S1463
Interdisciplinary - Other
ESTRO 2026
limited-angle CBCT (LA-CBCT) scanning is critical for non-coplanar SRS/SBRT, as it helps avoid potential collisions while ensuring treatment delivery accuracy. However, severe data occlusion outside the scanning range makes high-fidelity reconstruction extremely challenging. Moreover, the requirement for high- resolution volumetric CBCT further increases the computational burden due to GPU memory limitations when using deep learning methods. To address these challenges, this work proposes leveraging both image- domain and projection-domain prior information of the target patient and develops an efficient implicit neural representation (INR)–based sinogram inpainting framework for patient-specific, high-fidelity
Conclusion: We propose a novel dual-domain prior-guided sinogram inpainting technique for LA-CBCT image reconstruction, enabling high-quality and stable imaging for non-coplanar SRS/SBRT. Keywords: non-coplanar SRS/SBRT, limited-angle CBCT Digital Poster Highlight 4888 A 2025 IPEM survey of MRI use in UK external beam radiotherapy treatment planning Richard Speight 1 , Sarah Allwood-Spiers 2 , Michael Dubec 3,4 , Ben George 5 , Christopher Hamill-Taylor 6 , Stephen Hedley 7 , Lisa McDaid 8 , Maria Schmidt 9 , Jonathan Wyatt 7 , Trina Herbert 10 1 Medical Physics and Engineering, Leeds Cancer Centre, Leeds, United Kingdom. 2 Department of Clinical Physics and Bioengineering, , NHS Greater Glasgow and Clyde, Glasgow, United Kingdom. 3 Christie Medical Physics and Engineering, The Christie NHS Foundation Trust, Manchester, United Kingdom. 4 Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom. 5 MR-Linac Service Development, GenesisCare, Oxford, United Kingdom. 6 Tayside Cancer Centre, NHS Tayside, Dundee, United Kingdom. 7 Northern Centre for Cancer Care, Newcastle upon Tyne Hospitals NHS Foundation Trust, Newcastle, United Kingdom. 8 Radiotherapy, The Christie NHS Foundation Trust, Manchester, United Kingdom. 9 retired NHS Clinical Scientist, na, London, United Kingdom. 10 Radiotherapy, The Royal Marsden NHS Foundation Trust, London, United Kingdom Purpose/Objective: Magnetic resonance imaging (MRI) offers superior soft tissue contrast compared to computed tomography (CT) and is increasingly integrated into external beam radiotherapy (EBRT) planning. Following the 2018 Institute of Physics and Engineering in Medicine (IPEM) survey (Speight et al 2019), a 2025 national survey was conducted to assess changes in MRI access, utilisation, quality assurance (QA) compliance, and adherence to 2021 IPEM guidance (Speight 2021) across UK
image reconstruction. Material/Methods:
We develop a patient-specific sinogram inpainting framework that leverages both image-domain and projection-domain information acquired from a prior coplanar scan. An INR network is pretrained to encode the full-view projections at the coplanar position. During testing, the network is fine-tuned to synthesize full-view projections for limited-angle acquisitions under non-coplanar positions. The dual-domain prior guidance is achieved through two key mechanisms: (1) the fine-tune loss incorporates volumetric information obtained by warping the coplanar image to the non- coplanar geometry via rigid registration based on couch motion readouts, thus providing anatomical constraints for projection synthesis; and (2) the network initialization from the pretrained model embeds prior full-view projection knowledge, ensuring consistent data fidelity. The proposed approach is evaluated by comparing the generated sinograms with the ground-truth measurements, as well as by assessing the corresponding FDK reconstructions from both generated and ground-truth projections. Results: We conducted experiments on a head phantom scanned using a Varian TrueBeam system. A full-scan acquisition with 200° coverage at the initial coplanar position was used for training. Two non-coplanar setups involving a 10° couch rotation were considered. To evaluate the sinogram inpainting performance, we tested various limited-angle scenarios with angular coverages ranging from 15° to 90°. The proposed method achieved stable reconstruction quality across different angular ranges, in contrast to the original limited-angle-based reconstruction, which exhibited severe artifacts. Figure 1 illustrates the comparison of reconstruction quality between the two methods.
Made with FlippingBook - Share PDF online