Your mask image has over 7 million voxels in it... Are you working with diffusion data -- i.e., is the input image from TBSS? If so, you should use the "--T2" option in randomise. You should also check your mask, as it's unlikely you have data at each point within the mask you're using.
On Dec 11, 2012, at 7:51 PM, Eric Walden wrote:
I have run randomise using the following command:
randomise_parallel -i all_total_connect -o ./randomise/Duration -d Duration.mat -t Duration.con -n 5000 -v 5 -T -m $FSLDIR/data/standard/MNI152_T1_1mm_brain_mask.nii.gz -x
I get the attached picture which shows a cluster with T-stats from 2-5 (red to yellow). When I look at "Duration_tstat2_nii" in a particular voxel in the cluster I see a tstat of 4.93, which is highly significant. The t-stat one voxel up is 5.3 and one voxel down is 4.6, so there is a cluster.
When I look at "Duration_vox_p_tstat2.nii" the 1-p is 0.99899, so highly significant. However, when I look at "Duration_vox_corrp_tstat2.nii" the corrected 1-p is 0.
I understand that the corrected 1-p should be lower, but from 0.99899 to 0, within a clear cluster of t-stats above 2.00, this seems like to precipitous of a drop.
When I look at the same voxel in the file "Duration_tfce_p_tstat2.nii" (shown in green on the picture and thresholded to 0.99), I have 1-p = 0.990103. But when I look at "Duration_tfce_corrp_tstat2.nii" the 1-p value drops to 0.188447, which again seems like too much of a drop.
Any idea why this might be the case. There is clearly a cluster of significant voxels, but corrp makes them ridiculously insignificant.
<randomise.png>