Session Item

Recent developments in artificial intelligence (AI) and automation has enabled fast solutions for online adaptive radiotherapy (oART). These include AI-driven deformable image registration between original planning computed tomography (pCT) to the daily cone-beam CT (CBCT), creating a synthetic CT (sCT) for automated plan re-optimization on the anatomy of the day. However, sCTs are inherently prone to uncertainties, both geometrical and related to assignment of Hounsfield units (HU). To minimize these uncertainties, this study assesses the feasibility of using CBCT for direct dose calculations in AI-driven CBCT-based oART. 

In order to ensure consistency between CT and CBCT, HU to relative electron density (RED) calibration curves were acquired for the CT-specific Pelvis mode and CBCT Pelvis Large acquisition mode, using a dedicated electron density phantom. The CBCT images were iteratively reconstructed using the clinical (iCBCT_clin) as well as an improved reconstruction model in development mode with motion artifact reduction (iCBCT_dev). Using the acquired HU-RED calibration curves, dose calculations were performed on CT and CBCT scans of an anthropomorphic pelvic phantom. Furthermore, dose calculations were performed on the pCT, sCT, iCBCT_clin and iCBCT_dev of twelve patients. Dose calculations on both CBCT reconstructions were compared to the corresponding results for the pCT and sCT, using the specific calibration curves. Resulting dose distributions were evaluated by gamma analysis, using a 2% / 2 mm criteria, as well as comparison of dose volume histogram (DVH) parameters. 

Calibration curves revealed good agreement between CT and CBCT, with a median root-mean square deviation of 24 HU. For the anthropomorphic phantom, the CT to CBCT dose comparisons resulted in gamma passing rates greater than 98.8% for all acquisition modes. For the patient data, dose profiles and DVH comparisons between CT and CBCT dose calculations were in general in agreement. Dose distributions calculated on the iCBCT_dev images had a median gamma passing rate of 93.5% and 97.1% for the pCT and sCT dose comparisons, respectively. Corresponding iCBCT_clin calculations had a median gamma passing rate of 88.7% and 93.1% for pCT and sCT, respectively (Table 1). The median gamma passing rates differences were not statistically significant. Overall, both DVH and gamma analysis indicated improvements in dose calculations using the artifact improved CBCT reconstruction. 

This study has demonstrated the feasibility of using CBCT images for direct dose calculations in an oART workflow; comparable to pCT and sCT in terms of calibration data and resulting dose calculations. Further improvements to CBCT reconstructions may support this even further, as demonstrated for the motion artifact reduced CBCT reconstruction. However, further studies with an extended pool of patient cases are required to confirm these initial results.