Low dose positron emission tomography emulation from decimated high statistics: A clinical validation study

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Radiochemistry

Background:

• Positron emission tomography (PET) data handling offers flexibility for retrospective analysis.
• Emulating scans with different statistical compositions from identical radiotracer distributions aids algorithm evaluation.
• Previous validation of this emulation technique relied on simulations and phantom data, not human subjects.

Purpose of the Study:

• To validate low-dose (LD) PET scan emulation using actual human subject data.
• To compare emulated LD PET scans derived from standard-dose (SD) scans with actual LD scans.
• To assess the impact of randoms fractions on PET image statistics at varying activity levels.

Main Methods:

• Eleven non-small cell lung cancer patients underwent two independent FDG-PET/CT scans: one LD and one SD.
• Images were reconstructed using filtered back-projection (FBP) and OSEM with time-of-flight and point-spread function modeling (PSFTOF).
• Count-matched emulations (ELD) were reconstructed from SD data, and noise/reproducibility were evaluated in liver and lesion volumes of interest.

Main Results:

• Randoms fraction estimates differed significantly between LD (17.4%) and SD (42%) data.
• Distributions of metabolic volumes, means, and maxima in lesions were similar between LD and SD sets.
• Statistical tests showed no significant difference (P > 0.05) between LD and ELD reconstructions for liver and lesion VOIs, regardless of reconstruction method.

Conclusions:

• Clinical LD PET scans can be accurately emulated by statistically decimating SD scans.
• This validation using human subject data supports the use of emulated LD PET for research.
• The method provides a reliable approach for studying low-count conditions and algorithm stability in PET imaging.