Introduction: Clinical Treatment Planning Systems (TPS) for proton pencil beam scanning (PBS) typically do not consider treatment delivery time, limiting advanced applications like FLASH therapy, 4D dose calculation, and in vivo verification that depend on accurate temporal modeling.We developed a machine-learning framework to predict machine-specific delivery timing using only standard DICOM-RT plan data.Methods: A component-based model (predicting spot delivery, spot transition, and layer switch times) was developed using Random Forest regressors.The framework was trained on machine log files and validated on two distinct systems: an IBA ProteusPlus and a Varian ProBeam, incorporating machine-specific pre-processing to handle proprietary logic like spot reordering.Results: The models achieved high accuracy for spot delivery (R 2 > 0.98) and spot transition (R 2 > 0.95) time prediction on both systems.Energy layer switching time was the primary source of error, leading to an underestimation of total field time (~3-5%).Despite this, Gamma analysis for predicted dose rate maps against logfile-based maps showed excellent agreement, with pass rates consistently meeting or exceeding 97% (0.5%/2mm criteria).Conclusions: This work validates a robust, adaptable framework for predicting PBS delivery timing.By enabling time-aware plan evaluation, this model provides the foundation for optimizing treatment efficiency and enabling next-generation, dose-rate-dependent treatment modalities.
Meijers et al. (Sat,) studied this question.