S1850
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
Purpose/Objective: This study aimed to validate all components of an automated synthetic CT (sCT)-based CBCT dose calculation software for breast cancer, focusing on image accuracy, contour quality, and dosimetric consistency. Material/Methods: This retrospective analysis included 20 breast cancer patients previously treated on a TrueBeam linac after surgery (right/left, lumpectomy/mastectomy). For each patient, five extended CBCTs acquired during treatment were randomly selected. Prescriptions were 63.2 Gy to the tumor bed, 52.2 Gy to the breast and internal mammary chain (IMC), and 49.3 Gy to the clavicular lymph nodes (CLN) in 29 fractions. A 7 mm CTV–PTV margin was applied. Treatments used VMAT with 2–3 partial arcs and 6 MV FF energy.The Adaptbox software (v2.3.2, Therapanacea) is an AI-driven platform that generates sCT from CBCT using a deep learning model. Organs at risk (OARs) and clinical target volumes (CTVs) are delineated using AI segmentation and elastic propagation, respectively. Dose calculation is performed on both the planning CT (pCT) and the sCT using a proprietary collapsed cone algorithm.sCT accuracy was assessed by comparing Hounsfield units (HUs) between sCT and pCT within OARs. Physician-defined target volumes and OARs were delineated on each CBCT. AI-generated contours (breasts/chest wall, lungs, heart) and elastically propagated contours (IMC, CLN) on the sCT were compared to physician-defined structures using the Dice Similarity Coefficient (DICE). Dosimetric accuracy was evaluated by comparing Adaptbox-calculated doses with Eclipse AAA (v15.6) on the pCT using identical VMAT plans. Dose–volume histogram (DVH) metrics and 3D global gamma analyses (2%/2 mm and 3%/3 mm, 10% threshold) were used for evaluation. Results: Median HU values were comparable between sCT and CT. For the breast ROI, HUs ranged from − 115 to 73 HU on CT and − 106 to 54 HU on sCT, with the largest discrepancies in patients with implants. AI-generated OAR contours showed high accuracy (DICE > 0.9). For elastically propagated targets, mean DICE values were 0.87 ± 0.07 (CLN) and 0.70 ± 0.14 (IMC). Target DVH differences were within 2%. The highest mean relative deviation for OARs (13.1%, 0.6 Gy) was observed in the contralateral breast. Mean gamma pass rates were 98.8 ± 2.4% (3%/3 mm) and 94.4 ± 5.1% (2%/2 mm), with discrepancies mainly near the skin.
Conclusion: Adaptbox enables accurate, fully automated CBCT- based sCT dose calculation for breast cancer, demonstrating robust performance in image fidelity, contouring precision, and dose consistency. Keywords: Adaptbox, breast, deep learning
Digital Poster 1652
Evaluation of single-arc RapidArc Dynamic vs standard two-arc VMAT in prostate radiotherapy Paula Navarro-Palomas 1 , Esther Bailon 1 , Maria Salgado 2 , Gerard Meca 1 , Elías Gomis 1 , Joel Mases 1 , Sara Moreno 1 , Artur Latorre-Musoll 1 , Jordi Saez 1 1 Servei d’Oncologia Radioteràpica, Hospital Clínic de Barcelona, Barcelona, Spain. 2 Servei d’Oncologia Radioteràpica, Institut del Càncer i Malalties de la Sang (ICAMS), Barcelona, Spain Purpose/Objective: In prostate radiotherapy, treatment time is a critical factor for maintaining reproducible bladder and rectal filling conditions, particularly in patients who struggle to maintain a comfortably full bladder. At our centre, prostate treatments are routinely delivered with two full arcs, resulting in a beam-on time of 2 minutes. RapidArc Dynamic (RAD) optimizes collimator rotation, potentially enhancing dose conformity even with a single full arc.This exploratory study aimed to evaluate whether single-arc RAD plans, which are capable of reducing beam-on time by half, could maintain target coverage and organ-at-risk (OAR) sparing compared with standard two-arc VMAT. Material/Methods: Nineteen localized prostate cancer patients, including a few post-prostatectomy cases, were retrospectively re-planned in Eclipse v18.1 (Varian Medical Systems) using the AAA algorithm. Two plans were generated per patient: (i) Our standard practice (STD), for reference, consisting of two 6MV full arc plans with fixed collimators at 30º/330º and (ii) The RAD plan, consisting of a single 6MV-FFF full arc plan with optimized collimator rotation (without stop angles). The same optimization objectives, which have been validated to provide clinically acceptable solutions, were applied to all plans for each patient, with no
Made with FlippingBook - Share PDF online