Optimizing cone-beam CT image quality and dose calculation accuracy for adaptive radiotherapy

Since the introduction of cone-beam computed tomography (CBCT) it has been shown that CBCT leads to improved patient treatment in radiotherapy. Despite the advances in CBCT, its application is still largely limited to patient positioning. This is mostly due to the inferior image-quality compared to conventional CT. The main reason is the increased scatter due to the wider cone angle. This and other effects such as image-lag, beam-hardening, and patient motion cause image artifacts like cupping, streaks, and blurring, leading to Hounsfield unit (HU) inaccuracies. These inaccuracies limit the further application of CBCT, especially in the field of adaptive radiotherapy (ART).
The aim of this thesis was twofold. The first objective was to investigate and improve the accuracy of dose calculations on CBCT. The use of an anti-scatter grid and an iterative scatter correction algorithm was explored. Further investigations into the possible causes for the limited dose calculation accuracy for lung were performed and the HU stability over time was studied. The second objective was to improve CBCT image quality. Here, the combination of dual-energy imaging and iterative reconstruction with total variation regularization was investigated. Moreover, a novel deep learning-based reconstruction method that combines model knowledge with a learned component was developed.
The presented results show that the improved CBCT image quality and dose calculation accuracy facilitates further applications of CBCT for ART.