 |
 |
Dear Oskar, threshold-free cluster enhancement (TFCE) is using the advantages of both voxel level and cluster size inferences. Large clusters and/or large voxel peaks contribute to the TFCE value. The advantage of TFCE over cluster statistics is that you don't need to define any initial threshold for the voxel-level inference. The result is just one inference based on the distribution of the TFCE values (estimated by permutations because the distribution is unknown). Thus, there is no additional cluster inference possible. One issue that is maybe somewhat misleading in my toolbox is that I also provide non-parametric inferences for T-statistics. These inferences should be similar to the results of the SnPM toolbox. However, I don't provide any additional inferences for cluster-size because in my experience TFCE is almost in any case superior to cluster statistics. Maybe, this will be added in future. For the filenames see below for my explanation... On Fri, 24 Jan 2014 10:32:22 +0000, Oskar W <[log in to unmask]> wrote: >Hi all, >I am using the TFCE toolbox by Christian Gaser for a VBM analysis, and I have run a two group t-test with a contrast [1 -1], to find voxels where group1 > group2. Now I wonder how to interpret the output files. I get, among others, files named > >T - non-parametric T values >T_log_p - logarithm of p-values for non-parametric T values (can be easily used to overlay a thresholded image using log p-threshold) >T_log_pFWE - logarithm of p-values (FWE corrected for multiple comparisons) for non-parametric T values >TFCE - non-parametric TFCE values >TFCE_log_p - logarithm of p-values for non-parametric TFCE values >TFCE_log_pFWE - logarithm of p-values (FWE corrected for multiple comparisons) for non-parametric TFCE values > >I would preferably report clusters where group1 > group2 with p-value of 0.95 for some type of multiple correction. How can I use any of these files to obtain this? Best, Christian > >Cheers, Oskar W >M.Sc. Eng.Phys. >Lund University