Feasibility of CT-Only 3D Dose Prediction for VMAT Prostate Plans Using Deep Learning

Current radiotherapy planning workflows start with segmentation of the organs at risk (OARs) together with target volumes (TVs) in order to determine a patient specific optimal treatment plan and its corresponding 3D dose distribution. This is a time-consuming optimization process including many manual interventions. Despite strong resemblance between patients treated for the same indication, the optimization is almost always performed without 3D prior knowledge. Automated segmentation of OARs and TVs and automated generation of dose distributions are thus expected to be more time-efficient. We investigate the feasibility of CT-only dose prediction and the profitability of additional isocenter and contour information. To evaluate the network’s performance, a 5-fold cross-validation is performed on 79 prostate patients, all treated with volumetric modulated arc therapy.