S2162
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
6 Department of radiation oncology, Institut du Cancer de Montpellier, Montpellier, France Purpose/Objective: Cone-beam CT (CBCT) imaging is widely used in radiotherapy for daily patient setup; however, its clinical application remains limited by suboptimal image quality, spatial resolution, and field of view. Extending CBCT beyond positioning could be key to realising the full potential of adaptive radiotherapy. Achieving this requires improved image fidelity to support accurate organ-at-risk delineation, dose calculation, and treatment replanning. This study presents and clinically evaluates an AI–driven approach to generate synthetic-CT (sCT) images from CBCT, addressing major technical limitations and demonstrating feasibility for enhancing adaptive workflows in lung cancer management. Material/Methods: Developing a 3D multimodal translation model for medical imaging presents challenges, including the computational burden of volumetric data, limited availability of high-quality paired datasets, and the need to preserve quantitative accuracy for clinical use. To overcome these, a 3D latent diffusion model was designed. Its variational autoencoder (VAE) component enables 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 mechanism was implemented to ensure anatomical consistency and quantitative reliability of the generated images. The unsupervised model was trained on approximately 34000 CT scans (50% pelvis, 37.5% head and neck, 12.5% thorax) to ensure anatomical diversity, followed by fine-tuning with 286 thoracic cancer patients (2097 paired datasets). For clinical evaluation, an independent cohort of 20 lung cancer patients treated across three European cancer centres was analysed. Planning-CTs were deformably registered to CBCTs (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 for target volumes between sCTs and wCTs were minimal (0.14–0.40% - Table 1), with the largest observed for Dmax (0.40%) and the lowest for D50% (0.14%). Median gamma pass rates reached 99.75% (2%/2 mm) and 99.94% (3%/3 mm) at a 1% low- dose threshold (Table 2), confirming excellent dosimetric agreement.
tuned for cross-modality translation using limited paired samples. A ControlNet-based conditioning strategy was employed to achieve anatomically consistent and quantitatively reliable image synthesis. The unsupervised model was trained on approximately 34000 CT scans (50% pelvis, 37.5% head and neck, 12.5% thorax) to ensure anatomical diversity, followed by fine-tuning with 248 breast cancer patients (1126 paired datasets). For clinical evaluation, an independent cohort of 20 breast cancer patients from three European cancer centres was used. Planning-CTs were deformably registered to CBCTs to obtain warped-CTs (wCTs), accounting for anatomical and positioning variations. Treatment plans were re-optimised and recalculated on both wCTs and sCTs for dosimetric comparison. Results: Median relative differences in DVH parameters for target volumes between sCTs and wCTs were minimal (0.09–0.19% - Table 1), with the largest difference observed for D98% (0.19%). Median gamma pass rates reached 99.74% (2%/2mm) and 99.96% (3%/3mm) with a 1% low-dose threshold (Table 2), confirming excellent dosimetric agreement. Conclusion: This study demonstrates the feasibility and accuracy of AI-based CBCT-to-CT conversion for adaptive breast radiotherapy. The resulting sCTs provide dosimetric precision comparable to conventional CTs, supporting their integration into a comprehensive adaptive workflow including organ-at-risk delineation, dose simulation, and replanning to enhance personalised and efficient radiotherapy. Keywords: Synthetic-CT, Adaptive, Breast Digital Poster 3785 Diffusion-based synthetic-CT generation from thorax CBCT Lorenzo Colombo 1 , Audrey Duran 2 , Quentin Spinat 3 , Pierre Olléon 3 , Olivier Teboul 3 , Nikolaos Paragyios 4 , Pauline Maury 5 , Pascal Fenoglietto 6 1 Clinical affairs, TheraPanacea, Paris, France. 2 AI engineering, TherePanacea, Paris, France. 3 AI engineering, TheraPanacea, Paris, France. 4 CEO, TheraPanacea, Paris, France. 5 Department of radiation oncology, Institut Gustave Roussy, Villejuif, France.
Made with FlippingBook - Share PDF online