual arc setup on a Varian TrueBeam. Up to 14 critical structures had been contoured (minimum 11), with at least one structure partially overlapping the PTV. Plans had PO-1856 Minimizing dose to brain structures by knowledge-based planning S. Koch 1 , C. Stevelink 1 1 Medisch Spectrum Twente, Radiotherapy, Enschede, The Netherlands