A Novel Stochastic Dose Rate Predictive Model for Synchrotron-Based Pencil Beam Scanning Systems
Abstract
Purpose
Unexpected toxicity during proton therapy (PT) cancer treatments may be associated with the linear energy transfer (LET) or the instantaneous dose rate. However, millisecond timescale dose rate determination becomes unfeasible in modern systems due to the inherent variability of beam delivery sequences across realizations. A novel predictive approach is presented to enable fast and accurate prospective dose rate estimation in a synchrotron-based pencil beam scanning (PBS) PT system using stochastic modeling of dose delivery.
Methods
Dose rate calculation requires both three-dimensional spot dose information and the irradiation timing. These doses are obtained using Monte Carlo simulations with two approaches: a per-spot dose calculation, and a virtual-spot approach in which per-spot doses were derived from per-energy-layer dose calculations. In the virtual-spot approach, three-dimensional per-spot dose distributions were derived using virtual proton trajectories and spot spreading in water. To account for dose delivery unpredictability, N=100 treatment realizations are obtained per approach using a stochastic model to accurately determine the expected per-voxel temporal sequence and the corresponding dose rate variability. Compared with the per-spot approach, the virtual-spot approach reduces computational time and memory requirements by a factor 100th. Although this method has been tested on patients treated with a Hitachi Synchrotron, is adaptable to other systems.
Results
Voxel-wise dose rate histograms are computed for both approaches from which mean, maximum and standard deviation are derived. Differences among both approaches are below 5% in 95% of voxels. Maximum dose rates range from 20-25 Gy/s over spot timescales (1-2 milliseconds) and increases by a 1.2 factor at intra-spot scales (0.5-1 milliseconds).
Conclusion
The virtual-spot approach is a valid method to predict the dose rate in PBS-PT plans enabling both prospective and retrospective analyses without machine LOG files and can be integrated into dosimetric and radiobiological modeling frameworks for enabling dose and risk assessments.