S2163
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
during clinical treatment. Organs-at-risk (bladder, bowel, rectum) were automatically delineated by AI and target volumes automatically propagated without manual adjustment. For each fraction, an adapted plan was automatically generated based on a standardized clinical goal template. Geometric accuracy between automated and clinically accepted contours was assessed with the Dice Similarity Coefficient (DSC) and 95% Hausdorff distance (HD95). Dosimetric plan quality was evaluated by comparing target coverage of the simulated automated plan versus the clinically delivered plan, with the dose volume histograms (DVHs) both calculated using the clinically accepted contours as reference. Results: The geometric assessment using DSC and HD95 is shown in Table 1. Figure 1 illustrates the target coverage in terms of V95%. Dosimetric evaluation demonstrated median and range V95% of 99.9% [9.4- 100.0], 99.6% [92.5-100.0] and 99.8% [77.6-100.0] for respectively mesorectum inferior, superior and lymph node pelvic. Clinical dose constraints (V95% ≥ 98%) were met in 35 out of 50 automated plans and in 43 out of 50 clinically accepted plans.
Conclusion: This work demonstrates the feasibility and dosimetric accuracy of AI-generated sCTs from CBCT images in lung cancer radiotherapy. The results highlight the potential of this approach to improve image quality, enable accurate dose calculation, and streamline adaptive workflows for personalised and efficient treatment delivery. Keywords: Synthetic-CT, adaptive, thorax Can AI be trusted for fully automated CBCT-guided online adaptive radiotherapy in rectal cancer? A. Maasland 1 , R. de Jong 2 , J. Visser 3 , N. van Wieringen 3 , N. Bijker 3 , E.D. Geijsen 3 , A. Bel 3 , D. den Boer 1 1 Department of Radiation Oncology, Cancer Center Amsterdam, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, Netherlands. 2 Department of Radiation Oncology, Leiden University Medical Center, Leiden, Netherlands. 3 Department of Radiation Oncology, Cancer Center Amsterdam, Amsterdam UMC, University of Amsterdam, Amsterdam, Netherlands Purpose/Objective: The large interfractional shape changes of target volumes in rectal cancer illustrate the need for online adaptive radiotherapy (oART) [1, 2]. Current online adaptive workflows remain semi-automated due to manual contour adjustments, which prolong Digital Poster Highlight 3799 treatment times and increase staff workload. A fully automated workflow could reduce both [3]. This study investigates the geometric and dosimetric performance of a fully automated, AI-driven oART workflow without manual adjustments for rectal cancer on a CBCT-guided system. Material/Methods: Retrospective data were collected from locally advanced rectal cancer patients receiving CBCT-guided oART between January and May 2024 (Ethos Therapy™, Varian a Siemens Healthineers Company, USA). Ten patients, treated with twenty-five fractions of 2 Gy, were included. For each patient, fractions 1, 7, 13, 19 and 25 were selected to account for weekday variation and represent the course, resulting in a total of fifty treatment sessions. The fully automated workflow was simulated in the emulator software (v2.0) for both the planning CT and CBCTs acquired
Made with FlippingBook - Share PDF online