Improved compact representation method for Monte Carlo phase spaces.

Phase space files used in particle Monte Carlo simulations can be large, and thereby awkward to distribute and slow to load into memory. To circumvent this, compact source models and phase space representations have been developed. In this study, the aim is to increase the versatility of a previously presented method, based on principal component analysis whitening, by evaluating several modifications to the method on simulated cone-beam computed tomography projection images. Scaling the phase space components before whitening was seen to give the largest improvement and allowed for accurate modelling of phase spaces with a shifted detector and/or a bowtie filter present, configurations where the original method struggled. The modified method reproduced results using the original phase space to within a few percent both in energy fluence on the detector and in simulated projection and scatter images of a patient pelvis, without obvious position-dependence of the errors. Hence, the modified method was deemed sufficiently accurate for most applications relying on cone-beam computed tomography simulations. The final aim is to use the more versatile method for a broader range of phase space-based applications such as linear accelerator beam simulations and room shielding calculations.