ESTRO 2024 - Abstract Book

S3488

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

ESTRO 2024

For the tested cases, the two-step BGO is able to identify a combination of angles that leads to plans that are of a comparable dosimetric quality compared with manual plans while eliminating the complex and time-consuming beam geometry set-up step. It is noteworthy that the initial ML prediction may be close to the planner’s manual choice, but it may not yet guarantee a dosimetrically acceptable result as small changes in the limiting angles for the arc and AvS are relevant. In the second step, the generalized simulated annealing algorithm for beam angle optimization, searches effectively through the possible thousands of beam angle combinations within the search space. We discuss the applicability and limits of the method.

Table 1. Dosimetrically comparable plan to a manual plan is obtained after steps 1 and 2 of the beam geometry optimization method for an example patient.

Conclusion:

We show that dosimetrically comparable plans to manual ones can be obtained in an automated fashion, combining an initial beam geometry selection with a dosimetric refinement of the beam angles. The search heuristics makes the method powerful in identifying suitable angle combinations. The work paves the way to automation of the beam geometry setup that has the potential for significant savings of planner’s time.

Figure 1. Comparison of dose distributions between the automated and the manual plan for an example patient.

Keywords: Beam geometry, VMAT, automation

544

Digital Poster

Evaluation of a deep-learning-based synthetic-CT generating software for hippocampal-sparing-WBRT