Abstract: Boron dose is an important component of the actual therapeutic dose of boron neutron capture therapy (BNCT). In order to accurately evaluate the actual therapeutic effect of BNCT, it is important to know the real-time distribution of boron dose during the treatment. However, there is still a lack of practical methods for monitoring boron dose in clinical. Accordingly, this work proposed for the first time using generative adversarial network (GAN) to predict the real-time three-dimensional distribution of boron dose based on the three-dimensional distribution detected during treatment. Based on an anthropomorphic male phantom with Chinese phy-siological features, 19 cases of glioma were constructed for analysis basing Monte Carlo simulations. The results show that this boron dose prediction method based on GAN can achieve quantitative prediction of the three-dimensional distribution. The value of SSIM between the results generated by Monte Carlo simulation and GAN are all about 0.98 within complex tumor structure and non-uniform distribution of boron concentration in the target area.