S2175
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
Figure 1: Comparison of liver images, motion field Jacobians, and their distributions in forward and inverse directions.
while the inverse motion field maps back to the planning CT. Both fields are jointly optimized by minimizing their inverse consistency error while enforcing local constraints, illustrated with an incompressibility constraint on the liver using the Liver-DIR-QA database[2]. Robustness is evaluated by synthetically undersampling the projection data to simulate CBCT artifacts[3]. By penalizing inverse consistency errors, physical properties defined on the CT are propagated to the daily CBCT, removing the need for clinically approved contours on daily images. The method is compared with a baseline model that estimates both transformations independently, and an inverse-consistent model without incompressibility constraint. Performance is assessed using the Jacobian determinant, and the target registration error using 83 anatomical landmarks. Results: Figure 1 shows large volume variations for the baseline model inside the liver for the forward and inverse motion field, with an inter-quartile range (IQR) of 0.20 respectively. Enforcing inverse consistency narrows the distribution of the Jacobian determinant (IQR = 0.11). The proposed model defines an incompressibility region in the liver on the planning CT, which, through inverse consistency, also carries over to the follow-up image (IQR = 0.02). Reducing the image quality, Figure 2 shows improved landmark alignment for sparsely sampled images ( ≤ 120 projections). Compared to the baseline model, the mean TRE reduces by -0.03, 0.05, 0.35, 0.68 and 1.79 mm over the five projection steps.
Figure 2: synthetic CBCT reconstructions and landmark error distributions of the image registration models. Conclusion: The proposed inverse-consistent registration model produces anatomically plausible deformations without requiring daily CBCT contours. By transferring physical constraints from the CT image through inverse consistency, the method improves DIR regularity and robustness to noise, supporting more accurate and
efficient ART. References:
1) Groot Koerkamp, M. L., Bol, G. H., Kroon, P. S., et al. (2024). Bringing online adaptive radiotherapy to a standard C-arm linac. Physics and Imaging in Radiation Oncology, 31, 100597. https://doi.org/10.1016/j.phro.2024.1005972) Zhang, Z., Criscuolo, E. R., Hao, Y., McKeown, T., & Yang, D. (2025). A vessel bifurcation liver CT landmark pair dataset for evaluating deformable image registration algorithms. Medical Physics, 52(1), 703–715. https://doi.org/10.1002/mp.175073) Zachiu, C., de Senneville, B. D., Tijssen, R. H. N., et al. (2017). Non- rigid CT/CBCT to CBCT registration for online external beam radiotherapy guidance. Physics in Medicine and Biology, 63(1), 015027. https://doi.org/10.1088/1361- 6560/aa990e Keywords: adaptive image registration, inverse consistency
Made with FlippingBook - Share PDF online