Reduced variance of MBF estimates by dynamic ⁸²Rb PET

Objectives Dynamic ⁸²Rb PET has potential for routine clinical assessment of myocardial blood flow (MBF) and flow reserve. Kinetic analysis has typically used 1- or 2-compartment (1C,2C) models. 1C models produce biased MBF estimates due to model mismatch but with lower uncertainty; 2C models better match the underlying physiology but MBF estimates have greater uncertainty. We propose a method to apply a 2C model with preserved accuracy but greatly reduced variance in estimated flow.

Methods We employ a 2C model with 5 parameters: fractional blood volume (fv), and rate constants for exchange between capillary/interstitial (k1, k2) and interstitial/cellular (k3, k4) spaces. Earlier studies (Ziegler et al,Circ.Res 1971;29:208) showed that in a given heart the rate constant for ⁸⁶Rb exchange across the sarcolemma is ~independent of flow within the 1^st min after injection. With the assumption that this can be extrapolated to study durations of ~6 min and hyperemic flows, we studied a two-step fitting procedure: jointly fitting rest/stress dynamic PET data to a 2C model with common values for k3, k4, as well as fv, followed by individual fits of the rest and stress data for k1, k2 with fv, k3, k4 held fixed to the values determined from the joint fit. Simulations were performed using a clinical PET-based input function. Tissue uptake curves were modeled (3-compartments) for 8 MBF combinations: rest=(0.5/1)/stress=(1.5/2/2.5/3)ml/min/g, 3 noise levels (2.5/5/10)%, 2000 realizations each, and fit using proposed method.

Results Compared to standard 2C fits, MBF interquartile range (IQR) was reduced over the flow range by 41-73% and median absolute deviation (MAD) reduced by 37-68% at 2.5% noise; at 10% noise MBF IQR was reduced by 40-73% and MAD reduced by 16-39%. MBF MAD was also reduced by 52-71% (10% noise) relative to 1C fits after mapping k1 to MBF with a published empirical relation (Yoshida et al,JNM 1996;37:1701) and accounting for its parameter uncertainties.

Conclusions Initial feasibility of the proposed method was demonstrated in simulations. Ongoing work will assess its accuracy in animal studies with MBF measurement by radiolabeled microspheres