Enhanced whole-body PET parametric imaging using hybrid regression and thresholding driven by kinetic correlations

Whole body PET/CT, a well established imaging method in nuclear medicine for the clinical evaluation of a wide variety of metastatic cancer malignancies, commonly involves static scanning over multiple beds. Recently, we proposed a clinically feasible transition of whole-body PET/CT imaging to the dynamic domain, by acquiring (i) an initial 6min dynamic scan over the heart, followed by (ii) an optimized sequence of whole-body PET scans, allowing for quantitative whole body parametric imaging. Comparative evaluation of parametric and SUV images indicated enhanced contrast-to-noise ratio (CNR) but also higher noise for the parametric images. The objective of this study is to further improve parametric image CNR to enhance tumor detectability, by limiting noise in the estimates, while enhancing contrast and quantitative accuracy of parametric images. For this purpose, we utilize the weighted correlation coefficient (WR) of the kinetic model (Patlak) fits at each voxel to determine the cluster of voxels, where (i) advanced, as opposed to conventional, statistical parameter estimation, (ii) spatial smoothing or (iii) thresholding is applied. Thus, we facilitate the integration of whole body parametric imaging into the clinic as a competitive alternative to SUV. Through quantitative analysis on selected tumor regions of the resulting images, we show enhanced CNR when ridge regression is applied only to voxels associated with high WR, while ordinary least squares (OLS) and WR driven post-smoothing is performed to the rest. This hybrid regression method yields reduced mean squared error in tumor regions, compared to OLS. In addition, by setting the WR threshold level in the range [0.85 0.9], CNR is further enhanced for tumor regions of high WR. Finally, for the same type of tumors, hybrid regression also achieves higher CNR, compared to SUV, when the last two dynamic frames are omitted, allowing for shorter acquisition times.