Permutation testing made practical for functional magnetic resonance imageanalysis

We describe an efficient algorithm for the step-down permutation test, appl ied to the analysis of functional magnetic resonance images. The algorithm' s time bound is nearly linear, making it feasible as an interactive tool. R esults of the permutation test algorithm applied to data from a cognitive a ctivation paradigm are compared with those of a standard parametric test co rrected for multiple comparisons. The permutation test identifies more weak ly activated voxels than the parametric test, always activates a superset o f the voxels activated by this parametric method, almost always yields sign ificance levels greater than or equal to those produced by the parametric m ethod, and tends to enlarge activated clusters rather than adding isolated voxels, Our implementation of the permutation test is freely available as p art of a widely distributed software package for analysis of functional bra in images.