Improving Monte Carlo dose distribution

Description

Recent developments on asymptotic Variance Reduction Techniques (aVRT) open new ways to improve the efficiency of Monte Carlo (MC) simulations in the field of medical physics. In radiation therapy, MC simulations are used to calculate directly the absorbed dose in patients or to determine the dose point-kernels for analytical dose engines. This is particularly true for new promising radiotherapy protocols, which are mainly oriented towards very high-dose-rate treatment and thus require extremely precise dose distributions. Our overall goal is to improve the statistical quality of the MC dose distributions using post-processing provided by a deep learning approach. A learning system was built using U-Net. However, this system suffers from poor generalization properties when applied on other type of data. Hence, the system only works for one MC simulation setup and any change will decrease the performance. In this paper, to adapt the knowledge from a related database that has already been learned in a new type, we propose a color transfer by minimizing the Wasserstein distance between two histograms of dose. The results show that our method is promising for recovering complex dose distributions without depending on a particular medical application as the other works.