Clinical validation of high-resolution image reconstruction algorithms in brain 18F-FDG-PET: effect of incorporating Gaussian filter, point spread function, and time-of-flight

ObjectivesAccurate estimation of radiopharmaceutical uptake in the brain is difficult because of count statistics, low spatial resolution, and smoothing filter. The aim of this study was to assess the counting rate performance of PET scanners and the image quality with different combinations of high-resolution image reconstruction algorithms in brain F-18-2-fluorodeoxy-d-glucose (F-18-FDG)-PET.Materials and methodsUsing 23 patient studies, we analyzed the coincidence rates of true and random, random fraction, and the noise equivalent counts per axial length (NECpatient) in brain and liver bed positions. The reconstruction algorithms were combined with baseline ordered subsets expectation maximization, Gaussian filter (GF), point spread function (PSF), and time-of-flight (TOF). The image quality of the brain cortex was quantitatively evaluated with respect to spatial resolution, contrast, and signal-to-noise ratio (SNR).ResultsThe true coincidence rate in the brain was higher by 1.86 times and the random coincidence rate was lower by 0.61 times compared with that in the liver. In the brain, random fraction was lower and NECpatient was higher than that of the liver. Although GF improved the SNR, spatial resolution and contrast were reduced by 12 and 11%, respectively (P<0.01). PSF improved spatial resolution and SNR by 11 and 53%, respectively (P<0.01), and TOF improved SNR by approximate to 23% (P<0.01).ConclusionWe have demonstrated that a high-resolution image reconstruction algorithm for brain F-18-FDG-PET is promising without the use of a GF because of high true coincidence counts and that combined with PSF and TOF is optimal for obtaining a better SNR of the image.