ESTRO 2021 Abstract Book

S728

ESTRO 2021

may be preferable is this setting.

Poster discussions: Poster discussion 28: 4DCT for planning

PD-0890 A novel approach to validate the accuracy of Mid-Position images calculated from 4DCT data F. Ghareeb 1 , J. Stroom 1 , M. Pereira 2 , D. Boukerroui 3 , M.J. Gooding 3 , C. Greco 1 1 Champalimaud Centre for the Unknown, Department of radiation oncology, Lisbon, Portugal; 2 Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisboa, Portugal; 3 Mirada Medical Ltd., Science, Oxford, United Kingdom Purpose or Objective The Mid-Position (MP) CT is calculated using a deformable image registration (DIR) of 4DCT data to create a time-weighted mean position of the anatomy during the breathing cycle. In addition, the deformation vector fields (DVFs) can be used to estimate motion amplitudes (MA) for relevant ROIs. Several 4DCT validation datasets are publicly available to test DIR algorithms. Although the accuracy of DVFs between phases was verified by experts using anatomical landmarks, these data provide a limited number of cases and are subjected to human and registration accuracy. Furthermore, the landmarks are spares and using implanted seeds as landmarks may introduce a bias to the registration algorithm. The purpose of this work was to devise an approach to validate the MPCT based on synthetic data where the ground truth MPCT can be known, and the MAs within the ROIs can be accurately estimated. Materials and Methods 10 synthetic 4DCT datasets were created based on 10 treatment cases (4 lungs, 5 livers and 1 pancreas tumors), each case consisted of 10 phases (pixel spacing ranged from 0.84 to 1.14 mm and slice thickness of 2 mm for all cases). For each case, the reference phase (maximum exhale phase) was registered to the other nine using MATLAB Demons DIR algorithm to acquire 9 DVFs. To eliminate the registration accuracy dependency, the original phases (except the reference) were deleted and a new synthetic 4DCT was created by applying the 9 DVFs to the reference phase. Since these DVFs have the same starting point, they can be averaged to a single mean DVF. The ground truth MPCT dataset of the synthetic 4DCT was created by applying the mean DVF (of the 9 DVFs and 1 zero-motion reference DVF) to the reference phase. The MPCTs for the synthetic 4DCTs were then also calculated by an automated research prototype (developed by Mirada Medical Ltd) and were compared to the ground truth MPs (slice by slice) using structural similarity (SSIM) index and 2D correlation coefficient (CC). The CTV was delineated on the ground truth MPCT for each synthetic 4DCT and copied to the corresponding Mirada MPCT. For each CTV, the mean CT HU values and the mean peak-to-peak MAs (derived from the DVFs in the AP, LR and SI directions) were compared. Results The minimum SSIM and CC were 0.9999 and 0.9994, respectively, indicating a good match between the ground truth and Mirada constructed MPCTs, as can be noticed from Fig1. Peak-to-peak mean MA and CT values results for CTVs are shown in Table.1. Differences in the MAs were less than 1 mm for most of the cases, and the difference in the mean CT HUs within the CTVs was negligible.