Fast Monte Carlo-Based Inverse Planning for Prostate Brachytherapy by Using Deep Learning

Inverse planning is an essential tool for optimizing the delivered radiation dose on low-dose-rate (LDR) prostate brachytherapy. Clinical inverse systems use TG-43 computation formalism in order to perform a fast optimization. However, this method approximation that often leads overestimation, resulting suboptimal plans. Alternatively, Monte Carlo simulation (MCS) can be used obtain accurate distribution, but considerably increasing estimation time. We propose LDR brachytherapy uses deep convolutional neural network (DCNN) trained graphics processing unit (GPU)-based MCS generated database estimate distribution and organs at risk. Segmentations of seeds’ positions are given as DCNN input. The mean percent error test set was −1.19±0.94% within prostate. dosimetric parameters each organ every configuration optimization loop. final DCNN-based plans were good agreement compared same recalculated with full MCS. proposed based capable reach equivalent level accuracy runtime less than 1 min using conventional GPU card.