Artefactual subcortical hyperperfusion in PET studies normalized to global mean: lessons from Parkinson's disease

Recent studies of Parkinson's disease (PD) report subcortical increases of cerebral blood flow (CBF) or cerebral metabolic rate of glucose (CMRglc), after conventional normalization to the global mean. However, if the global mean CBF or CMRglc is decreased in the PD group, this normalization necessarily generates artificial relative increases in regions unaffected by the disease. This potential bias may explain the reported subcortical increases in PD. To test this hypothesis, we performed simulations with manipulation and subsequently analysis of sets of quantitative CBF maps by voxel-based statistics.

CBF images from 49 healthy volunteers were randomly assigned to four sets of two groups of 20 subjects. For each set, CBF images in one group were manipulated, while the other group served as controls. In the first simulation, an isolated cortical decrease was simulated. In the second simulation, focal increases in the globus pallidus and thalamus motor nuclei were simulated in addition to cortical decreases, thus emulating the putative metabolic pattern in PD. The data were analyzed with both the SSM method and a univariate statistical approach with normalization to either the global mean or to the white matter mean.

In Simulation I, global normalization robustly created artefactual subcortical increases, irrespective of analysis methodology. Simulation II demonstrated that an increased signal from the small subcortical structures involved in PD can probably not be detected with present instrumentation and typically-used sample sizes.

Imposing focal decreases on cortical CBF in conjunction with global mean normalization gives rise to spurious relative CBF increases in all of the regions reported to be hyperactive in PD. Since no PET study has reported absolute increases of CBF or CMRglc in any subcortical region in PD, we conclude that the relative increases reported in some studies most likely arise from biased normalization to the global mean.