Automated daily dose accumulation workflow for treatment quality assurance during online adaptive radiotherapy with a 0.35T MR-linac.

Purpose: This study assesses a novel, automated dose accumulation process during MR-guided online adaptive radiotherapy (MRgART) for prostate cancer, focusing on inter-fractional anatomical changes and discrepancies between delivered and planned doses.

Methods: A retrospective analysis was conducted on seven prostate cancer patients treated with a five-fraction stereotactic body radiation therapy (SBRT), using a 0.35T MRIdian MR-LINAC system. Daily plans were adapted when dose thresholds were exceeded. Planning MRI (pMRI) and daily MRIs (dMRIs) were imported into MIM software for automated and manual dose accumulation procedures. Rigid and deformable image registrations were followed by dose accumulation to compare delivered and planned doses. Manual and automated image registrations were compared by calculating the Hausdorff distance (HD), Jaccard, and DICE metrics.

Results: Moderate discrepancies in dosimetric parameters for the planning target volume (PTV) were observed between auto-accumulated and planned doses, such as $D_{95\%}$ and $D_{0.03\mathrm{cc}}$ , with average differences of $-0.60\pm 0.61$  Gy and $-1.31\pm 0.42$  Gy, respectively. Volume differences of $V_{34.4\mathrm{Gy}}$ and $V_{36.25\mathrm{Gy}}$ indicated that auto-accumulated doses consistently had lower numbers compared to planned doses, with mean discrepancies of $-1.80\%\pm 1.05\%$ and $-2.82\%\pm 1.72\%$ , respectively. Organs at risk (OAR) dosimetric parameters exhibited higher dose volumes in auto-accumulated doses, with moderate differences (planned [cc] vs. auto-accumulated [cc]) observed in parameters such as urethra PRV $V_{8.4\mathrm{Gy}}$ at $\left(4.33\pm 1.90\mathrm{vs}.4.34\pm 1.90\right)$ , rectum $V_{24\mathrm{Gy}}$ at $\left(1.17\pm 1.53\mathrm{vs}.1.74\pm 1.91\right)$ and rectum $V_{28.2\mathrm{Gy}}$ at $\left(0.38\pm 0.55\mathrm{vs}.0.59\pm 0.71\right)$ . The comparison between manually and auto-accumulated doses revealed negligible variations, as also indicated by strong concordance in geometric indices and t-test p-values above 0.7.

Conclusion: The automated workflow, developed in collaboration with MIM Software Inc., demonstrates high accuracy compared to manual accumulation. The moderate differences observed between planned and accumulated doses emphasize the need for accurate dose accumulation for adaptive plans.