Hi. I am using randomise in a group analysis of DTI data, with TFCE, followed by "cluster" to identify significant clusters at p < 0.05 (FWE corrected). All of my FA images are registered into MNI space, and prior to running randomise I smoothed with a 2 mm Gaussian kernel, under the assumption that this will both increase the SNR and decrease the variability due to spatial misregistration of the images into MNI space. I am wondering how to interpret very small clusters from "cluster" (my smallest cluster is 7 voxels), since the registration and smoothing necessarily creates spatial coherence which is much larger than this (probably on the order of 100 voxels - note: voxels in MNI space are 1 mm isotropic). Can/should I ignore clusters below a certain threshold, and if so, how can I objectively determine this cut-off?
Thanks,
Mark Wagshul
|