S1762
Physics - Dose prediction/calculation, optimisation and applications for particle therapy planning
ESTRO 2026
previous beamlet-free optimization strategy [3] has shown the potential to eliminate explicit beamlet pre- computation, offering a promising route toward more efficient MC-based optimization. Nevertheless, this approach remains limited in terms of modularity and accuracy.We propose a novel beamlet-free optimization framework designed to further enhance computational efficiency and reduce memory requirements, while maintaining clinically acceptable accuracy. Material/Methods: We consider a voxel wise dose prescription . The optimization problem is defined as:
where is the vector of spot weights,
is the
beamlets matrix, and is a weighting matrix allowing differential enforcement of the prescription across voxels. The gradient of the objective is derived as The optimization is structured into three steps: (1) sampling a limited set of protons per spot via MC simulation(2) computing the intermediate terms and , obtained as scalar products quantifying, respectively, beamlet-to-beamlet and beamlet-to-prescription correlations.(3) updating spot weights using the Adam optimizer, as the proposed formulation is compatible with stochastic gradient optimization and proton batch subsampling. Results: We successfully computed the beamlet correlation matrix ( , Figure 1) and the beamlet to dose prescription correlation vector ( ), demonstrating the feasibility of integrating these into the optimization process. This formulation allows the algorithm to capture beamlet interdependencies and their relative contributions to the dose distribution.
A full optimization workflow was implemented and executed from scratch on a water phantom, using 729 spots and 2,500 simulated protons per beamlet. The resulting dose distribution and DVH (Figure 2) confirmed the ability of the method to achieve the prescribed 50 Gy target dose.
Conclusion: This study demonstrates the feasibility of a more complete and modular beamlet-free optimization framework that reduces both MC computation time and memory requirements. The proposed method enables efficient Monte Carlo–based plan optimization with reduced particle statistics, while avoiding the resource-intensive precomputation and storage of beamlets thereby paving the way for scalable clinicla application References: [1] Souris K, Lee JA, Sterpin E. Fast multipurpose Monte Carlo simulation for proton therapy using multi- and many-core CPU architectures. Med Phys. 2016;43: 1700.[2] Wuyckens S, Janssens G, Chocan Vera M, Sundström J, Di Perri D, Sterpin E, et al. Proton arc
Made with FlippingBook - Share PDF online