ESTRO 2021 Abstract Book

S1261

ESTRO 2021

Conclusion We have developed a framework to assess the geometric and dosimetric impact of DIR uncertainty for dose accumulation of proton therapy of H&N tumors. Knowing the localization and the impact of the dosimetric uncertainties of the cumulative treatment dose could be used to improve patient outcome models. PO-1536 Contour-based deformable image registration for dose accumulation in MR-guided radiotherapy U. Bernchou 1,2 , R.L. Christiansen 1,2 , H.R. Jensen 2 , F. Mahmood 1,2 , A. Bertelsen 2 , T. Schytte 3,4 , C. Brink 1,2 1 University of Southern Denmark, Department of Clinical Research, Odense, Denmark; 2 Odense University Hospital, Laboratory of Radiation Physics, Department of Oncology, Odense, Denmark; 3 University of Southern Denmark, 2. Department of Clinical Research, Odense, Denmark; 4 Odense University Hospital, Department of Oncology, Odense, Denmark Purpose or Objective The soft tissue contrast provided by the MR-linac has spurred an increasing use of online plan adaptation in radiotherapy (RT) of pelvic cancer patients. Accurate accumulation of dose from such adapted plans is needed to access the plan quality over the treatment course and can potentially guide the online adaptation process. A vital step for accurate accumulation of dose is ensuring a correct mapping of the dose delivered to individual voxels throughout the treatment course in a common frame of reference. This step is usually performed using deformable image registration (DIR), which is challenging for highly deformable organs such as the bladder and rectum. The purpose of the current study was to compare the DIR performance of the open-source software toolbox Elastix and a commercial solution from MIM Software as potential algorithms for dose accumulation of these organs during MR-guided RT. Materials and Methods The study included 12 patients with prostate cancer referred for 78 Gy in 39 fractions. Bladder and rectum contours were delineated on T2-weighted MRI scans performed for RT planning and during the treatment course at fraction 10, 20, and 30 on a Philips Ingenia 1.5 T system. Contours were propagated from the MRIs acquired during the treatment course to the planning MRI using the DIR algorithms provided by Elastix or MIM. As would be the case for dose accumulation on an MR-linac, manual delineations were available in both datasets and were used as DIR input to steer the deformation - a contour-based approach. The propagated structures were compared against manually delineated structures using dice similarity coefficient (DSC), mean surface distance (MSD), and Hausdorff distance (HD). The difference in performance between the two algorithms as well as the difference in performance of each algorithm compared to the inter-observer variation (IOV) based on time-separated manual re-delineations were tested using Wilcoxon signed-rank tests. Statistical significance was set to 0.003 to account for multiple testing (Bonferroni correction).