Sources of Variability in Normative Cerebral FDG-PET imaging

Abstract

PurposeQuantitative interpretation of brain [{superscript 1}F]FDG-PET increasingly relies on comparisons with normative datasets. However, normative values may be influenced by technical and biological factors, limiting their generalizability. We investigated the effects of scanner manufacturer, reference region, age, and sex on regional [{superscript 1}F]FDG uptake in cognitively normal (CN) adults and generated covariate-adjusted normative reference data. MethodsA total of 449 CN participants from the Alzheimers Disease Neuroimaging Initiative (ADNI) were included. Regional SUVr were calculated using three reference regions (whole cerebellum, pons, cortical gray matter) and converted to Z-scores. Linear regression models were used to estimate standardized regression coefficients ({beta}), and 10-fold cross-validation was performed to quantify the out- of-sample predictive contribution of each covariate using incremental explained variance ({Delta}R{superscript 2}). ResultsScanner manufacturer introduced large, spatially structured biases. Compared with Siemens systems, GE and Philips scanners yielded lower Z-scores in frontal and medial temporal regions, with effect sizes approaching one standard deviation in selected regions ({beta} up to -0.85). Age showed region- specific associations with subcortical nuclei, medial temporal structures, and the posterior cingulate cortex, and was the strongest biological predictor in cross-validation ({Delta}R{superscript 2}{approx}0.11). Sex effects were negligible ({Delta}R{superscript 2}<0.001). Cortical gray matter normalization minimized biological and technical confounding, and the AD meta-ROI demonstrated high robustness across manufacturers and normalization strategies. ConclusionScanner manufacturer and age are the major sources of variance in brain [{superscript 1}F]FDG-PET quantification in CN subjects. Cortical gray matter provides the most stable reference region and supports harmonized, covariate-adjusted normative datasets for clinical and research applications.