ESTRO 2025 - Abstract Book

S3545

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTRO 2025

Various prior distributions (normal, uniform) and initial values were tested to analyze their impact on the convergence of the parameters (Fig. 1). The results were compared according to the root mean square error (RMSE) with data for T1 cells exposed to C12, Ne20, and He4 particles extracted from the PIDE. After fixing the prior distribution as normal, a grid search was performed, varying in MCDS the cell diameter (5–15 µm) and the DNA content (5.5–6.5 Gbp), and computing the corresponding survival fractions.

Fig.1: Example of Bayesian fitting to infer the model parameter (see ref. [2]).

Results: The Bayesian analysis showed that convergence is achieved regardless of the selected prior, with the RMSE varying between 0.0016 and 0.0017, confirming the robustness of the Bayesian approach for this type of adjustment. With the grid search, it was possible to optimize the cell parameters, resulting in a 9% reduction in the RMSE (Fig. 2), with a DNA content slightly larger than the one suggested in PIDE (6 GBp).

Fig.2: Result of the optimization procedure. For variables explanation see [2].