S2164
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
Conclusion: The fully automated oART workflow for rectal cancer on the Ethos system demonstrated target coverage in 16% fewer sessions compared to clinically acceptable plans. This reduction was mainly attributed to inaccuracies when air pockets were present and differences in contour definition between the AI model and institutional practice. These findings highlight the need for further advancements in AI-based oART to achieve reliable clinical use. The study remains limited by using clinically accepted contours that may not represent the gold standard. References: [1] de Jong R, Visser J, van Wieringen N, Wiersma J, Geijsen D, Bel A. Feasibility of Conebeam CT-based online adaptive radiotherapy for neoadjuvant treatment of rectal cancer. Radiation Oncology. 2021;16:1–11.[2] Intven MPW, de Mol van Otterloo SR, Mook S, Doornaert PAH, de Groot-van Breugel EN, Sikkes GG, et al. Online adaptive MR-guided radiotherapy for rectal cancer; feasibility of the workflow on a 1.5T MR-linac: clinical implementation and initial experience. Radiother Oncol. 2021;154:172– 8.[3] Azzarouali S, Goudschaal K, Visser J, Daniels L, Bel A, den Boer D. Minimizing human interference in an online fully automated daily adaptive radiotherapy workflow for bladder cancer. Radiat Oncol. 2024;19(1):138. Keywords: Online adaptive, rectal cancer, AI Digital Poster 3813 Diffusion-based synthetic-CT generation from pelvis male CBCT Lorenzo Colombo 1 , Audrey Duran 2 , Quentin Spinat 2 , Pierre Olléon 2 , Sami Romdhani 2 , Olivier Teboul 2 , Nikolaos Paragyios 3 , Pascal Fenoglietto 4 , Moritz Westermayer 5,6 , Marie-Claude Biston 5,6 1 Clinical affairs, TheraPanacea, Paris, France. 2 AI engineering, TheraPanacea, Paris, France. 3 CEO, TheraPanacea, Paris, France. 4 Departement of radiation oncology, Institut du Cancer de Montpellier, Montpellier, France. 5 Department of medical physics, Centre Léon Bérard, Lyon, France. 6 CNRS UMR5220, CREATIS Inserm 1044, Lyon, France Purpose/Objective: Cone-beam CT (CBCT) imaging is routinely used in radiation therapy for patient positioning, but its broader clinical application remains limited by suboptimal image quality, spatial resolution, and field of view. Extending CBCT beyond setup verification could substantially advance adaptive radiotherapy. Achieving this requires enhanced image fidelity to support accurate organ-at-risk delineation, precise dose computation, and adaptive treatment
replanning. This study presents and clinically validates an artificial intelligence (AI)–based framework that generates high-quality synthetic CT (sCT) images directly from CBCT scans. By addressing key CBCT limitations, the proposed approach demonstrates strong potential to enable fully CBCT-driven adaptive radiotherapy for pelvic male cancer patients. Material/Methods: Developing a 3D multimodal translation model for medical imaging presents challenges, including the computational demands of volumetric data, the scarcity of high-quality paired datasets, and the need to maintain quantitative accuracy for clinical use. To address these, a 3D latent diffusion model was designed. Its variational autoencoder (VAE) component achieves strong data compression while maintaining reconstruction fidelity and computational efficiency. The diffusion model was first trained in an unsupervised manner on unpaired data, then fine- tuned for cross-modality translation using limited paired samples. A ControlNet-based conditioning strategy was employed to ensure anatomical consistency and quantitative reliability of the generated images. The unsupervised model was trained on approximately 34,000 CT scans (50% pelvis, 37.5% head and neck, 12.5% thorax) to ensure anatomical diversity, followed by fine-tuning with 325 pelvic male cancer patients (3,406 paired datasets). For clinical evaluation, an independent cohort of 20 pelvic male cancer patients treated across three European cancer centres was selected. Planning CTs were deformably registered to CBCTs (warped CTs, wCTs) to account for anatomical and positional variations. Treatment plans were re-optimised and recalculated on both wCTs and sCTs for dosimetric comparison. Results: Median relative differences in dose–volume histogram (DVH) parameters between sCTs and wCTs were minimal (0.12–0.21% - Table 1), with the largest difference for D2% (0.21%) and the lowest for Dmean and D98% (0.12%). Median gamma pass rates reached 99.99% (2%/2 mm) and 100% (3%/3 mm) at a 1% low- dose threshold (Table 2), confirming excellent dosimetric agreement.
Conclusion: This study demonstrates the feasibility and accuracy of AI-generated sCTs from CBCT images in pelvic male cancer radiotherapy. The results confirm their potential to enhance adaptive planning, streamline clinical workflows, and support fully CBCT-based personalised radiotherapy.
Made with FlippingBook - Share PDF online