Real Time Mixed Helium-Carbon Imaging and Range Prediction

Background and Aims: This work presents an event-centric reconstruction and image formation strategy for a one-dimensional coded-mask gamma camera developed for prompt-gamma imaging in proton therapy within the SiFi-CC collaboration.The method aims to improve the accuracy of reconstructed prompt-gamma depth profiles by jointly addressing interaction reconstruction and imagelevel optimization under realistic detector conditions.Methods: The detector is based on scintillating LYSO fibers with dual-ended optical readout, providing redundant signal information for each gamma interaction.Instead of conventional channel-wise evaluation, coincident detector responses are aggregated into event representations that capture spatial, temporal, and statistical correlations.A graph-based learning model is used to infer interaction parameters from this representation, enabling reliable reconstruction of interaction locations and deposited energies even in the presence of inactive or malfunctioning detector channels.The reconstructed interaction information is subsequently used for prompt-gamma depth profile using a maximum-likelihood expectation-maximization algorithm.To fully exploit the improved event reconstruction, image reconstruction parameters are optimized using a multi-objective genetic algorithm based on the NSGA-III strategy implemented in the Optuna framework.Results: The optimized reconstruction achieves sub-millimeter accuracy in determination of beam range, with a root mean square error of approximately 0.5 mm on a dataset recorded using 10 10 protons. Conclusions:The results show that combining event-level graphbased reconstruction with systematic image optimization substantially enhances the performance of coded-mask prompt-gamma imaging systems.