“Pseudo-reference-based Input Function Shape” (pRef-IFS): Towards a True image-derived input function for PET kinetic modeling

Abstract

PET absolute quantification requires the metabolite-corrected plasma input function (C_P), especially when an ideal reference region is unavailable. We propose “pseudo-Reference-based Input Function Shape” (pRef-IFS), a novel non-invasive method to estimate C_P from PET data in two steps: (1) generating the unscaled C_P shape from a pseudo-reference region (pRef) described by a one-tissue (1T) compartment model, with the pRef clearance rate $k_2^{\prime }$ either estimated with simplified reference tissue models (SRTM) or assumed a priori; (2) scaling C_P with an early image-derived blood time-activity curve (ID-BTAC). We here explored the first step (assuming accurate step-2 scaling), evaluating the impact of violating 1T assumptions and inaccurate $k_2^{\prime }$ estimates on pRef-IFS recovery with both synthetic and human data. The distribution volumes (V_T) were the main outcome measure. With 1T kinetics, SRTM $k_2^{\prime }$ was unbiased, irrespective of pRef specific binding; pRef-IFS had small V_T bias in real data: ¹¹C-LSN3172176 (cerebellum pRef: −2 ± 5%; thalamus: 11±8%); ¹⁸F-SynVesT-1 (centrum semiovale: −1 ± 14%; cerebellum: 6 ± 4%). With two-tissue kinetics, pRef-IFS performed poorly, especially with ¹⁸F-FPEB, while ¹⁸F-ASEM produced small biases (cerebellum: −4 ± 10%). If good scaling with ID-BTAC can be achieved, pRef-IFS promises to be fully non-invasive for radiotracers reasonably described by 1T kinetics, even with modest specific binding in the pRef.