ESTRO 2024 - Abstract Book

S3943

Physics - Image acquisition and processing

ESTRO 2024

CBCT scans were acquired for twenty-one breast, lung, and pelvis cancer patients (seven patients in each group) undergoing radiotherapy. For breast patients, these were acquired in breath-hold. CBCTs were acquired on a Halcyon system equipped with the HyperSight imaging solution and on a TrueBeam system with conventional CBCT technology (Varian), as part of a clinical imaging trial (NCT05524454). HyperSight images were acquired within the first week of treatment. TrueBeam CBCTs were collected from the treatment fraction on the same day for all but three patients. The planning CTs were also collected for comparison to the CBCTs. The HyperSight CBCT on the Halcyon system had a large imaging panel (86x43 cm2) providing CBCT images with a field-of-view of 53.8 cm and a length of 24.5 cm in the cranial-caudal direction, whereas conventional TrueBeam CBCT images had a field-of-view of 46.5 cm and length of 17.5 cm. Furthermore, the HyperSight CBCT scans could be reconstructed with either IR, with additional metal artifact reduction if needed, or the Feldkamp-Davis-Kress (FDK) algorithm for filtered back-projection, while the TrueBeam CBCT scans could only be reconstructed with FDK. The planning CT was rigidly registered to each of the CBCTs, and then regions-of-interest (ROIs; 0.1-4.0 cm3) were placed on the planning CT in homogeneous regions of the organs-at-risk, as well as muscle and fat. The ROIs were rigidly copied (small adjustments for minor anatomical changes were allowed) to the three CBCT scans (HyperSight with IR and FDK, and TrueBeam with FDK). From the CT numbers inside the ROIs, boxplots were created for the mean (quantifying accuracy) and standard deviation (quantifying image noise) for the four image sets. In Figure 1, a patient example illustrates the image quality of the four image sets. The fast HyperSight acquisition allowed for imaging within one breath-hold for all breast patients, and the large imaging panel ensured that all patients were fully within field-of-view. The HyperSight CBCT had qualitatively better image quality and reduced artefacts compared to the TrueBeam CBCT. Quantitatively, a reduction in image artefacts and improvement in CT number stability can be seen for HyperSight IR in the boxplots of the mean and standard deviation of the CT numbers (Figure 2). The results for the HyperSight IR images were close to those for the planning CT, while the standard deviation for TrueBeam CBCT was much higher (median standard deviation for muscle and fat of ~25 HU for TrueBeam, compared to ~12 HU for HyperSight IR). Note especially the larger inter-quartile range of the mean CT numbers for HyperSight FDK (~60 HU) and TrueBeam (~55 HU), compared to the planning CT and HyperSight IR (both ~18 HU). Visual inspection showed that the increased standard deviation and inter-quartile range of the mean CT numbers for the TrueBeam images were caused by 1) image artefacts (e.g., motion artefacts, or streak artefacts around bone, resulting in higher CT numbers than expected), or 2) CT numbers being systematically offset for a large region. These issues were not observed in HyperSight IR images. The increased spread in mean CT numbers for FDK reduces the dose calculation accuracy when a linear CT-number to-mass-density conversion curve is used for dose calculation on these CBCT images. CT numbers for muscle and fat are expected to be around 50 HU and -100 HU, respectively. This was found to be the case for the planning CT and HyperSight IR, while for HyperSight and TrueBeam FDK, larger deviations from expected values were seen. CT numbers also differed per anatomical site. E.g., for TrueBeam, the median CT number for muscle was 92 HU in breast cancer patients (often affected by bone streak artefacts) and -3 HU for lung cancer patients, while it was 40 and 44 HU, respectively, for HyperSight IR. Results: