Enhanced formation of parametric images using fast regressive GLLS for noisy functional imaging

Parametric images derived in functional imaging can visualize the spatial distribution of physiological parameters in vivo. However, the high level of noise intrinsic in single photon emission computed tomography (SPECT) may lead to physiologically meaningless parameter estimates such as negative kinetic rate constants using the generalized linear least squares (GLLS) method for compartmental model fitting. In this study, an enhanced GLLS method using fast regressive adjustment of parameters was investigated for improving the reliability of GLLS applied to dynamic SPECT data. Monte Carlo simulation data were used to systematically evaluate accuracy and reliability of derived parametric images. The simulation and experimental results demonstrate that the enhanced GLLS method can achieve more reliable parametric images, while largely preserving computational efficiency.