Figure 1 shows the phantom images reconstructed with different iterations and with HS/UHS, while Figs. 2–4 include an overview of axial slices and MIPs of the phantom images reconstructed with different number of iterations and reconstruction times.
Hot phantom
The RC for UHS was highest for all the spheres from biggest to smallest with 8 iterations, 50-minutes acquisition and 2mm FWHM gaussian filter (0.87, 0.70, 0.63, 0.58, 0.52 and 0.53). For the same gaussian filter, the RC was lowest for 2 iterations (0.84, 0.68, 0.63, 0.53, 0.45 and 0.36). For the 5-minutes acquisition the RC for the spheres were 0.95, 0.69, 0.76, 0.49, 0.49, 0.36 and 0.97, 0.74, 0.72, 0.47, 0.44, 0.22 for 8 iterations and 2 iterations respectively (Fig. 5–6).
The SNR was found to be highest for UHS 2 iterations, 50-minutes acquisition, with 6 mm FWHM gaussian filter (49.51, 38.83, 33.76, 29.67, 26 and 17.57). For the same filter, the SNR was lowest for 8 iterations (27.21, 21.47, 18.82, 16.98, 15.82 and 11.49). For 2 mm FWHM gaussian filter, which is more clinically relevant due to sharper images the SNR for 2iterations were 27.76, 21.86, 19.42, 17.64, 16.62 and 11.65 (for HS 22.13, 17.36, 15.72, 13.57, 13.24 and 10.2 respectively). The SNR for the UHS 5-minutes acquisition was 8.81, 6.47, 6.30, 5.28, 4.59 and 3.94 respectively. The SNR diagrams can be found in Fig. 7.
The percent background variability for UHS-50minutes acquisition was highest for all the spheres, from largest to smallest, for the 8 iterations with 2 mm FWHM gaussian filtering (76.71, 82.03, 84.78, 77.83, 78.61 and 82.46%) and for the same filter lowest for 2iterations (27.59, 27.08, 27.36, 26.44, 30.11 and 33.51%). For the same parameters but with HS reconstruction the percent background variability was (97.53, 104.49, 104.37, 98.46, 90.56 and 87.43%) and (35.00, 36.05, 33.99, 33.41, 35.97 and 36.18%) for 8 and 2 iterations respectively. For the UHS 5-minutes acquisition the percent background variability for 2 iterations and 2 mm FWHM gaussian filter was 99.54, 95.72, 93.78, 75.5, 111.99 and 70.48%. Lung error was measured 9% at the lowest for UHS 50-minutes acquisition, 8 iterations with 2 mm FHWM gaussian filtering and 18% for 2 iterations. For the 5-minutes acquisition it was measured 5% and 6% respectively. For the HS 50-minutes acquisition it was measured 13% and 19% for the 8 and 2 iterations respectively.
To test the performance of the product of iterations and subsets on RC at low count statistics we reconstructed the phantom data for the 5-minutes acquisition also with 10 iterations and 1 subset. The RC were 1.07, 0.88, 0.84, 0.46, 0.47 and 0.27 in contrast to 0.97, 0.74, 0.72, 0.47, 0.44 and 0.22 for 2 iterations for the biggest to the smallest sphere.
Cold phantom
Similar to the hot phantom, the RC was highest for all spheres with 8 iterations, 50-minutes acquisition and 2 mm FWHM gaussian filter (0.90, 0.71, 0.62, 0.51, 0.52 and 0.32). For the same gaussian filter, the RC was lowest for 2 iterations (0.86, 0.67, 0.59, 0.47, 0.46 and 0.26). For the 5-minute acquisition the RC for the spheres were and 0.83, 0.55, 0.51, 0.39, 0.34, 0.12 and 0.80, 0.53, 0.48, 0.37, 0.31 and 0.10 for 8 iterations and 2 iterations respectively (Fig. 5–6).
The SNR was highest for UHS 2 iterations, 50-minutes acquisition, with 6 mm FWHM gaussian filter (564.00, 361.27, 306.54, 233.08, 206.83 and 115.37). For the same filter, the SNR was lowest for 8 iterations (318.39, 250.35, 214.29, 164.30, 151.18 and 91.21). For 2 mm FWHM gaussian filter, which provides higher resolution the SNR for 2 iterations were 327.86, 255.27, 222.68, 178.26, 174.60 and 100.47. The SNR for the 5-minutes acquisition were 305.84, 203.09, 183.55, 139.75, 117.83 and 36.89. For the HS mode the SNR for 2 iterations, 50-minutes acquisition and a 2 mm FWHM gaussian filter was 217.62, 172.25, 147.47, 120.79, 125.80 and 75.78 (Fig. 7).
The percent background variability for UHS 50-minutes acquisition was highest for the 8 iterations with 2 mm FWHM gaussian filtering (475.3, 442.56, 545.12, 266.75, 214.96 and 144.3%) and for the same filter lowest for 2 iterations (27.59, 27.08, 27.36, 26.44, 30.11 and 33.51%). For the same parameters but with HS reconstruction the percent background variability was (577.45, 730.61, 547.26, 233.73, 401.87 and 244.38%) and (115.14, 123.05, 126.9, 78.78, 121.35 and 66.19%) for 8 and 2 iterations respectively. For the UHS 5-minutes acquisition the percent background variability for 2 iterations and 2 mm FWHM gaussian filter was 603.38, 701.46, 454.39, 406.59, 263.51 and 145.73%. Lung error was measured 105% at the lowest for UHS 50-minutes acquisition, 8 iterations with 2 mm FHWM gaussian filtering and 120% for 2 iterations. For the 5-minutes acquisition, it was measured 308% and 334% respectively.
The 10 iterations 1 subsets produced the following RC for a 5 minutes acquisition: 0.78, 0.54, 0.45, 0.4, 0.35 and 0.12 in contrast to 0.8, 0.53, 0.48, 0.37, 0.31 and 0.1 for the 2 iterations 5 subsets for the biggest to the smallest sphere.
Patient data
As a convention, the values presented are in the following sequence: 20, 10, 5 and 1 reconstruction times. The SNR with UHS, for 2 iterations was 19%, 20%, 24% and 31% higher compared to 4 iterations (Fig. 8). The reconstruction with 2 iterations and UHS produced 2%, 10%, 17% and 61% higher SNR compared to HS. Between reconstruction matrices of 220 x 220 and 440 x 440, the SNR percentage difference was lower than 10% for all reconstruction times for 2 iterations and UHS. The background liver VOI mean did not show any differences between 2 and 4 iterations for both UHS and HS. The coefficient of variation (COV) was measured 6%, 15%, 16% and 22% lower for UHS to HS for 2 iterations and 13%, 22%, 20% and 26% for 4 iterations. The COV was 31%, 38%, 48% and 64% lower for UHS 2 iterations compared to 4 iterations. The reconstruction with UHS 220 x 220 matrix exhibited 6%, 5%, 7% and 16% lower COV compared to the 440 x 440 matrix.