Room: Exhibit Hall | Forum 9
Purpose: Monte Carlo (MC) simulations are a powerful and useful tool for improving image quality in x-ray imaging modalities. An accurate x-ray source model is essential to a MC model for CBCT but can be difficult to implement on a GPU while maintaining efficiency and memory constraints. A statistical analysis of the primary photon distribution from a MC x-ray tube simulation is completed and used to create a compact source model for use in CBCT MC simulations.
Methods: MC simulations of an x-ray tube were carried out using a modified version of BEAMnrc. A phase space (PHSP) file was collected at the exit of the tube and photons sorted into three categories: primary, tube scatter, or off-focal radiation. A PCA whitening transforms was calculated using the covariance matrix of the primary PHSP components (energy, position, and direction) and used to transform the PHSP data. A set of 5 quantile functions were created from the transformed PHSP data and sampled using MC techniques to simulate an energy fluence image.
Results: The statistical analysis showed the primary PHSP had correlations between position and direction which were reduced by using the PCA whitening transform allowing the creation of 5 quantile functions from the marginalized probability distributions functions of the transformed PHSP components. The mean absolute error of the energy fluence image when compared to the BEAMnrc data was 1.2%. When the data is untransformed the error was 26.6%.
Conclusion: Our automated method reduces the memory required to accurately simulate a x-ray source on a GPU MC system. Instead of loading a pre-computed phase-space or multivariate quantile function a set of 5 quantile functions and a transform can be loaded and sampled from. This method reduces the memory requirements by greater than 5 orders of magnitude, going from gigabytes of data to kilobytes.
Funding Support, Disclosures, and Conflict of Interest: Funding was received from Elekta, no Elekta products or materials were tested or evaluated in the submission.