A Feasibility Study for Predicting 3D Radiotherapy Dose Distribution of Lung VMAT Patients

Accurate dose prediction has been proved to be able to improve radiotherapy planning efficiency. Recently, deep neural networks have been used in this area and made some progress. However, existing deep-learning-based methods could not predict dose distribution accurately for tumors at various locations, i.e. lung cancer. This article proposes a new deep neural network CAD-UNet that combines 3D U-net, dense connection, and SE-net architecture. Spatial distance information is used as a special input channel in addition to contour information. Dice similarity coefficients of planning target volume (PTV) region was added to the mean squared error (MSE) loss function. A cohort of 192 VMAT plans for lung cancer patients was selected for this study. The trained CAD-UNet and HD-UNet were tested on the test cases. The dose parameters derived form predicted dose distribution were used to generate new plans in the treatment planning system (TPS). The results showed that CAD-UNet can successfully predict dose distribution of lung cancer cases in VMAT, outperforming HD-UNet in PTV region homogeneity. Regenerated plans based on predicted dose showed improvements in DVHs of organs-at-risk (OAR). Those improvements showed that CAD-UNet has the potential to guide dosimetrist in the radiotherapy planning stage.