(Thursday, 7/18/2019) 10:00 AM - 12:00 PM
Room: 301
Purpose: To improve the accuracy and the efficiency of predicting patient-exit planar dosimetry images for MV electronic portal imaging devices (EPIDs).
Methods: A planar dose prediction model for aS-1000 EPID is developed based on parameterizing Monte-Carlo generated phase-space fluence data and an analytical modeling the particle scattering behaviors. As an alternative to sampling of particles, a dimensionality-reduction model is applied to the linac phase-space fluence data resulting in few particle beamlets featuring similar energies, distance from central axis, and azimuthal directions in space. While beamlets are attenuating through linac geometries and patient CT voxels, effective path lengths and center of masses of the beamlets are calculated to be used, with beamlet energy, in look-up corresponding scattering model. For each scattering kernel, a virtual source position is calculated/optimized so that it can accurately evaluate the scattered particle map at an arbitrary particle scoring plane.
Results: Fluence maps generated at an arbitrary vertical EPID plane by this method perfectly regenerates a fluence map that would have been created by sampling from the phase-space file. The scattering kernels, at distance larger than 10 mm from a scattering exit, are found to follow the inverse square law from the virtual source. The 2D Gamma passing rate is >99% for a 10x10 cm2 open field, at 1% 1mm criterion, between a predicted and measured EPID images in air and in 40 cm water phantom. Extremely large and small open fields experience output inaccuracies at the edges by an amount <6%.
Conclusion: In this work, we emphasize the possibility of phase-space data reduction and the analytical modeling of the particle scattering during the computation of EPID dosimetry images. This model features the accuracy of slow Monte-Carlo models and the speed of approximation-based analytical methods to predict EPID dose images.
Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by VARIAN Medical Systems. We gratefully acknowledge the support of Nvidia GPU card.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
