Dual-Tracer PET Image Separation by Deep Learning: A Simulation Study

Multiplexed positron emission tomography (PET) imaging provides perfectly registered simultaneous functional and molecular of more than one biomarker. However, the separation multiplexed PET signals within a single scan is challenging due to fact that all tracers emit positrons, which, after annihilating with nearby electron, give rise 511 keV photon pairs are detected in coincidence. Compartment modelling can separate single-tracer from based on differences bio-distribution kinetics radioactive decay. compartment-modelling-based method requires staggered injections assumes each tracer’s input function known. In this paper, we propose deep-learning-based simultaneously dual-tracer without explicitly knowing functions. We evaluate proposed [18F]FDG [11C]MET simulations compare its performance method, assessing dependence time interval between tracer as well amount training data. It shown implicitly denoises separated images offers reduced variance compared compartment modelling.