FAST (Fast Analytical Simulator of Tracer)-PET: an accurate and efficient PET analytical simulation tool

Introduction: Simulation of Positron Emission Tomography (PET) images is an essential tool in the development and validation of quantitative imaging workflows and advanced image processing pipelines. The most widely used Monte Carlo simulation tool, Geant4 Application for Tomographic Emission (GATE), suffers from high computational demands. Alternative analytical simulators either omit time of flight (TOF) implementations or still require long processing times. In this study, we develop and validate the quantitative accuracy of FAST-PET, a novel analytical framework to simulate PET images, as well as compare its performance against GATE as the gold standard.

Methods: An overview of the workflow is shown in Figure 1a. FAST-PET simulates PET images by performing precise forward projection, scatter, random estimation, and reconstruction that match the Siemens Vision-600 PET/CT scanner geometry and statistics using the Siemens reconstruction software e7 tools. Although the same process should be applicable to other scanner models, we focus on the simulation of the Biograph Vision-600 here. FAST-PET is calibrated according to a National Electrical Manufacturers Association (NEMA) Image Quality (IQ) phantom scan, so it only requires attenuation and activity maps with user-specified acquisition time and reconstruction settings as inputs to simulate PET images. In the first part of the validation, a physical NEMA IQ phantom was scanned for 10 minutes on the Biograph Vision-600, and the same phantom was simulated with both FAST-PET and GATE. In all cases, five 120-second frames and five 5-second frames were reconstructed using ordinary Poisson OSEM algorithm, with 8 iterations and 5 subsets, PSF correction, TOF, and no post-reconstruction filter. The activity mean maps were generated by taking the voxel-wise mean activity across the five noise realizations. For each acquisition time, its intensity distributions within regions of interest (ROIs) in the background and within 6 spheres were plotted in histograms to compare the three methods. To validate performance against clinical images, we simulated FAST-PET and GATE images of 5-minute scans for 7 patients given their FDG-PET/CT images as ground truth. The reconstruction settings followed the previous description except with 4 iterations, 5 subsets, and 4 mm FWHM Gaussian post-reconstruction filter. Several normal organs (spleen, liver, stomach, pancreas, left and right lung) were segmented by MOOSE (Sundar, 2022), whereas tumors were segmented manually. The concordance correlation coefficients (CCC) of the mean intensities and coefficients of variation within normal organs and tumors were calculated to show their agreement.

Results: Figure 1b depicts representative slices of experimental and simulated (FAST-PET and GATE) 5-second and 120-second scans of the NEMA phantom. The distributions of mean activity shown in Figure 1c exhibit notable similarity among all three methods for both long (120s) and short (5s) acquisition times, indicating that images produced by both simulation methods closely resemble real scan images in quantitative characteristics under both low and high noise conditions. Figure 1d depicts representative FAST-PET and GATE slices of clinical patient simulations with diagonal and opposite diagonal profiles. Their agreement indicates similarity between both simulated images. Scatter plots and CCC values in Figure 1e confirm the agreement between GATE and FAST-PET in terms of both mean activity and variability within ROIs. Critical to this effort, FAST-PET significantly outperforms GATE in efficiency, simulating a PET image in about 2.5 minutes compared to GATE's 56 hours on a 24-core, 3GHz Intel computer, underscoring its speed and user-friendliness in simulating PET images.

Conclusions: FAST-PET has been developed and validated as an analytical simulation tool, designed to produce PET images that mirror those acquired from actual scanners and GATE simulations, while markedly reducing the processing time.

• Download figure
• Open in new tab
• Download powerpoint