Scanner dependent noise properties of the Q. Clear PET image reconstruction tool

In this paper we compare the noise properties of the newly introduced Q.Clear image reconstruction algorithm and conventional OSEM using the ACR phantom measured on the GE Discovery 600 and 690 PET/CT scanners with various count statistics. In the D600 measurement the SNR decreases from 31.5±1.3 and 26.9±1.6 to 5.3±0.4 and 5.5±0.3 for the Q.Clear (with regularization strength parameter beta=350) and the OSEM images, respectively as the statistics decreases from 100% to 5%. The average mean hot concentration recovery values of the three biggest rods over all time slices are 0.57±0.01, 0.66±0.04 and 0.8±0.02 for Q.Clear (with beta=350) and 0.63±0.03, 0.71±0.05 and 0.83±0.03 for OSEM, respectively. The SUV_max related maximum hot activity concentration recovery calculated from the maximum concentration averaged for these 3 rods increased from 0.78 and 0.88 to 1.1 and 1.5, respectively for the two algorithms. In the D690 measurement the SNR decreases from 26.6±1.4 and 31.0±1.3 to 3.8±0.4 and 6.3±0.5 for Q.Clear and the OSEM, respectively. The mean hot concentration recovery values are 0.66±0.04, 0.8±0.02 and 0.92±0.01 for Q.Clear and 0.65±0.04, 0.78±0.02 and 0.9±0.02 for OSEM, respectively. The maximum hot activity concentration recovery increased from 0.98 and 0.93 to 2.3 and 1.4, respectively for the two algorithms. In summary for the two scanners the SNR of Q.Clear matches the SNR of OSEM with different regularization strength applied. Although the mean concentration recovery values were not significantly different in either measurement (with beta=350) the maximum recovery was affected by noise and was closer to one with Q.Clear in the case of D600 and with OSEM for the D690. Q.Clear resulted in significantly better cold contrast in bone for both scanners (0.053±0.004 vs. 0.106±0.01 and 0.0±0 vs. 0.144±0.004) and in air for the D690 (0.305±0.004 vs. 0.380±0.005).