DoseNet: An Ensemble-Based Deep Learning Method for 3D Dose Prediction in IMRT

To improve the accuracy of dose distribution prediction in radiotherapy, this study incorporates the flexible application of radiotherapy background knowledge into the task of deep neural network dose prediction, named DoseNet. In response to the problem of introducing the distribution characteristics of high energy radiation into the tissue, an assisted prediction of target conformal dose distribution is designed. To address the problem of non-uniform distribution in the dose prediction results of DoseNet, a sub-optimal model using the ensemble method of dose metric was designed. Finally, we found that using the DoseNet ensemble method for dose prediction accurately predicted patient-specific doses for rectal cancer treatment plans.