S2054
Physics - Image acquisition and processing
ESTRO 2026
University of Copenhagen, Copenhagen, Denmark. 3 Health Technology, Technical University of Denmark, Kgs. Lyngby, Denmark. 4 Radiology, Copenhagen University Hospital - Herlev and Gentofte, Herlev, Denmark. 5 Danish Centre for Particle Therapy, Aarhus University Hospital, Aarhus, Denmark. 6 Oncology, Copenhagen University Hospital - Rigshospitalet, Copenhagen, Denmark. 7 Oncology, Aalborg University Hospital, Aalborg, Denmark. 8 Clinical Medicine, Aalborg University, Aalborg, Denmark. 9 Oncology, University Hospital of Southern Denmark, Vejle, Denmark. 10 Oncology, Zealand University Hospital, Næstved, Denmark Purpose/Objective: Conventional CT has limited soft tissue contrast, which is why MRI remains the preferred modality for prostate cancer delineation. However, MRI introduces additional workflow complexity, co-registration uncertainties, and resource demands, highlighting the need for a single CT scan for both delineation and dose calculation. Photon-counting and dual-energy CT (PCCT and DECT) represents potential alternatives, both offering improved soft tissue contrast through virtual monoenergetic images (VMI). Additionally, PCCT provides higher spatial resolution and eliminates electronic noise.To evaluate the clinical relevance of these benefits, structured methods are needed to assess DECT and PCCT reconstructions in their ability to visualize key anatomical structures critical for defining the clinical target volume (CTV). The purpose of this study was to develop an expert-informed image quality rating scale to benchmark novel CT imaging performance for prostate delineation. Material/Methods: Five patients with prostate cancer underwent MRI, contrast-enhanced DECT and contrast-enhanced PCCT. VMIs [40, 70] keV were reconstructed using Qr and Br kernels at varying sharpness levels. Eight anatomical structures essential for CTV delineation (as defined by ESTRO ACROP guidelines) were selected. For each structure, three representative CT images were chosen to illustrate each of five Likert scale levels (1 = impossible differentiation and major image noise and/or beam hardening to 5 = sharply demarcated border and no or negligible image noise and/or beam hardening).In total, 120 images were evaluated by ten radiation oncologists and one radiologist from 6 centers. Each expert independently selected the image best representing each scale level. Consensus was defined as ≥ 80% agreement; if unmet, three new images were re-evaluated for that scale level until consensus was reached. Results: A structured 5-point image quality scale was successfully developed for each anatomical structure, providing a standardized framework for assessing
regions (Table 1): MAE from 44 HU (thorax) to 52 HU (head&neck), PSNR above 31 dB, and MS-SSIM > 0.9, indicating high structural and contrast fidelity.
A demonstration video is available at: https://github.com/vboussot/SlicerImpactSynth Conclusion: Slicer IMPACT-Synth provides a transparent and extensible platform for sCT generation in radiotherapy, delivering accurate synthetic images together with dedicated QA tools to verify their anatomical and dosimetric consistency. It enables daily dose recalculation, and combined with the upcoming IMPACT-Reg plugin, will support dose accumulation. Generated sCTs can also be used for automatic segmentation using pretrained CT models, facilitating MR and CBCT contouring and paving the way for trustworthy, clinically integrated AI workflows in adaptive radiotherapy. References: [1] Thummerer, Adrian, et al. "SynthRAD2025 Grand Challenge dataset: Generating synthetic CTs for radiotherapy from head to abdomen." Med Phys (2025).[2] Hémon, Cédric, et al. "Modeling dose uncertainty in cone-beam computed tomography: Predictive approach for deep learning-based synthetic computed tomography generation." PhiRO (2025).[3] Boussot, Valentin et al. "KonfAI: A Modular and Fully Configurable Framework for Deep Learning in Medical Imaging." arXiv (2025).[4] Boussot, Valentin, et al. ”Why Registration Quality Matters: Enhancing sCT Synthesis with IMPACT-Based Registration.” arXiv (2025).[5] Boussot, Valentin, et al. "IMPACT: A Generic Semantic Loss for Multimodal Medical Image Registration." arXiv (2025). Keywords: Synthetic CT, Quality Assurance, 3DSslicer Image quality assessment of photon-counting and dual energy CT for prostate radiotherapy planning Louis MD Teller 1,2 , Cecilie V Henneberg 1,3 , Henriette Lindberg 1 , Vibeke Løgager 4 , Stine E Petersen 5 , Vicki T Taasti 5 , Weronika E Olech 4 , Henriette K Mortensen 1 , Christina B Jakobsen 6 , Jimmi Søndergaard 7 , Andreas Carus 7,8 , Christine V Madsen 9 , Dorthe Yakymenko 10 , Lene S Mouritsen 1 , Line H Dohn 1 , Nicoline Raaschou- Jensen 1 , Claus P Behrens 1,3 , Felix C Müller 4 , Michael B Andersen 4,2 , Gitte F Persson 1,2 , Jens Edmund 1 1 Oncology, Copenhagen University Hospital - Herlev and Gentofte, Herlev, Denmark. 2 Clinical Medicine, Digital Poster 1230
Made with FlippingBook - Share PDF online