An evaluation of the effects of PET timing resolution versus NEC on image quality

56 Objectives: Surprisingly it is unknown what sensitivity gains in noise equivalent counts (NEC) are equivalent to an improvement in time-of-flight (TOF) resolution, or vice-versa. For example, when using a standard clinical protocol, one system may have an NEC of 158 kcps and a TOF resolution of 385 ps, while a second system may have an NEC of 49 kcps and a TOF resolution of 325 ps, and it is unknown a priori how the relative task performance will compare if all other factors are equal. We present initial evaluations with task-based metrics for lesion detection using simulations of a phantom that can potentially be used in practice. Methods: We used SimSET to produce calibrated simulations of a PET/CT scanner (5-ring GE DMI). Calibrated values were the prompt, random, scattered, and true coincidence rates used in NEC calculations. The simulated ACR PET phantom was modified to allow for high uptake regions (i.e. the 9hot rods9 region) to be in a warm background, as has been demonstrated using porous plastics (Wollenweber 2014). We used the same image reconstruction algorithms as used on the scanner but modified to allow for different TOF resolutions. Parameters varied included total NEC (38, 72, 107, 143, 179 Mcts), TOF resolution (193, 386, 550 ps), contrast ratio (1.23, 1.45), and rod diameter (4.8, 6.4, 7.9, 9.5, 11.1, 12.7 mm). Regions of interest (ROIs) were measured using the known true locations for the rods and the background regions. Detectability (d9) was calculated using the approach (described by Wollenweber et al., 2016) and Difillipo (2019), which is essentially a t-test between rod and background ROI values with the noise determined by the replicated rod and background ROIs. The d9 values were converted to the area under the receiver operating curve (AUC) values using the Gaussian approximation approach. Error bars were determined by using separated axial regions for repeated calculations of the AUC values. To determine the relative impact of PET timing resolution versus NEC on image quality, the AUC values were used to generate iso-contour plots for fixed AUC values as a function of timing resolution and NEC. Results: For the 4.8 mm rods the AUC ranged from 0.57 to 0.79 for the variations studied. There was an increase in AUC with increasing counts and also with improved timing resolution. This relative behavior in AUC was consistent and with overall increases in AUC as contrast and road diameter values increased. As the rod diameter exceeded 7.9 mm, the AUC quickly approached values of 1.0. Thus the range where there were meaningful differences in AUC as a function of contrast was relatively narrow. The coefficient of variation for the AUC results was typically 15%. In comparing the impact of PET timing resolution versus NEC on detectability, the 4.8 mm rods showed a roughly linear trade-off in PET timing resolution versus NEC for AUCs ranging from 0.575 to 0.75. Within the linear regime the ratio of NEC /TOF ranged from 0.07 to 0.18 Mcts/ps. As a specific example, for a contrast of 1.45 with 4.8 mm rods, a detection AUC of 0.70 is achieved with an (NEC rate, TOF resolution) performance of either (89 kcps, 200ps) or (121 kcps, 400 ps). In other words a 26% increase in NEC rate was equivalent to a 100% improvement in TOF resolution. This linear behavior became less regular as the AUC approached 1.0, i.e. when rod diameter and/or contrast levels increased. In these regimes, e.g. with AUC > 0.8, the AUC was largely dependent on NEC and independent of TOF resolution. Conclusions: The relative effects of PET timing resolution versus NEC on image quality as determined by detectability are linear for lower AUC values, with NEC sensitivity having a larger impact than TOF resolution. As AUCs approach 1.0 the tradeoffs become non-linear with relatively little effect from TOF resolution. These trade-offs have implications for both PET scanner design and system performance for clinical tasks.