Hi Jamie,
I suspect you were using FLAME 1+2 for estimation in a higher level analysis. Feel free to read up on why this happens in the technical report TR03MW1 available at http://www.fmrib.ox.ac.uk/analysis/techrep/. In brief, when using FLAME 1+2 there is a second (much slower but more accurate) stage that is used to refine the initial results obtained from FLAME1, i.e. this gives you the option to get slightly 'better' estimates if you're willing to wait substantially longer.... This - for the sake of keeping run time low - is only run for voxels around the inital Z-stats level (2 or 2.3 in your case) as these are the voxels that matter (in the sense that the estimated pattern of activity might change). Estimated Z-stats should change only by small amounts, if the amount of change is large then there is something more fundamental wrong with your data (e.g. many outliers ).
hth
Christian
On Jan 14, 2012, at 2:46 , Jamie Feusner wrote:
> I have a question about how Z-score values for specific voxels within a cluster appear to change depending on the Z threshold used in FEAT analyses.
>
> I ran an fMRI analysis using a simple blocked design and initially used cluster-based thresholding with a Z threshold of 2.0. I then reran the analysis using a threshold of 2.3, without making any other changes to the design file. When looking at the overlaid images, the clusters for Z>2.3 seem to make sense in terms of being in the same location and of a slightly smaller extent relative to Z>2.0. However, when looking at the thresh_zstat maps in FSLView, the Z values for voxels in the significant clusters sometimes differ between the 2.3 and 2.0 maps. I then compared the unthresholded zstat.nii images of the 2.3 and 2.0 results in FSLView, and all the voxels across the brain have the same Z values between the two versions except for the voxels within the clusters that were significant, which had either higher, lower, or the same Z values for the 2.3 results. The histograms for the thresh_zstat images also were different.
>
> I am confused about this because I thought that the Z values assigned to each voxel should be the same regardless of what Z-threshold is used for the cluster-based thresholding performed in FSL. I’m sure there must be something that I don’t understand, perhaps about the cluster-based thresholding, so if anyone has an explanation I would greatly appreciate it.
>
> Thanks,
>
> Jamie Feusner
|