Single-scan Bayes estimation of cerebral glucose metabolic rate: comparison with non-Bayes single-scan methods using FDG PET scans in stroke.

Abstract

Three single-scan (SS) methods are currently available for estimating the local cerebral metabolic rate of glucose (LCMRG) from F-18 deoxyglucose (FDG) positron emission tomography (PET) scan data: SS(SPH), named for Sokoloff, Phelps, and Huang; SS(B), named for Brooks; and SS(H), named for Hutchins and Holden et al. All three of these SS methods make use of prior information in the form of mean values of rate constants from the normal population. We have developed a Bayes estimation (BE) method that uses prior information in the form of rate constant means, variances, and correlations in both the normal and ischemic tissue populations. The BE method selects, based only on the data, whether the LCMRG estimate should be computed using prior information from normal or ischemic tissue. The ability of BE to make this selection gives it an advantage over the other methods. The BE method can be used as a SS method or can use any number of PET scans. We conducted Monte Carlo studies comparing BE as a SS method with the other SS methods, all using a single scan at 60 min. We found SS(H) to be strongly superior to SS(SPH) and SS(B), and we found BE to be definitely superior to SS(H).