Commentary: Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates

摘要

Figure 1. Impact of using different image resolutions onto family-wise error (FWE) corrected p-values with the analysis of resting-state fMRI data using a fake on-off block design. The statistical model included an experimental condition with 18 blocks with a block length of 20 s using an onset every 40 s. The same statistical analysis was performed with different image resolutions obtained within the normalization step in SPM12 followed by a subsequent spatial smoothing of 8-mm full width at half maximum. (A) Orthogonal brain sections of an individual subject showing a cluster of brain activity that becomes significant when using image upscaling. (B) Systematic decrease of FWE corrected p-values with increasing image resolution for each participant. The plot shows p-values for a randomly selected cluster with p < 0.1 for each participant. On average, p-values became smaller than the half size of the original value (see line plotted in red color). (C) Systematic decrease of the estimated smoothness when using image upscaling.