ESTRO 2024 - Abstract Book

S3487

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

[1] Yagihashi, T., Inoue, T., Shiba, S. et al. Impact of delivery time factor on treatment time and plan quality in tomotherapy. Sci Rep 13, 12207 (2023).

[2] De Kerf, G., Van Gestel, D., Mommaerts, L. et al. Evaluation of the optimal combinations of modulation factor and pitch for Helical TomoTherapy plans made with TomoEdge using Pareto optimal fronts. Radiat Oncol 10, 191 (2015).

[3] Van Gestel, D., De Kerf, G., Wouters, K., Crijns, W., Vermorken, J. B., Gregoire, V., & Verellen, D. Fast Helical Tomotherapy in a head and neck cancer planning study: is time priceless?. Radiat Oncol, 10, 261 (2015).

437

Digital Poster

Two-step beam geometry optimization for VMAT gantry angles in breast treatments

Mikko Hakala 1 , Luca Cozzi 2 , Elena Czeizler 1

1 Varian Medical Systems, a Siemens Healthineers Company, Helsinki, Finland. 2 Varian Medical Systems, a Siemens Healthineers Company, Steinhausen, Switzerland

Purpose/Objective:

We developed a method to set up the start and stop gantry angles and avoidance sector (AvS) limits for volumetric modulated arc therapy (VMAT) breast treatments in an automated fashion. VMAT plan generation conventionally requires repetitive human tasks where the gantry angles are selected based on patient anatomy. The angles may need to be fine-tuned by the planner with multiple runs of the optimizer until the clinical goals are satisfactorily reached. We report on a method developed to automate this optimization step. The approach we developed in this study could also be applied to optimize the static beam positions for IMRT plans.

Material/Methods:

Our beam-geometry optimization (BGO) approach takes into consideration both geometric and dosimetric components when automatizing the task. In the first step (geometric-based selection), the initial gantry angles are obtained either from a template, from machine-learning (ML) predictions, or manually. A search range around these initial positions is specified for each angle. In the second step (dosimetric-based optimization), the angles are refined within the range-bound search space. We applied the method for free breathing and deep inspiration breath hold patients for left- and right-sided breast treatments. The patients were manually planned for reference. In the first step of the method, two machine learning models were trained: one for VMAT start and stop angles and one for the AvS limits. Input features into the neural network were grouped projections of contoured 3D structures as seen from the beams-eye-view around the patient. In the second step of the method, the refinement of beam angles is based on the generalized simulated annealing algorithm, where the objective function to minimize is the optimizer cost. As clinical goals, we monitor mean dose to heart, to ipsi- and contralateral lungs, and to contralateral breasts in addition to target doses. Optimization is done in a multidimensional search space and the search space boundaries can be inputted from the user or defined based on ML or other models.

Results: