ESTRO 2023 - Abstract Book

S826

Monday 15 May 2023

ESTRO 2023

With the growth of interest in Knowledge-Based Planning, more and more information is available to the treatment planner to reach a clinically acceptable plan. Knowledge of spatial dose distribution has become a major challenge in treatment planning optimization as it allows for better shaping of the dose with the help of voxel-based dose mimicking constraints. This is especially significant in regions where PTVs and OARs overlap. In a clinical workflow, the spatial dose information may come from an AI model trained to predict a clinically acceptable dose for a new patient, reducing the usual back and forth steps between optimization and clinical evaluation of the dose. When replanning is needed (e.g. because of machine failure, QA not passing, etc.) the initial dose is already available, which makes dose mimicking the most natural way to replicate what was first intended. The aim of this study is to define an optimization framework that minimizes organ-wise voxel-wise constraints with a single template and to investigate whether this approach is able to reproduce clinically accepted doses without manual interaction from the planner. Materials and Methods We retrospectively optimize 94 prostate cases from a major European cancer center with 80Gy prescription on the prostate (PTV2) and 56Gy on the seminal vesicles (PTV1), using the clinical dose as our mimicking target. The structures we define in the dose mimicking template are the PTV1&2, rectum, bladder, femoral heads, overlaps of bladder and rectum with PTV2, ring on PTV2. We penalize any dose difference on the PTVs and only overdose for OARs. We also include a global constraint on all OARs to reduce the dose if possible. Mimicked dose is obtained with a dose engine using a Collapsed Cone Convolution algorithm and normalized to match the D50 of the PTV2. Results Table 1 depicts average differences between clinical and mimicked dose for some points of the DVH curves for all organs:

Figure 1 depicts DVH for one patient of the cohort and dose slices at the isocenter:

Conclusion This study shows the feasibility of using a single template dose mimicking optimization framework to successfully replicate or improve the target doses on PTVs and OARs for a large dataset. This is achieved without any fine-tuning, meaning minimal effort is requested from the planner once the initial dose is validated. In the future this framework will be extended to adaptive planning as patient-specific tradeoffs, that may have been made for the original plan, can be used in case replanning is suggested.

PD-0978 Electron modulated arc therapy for breast boost: development of a treatment planning process G. Guyer 1 , S. Mueller 2 , D. Frei 2 , W. Volken 2 , K. Loessl 2 , D.M. Aebersold 2 , P. Manser 3 , M.K. Fix 3