Evaluation of semiempirical VMAT dose reconstruction on a patient dataset based on biplanar diode array measurements

We report the results of a preclinical evaluation of recently introduced commercial tools for 3D patient IMRT/VMAT dose reconstruction, the Delta4 Anatomy calculation algorithm. Based on the same initial measurement, volumetric dose can be reconstructed in two ways. Three-dimensional dose on the Delta4 phantom can be obtained by renormalizing the planned dose distribution by the measurement values (D4 Interpolation). Alternatively, incident fluence can be approximated from the phantom measurement and used for volumetric dose calculation on an arbitrary (patient) dataset with a pencil beam algorithm (Delta4 PB). The primary basis for comparison was 3D dose obtained by previously validated measurement-guided planned dose perturbation method (ACPDP), based on the ArcCHECK dosimeter with 3DVH software. For five clinical VMAT plans, D4 Interpolation agreed well with ACPDP on a homogeneous cylindrical phantom according to gamma analysis with local dose-error normalization. The average agreement rates were 98.2% ± 1.3% (1 SD), (range 97.0%-100%) and 92.8% ± 3.9% (89.5%-99.2%), for the 3%/3 mm and 2%/2 mm criteria, respectively. On a similar geometric phantom, D4 PB demonstrated substantially lower agreement rates with ACPDP: 88.6% ± 6.8% (81.2%-96.1%) and 72.4% ± 8.4% (62.1%-81.1%), for 3%/3 mm and 2%/2 mm, respectively. The average agreement rates on the heterogeneous patients' CT datasets are lower yet: 81.2% ± 8.6% (70.4%-90.4%) and 64.6% ± 8.4% (56.5%-74.7%), respectively, for the same two criteria sets. For both threshold combinations, matched analysis of variance (ANOVA) multiple comparisons showed statistically significant differences in mean agreement rates (p < 0.05) for D4 Interpolation versus ACPDP on one hand, and D4 PB versus ACPDP on either cylindrical or patient dataset on the other hand. Based on the favorable D4 Interpolation results for VMAT plans, the resolution of the reconstruction method rather than hardware design is likely to be responsible for D4 PB limitations.