Super-resolution surrogate model for accelerated hadrontherapy simulations

The WP6-Spoke2 of ICSC, the Italian Research Center on HPC, Big Data, and Quantum Computing, focuses on integrating machine learning models to enhance the Geant4 simulation toolkit. Geant4 is a widely used framework in medical physics, capable of modeling particle interactions at micrometer and sub-micrometer scales. However, its computational cost scales linearly with the complexity of the system, which limits its practical use. Hadrontherapy example has been identified as a key application area, as it allows accurate calculations of dose and linear energy transfer distributions in water and other materials. To support this, we first generate a high-fidelity dataset using a densely voxelized water phantom, serving as a benchmark for primary and secondary particle interactions. We then developed and trained a generative model to reproduce the spatial correlations in this dataset. The objective is to create a super-resolution surrogate model that enhances the spatial resolution of low-granularity hadrontherapy simulations, substantially reducing computational demands while maintaining the accuracy of essential physical quantities such as LET. Although originally developed for Geant4, the model is designed to be adaptable to other simulation frameworks and experimental datasets.