Automated deep learning segmentation and planning for left-sided breast radiotherapy with minimised adaptations based on dose, TCP and NTCP criteria

Four scenarios utilizing varying deep learning segmentations and planning adaptations were compared.-Setups were evaluated on DVH goals and radiobiological metrics.-43 % of plans generated in a fully deep learning workflow met all clinical goals.-Edits to target delineation influenced the plan, to organs-at-risk not at all.-New workflow reduces active time (50 %) and in between steps (30 %) on a plan.