S2069
Physics - Image acquisition and processing
ESTRO 2026
Mini-Oral 2791 Introduction of the world's largest temporal FDG- PET/CT oligometastatic disease dataset and required tools Maksym Fritsak 1 , Hubert S. Gabry ś 1 , Sebastian M. Christ 1 , Maximilian Rokuss 2 , Nicolas Martz 1,3 , Murong Xu 4 , Bjoern Menze 4 , Martin Huellner 5 , Matthias Guckenberger 1 , Stephanie Tanadini-Lang 1 1 Department of Radiation Oncology, University Hospital Zurich, Zurich, Switzerland. 2 Department of Medical Image Computing, German Cancer Research Center, Heidelberg, Germany. 3 Department of Radiation Oncology, Institut de Cancérologie de Lorraine, Vandœuvre - Lès - Nancy, France. 4 Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland. 5 Department of Nuclear Medicine, University Hospital Zurich, Zurich, Switzerland Purpose/Objective: Oligometastatic disease (OMD) represents an intermediate cancer state with potential for curative treatment, unlike poly-metastatic disease, which is typically managed palliatively. Defining OMD solely by the number of metastases at one single time point may be too simplistic and exclude patients amenable for curative-intent therapy. This work aimed to establish a longitudinal FDG-PET/CT database to study and understand the temporal course and dynamics of OMD and develop tools for its quantitative analysis. Material/Methods: The database includes 877 patients ( ≥ 2 PET/CTs) diagnosed with OMD before 2023 at University Hospital Zurich, based on FDG-PET/CT imaging (Fig. 1), defined by the presence of one to five distant metastases; 349 patients have ten or more PET/CT scans acquired during their course of disease. Patients were selected irrespective of cancer diagnosis or cancer treatment strategy. A vendor-specific strategy was developed to convert PET images from native units to body-weight–normalized SUV [1]. The image biomarker extraction tool Z-Rad [2] was developed in compliance with Image Biomarker Standardization Initiative (IBSI), supporting multiple imaging modalities. Different normal anatomy segmentation models were validated, as well as different cancer lesion segmentation models, leading to the selection of the best normal anatomy segmentation model and the development of a new lesion segmentation model. Results: The vendor-specific SUV conversion strategy [1] covered major vendors: GE, Siemens, and Philips. Validation on 1,658 FDG-PET scans from different vendors demonstrated consistency with vendor DICOM conformance statements, whereas QIBA strategies, commercial, and public implementations showed deviations of up to 32%. Z-Rad [2], an open-
optimised clinical protocols, modifications applied, and associated scan time reductions is provided in Table 1 and some example images of the CS-SENSE and AI CS-SENSE optimised images are shown in figure 1. Optimised sequences maintained or improved image quality, with no clinically significant artefacts observed, including in patients with metallic implants. Geometric distortion remained within 2 mm for all sequences. Clinical review confirmed that all sequences were suitable for target delineation.
Conclusion: Integration of AI based Compressed SENSE into radiotherapy MRI protocols is feasible, robust, and clinically valuable. AI-based acceleration enables faster imaging without compromising quality or geometric fidelity, supporting departmental efficiency. This is likely to improve patient experience, decrease motion artefacts and increase patient throughput. Keywords: MRI, AI, optimisation
Made with FlippingBook - Share PDF online