S2213
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
adapted based on the DVFs derived from these volumetric images to minimise the dosimetric errors from the treatment plans with a real-time dose calculation algorithm. Deformable MLC tracking was compared against rigid MLC tracking (translational adaptation only) and no tracking (Figure 1). Dosimetric performance was quantified by comparing the simulated delivered dose to the planned dose using 3D gamma analysis (2%/2 mm), normalized root- mean-square error (NRMSE), and PTV D ₉₈ %. The motion model error was quantified as the distance between predicted and ground truth fiducial marker positions.
Conclusion: Deformable dose-guided MLC tracking based on motion-model-derived volumetric imaging substantially improves dose delivery accuracy in the presence of respiratory motion and anatomical deformation. The demonstrated feasibility on patient data and the observed dosimetric benefit represent an essential step toward the clinical implementation of real-time dose-guided radiotherapy. Keywords: motion model, 4DCBCT, 4d dose accumulation Digital Poster 1089 AI-powered workflow for real-time pulmonary tumor detection on stereoscopic DRRs Motchy SALEH Grand Est, CRAN, Nancy, France. Grand Est, Chr Metz Thionville, Metz, France Purpose/Objective: Real-time localization of pulmonary tumors is a key requirement for motion-adapted radiotherapy. This work aims to establish a fully automated workflow for generating training datasets enabling artificial intelligence (AI)-based real-time tumor detection and segmentation in stereoscopic digitally reconstructed radiographs (DRRs) acquired with ExacTrac imaging. Material/Methods: Data from 100 lung cancer patients were included, each with one planning 4D-CT and an average of five 4D-CBCT acquisitions, yielding approximately 500 CBCT datasets. Expansion to 200 patients is planned by February 2026. A Python-based pipeline was
Results: Deformable MLC tracking improved dose accuracy compared to both rigid and no tracking, achieving significantly higher gamma passing rates and lower NRMSE values (p < 0.001). Rigid tracking also outperformed no tracking (p < 0.001). While DVH analysis of PTV D ₉₈ % followed the same trend, differences between deformable and rigid tracking were not statistically significant (p = 0.15; Table 1). The benefit of deformable tracking was more pronounced in lung cases, with mean gamma passing rates of 98% versus 85% for rigid tracking, compared to 93% versus 89% in liver cases. Overall, the dosimetric benefit correlated with both higher anatomical deformability - quantified by the mean Jacobian determinant ( , ) and mean respiratory amplitude ( ). The motion model demonstrated submillimeter geometric accuracy , with a mean error of 0.1 ± 2.9 mm.
Made with FlippingBook - Share PDF online