A Flexible Image Segmentation Prior to Parametric Estimation

A flexible method based on spatial and temporal pixel variance to compute parametric images in positron emission tomography (PET) is reported. For [(18)F]fluorodeoxyglucose and [(15)O]water brain studies, images were segmented based on coefficients of variation and correlation coefficients of neighboring pixels, and kinetic parameters were estimated by dynamic (DYN) and autoradiographic (ARG) fitting. For comparison, regional glucose metabolism (rCMRGlc) and blood flow (rCBF) in both DYN and ARG were estimated from segmented and usual images. The maximal relative error was found to be 4, 10 and 17% for ARG and DYN rCMRGlc and DYN rCBF, respectively.