The use of simulations to study the effect of acquisition schemes and quantification methods using positron emission tomography: an 18F-Fallypride study.

Introduction: This computational study evaluates the accuracy of kinetic models and acquisition schemes in dynamic PET imaging using simulations of 18F-fallypride PET in the human brain on the real-world data.

Methods: We employed a 2-tissue 4-k model to generate ideal tissue curves for three regions (putamen, thalamus, and temporal cortex) and a reference region (cerebellum), incorporating a simulated metabolite-corrected input function. Realistic measurements were simulated over a 240-min PET scan by defining acquisition protocols (frame timings and durations), modeling tracer decay, and adding noise. Distribution volume ratios (DVRs) were calculated using the Logan reference analysis and the simplified reference tissue model (SRTM), the relative error in DVR was also assessed across various acquisition protocols. Rate constants from the 2-tissue model were varied, and Bland-Altman analysis was quantified to determine bias relative to ground-truth DVR.

Results: Results indicate that, under low noise conditions, the Logan reference method performed optimally with a protocol involving a 60-min dynamic scan, a 60-min break, a 30-min scan, another 60-min break, and a final 30-min scan. In noisier conditions, the SRTM yielded the best results with a 150-min effective scan time incorporating three breaks.

Conclusion: These findings highlight the impact of noise and acquisition strategy on model performance, informing optimal PET imaging protocols.