S2058
Physics - Image acquisition and processing
ESTRO 2026 Figure 2: Comparison of registrations using spCBCTs (a, d, g), the sCBCTs generated by the Vox2Vox model (b, e, h) and the clinical CBCT from Varian Medical Systems (c, f, i). The registered CBCTs are overlayed with the corresponding planning CT scan. The overlaying CT scan is displayed at the HU range [- 800,2000] in red with transparency. The CBCTs are displayed at the HU range [-1000, 2000]. Conclusion: Generative DL models can synthesize sCBCTs from spCBCTs, demonstrating potential for imaging dose reduction in IGRT. While image quality was not yet clinically sufficient, likely due to limitations in open- source reconstruction, further work with advanced reconstruction methods and larger datasets may enable clinically viable ultra low-dose CBCT generation. Keywords: Generative Deep Learning Models, dose- reduced CBCT Optimizing Material Decomposition and Dose Efficiency in Dual-Energy X-ray Imaging on a Flat- Panel Detector Nawal Alqethami 1 , Mohamed Ali Bnouni 1 , Wentao Xie 1 , Vanessa da Silva Mendes 2 , Michael Reiner 2 , Philipp Steininger 3 , Prasannakumar Palaniappan 1 , Marco Riboldi 1 1 Department of Medical Physics, Ludwig-Maximilians- Universitat Munchen, Munich, Germany. 2 Department of Radiation Oncology, University Hospital, LMU Munich, Munich, Germany. 3 medmedPhoton GmbH, medmedPhoton GmbH, Salzburg, Austria Digital Poster 1371 Purpose/Objective: This study evaluates linear and cubic models for calibrating a flat-panel dual-energy (DE)X-ray detector, comparing their ability to suppress bone and enhance tumor visibility to theweighted logarithmic subtraction (WLS) method. Image quality and radiation dose werealso assessed to determine the overall efficiency of each model. Material/Methods: To calibrate the flat-panel detector, planar low-energy (LE, 60 kVp, 15 mA, 17 ms) and high-energy (HE, 120 kVp, 15 mA, 12 ms) images were acquired with a total exposure of 0.435mAs. Variable thicknesses of aluminum sheets ([0 to 3.6] cm at 0.2 cm steps, simulating bone)and solid water plates ([0 to 18] cm at 1.0 cm steps, simulating soft tissue) were used in allpossible combinations, resulting in 361 DE measurements (including both HE and LE). Meanpixel intensities around the flat-panel focal spot were calculated to reduce noise and establishbaseline incident intensities for material decomposition. To test
Made with FlippingBook - Share PDF online