Dosimetric Evaluation of Dose Calculation Uncertainties for MR-Only Approaches in Prostate MR-Guided Radiotherapy

Purpose: Magnetic resonance imaging guided radiotherapy (MRgRT) allows treatment plan adaptation on the MRI of the day. For dose calculations, a structure-specific bulk relative electron density (RED) overwrite derived from a planning computed tomography (CT) poses as one possible treatment workflow. However, this approach introduces uncertainties due to assignment of mean densities and requires a planning CT. The aim of this study was to investigate the uncertainty of the used patient-specific (PSCT) dose calculation in contrast to the correct calculation on a CT and compare to MR-only workflows using population-based bulk ED (PBCT) and artificial intelligence-based pseudo-CTs (AICT).Methods: Twenty primary prostate cancer patients treated on the 1.5 T MR-Linac were chosen from the clinical database, based on best visual congruence between the planning CT and daily MRI. CT-based reference dose distribution was compared to different pseudo-CT approaches. 1) For PSCT, mean REDs for the femur, pelvis, sacrum, rectum, bladder, and patient were assigned based on individual mean CT densities. 2) Population-based mean REDs were derived based on 50 previous, independent patients and assigned to the structures for the PBCT approach. 3) An AI model for pseudo-CT generation was trained using end-to-end ensembled self-supervised GANs and used to create AICTs from T2w-MRIs. For comparison, the CT was registered to the MRI, structures rigidly propagated, and treatment plans recalculated. Differences of DVH parameters were analyzed, and dose distributions were compared using gamma analysis.Results: All approaches were able to reproduce the dose distribution accurately, according to a gamma criterion of 3%/3 mm, with pass rates greater than 98%. Applying a 2%/2 mm criterion, the median gamma pass rates for PSCT, PBCT, and AICT resulted in 98.6%, 98.2%, and 99.0%, respectively. The median differences for PTV D-98% resulted in 0.13 Gy for AICT, -0.31 Gy for PBCT, and -0.32 Gy for PSCT. The OAR-related DVH parameter showed similar results between the three investigated methods.Conclusion: In this study, a detailed analysis of uncertainties of MR-only treatment planning concepts for pelvic MRgRT was performed. Both a PBCT and an AICT approach, which bypass the need for a planning CT, may be considered clinically acceptable while reducing imaging dose and registration issues.