Evaluation of digitally reconstructed radiographs (DRRs) used for clinical radiotherapy: a phantom study.

Digitally reconstructed radiographs produced from a commercial CT simulator have been evaluated using an in-house test phantom. The phantom consists of a polystyrene cubic block of dimension 15 cm. It contains four test patterns to measure contrast detail, modulation transfer function, ray line divergence accuracy, and spatial distortion. A total of six CT data sets, which vary by CT slice thickness and separation between slices, as well as CT slice reconstruction area have been used to analyze digitally reconstructed radiographs produced by the system. Results show that contrast detail is independent of slice thickness and separation but dependent upon slice reconstruction area for small object diameters (< or = 1.5 mm). Half field (24-cm diameter) reconstruction images provide lower threshold contrasts than full field (48-cm diameter) scans. The modulation transfer function for each data set was calculated and the spatial frequency at which the modulation transfer function is 50% (f50) of the maximum indicates that high contrast resolution depends on slice thickness for both the full and half field reconstructions. For full field scans, f50 values were 0.19, 0.10, and 0.10 line pairs/mm for the 2 mm/2 mm (slice thickness/separation), 5 mm/3 mm, and 5 mm/5 mm data sets, respectively. Similarly, half field f50 values were 0.19, 0.10, and 0.10 line pairs/mm for these same three thickness/separation data sets, respectively. The error in the ray tracing component of the digitally reconstructed radiograph algorithm for source to skin distances between 60 and 200 cm was 1.0 mm while the spatial linearity error was < or = 2.5 mm. Errors in CT simulator collimator and table rotations were calculated by measuring the angle between a grid pattern inlayed onto two orthogonal faces of the phantom and a graphical grid superimposed onto the digitally reconstructed radiograph by the CT simulator software. Measured angular differences were < or = 1.0 degree. The largest error in shifting the CT simulator field isocenter was 2.2 mm and occurred on the 5-mm slice thickness and separation CT data sets.