S1848
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
simultaneously to PTV1 using VMAT (two arcs, 6 MV FF) on a TrueBeam STx. The Adaptbox workflow generates sCT from CBCT using a deep learning model. Organs at risk (OARs) and clinical target volumes (CTVs) are then delineated using AI segmentation and elastic or rigid propagation, respectively. Finally, dose calculation is performed on both planning CT (pCT) and sCT using a proprietary collapsed cone algorithm.For contour evaluation, physician-defined target volumes and OARs were delineated on the first-fraction CBCT for each patient. These physician-defined contours were compared with AI-derived contours on the sCT generated by Adaptbox using the Dice Similarity Coefficient (DICE). Dosimetric accuracy was assessed 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: Bladder contours achieved the highest accuracy (DICE = 0.91 ± 0.05), while rectum reached 0.77 ± 0.08, mainly due to guideline variability. Rigid propagation yielded the best prostate accuracy (DICE = 0.91 ± 0.07), whereas elastic propagation improved SV delineation (DICE = 0.69 ± 0.14). Target DVH differences were within 2%, with the largest deviations in high-dose regions, notably rectum D0.1% (2.13 ± 0.70 Gy). Mean gamma pass rates were 99.99 ± 0.03% (3%/3 mm) and 99.81 ± 0.45% (2%/2 mm).
Conclusion: Both AP methods successfully generated high-quality, clinically acceptable plans for breast cancer radiotherapy with minimal human intervention. These findings demonstrate that automated end-to-end VMAT planning is feasible, can improve plan consistency, efficiency, and quality, supporting integration into routine clinical workflows. References: [1] M. Hussein et al., “Automation in IMRT planning – recent innovations,” Br. J. Radiol., vol. 91, p. 20180270, 2018.[2] P. Meyer et al., “Automation in RT planning: clinical use and trends,” Cancer/Radiother., vol. 25, pp. 617–622, 2021.[3] M. S. Thomsen et al., “Dose constraints for whole breast RT in DBCG HYPO trial,” Clin. Transl. Radiat. Oncol., vol. 28, pp. 118–123, 2021.[4] A. Munshi et al., “Dose fall-off with VMAT and 3D-CRT,” Cancer Radiother., vol. 23, pp. 138–146, 2019.[5] L. L. Puckett et al., “Consensus quality measures and dose constraints for breast cancer,” Pract. Radiat. Oncol., vol. 13, pp. 217–230, 2023. Keywords: auto-planning, breast cancer Evaluation of the contour generation and dose calculation algorithm of a AI-powered offline adaptive software for prostate cancer Jessica Prunaretty, David Azria, Pascal Fenoglietto Radiotherapy, Institut du Cancer de Montpellier, Montpellier, France Purpose/Objective: The Adaptbox software (v2.3.2, Therapanacea) is an AI- powered platform integrating CBCT-based synthetic CT (sCT) generation, auto-contouring, and accelerated dose calculation for offline adaptive workflows. While sCT image quality has been previously assessed [1], this study aimed to validate AI-generated contours and the dose engine for prostate cancer. Material/Methods: Twenty prostate cancer patients from the RCMI-GI clinical trial were retrospectively analyzed. Specific contouring definitions were applied: seminal vesicles (SVs) were delineated in their proximal 2 cm from the prostate attachment, and the rectum was defined 2 cm above and below the prostate–SV junction. PTV2 encompassed the prostate, and PTV1 included PTV2 plus an anisotropic margin around the SV. Treatments delivered 80 Gy in 40 fractions to PTV2 and 56 Gy Digital Poster 1632
Conclusion: Adaptbox demonstrates robust performance for contour generation and dose calculation in prostate cancer. However, rectum and SV delineations remain sensitive to guideline variability, and minor deviations persist in high-dose regions. References: [1] T. Roque et al. PO-2351 Can artificial intelligence bring cone beam CT acquisitions to planning CT quality?, Radiotherapy and Oncology, Volume 182, Supplement 1, 2023, S2115-S2116 Keywords: offline adaptive workflow, adaptbox, deep learning
Made with FlippingBook - Share PDF online