Last vertex splitting: a new retroactive Monte Carlo splitting technique applied to LINAC out-of-field dose...

We propose a new variance reduction technique called last vertex splitting (LVS) designed to reduce computation time in Monte Carlo (MC) simulations for particles traversing high-attenuating media, such as the collimator and other beam-limiting devices in a LINAC head. Combined with a hybrid version of the track length estimator (hTLE), the LVS method accelerates out-of-field (OOF) dose calculations by optimizing photon tracking and interaction modeling. Our analysis indicates that the residual bias introduced by the method remains below one percent, with an estimated efficiency speed-up of ×4.5 for a 10×10 cm²field, to ×10 when combined with hTLE. Typically, this approach enables the rapid generation of extensive MC datasets, facilitating the training of deep learning algorithms to predict OOF doses more efficiently. Beyond radiotherapy applications, the LVS method can be adapted for scenarios requiring computationally efficient simulations of particle transport in complex geometries, such as in radiation shielding assessments.