Hi Dan,
As a proxy for the p-value, for an image thresholded at a certain p-level, one approach could be to take the voxel with the minimum p-value in each cluster as characterising the cluster's p-value. e.g if you threshold at a ( 1-p ) value of 0.95, then if the minimum p-value in a cluster is 0.96 you could characterise the corrected p-value for that cluster as being no more than 0.96. The p-threshold should be chosen, before the analysis, as being the threshold of significance for clusters. Otherwise randomise would need to be modified to calculate the cluster null-distribution for TFCE images, which would ( among other things ) require a TFCE threshold value to be determined.
Hope this helps,
Kind Regards
Matthew
> Hi Matthew,
>
> Thanks for your reply. For my clarification, is there any way to extract the cluster-wise p-values from a randomise analysis with TFCE? Or is a discrete cluster-wise p-value unobtainable in the TFCE framework, because of the involvement of voxel-wise statistics in cluster formation (or other reasons)?
>
> Thanks,
>
> Dan
|