ESTRO 2025 - Abstract Book

S3625

Physics - Quality assurance and auditing

ESTRO 2025

Material/Methods: A head and neck treatment plan was artificially manipulated to create 8 different error types (Fig. 1). For each error type 3-4 severity grades (e.g., grade 3=local dose deviation >3%) were generated, totaling 29 scenarios. Clinical PB-PSQA included absolute point dose measurements in water (Semiflex, PTW), 2D relative dose measurements (Lynx, IBA), and manual plan parameter checks. Measurements were taken at both ideal (expected location of introduced error) and standard-clinical (arbitrarily selected) locations, accepting point dose deviations within ±4% and global 2D Gamma pass rates of Γ(3%/3mm, cut-off 10%)>95%. The phantom-less PSQA included log-file-based QA (LFQA) with (I) spot-level statistics, accepting spot position/weight differences of up to 2mm/0.05MU, respectively, and (II) dose reconstruction on the planning-CT with local 3D Γ(2%/2mm, cut-off 95%)>95%. Script-based plan parameter comparisons were performed, reproducing the currently employed manual checks. Error detection sensitivity was defined as the percentage of detected error scenarios for both PL-PSQA and PB-PSQA.

Results: At arbitrary measurement locations, PB-PSQA (dose measurements, manual checks) detected 52% of scenarios, while PL-PSQA (LFQA, automated checks) detected 90% (Fig. 1B). When excluding the manual/automated checks, sensitivity dropped to 10% and 76%, respectively. Hence, LFQA was superior to phantom-based measurements under clinical PSQA conditions. When evaluated at the expected error locations (not clinically realistic), PB-PSQA also achieved an error detection sensitivity of 90%. In this case, point dose measurements aligned with the TPS-calculated and log-file results (Fig. 2A). Nevertheless, Lynx 2D dose measurements failed to detect most scenarios due to more relaxed passing criteria