S1868
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
Monaco. The automatically generated plans (a_plan) were compared with the original clinical plans (c_plan) and the dose predictions (p_plan). To assess the deliverability of the a_plan, the treatment plans available for the irradiations (n=49) were measured using ionization chamber array (MatriXX and MultiCube, IBA Dosimetry GmbH, Germany) and the measured dose distributions were compared to dose calculation with the gamma index (global gamma, 3%/3mm, 15% threshold). Furthermore, beam-on- times were recorded for each plan during quality assurance measurements. To evaluate the clinical acceptability and quality of the a_plans, three experienced medical physicists reviewed the a_plans relative to the c_plans.
dosimetric advantages by improving sparing of critical organs without compromising target coverage or plan quality. This study highlights the importance of incorporating bias dose planning in multi-phase breast radiotherapy, especially when sequential boosts are required and supports its suitability for adoption in routine clinical practice. References: 1. Onal C, Efe E, Guler OC, Yildirim BA. Dosimetric Comparison of Sequential Versus Simultaneous integrated Boost in Early-stage Breast Cancer Patients Treated with Breast-conserving Surgery. In Vivo. 2019 Nov-Dec;33(6):2181-2189. doi: 10.21873/invivo.11720.2. Prokofev I, Salim N. Off- isocentric VMAT technique for breast cancer: Effective dose reduction to organs at risk and its applicability based on patient anatomy. J Appl Clin Med Phys. 2024 Mar;25(3):e142373. Krug D, et al. (2021). Advanced dose summation and optimization in sequential breast radiotherapy. Phys Imaging Radiat Oncol, 17:9–15.4. Fogliata A, et al. (2010). Dosimetric evaluation of breast treatment plans with different radiotherapy techniques. Radiat Oncol, 5:106. Keywords: Breast, Bias dose plan, sequential plan, VMAT Clinical Validation of Deep Learning-based Automated Treatment Planning Workflow Tuomas Virén 1 , Nina Pesonen 1,2 , Janne Heikkilä 1 , Juuso TJ Honkanen 1 , Miitu KM Honkanen 1 , Jan Seppälä 1 , Henri Korkalainen 1,2 1 Center of Oncology, Kuopio University Hospital, Kuopio, Finland. 2 Department of Technical Physics, University of Eastern Finland, Kuopio, Finland Purpose/Objective: Radiotherapy (RT) treatment planning is a laborious task prone to human errors. Automated treatment planning could reduce the staff workload and improve the consistency of RT treatments. In this study, we developed and validated an automated treatment planning workflow for left-sided breast cancer RT. Material/Methods: A total of 53 women with left-sided breast cancer were enrolled in this study. Automated treatment planning workflow was developed using Monaco (version 6.1.3, Elekta AB, Sweden) scripting API (Figure 1). First, an in- house deep learning-based model was used to predict a clinically achievable dose distribution for each patient in MIM (v. 7.3.3, MIM software inc., USA with an in-house Python extension). Second, dose constraints were derived from the dose predictions to be used as a starting point for automated optimization. Finally, treatment plans were automatically generated in Digital Poster 2035
Figure 1. Deep learning-based automated treatment planning workflow. Results: The a_plans were dosimetrically comparable with the c_plans (Figure 2). Furthermore, the dose predictions were in-line with both clinical and automated plans. Gamma passing rates of the a_plans were similar to the c_plans but beam-on-times were slightly longer as compared to those of the c_plans (Figure 2). Based on medical physicists’ evaluations, 83% percent of the a_plans were considered clinically acceptable and 56% were rated equal to or higher quality when compared to the c_plans.
Made with FlippingBook - Share PDF online