S2083
Physics - Image acquisition and processing
ESTRO 2026
two-step affine + deformable (SyN) registration, providing access to a probabilistic WMT atlas (HCP- 842, 52 tracts)[1], allowing voxel-wise dose mapping in a reference space.To assess robustness, we conducted a sensitivity analysis on the impact of registration uncertainties on WMT doses in a retrospective cohort of 108 patients with brain metastases: (1) Interpolator impact analysis: dose maps were propagated using four interpolation schemes (linear, B-spline, Lanczos, Gaussian) at both affine and SyN stages, resulting in 16 combinations; (2) Local perturbation analysis: synthetic deformation fields (1 mm radial displacement within a 12 mm shell around the PTV) simulated misregistrations in steep dose gradients.For prospective validation, data from nine patients with pre-SBRT diffusion MRI (b=1000, 64 directions, 2 mm isotropic) were analyzed. Atlas-derived WMTs were transferred into the individual CT space after applying the inverse deformation field from ICBM to CT, allowing direct spatial comparison with diffusion- based tractography. Two reference tractography methods, TractSeg [2] and DeepWMA [3] were used for benchmarking, and overlap was quantified using Dice coefficient for 18 main WMTs.See Figure1.
Conclusion: This fully automated atlas-based pipeline provides a dual benefit: (1) in the standardized ICBM space, it enables large-scale, voxel-based analyses of dose– side-effect correlations; (2) in the patient CT space, it allows direct integration of critical WMTs for tract- aware SBRT planning. With TractoBrainRT, we deliver a ready-to-use, open-source tool to map WMTs without DWI — accessible to the entire radiotherapy community, supporting tract-specific dose guidance. References: [1] Yeh F-C. Population-based tract-to-region connectome of the human brain and its hierarchical topology. Nat Commun 2022;13:4933. https://doi.org/10.1038/s41467-022- 32595-4.[2] Wasserthal J, Neher P, Maier-Hein KH. TractSeg - Fast and accurate white matter tract segmentation. NeuroImage 2018;183:239– 53. https://doi.org/10.1016/j.neuroimage.2018.07.070.[ 3] Zhang F, Cetin Karayumak S, Hoffmann N, Rathi Y, Golby AJ, O’Donnell LJ. Deep white matter analysis (DeepWMA): Fast and consistent tractography segmentation. Med Image Anal 2020;65:101761. https://doi.org/10.1016/j.media.2020. 101761. Keywords: Atlas registration, white matter, brain RT Generalizability of deep learning networks for synthetic CT generation from limited FOV CBCT of the pelvic region in Carbon Ion Radiotherapy Maksym Hladchuk 1 , Giovanni Parrella 1 , Francesca Camagni 1 , Anestis Nakas 1 , Silvia Molinelli 2 , Alessandro Vai 2 , Alfredo Mirandola 2 , Mario Ciocca 2 , Andrea Pella 3 , Amelia Barcellini 4,5 , Viviana Vitolo 4 , Jessica Franzetti 4 , Ester Orlandi 4 , Chiara Paganelli 1 , Guido Baroni 1,3 1 Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, Milan, Italy. 2 Medical Physics, National Center for Oncological Hadrontherapy (CNAO), Pavia, Italy. 3 Bioengineering Unit, National Center for Oncological Hadrontherapy (CNAO), Pavia, Italy. 4 Clinical Unit, National Center for Oncological Hadrontherapy (CNAO), Pavia, Italy. Digital Poster 3902
Results: On the retrospective cohort, the pipeline achieved precise spatial registration with excellent cortical alignment (mean Dice = 0.89) with B-spline interpolator for both affine and SyN steps. Sensitivity analyses showed negligible effects on Dmean and Dmin (<0.05 Gy) for all WMTs and limited impact on Dmax (median Δ Dmax <0.1 Gy; 95th percentile <0.3 Gy). Local perturbations caused rare outliers (<2%) with Δ Dmax ≈ 1 Gy for tracts adjacent to the PTV. Atlas- based reconstructions achieved median Dice = 0.45 across all evaluated WMTs vs TractSeg and 0.31 vs DeepWMA, comparable to inter-tractography variability (DeepWMA vs TractSeg: Dice = 0.49; Figure2).
Made with FlippingBook - Share PDF online