Automated Treatment Planning in Radiation Therapy using Generative\n Adversarial Networks

Abstract

Knowledge-based planning (KBP) is an automated approach to radiation therapy\ntreatment planning that involves predicting desirable treatment plans before\nthey are then corrected to deliverable ones. We propose a generative\nadversarial network (GAN) approach for predicting desirable 3D dose\ndistributions that eschews the previous paradigms of site-specific feature\nengineering and predicting low-dimensional representations of the plan.\nExperiments on a dataset of oropharyngeal cancer patients show that our\napproach significantly outperforms previous methods on several clinical\nsatisfaction criteria and similarity metrics.\n