Machine Learning-Based Beam Delivery Time Model for Mevion S250i with Hyperscan Technology
Abstract
Purpose
Accurate prediction of beam delivery time (BDT) is essential for operational efficiency, 4D dose calculations, and advanced proton therapy techniques such as proton arc therapy. Despite its importance, no machine-specific BDT model currently exists for Mevion systems. This study aims to develop a machine-learning (ML) model to predict BDT for the Mevion S250i with Hyperscan technology.
Methods
We analyzed 1120 machine log files from 11 patients to extract features related to spot position, energy layer changes, Adaptive Aperture (AA) movements, and spot charge. Inter-spot time (ΔT) was used as the target variable. A Random Forest model was trained on 70% of the data with 5-fold cross-validation for hyperparameter tuning and tested on the remaining 30%. Explainable AI techniques, including SHAP analysis, were applied to identify feature contributions.
Results
ΔT exhibited a wide range, from milliseconds to seconds. Short intervals were primarily associated with spot drilling and spot scanning, while longer intervals were driven by AA or range shifter motion. The model achieved mean absolute errors (MAE) of 0.9 ms for short ΔT intervals (1000 ms). SHAP analysis revealed AA shifts as the dominant predictor for ΔT >50 ms, whereas spot shifts and spot charge were most influential for shorter intervals. When applied to two clinical scenarios, volumetric repainting and dynamic 4D dose recalculation, the predicted cumulative delivery times deviated by only -1.6% from machine logs.
Conclusion
This work introduces the first ML-based BDT model for the Mevion S250i system, accurately capturing machine-specific temporal dynamics. Explainable AI provided insights into operational factors influencing delivery time, including energy layer switching, AA adjustments, and spot positioning. The model demonstrated strong predictive performance across clinically relevant applications, supporting its potential use in interplay assessment, 4D dose calculation, and delivery time–based plan optimization.