Performance of a randomization test for single-subject (15)O-water PET activation studies

We adapted and implemented a permutation test (Holmes 1994) to single-subject positron emission tomography (PET) activation studies with multiple replications of conditions. That test determines the experimentwise alpha error as well as location and extent of focal activations in each individual. Its performance was assessed in five normal volunteers, using (15)O-H2O-PET data acquired on a high-resolution scanner, with septa retracted (3D mode), during functional activation by repeating words versus resting (four replications each). Calculated alpha errors decreased and the size of activated tissue volumes (voxels with P < or = 0.05) increased with increasing filter kernel size applied to the difference images. At a filter kernel of 12 mm Gaussian full width at half maximum, significant focal activations were seen bilaterally in superior temporal cortex, including Brodmann's areas 41 and 42, in all five subjects. Additional foci were detected in the precentral gyrus, left inferior frontal gyrus, supplementary motor area, and cerebellum of several subjects. The average CBF increase in activated voxels ranged from 17.6% to 28.7%. Activated volumes were smaller than those detected with a standard parametric test procedure. We conclude that the permutation test is a less sensitive procedure, having the advantage of not depending on unproven distributional assumptions, that detects strong activation foci in individual subjects with high reproducibility.