Scanner-dependent bias correction for accurate quantification of myocardial blood flow with rubidium-82 positron emission tomography imaging

Objectives: 3D PET-CT is becoming the standard for myocardial perfusion imaging due to its increased sensitivity and high image quality with lower patient doses. However, random and scattered photon counts are increased, requiring systems with accurate corrections, especially during the first-pass transit of the tracer where detector saturation and dead-times are highest. This is particularly important in dynamic imaging for absolute quantification of myocardial blood flow (MBF) where accuracy must be maintained over a wide range of activities and count-rates. The objective of this study was to assess the accuracy of MBF quantification with Rb-82 PET using a post-reconstruction activity bias correction method.

Methods: Dynamic images (40x15s) were acquired on 3D BGO and LBS PET cameras after injection of 1100 MBq (30 mCi) of Rb-82 into the cardiac insert of an anthropomorphic torso phantom. Left ventricle (LV) wall time activity curves were plotted as a function of scanner-recorded coincidence dead-time correction factors (DTF). A polynomial function was fit to the data from each camera to correct for underestimation of reconstructed image activity. Accuracy of the corrections was evaluated first using repeat scans of the torso phantom. The proposed corrections were then applied to data from 2 independent cohorts of 20 patients who underwent rest+stress imaging on the BGO and LBS scanners (N=40 scans/camera), with Rb-82 injected activities of 8 and 10 MBq/kg respectively. 6-min dynamic images were analyzed using a one-tissue compartment model, to compare MBF estimates with and without bias correction.

Results: The correction functions from the phantom scans were determined to be BCF = 0.14×DTF² + 0.0047×DTF + 1 for the BGO and BCF = 0.17×DTF² + 0.038×DTF + 1 for the LBS camera, with excellent fits to the normalized activity vs. dead-time factors (BGO: R²=0.99, LBS: R²=0.98). Before correction, LV wall activity in the phantoms was found to be underestimated by >70% in the early time frames when DTFs are highest. For the repeat torso scans on both cameras, the proposed correction functions removed all measurable activity bias from the time-activity curves (R²=0 for both cameras). For the patient scans, the bias correction increased the early-frame activity values (10-30%) as expected, resulting in decreases in the MBF estimates of 0.10 and 0.19 mL/min/g on average at rest and stress on the BGO system and 0.03 and 0.07 mL/min/g on the LBS camera (p < 0.005).

Conclusion: The accuracy of dynamic 3D PET myocardial blood flow measurements can be improved using a non-linear calibration of the reconstructed image activity as a function of scanner-recorded dead-time factors. While further studies are needed to validate the accuracy of the proposed correction in low-dose vs. high-dose patient studies, initial patient results suggest that MBF can be overestimated when activity bias during the tracer first-pass is not corrected Research Support: Canadian Institutes of Health Research.