Resampling can be used to compute the null distribution of any test statistic for the purpose of measuring the significance of the measured value. This study investigated how well the spatial and temporal correlations of simulated and experimentally observed fMRI time series were preserved under Fourier and wavelet resampling methods. The null distributions of a test statistic estimated by each resampling method were compared. In addition, both resampling methods were applied to locate activated voxels in an fMRI dataset and ROC analysis showed that wavelet resampling performed more accurately than Fourier resampling.
Copyright 2004 Wiley-Liss, Inc.