Objective-function-guided automated VMAT planning reduces OAR dose, low-dose exposure, and inter-planner variability in breast radiotherapy

www.nature.com/scientificreports OPEN Objective-function-guided automated VMAT planning reduces OAR dose, low-dose exposure, and inter-planner variability in breast radiotherapy Hannes Rennau & Guido Hildebrandt This study presents and evaluates an automated volumetric modulated arc therapy (VMAT) planning framework for breast cancer based on objective function value (OFV)–guided optimization. The primary objective is to systematically improve organ-at-risk sparing through automated and reproducible optimization of planning constraints while maintaining clinically acceptable target coverage. An OFV-guided optimization workflow was empirically developed using a separate sensitivity dataset and subsequently evaluated in 20 clinical breast cancer patients (13 left-sided, 7 right-sided). The automated Python-based framework iteratively adapts MaxEUD constraints during optimization until dose metrics converge, without manual intervention. Automatically generated plans were compared to clinically delivered VMAT plans using target coverage, dose–volume metrics, and monitor units as a surrogate for delivery efficiency. The automated approach consistently achieved significant reductions in mean organ doses and low-dose volumes (e.g., V2Gy , V5Gy ) while preserving PTV coverage. Mean heart dose decreased from 3.09 ± 1.06 Gy to 2.21 ± 0.69 Gy for left-sided cases and from 1.88 ± 1.06 Gy to 1.17 ± 0.25 Gy for right-sided cases (p < 0.001). Significant dose reductions were also observed for the ipsilateral lung, contralateral lung, and contralateral breast, accompanied by a 16.9 % reduction in monitor units (p < 0.001) and reduced inter-patient variability. In contrast, no statistically significant difference was observed for lung V20Gy . In conclusion, OFVguided VMAT optimization enables reproducible and systematic improvement of organ-at-risk sparing in breast radiotherapy. By reducing organ doses, low-dose burden, monitor units, and inter-patient variability without compromising target coverage, the proposed framework could provide a robust and standardized baseline for clinical VMAT planning and a consistent foundation for future data-driven and machine-learning–based optimization approaches. For several decades, the tangential 3D-conformal radiation therapy (3DCRT) technique has been the standard treatment approach for early-stage breast cancer. Several key factors have contributed to this technique remaining the standard of care, including: (i) safe dose delivery using only two angles, (ii) a wide safety margin to accommodate breathing and breast swelling, (iii) a steep dose gradient towards the ipsilateral lung, heart, and contralateral breast, and (iv) significantly fewer monitor units (MU) and a reduced low-dose burden compared to multi-field intensity-modulated radiotherapy (IMRT) or volumetric-modulated arc therapy (VMAT) techniques. However, the introduction of VMAT has significantly improved the sparing of nearby organs and treatment efficiency1. Despite these advantages, fluence-modulated techniques like VMAT increase the low-dose burden to the patient’s body. This may not always justify the elevated secondary cancer risk, especially for younger women who are more vulnerable. Studies2–4 have argued that radiation-induced cancer risk can be reduced by using classical 3DCRT, which results in less accelerator head scatter. According to5, most secondary tumors occur within 5 cm of the target volume at doses below 6 Gy. Therefore, maximal protection of nearby organs must be prioritized when choosing VMAT over 3DCRT. A recent summary by4 even suggested the superiority of 3DCRT Department of Radiation Oncology, University Hospital Rostock, Rostock 18057, Germany. Scientific Reports | (2026) 16:15875 | https://doi.org/10.1038/s41598-026-52616-2 email: 1 for early-stage cases, which motivated our investigation into whether the full potential of VMAT is currently being utilized. Modern linear accelerators with dynamic adaptations of gantry speed and dose rate continue to enhance VMAT plan quality. VMAT has been found superior to IMRT regarding target coverage and treatment time6–8. While9 found VMAT superior to 3DCRT for heart and LAD sparing, they reported higher doses to the contralateral side. VMAT is often favored for complex plans to reduce heart dose10 and is superior for target coverage and ipsilateral sparing11. Further studies12 showed that VMAT yields lower heart doses than forwardplanned IMRT (FIMRT). For bilateral breast cancer, hybrid approaches (h-VMAT) may be superior13, and VMAT remains the optimal choice for atypical anatomies like funnel chest14. Many institutions rely on predefined internal optimization criteria or a specific treatment protocol15–22. However, while plan quality is extensively compared, technical details concerning the automation protocols themselves are often limited. In17, limited optimization led to VMAT heart doses nearly twice as high as for 3DCRT. This suggests that optimization is often terminated once internal, institution-specific constraints are met, potentially leaving further dose reduction untapped. Current literature indicates that while average ipsilateral lung and heart doses are often similar between 3DCRT and VMAT, contralateral doses and low-dose volumes typically remain higher with VMAT. We hypothesize that these volumes can be significantly reduced to levels approaching 3DCRT through more rigorous optimization. Motivated by these inconsistencies, we developed a method to generalize VMAT planning to systematically lower mean organ doses. While deep inspiration breath-hold (DIBH) can further reduce doses11,23–27, our study investigates if substantial organs at risk (OARs) reductions are achievable via OFV-guided (objective function value) planning alone. Automated treatment planning has matured significantly, utilizing diverse strategies to reduce inter-planner variability and improve efficiency28,29. In modern treatment planning systems (TPS) such as RayStation and Eclipse, powerful scripting interfaces (Python) and the Eclipse Scripting API (ESAPI) allow for the implementation of complex automation frameworks. These methodologies range from rule-based heuristic scripts that mimic the decision-making process of an experienced planner to more advanced lexicographic optimization and automated multi-criteria optimization (MCO) navigation30,31. Scripting is also increasingly used to integrate Knowledge-Based Planning (KBP) and deep learning-based dose predictions into the clinical workflow, automatically translating predicted dose-volume histograms (DVH) into deliverable optimization objectives. The literature consistently shows that automated and scripted treatment planning achieves target coverage comparable to manual planning while improving efficiency and reproducibility32–36. presented a straightforward method for automatic treatment planning using Python in RayStation, with automatically generated plans being (...truncated)