S2378
Physics - Quality assurance and auditing
ESTRO 2026
Pathology and Spatially-Integrated Omics GSTeP facility, Fondazione Policlinico Universitario A. Gemelli, IRCCS, Rome, Italy. 8 Neuroscience, Catholic University of the Sacred Heart, Rome, Italy Purpose/Objective: To enable the implementation of an MRI-only online adaptive radiotherapy workflow for prostate cancer, where no simulation CT is acquired, an alternative for electron density estimation is essential. Synthetic CTs (sCTs) generated by AI models represent a potential solution. However, beyond model performance, clinical integration requires a stringent, robust quality assurance (QA) framework to assess the reliability, dosimetric accuracy and safety of each sCT method. This study designed and validated an automatic QA pipeline to determine whether a given sCT can be safely adopted within an oAMRgRT workflow for prostate cancer. Material/Methods: Sixteen prostate SBRT patients previously treated with the MRIdian ViewRay system were retrospectively analyzed. The “clinical sCT”, generated with a cGAN model previously selected and internally validated at our institution, was used as reference for dose recalculation. Two additional approaches were generated for QA: a bulk-density sCT, assigning predefined densities to key structures (skin, PTV, prostate, bladder, rectum, femurs and couch top) and an alternative sCT produced by an independent AI model trained on a different dataset. Paired comparisons were performed between the clinical sCT (reference) and each alternative to quantify dosimetric agreement and assess clinical reliability. The automated QA pipeline extracted DVH metrics, computed ROI-wise dose differences and checked for any violated dose guidance ( Δ dose), performed a 3%/2mm gamma-analysis ( γ ), and recorded processing time per patient. Results were summarized using a four-level traffic-light classification: pass, low attention, high attention, fail (Figure 1).
Results: The QA analysis showed excellent agreement between the reference and the alternative sCT, with median dose-guidance differences below 0.2 Gy (Figure 2) and gamma passing rates above 99%. Larger deviations in dose guidance, particularly for the PTV, were observed for the bulk-density sCT, but gamma passing rates were also above 93%. Results confirmed the dosimetric equivalence of the clinical sCT with respect to the other electron-density assignment methods, supporting its suitability for oAMRgRT. The sCT QA outcome was “pass” in 75% of cases, with “low attention” and “high attention” in 6.25% and 18.75% of cases, respectively, and no failures. The average QA processing time was 120 s (IQR =86s -157s) per patient.
Made with FlippingBook - Share PDF online