ESTRO 2020 Abstract book

S955 ESTRO 2020

an error-induced experiment with a 3D-printed heterogeneity phantom. This phantom produced the male pelvic anatomy. Subsequently, the following four kinds of errors were introduced: patient’s position error (−10–10 mm), couch rail setting error or change of body shape (140– 430 mL), and increased rectal gas (80–400 mL). The error- free EPID exit image was compared to those induced with error. The 2D EPID exit images acquired in all fractions for each patient were retrospectively analyzed during clinical care by calculating the gamma passing rate (GPR) between the first and other fractions to validate the feasibility of this system. A result higher than the criteria (GPR = 95%) was considered as an error. In addition, CBCT images acquired at fractions with error were also analyzed to explore the reason of errors. Results Large body shrinkage and increased rectal gas greatly changed the GPRs with criteria of 0.5%/0.5 mm (especially for rectal gas) and 1%/1 mm, although GPRs with 2%/2 mm and 3%/3 mm did not have major effects on body shrinkage changes. For example, GPR with 1%/1 mm in 80 mL and 240 mL air gaps were 87.5% and 65.2%, respectively, and those with 140 mL and 430 mL body shrinkage were 100% and 99%, respectively. Table 1 shows the summary of the retrospective analysis. In a total of 1,054 fractions, 10 fractions had errors (approximately 1.0% of all fractions). Of these 10 fractions, five had large body shape shrinkage (>200 mL) (Figure 1), and three had greatly increased rectal gas (>10 mL). Table 1 Summary of the number of fractions according to gamma passing rates in 1054 fractions

Purpose or Objective Deformable image registration (DIR) facilitated dose reconstruction and accumulation can be applied to assess delivered dose in the presence of anatomical changes and verify the validity of a treatment plan during treatment. However, displacement field errors (DFEs) in the forward registration or inverse consistency (IC) errors in the backward registration introduce uncertainties in the reconstructed dose. This study uses an in silico model based on clinically observed deformations as ground truth to investigate the dose accumulation uncertainty (DAU) in head-and-neck radiotherapy (HNRT). Material and Methods A planning CT ( pCT ), cone beam CT (CBCT) from week one of treatment and three later CBCTs were used to generate three corresponding in silico reference CBCTs serving as ground truth (GT) for 12 HNRT patients. To assess the DAU, the pCT was first matched to each GT CBCT using the same reference DIR algorithm (B-spline) to generate a deformed pCT ( dCT ). The treatment plan was then recalculated on the dCT and the resulting dose distribution mapped back to the pCT space using the backward deformation vector field (DVF) to produce a fraction-specific GT reconstructed dose (RD). The process was repeated using the DIR algorithm under investigation (demons) to generate a demons RD which was compared to the GT RD to calculate the voxel-specific dose error Δ tot . Inverse consistent c and inconsistent i voxels were identified by successive application of the forward and backward demons DVF, and comparing the net shift to the dose calculation grid size. The fraction-specific RD uncertainties for each structure S and both voxel distinctions (i.e., uSc and uSi ) were estimated using the 95% percentile range of Δ tot . The feasibility of incorporating the DAU as a confidence interval in dose-volume histograms (DVHs) of the delivered dose was demonstrated for various structures. For each structure, the voxel-specific uncertainties in the accumulated dose (AD) over all fractions uAS were first calculated by summing uSc or uSi in quadrature. Subsequently, the 95% level of confidence for each dose level bin in a DVH of the AD was calculated by averaging the uAS of all voxels with an AD equal to or larger than that dose level. Results Overall, the IC rate of voxels was 98.5% (Fig. 1) with uSc per fraction equal to [-2.3%; +2.1%], [-10.2%; +15.2%] and [-9.5%; +12.5%] relative to their planned dose for target structures, critical OARs and non-critical OARs, respectively. Inverse in consistent voxels generally showed a higher level of uncertainty. At the end of treatment, uAS was [-0.4%; +0.4%] for target structures. The impact of DAU is demonstrated in the DVH of Fig. 2.

Fig.1 Typical two example cases showing the gamma maps between first fraction and other fractions. Conclusion Our result showed that PerFRACTION with EPID exit images effectively detected the body shrinkage and rectal gas during the treatment course. In addition, this system could detect a certain number of (approximately 1.0%) clinical data errors. PO-1647 Quantifying the dose accumulation uncertainty after DIR in head-and-neck radiotherapy N. Lowther 1 , S. Marsh 2 , R. Louwe 1 1 Wellington Blood & Cancer Centre, Department of Radiation Oncology, Wellington, New Zealand ; 2 University of Canterbury, School of Physical and Chemical Sciences, Christchurch, New Zealand