Deep learning-based synthetic CT (sCT) generation models are often based on T1-weighted gradient echo sequences. However, these sequences are generally not used for tumor/organs-at-risk delineation. In this study, we trained a U-Net type neural network using T2-weighted turbo spin-echo images from the clinical radiotherapy treatment planning protocol originally used for tumor/organs-at-risk contouring. The use of clinical images preserves scan times and facilitates soft tissue delineation on the source images of the sCTs, avoiding registrations. We showed that sCTs generated by the trained model provide accurate dosimetric results while limiting CT-induced streaking dental artefacts.
This abstract and the presentation materials are available to members only; a login is required.