Generalized ridge regression versus simple ridge regression for generation of kinetic parametric images in PET

Weighted linear regression (WLR) is computationally efficient for generating parametric images in dynamic PET studies. However, due to high noise level of pixel kinetics, parametric images estimated by WLR usually have high variability. The authors have shown earlier that, for image-wise model fitting, the incorporation of simple ridge regression and spatial constraint (SRRSC) can improve the stability and signal-to-noise ratio of the estimated parametric images. In this study, the authors investigate the use of generalized ridge regression with spatial constraint (GRRSC) instead of SRRSC, and evaluate the amount of further improvement due to GRRSC. Computer simulation of O-15 water kinetics in a Hoffman brain phantom and human PET study were used as the data for evaluation. Results showed that both GRRSC and SRRSC improved parametric images quality for studies with high or middle noise level of dynamic images. GRRSC provide no significantly different parametric images compared with SRRSC. For its lower computational burden and its simplicity, SRRSC is suggested to be the choice method for generating parametric images for O-15 dynamic studies