Abstract
The brain's resting-state energy consumption is expected to be mainly driven by spontaneous activity. In our previous work, we extracted a wide range of features from resting-state fMRI (rs-fMRI), and used them to predict [(18)F]FDG PET SUVR as a proxy of glucose metabolism. Here, we expanded upon our previous effort by estimating [(18)F]FDG kinetic parameters according to Sokoloff's model, i.e., Ki (irreversible uptake rate), K1 (delivery), k3 (phosphorylation), in a large healthy control group. The parameters' spatial distribution was described at a high spatial resolution. We showed that while K1 is the least redundant, there are relevant differences between Ki and k3 (occipital cortices, cerebellum and thalamus). Using multilevel modeling, we investigated how much of the regional variability of [(18)F]FDG parameters could be explained by a combination of rs-fMRI variables only, or with the addition of cerebral blood flow (CBF) and metabolic rate of oxygen (CMRO(2)), estimated from (15)O PET data. We found that combining rs-fMRI and CMRO(2) led to satisfactory prediction of individual Ki variance (45%). Although more difficult to describe, Ki and k3 were both most sensitive to local rs-fMRI variables, while K1 was sensitive to CMRO(2). This work represents the most comprehensive assessment to date of the complex functional and metabolic underpinnings of brain glucose consumption.