Quantitative myocardial perfusion PET parametric imaging at the voxel-level

Objectives To assess performance of a novel approach to obtain quantitative parametric images in cardiac PET.

Methods We have proposed a parametric PET image formation method utilizing spatial constraints via what we term physiological clustering (PC): this iterative kinetic-driven-clustering method observes that ideal clustering depends on underlying physiology of functional regions, and in turn, physiological processes are quantified by kinetic parameter estimation. 30 noise-realizations of Rb-82 datasets were generated based on realistic kinetic parameters and input function. Transmural/non-transmural perfusion defects were simulated. We also validated our approach on n=5 Rb-82 patient studies. Voxel-wise kinetic parameters were obtained using (i) conventional post-smoothing (PS) on reconstructed images (Butterworth, order 4, cut-off 0.2-1 cycles/cm), and (ii) PC, where regression was performed on non-smoothed images via a quadratic penalty that penalizes deviations from mean kinetics as computed using PC of dynamic images.

Results Enhanced noise-bias performance was obtained for both transmural (T) and non-transmural (NT) defects. Specifically, for T, at optimal performance (minimum RMSE), PS reduced noise by 27% with 3.9% increase in bias (38% RMSE reduction). PC reduced noise by 33% with 1.2% increase in bias (46% RMSE reduction). For NT, PS reduced noise by 28% with 4.7% increase in bias (37% RMSE reduction). PC reduced noise by 33.4% with 3.0% increase in bias (47% RMSE reduction). PC, compared to PS, resulted in higher CNR (13.7 as opposed to 7.2 for T; and 11.3 as opposed to 5.7 for NT) at optimal performance. In patient studies, PC resulted in significantly higher CNR at matched K1 value (x 2 improvement at converged values).

Conclusions Parametric imaging based on PC clearly outperformed conventional estimation method in transmural and non-transmural perfusion defects.

Research Support 2014 Bradley-Alavi Fellowship, SNMMI