 |
 |
Hello, I ran randomise with TFCE interference for a paired t-test with 16 subjects. Functional MRI were collected at 3.75mm^3 isotropic voxels (smoothing 4mm FWHM) and ultimately warped to 2mm^3 MNI space. Significant results included a contrast yielding a cluster of 200 voxels, as well as other contrasts, in which clusters containing only 3 voxels survive correction (i.e., the 1-p>0.95 in the _tfce_corrp_stat* output). The location of these small clusters is consistent with broader clusters that are significant when parametric statistics are used instead. I understand that TFCE was designed to preserve the power of cluster-wise inference whilst being independent of an arbitrary cluster-forming threshold, and thus can preserve even localized activations if they have strong support… Still, I am seeking advice whether the small number of significant voxels (n=3) is considered worrisome when choosing TFCE as the primary analysis. I’d be grateful for any advice.