Session Item

Following the 2018 ESTRO physics workshop, a working group was established for studying problems related to Electronic Portal Imaging Device (EPID) transit in-vivo dosimetry implementation. In this study, a QA measurement protocol, to evaluate the accuracy and the sensitivity of any commercial EPID software, has been developed and tested at multi-institution level.

The accuracy of the EPID algorithm was evaluated in two homogeneous slab phantoms, with 10 cm and 30 cm thickness for 2x2, 5x5, 10x10 and 20x20 cm2 square fields. The field isocenters were set at the phantom center. For Forward-Projection (FP) algorithms, 2D gamma agreement index (GAI) was computed between the EPID images, and the FP algorithms prediction at 2%G/2mm, 3%G/2mm 5%G/2mm. For Back-projection algorithms (BP) the 2D GAI was evaluated between the treatment planning system (TPS) 2D computed dose distribution in the coronal plan containing the isocenter, and the corresponding 2D planar dose reconstructed by the EPID software. The EPID algorithms accuracy was further evaluated for clinical plans in the CIRS lung phantom for static fields conformal technique (3DCRT) and intensity modulated arc technique (VMAT). The sensitivity analysis was carried out in the lung phantom. Three classes of errors have been directly implemented: setup errors were simulated by translating the phantom 5, 10 and 15 mm away from the isocenter; anatomical variations were simulated by applying 1-2 and 3 cm bolus on the phantom top, and finally, delivery errors were simulated by changing the collimator angle by 5, 10 and 15 degrees. All errors were applied both to the 3DCRT and VMAT plans. Each error was simulated in the TPS system. Errors producing dose differences in the PTV; ΔD50%>3.5% or ΔD98%>7% or ΔD2%> 7% were considered relevant. The sensitivity of EPID algorithms to flag relevant errors was evaluated.

Three centers, two using FB, one using BP, produced full results, another center, equipped with a FB software, produced partial results for lack of the anthropomorphic phantom. The accuracy in the homogeneous phantom was evaluated by 6 FP and 3 BP algorithms at different beam energies. All algorithms produced acceptable results at 5%/2mm with GAI> 85%, for all field s except the 2x2 where lower accuracy is shown (FIG1a). In the lung phantom 2 FP and 1 BP algorithm was tested, showing a greater accuracy for the clinical VMAT plans, where all algorithms produced GAI> 95% for the most stringent 2%/2mm criterion (FIG1b.). The sensitivity analysis showed higher sensitivity of intercepting relevant errors for the 3DCRT plans with respect to the VMAT plans for all tested algorithms (FIG2).

FIG1

FIG2

We’ve developed a system to evaluate the implementation of EPID in-vivo dosimetry algorithms. Not all systems seem to provide satisfactory results in accuracy and sensitivity.