A fast algorithm for estimating FDG model parameters in dynamic PET with an optimised image sampling schedule and corrections for cerebral blood volume and partial volume

Abstract

The generalized linear least squares (GLLS) method for parameter estimation of nonuniformly sampled biomedical systems is a computationally efficient and statistically reliable way to generate parametric images for tracer dynamic studies with positron emission tomography (PET). However, previous work on GLLS in FDG-PET has been mainly based on a conventional sampling schedule (CSS) with twenty or more dynamic image frames, and with a standard four-parameter model which ignores the effects of cerebral blood volume (CBV) and partial volume (PV) on the tissue uptake measurements. In order to reduce image storage requirements and obtain more reliable parameter estimates, the authors derived a new OISS5-GLLS algorithm based on an optimal image sampling schedule involving a much smaller number of image frames with a five-parameter FDG model for correcting CBV and PV error effects, and validated this algorithm through computer simulations and clinical FDG-PET studies. The results showed that the OISS5-GLLS could provide reliable parameter estimates in dynamic FDG-PET studies, while greatly reducing computational complexity and image storage requirements.