Comparison among ⁸²Rb PET/CT myocardial flow reserve methods

Objectives: An important advantage of ⁸²Rb PET methodology over conventional SPECT imaging is the ability to compute absolute values of myocardial blood flow at rest & during stress, the ratio of which is myocardial flow reserve (MFR). However, several technical issues, including non-linear ⁸²Rb extraction fraction, necessitate the adoption of mathematical modeling assumptions, which have resulted in the development of multiple approaches & consequently different algorithms to compute MFR. Our study was undertaken to compare the results of 2 different methods used to compute MFR.

Methods: We conducted a retrospective analysis of ⁸²Rb PET/CT rest & regadenoson-stress data for 73 pts (age = 71±10 yrs; 36 male pts, 37 females pts) who were evaluated for known or suspected CAD. At rest, 0.94-1.22 GBq of ⁸²Rb was infused over 20-30 seconds from a ⁸²Sr-⁸²Rb generator; at peak pharmacologic stress, when hemodynamic steady state was achieved, a similar activity was infused for stress data acquisition. CT data were used to correct for attenuation using the manufacturer’s iterative reconstruction software, which also corrected for scatter & random events & normalized for detector inhomogeneities. MFR Method1 used a retention model, while MFR Method2 used a 1-tissue compartment, one rate constant (microsphere analog) model with spillover effects to describe tracer uptake in myocardial tissue. Bolus quality control (QC) algorithms available with Method2 were used to analyze automatically derived blood pool & tissue curves to detect technical problems with first-pass data used to compute MFR. Both methods were applied to the same data sets. Data for which algorithms produced unphysiologic MFR values (close to 0.0 or > 8.0), or for which first-pass bolus technical problems were detected were excluded. As global MFR < 2.0 is generally considered an abnormal result, the % of pts classified abnormal by the 2 methods were compared.

Results: MFR values of the 2 methods were not compared for 9 pts: unphysiologic MFR values were obtained by Method1 for 2 pts & by Method2 for 2 different pts, & the Method2 bolus QC algorithms indicated technical problems (pt motion) with an additional 5 pts. Of the remaining 64 pts, MFR values were normally distributed (Kolmogorov-Smirnov D = 0.06, p > 0.10), & MFR computations of the 2 methods were similar (2.02±0.61 versus 2.05±0.59, paired t-test p = 0.66). Linear regression analysis indicated significant correlation between the two MFR methods (Pearson r = 0.57, p < 0.0001), & Bland-Altman analysis indicated no bias (intercept = -0.09±0.28, p = 0.73) & no trend (slope = 0.02±0.07, p = 0.81); however, 95% confidence limits were large (-1.06 to +1.12). Both methods characterized 83% of cases the same as normal or abnormal, with “good inter-rater agreement” between methods (κ = 0.64), & McNemar’s test indicated no statistically significant difference between the results of the 2 methods (Δ = 6.8%, p = 0.27).

Conclusions: Although 95% confidence limits between the 2 methods were large, they produced essentially similar global MFR values & characterized abnormal pts with essentially the same frequency.

• Download figure
• Open in new tab
• Download powerpoint