Abstract
1617
Objectives: Dynamic rubidium-82 (Rb82) PET imaging enables quantification of myocardial blood flow (MBF), providing incremental diagnostic and prognostic value over myocardial perfusion imaging [1]. However, patient motion leads to misalignment of left ventricular regions of interest in the dynamic image causing artifactually increased blood-to-tissue spillover in the blood phase frames, which results in overestimation of MBF if the motion is not corrected. The objective of this study was to identify quality assurance (QA) metrics that were predictive of patient motion, and consequently uncertainty in flow estimates. Methods: 225 consecutive patients that underwent dynamic rest/regadenoson-stress Rb82 PET/CT imaging were evaluated. A 12 MBq/kg weight-based dose of Rb82 was administered using a CardioGen-82 infusion system (Bracco Diagnostics, Monroe Township, NJ). 7-minute 3D list mode scans were acquired from the start of the tracer infusion on a Biograph mCT PET/CT scanner (Siemens Healthcare, Knoxville, TN). Iteratively reconstructed dynamic images (16 frames x 5s, 6x10s, 3x20s, 4x30s, 1x80s) were processed with Corridor4DM (INVIA Medical Imaging Solutions, Ann Arbor, MI) generating 2 datasets: one without (NoMoCo) and one with image-based automated patient motion correction (AutoMoCo) [2]. For both datasets, a 1-tissue compartment model (1TCM) was fit to sampled left ventricle (LV) blood pool and tissue input functions providing estimates of MBF. LV blood-to-tissue spillover fraction (FV), FV coefficient of variation (COV(FV)), and R2 and χ2, describing the goodness-of-fit of the 1TCM to the input data, were computed as potential QA metrics. From the AutoMoCo dataset, the mean magnitude of the motion vectors taken across all dynamic frames was calculated for each patient and correlated to the relative improvement (NoMoCo-AutoMoCo) in COV(FV), R2 and χ2 with motion correction. Blood pool and tissue phase motion vectors were isolated and also correlated with the potential QA metrics. Results are reported as mean ± SD. Paired t-tests were used to compare continuous parameters. Correlation coefficients were compared using Fisher’s Z-test. P < 0.05 was considered significant. Results: The population mean MBF estimate was lower with motion correction (MBF = 1.61 ± 0.86 ml/min/g) vs. without (MBF = 1.65 ± 0.94 ml/min/g, p<0.001). Significant decreases in FV (0.24 ± 0.07 vs 0.27 ± 0.08, p < 0.001) and COV(FV) (0.007 ± 0.004 vs 0.012 ± 0.01, p < 0.001) were observed when motion correction was applied, while no significant changes were seen in R2 and χ2 with motion correction. A moderate correlation was observed for COV(FV) with the motion vectors (R2 = 0.35) (Figure 1), but there was no correlation of the motion vectors observed with R2 and χ2 (R2 = 0.0094, 0.039). After isolating the blood pool phase, where motion effects on MBF are more pronounced than in the tissue phase, the correlation of COV(FV) with motion did not increase significantly (R2 = 0.38, p = 0.35). Correspondingly, there was no correlation of COV(FV) with isolated tissue phase motion (R2 = 0.005), confirming that motion effects occur mainly in the blood pool phase. There was still no correlation observed for R2 and χ2 with isolated blood (R2 = 0.012, 0.046) or tissue phase (R2 = 0.012, 0.046) motion. Conclusions: The variation of the LV blood-to-tissue spillover parameter, COV(FV), is most predictive of patient motion, providing a useful QA metric and indicator of the subsequent uncertainty in the compartment model-derived MBF estimates for Rb82 PET images. References: [1] Murthy VL et al. Clinical quantification of Myocardial Blood Flow Using PET: Joint Position Paper of the SNMMI Cardiovascular Council and the ASNC. J Nucl Cardiol 2018. Feb;25(1):269-297. [2] Lee BC et al. Automated dynamic motion correction using normalized gradient fields for 82rubidium PET myocardial blood flow quantification. J Nucl Cardiol 2018. https://doi.org/10.1007/ s12350-018-01471-4.