Automation of breast cancer dosimetry with Vision Transformers

Abstract

A hybrid model combining a Vision Transformers encoder with convolutional layers is proposed for breast cancer dosimetry. Its predictions were compared with a baseline U-Net trained under the same conditions. The results suggest improved performance in OAR metrics, while maintaining acceptable accuracy in PTV dose prediction. However, no statistically significant differences were found, so further research is still needed to explore the full potential of Transformers in radiotherapy