Automating Dose Prediction in Radiation Treatment Planning Using Self-attention-Based Dense Generative Adversarial Network

Radiation treatment being a crucial step in cancer treatment, accurate radiation treatment planning is extremely important to reduce the effect of radiation on Organs-At-Risk (OAR). In order to determine the radiation treatment dose distributions received by the tumour and surrounding organs, this paper proposes a self-attention-based dense generative adversarial network (GAN). This is accomplished by adding a self-attention module to GAN which uses dense U-Net as generator. Using the DVH and dose scores, the performance is evaluated and compared to a few state-of-the-art models like simple GAN, self-attention-based GAN and dense GAN. The dataset was acquired from the OpenKBP 2020 AAPM Grand Challenge run by CodaLab. The DVH score and dose score of the self-attention-based dense GAN are 2.126 and 3.522, respectively. The proposed model performed better than other state-of-the-art models considered.