ESTRO 2025 - Abstract Book

S2870

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2025

3270

Digital Poster Automated Knowledge-Based (KB) plans generation using a benchmark model for left whole breast radiotherapy: multi-Institute application Elisabetta Cagni 1 , Alessia Tudda 2 , Roberta Castriconi 2 , Valeria Trojani 1 , Rita Buono 2,3 , Cintia De Almeida Ribeiro 4 , Francesca Dusi 5 , Marco Fusella 5 , Federica Guida 6 , Aldo Mazzilli 7 , Lorenzo Placidi 4 , Giulia Rambaldi Guidasci 8 , Claudio Fiorino 2 1 Medical Physics Unit, Department of Advanced Technology, Azienda USL-IRCCS di Reggio Emilia, Reggio Emilia, Italy. 2 Medical Physics Dept., IRCCS San Raffaele Scientific Institute, Milano, Italy. 3 Physics Department, Università Statale di Milano, Milano, Italy. 4 Department of Diagnostic Imaging, Oncological Radiotherapy and Hematology, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Roma, Italy. 5 UO Radioterapia, Policlinico Albano Terme, Padua, Italy. 6 Medical Physics Department, Veneto Institute of Oncology IOV–IRCCS, Padua, Italy. 7 Medical Physics Dept, University Hospital of Parma AOUP, Parma, Italy. 8 Radiotherapy Unit, Ospedale Isola Tiberina, Roma, Italy Purpose/Objective: Within the MIKAPOCo multicentric study, a knowledge-based plan prediction Benchmark Model (MBM) was previously developed, using Varian RapidPlan system (v.16.1), and validated for left-sided postoperative breast radiotherapy, using data from 393 patients treated with tangential fields (TF) across five institutions. The aim of current project (named AMBRA, Automated MIKAPOCo Breast Radiotherapy plans: optimization and evaluation Assessment) is to evaluate if MBM-generated plans are clinically acceptable and comparable to manual clinical plans across multiple institutions. Material/Methods: The MBM was shared with 6 institutions (internal, previously joining MBM training: INST3, INST5, INST9; external, new Institutions: INST1, INST6, INST11). An initial optimization template was created using only predicted model-generated endpoints, with endpoint priorities independently tuned at each institution on a test set of local patients. Local refinements were then combined to produce a unique, robustly tuned, template. Fully automated AMBRA plans were generated using the VITAT technique [1] for patients previously optimized with TF manual plans (CLIN). Performances of AMBRA and CLIN plans of 80 patients (N=20 for INST1 and INST5; N=10 for INST3, INST6, INST9, INST11), were compared focusing on heart and ipsilateral lung DMEAN, and PTV_D95% and PTV_D2%. Results: AMBRA plans generated dose distributions comparable to CLIN plans (Figure1). Across all validation cases (N=80), mean ipsilateral OAR doses were similar between AMBRA (DMEAN left lung =5,40±1,57 Gy and DMEAN heart =2,64±1,33 Gy) and CLIN (DMEAN left lung =5,54±1,72 Gy and DMEAN heart =2,30± 1,27 Gy), with no significant differences (p>0.05) across institutions. Significant differences (p<0.05) were only noted in heart DVH for INST1 (external) and, of minor entity, for lung DVH for INST9 (internal). Concerning INST1, differences were expected, due to the sparing of left arterial descending coronary in their CLIN plans, not included in MBM. PTV_D95% and PTV_D2% values (Figure 2) showed similar values over all institutions, with few small, although significant, differences only in external institutions.