Bregje In T Veld

Research Interests

• Accurate whole-heart segmentation on treatment-day cone-beam computed tomography (CBCT) is important for cardiac sparing and adaptive radiotherapy, but manual contouring in the CBCT domain is challenging because of low soft-tissue contrast, imaging artefacts, and inter-observer variability. In this study, we investigated whether a deep learning model trained on synthetic CBCT (sCBCT) images could be used to automatically segment the whole heart on real treatment-day CBCT scans. A labelled sCBCT dataset was generated from planning CT using a physics-based artefact simulation approach, and this dataset was used to train a three-dimensional autosegmentation model. Model performance was then evaluated on real CBCT images and compared with a state-of-the-art deformable image registration (DIR)-based contour propagation method. This work aims to assess the potential of synthetic-data-driven deep learning to improve the accuracy and efficiency of whole-heart segmentation on CBCT and to support more precise adaptive radiotherapy workflows.