cGAN Application: Dose Prediction For Radiotherapy Treatments

Abstract

In this work we present the application of a model based on conditional generative adversarial networks (cGAN) for the planning of stereotactic body radiotherapy (SBRT) treatments for prostate cancer. Inspired by previous approaches used in dose planning for other types of cancer, we propose an architecture adapted to the characteristics and clinical requirements of prostate cancer treatment. We evaluate the model’s performance using data from real cases treated at a radiotherapy center in C´ordoba, comparing the generated dose plans with the clinical plans approved by specialists. The results obtained on the test set show, on average, a RMSE of 2.98% over the PTV volume, 3.56% over the rectum volume and 2.32% over the bladder volume in the DVHs. This suggests that the use of generative adversarial networks could be a promising tool to improve efficiency in SBRT planning.