S1830
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
fully automated multicriterial plan generation: Erasmus-iCycle for IMRT head and neck cancer,» Medical Physics, 2013A. B. J. J. M. A. L. &. C. T. C. Y. Babier, «Knowledge-based automated planning with three-dimensional generative adversarial networks» Medical Physics, 2020J. W. J. C. Z. &. H. W. Fan, « Automatic treatment planning based on three- dimensional dose distribution predicted from deep learning technique,» 2019A. Shulman, Lignes directrices de planification VMAT - Programme de formation RCC IMRT2, Programme de formation RCC IMRT2L. RaySearch, Environnements de script RayStation 11A pour l'apprentissage automatique, RaySearch Laboratories, 2021 Keywords: Automation, Python, Optimization, RayStation Automated Evaluation of Organ Sparing in Prostate SBRT Using Dose-Gradient Metrics Geert De Kerf 1,2 , Michaël Claessens 1 , Dirk Verellen 1,3 1 Radiotherapy, Iridium Netwerk, Antwerp, Belgium. 2 Faculty of Medicine and Health Sciences, University of Antwerp, Antwerp, Belgium. 3 Centre for Oncological Research, University of Antwerp, Antwerp, Belgium Purpose/Objective: Traditional dose-volume histogram (DVH) metrics used in radiotherapy plan evaluation are limited by volume dependency and lack spatial context [1,2]. This study investigates Dose Gradient Curves (DGCs) [3] as a robust, volume-independent alternative for assessing organ sparing in prostate stereotactic body radiation therapy (SBRT), with a particular focus on integration into automated workflows. Material/Methods: Digital Poster 1340 Treatment plans of 154 prostate cancer patients were retrospectively analyzed. A benchmark set (BS) of 20 high-quality plans was established, and average DVH (aDVH) and DGC (aDGC) curves were derived for the bladder and anorectum. Plan quality of the remaining 134 plans was assessed using aDVH, aDGC, and expert-reviewed ground truth. A Δ AUC-based classifier was developed to automatically detect suboptimal organ sparing. The robustness of benchmark set size was evaluated by comparing subsets of five plans with extreme organ volumes. Results: DGC-based evaluation showed superior accuracy and precision compared to DVH-based methods, with reduced volume dependency. For the bladder, DGC analysis achieved 99% accuracy and precision, versus 87% and 94% for DVH. For the anorectum, DGC yielded 97% accuracy and 100% precision. The Δ AUC classifier achieved F1 scores of 97.1% (bladder) and
Volumetric Modulated Arc Therapy (VMAT) planning, enabling direct comparison between automated and manually generated plans. Material/Methods: This study represents the first evaluation of the Shulman method in combination with an automated radiotherapy planning workflow, benchmarked against conventional manual planning. The Shulman optimization method - which consists of creating the required optimization structures to facilitate the optimization process - was first implemented and scripted in a single case study using a dedicated scripting tool available in the RayStation treatment planning system, and subsequently applied to the remaining selected cases.Python scripts developed through the RayStation API automated key planning tasks, including the generation of optimization structures (e.g., PRVs and auxiliary volumes) and the definition of VMAT objectives, thereby streamlining the VMAT planning process.Data extraction and preprocessing were performed in Python (v3.8), while statistical analyses (paired t-test and Wilcoxon test) were carried out using statgraphics. A retrospective analysis of 30 head and neck (H&N) clinical cases (including larynx and cavum) was conducted to compare automated and manual plans in term of planning time, dosimetric parameters, dose coverage, heterogeneity index and organ-at-risk sparing. Results: The outcomes demonstrate improved dosimetric precision with the automated approach, showing a target volume coverage (D95%) of 98.7%± 1.2% compared to 97.9%± 1.8% for manual planning ( ρ = 0.002). Dose heterogeneity was also enhanced, with a heterogeneity index of 1.05± 0.03 for automated plans versus 1.08± 0.05 for manual ones. Notably, planning time was significantly reduced, by nearly 70%, from 78± 15.7 minutes (manual) to 22.5± 5.3 minutes (automated) ( ρ <0.001). The automated workflow also decreased inter-operator variability, promoting greater standardization. Organ-at-risk sparing was comparable or slightly improved in the automated plans. Conclusion: Automated dosimetry in RayStation, implemented through the Shulman method and dedicated scripting for head and neck cases, demonstrated clear advantages over manual planning, including higher dosimetric accuracy, greater plan consistency, and optimized workflow. A substantial time gain of nearly 70% was achieved. These results confirm that automation can significantly enhance efficiency while maintaining or improving plan quality, supporting its integration into clinical practice to deliver faster, standardized, and high quality treatment planning. References: P. W. J. D. M. L. P. B. S. F. D. &. H. B. J. M. Voet, «Toward
Made with FlippingBook - Share PDF online