A Vendor-Level Standardized Epid-Based Virtual Audit Framework for Radiotherapy
Abstract
Purpose
This study proposes and evaluates a framework for EPID-based virtual audit in radiotherapy using a model constructed from multi-institutional data and validated through comparison with machine-specific models.
Methods
Sixty-two datasets from 15 institutions were analyzed, covering Varian and Elekta, multiple photon energies, and both FF and FFF beams. For EPID response kernel estimation, opening density matrices (ODMs) were generated from DICOM-RT plan files. Corresponding EPID fluence images were acquired and spatially aligned with ODMs. Double-Gaussian EPID response kernel parameters were optimized by fitting kernel-applied ODMs to measured EPID fluence, modeling edge-smoothing effects. For gamma analysis, single full-arc VMAT treatment plans meeting predefined dose–volume constraints were created on a phantom geometry incorporating a C-shaped planning target volume and an organ at risk. The resulting DICOM-RT plan and EPID data were processed using identical workflows. The optimized EPID response kernel was applied to the ODMs to generate DICOM-RT plan-based fluence. Machine-specific parameters were first applied to verify reproducibility between DICOM-RT plan-based and EPID-derived fluences. Subsequently, averaged parameters were applied to evaluate the feasibility of a common parameter set. Due to differences in parameter trends, vendor-level analyses were performed for Varian (N = 33) and Elekta (N = 29) systems. Intensity scaling, resampling, and gamma analysis (3%/2 mm criterion, 10% intensity threshold) were performed.
Results
Using machine-specific parameters, the median pass rates (Interquartile Range) were 99.9% (99.7–100.0%) for Varian, and 98.9% (96.2–99.5%) for Elekta. Using averaged parameters, the median pass rates were 99.9% (99.7–100.0%) for Varian, and 98.5% (96.4–99.8%) for Elekta.
Conclusion
The proposed framework achieved reproducible fluence comparisons by applying a model constructed from multi-institutional data and validated against machine-specific models. Averaged parameters demonstrated consistent performance, supporting the feasibility of vendor-level standardized EPID-based virtual audit across multiple institutions.