PET image quantification using regularized image reconstruction

Objectives Regularized image reconstruction has the advantage of reaching convergence while suppressing noise. Time of flight imaging (TOF) increases the SNR of reconstructed images compared to non-TOF imaging. Point Spread Function (PSF) modeling has been shown to recover image sharpness. The objective of this study is to investigate the interplay of these three advances in image reconstruction on the SUVmax measurements of lesions using a phantom study.

Methods A NEMA IEC phantom with six spheres (10 - 37 mm diameter) and a lung insert was imaged on a GE 710 PET/CT scanner. Phantom background was filled with an activity concentration of 0.1 uCi/ml. Three sphere to background ratios were investigated (4, 8 and 10:1). For each case, ten noise realizations of equivalent integral counts were acquired for a total of thirty realizations. Seven reconstructions for each noise realization were performed: Time of flight(TOF) Regularized with Beta values (150, 350, and 550)(REG_TOF), non-TOF Regularized with Beta values(150, 350, and 550)(REG_NT), and TOF PSF OSEM(OSEM_TOF_PSF) using 3it, 24 ss, and a 6mm Gaussian filter, for a total of 210 reconstructions. All regularized reconstructions had PSF. For each case the recovery coefficient of each sphere is calculated as the ratio between measured SUVmax and the known sphere to contrast activity ratio. For each sphere to background ratio, sphere size, and reconstruction algorithm, the average recovery coefficient (ARC) of the ten noise realizations was calculated and the data was plotted.

Results General trends of the plots show that as sphere to background ratio increases there is a marked decrease in the underestimation of the lesion SUVmax with decreasing sphere sizes for all tested reconstruction algorithms. Furthermore, TOF_REG showed a marked positive bias in SUVmax for the 10 mm sphere with increasing sphere to background ratio. The bias was further increased with a lower regularization beta value as expected. Similar results with REG_NT were only observed with the lowest beta value of 150. For example for the 4:1 ratio, the ARCs for REG_NT (150, 350, and 550) were 0.93, 0.61, and 0.50; while for TOF_REG (150, 350 and 550) the ARCs were 1.15, 0.84, and 0.68; and for OSEM_TOF_PSF the ARC was 0.68. For the 8:1 ratio, the ARCs for REG-NT (150, 350, and 550) were 1.59, 1.07, and 0.73; while for TOF_REG (150, 350 and 550) the ARCs were 1.63, 1.33, and 1.07; and for OSEM_TOF_PSF the ARC was 0.75. For the 10:1 ratio, the ARCs for REG_NT (150, 350, and 550) were 1.65, 1.21, and 0.87; while for TOF_REG (150, 350 and 550) the ARCs were 2.02, 1.73, and 1.45; and for OSEM_TOF_PSF the ARC was 0.90.

Conclusions Regularized reconstruction in combination with TOF and PSF has the tendency to increase the SUVmax with increasing lesion to background ratio particularly for small lesions which could ultimately negatively impact patient management.