Impact of resolution modeling on accuracy and precision of lesion contrast measurements

Abstract

The finite spatial resolution of PET scanners leads to a loss of quantitative accuracy in the measurement of activity in small structures. We sought to study the impact of resolution modeling on the accuracy and precision of contrast measurements in whole-body imaging. Resolution models of differing degrees of accuracy were incorporated into the system model of a 3D iterative time-of-flight PET reconstruction algorithm. Phantom and patient data with 10-mm hot spheres at multiple locations were used to measure the accuracy and variability of contrast recovery coefficient (CRC) measurements. In both phantom and patient studies, the accuracy of CRC values in small structures at clinically reasonable noise levels increased significantly (40–70%, depending on the activity distribution) when a simple, spatially invariant resolution model was included in the reconstruction, compared with the CRC values without resolution modeling (∼0.3); more accurate models led to a modest (≤10%) further improvement in the quantitative accuracy. In simulated trues-only phantom data, the variability of CRC values with resolution modeling was comparable to or reduced from that without resolution modeling. In the measured phantom and patient studies, the precision of CRC values was generally not significantly affected by the inclusion of resolution modeling in the reconstruction algorithm. For whole-body imaging, a simple resolution model provides a marked increase in contrast recovery without a concurrent increase in the variability of the contrast measurement.