Hello,
I've recently discovered that randomise_parallel gives different outputs depending on whether I use version 4.1.9 or 5.0.4. The main difference seems to be masking certain voxels: when I examined voxelwise stats like "vox_p_tstat", I saw that large chunks of voxels had been masked out from the analysis in 4.1.9, such that all these voxels had a value of 0 (even though I did not provide any mask using the -m argument.) On the other hand, when I ran the same dataset through randomise_parallel in 5.0.4., a p-value was assigned to all brain voxels. This had a large impact on the p-values of my clusters (in clusterm_corrp_tstat), because the lack of masking made the clusters much larger.
I also double-checked that the voxels that had been masked out were not really supposed to have a value of 0 (p-value=1) because the newer version of the script assigned a broad range of p-values to these voxels.
Could anyone tell me more about this masking issue, and why 4.1.9 was performing it?
Thank you in advance!
Olga
|