btw, if you simply "touch" the contrast mat files in the analysis directory, it will only regenerate the contrasts. That should save you some time because it won't have to re-run all of selxavg. doug
Douglas N Greve wrote:
I think this is due to the high DOF and probably high t-value causing the significance in matlab to be computed as 0. This gets converted to Inf (infinity) when the -log10(p) calculation is made, then the Inf gets set to 0 when the volume is saved. The quick fix for you is to change fast_selxavg3.m (in $FREESURFER_HOME/fsfast/toolbox). There is a single line that reads:
fsigmat = -log10(pmat)
change this to
fsigmat = -log10(pmat + eps(0));
Let me know if this works and I'll update our source code.
doug
Qi Zhu wrote:
Dear Doug,
I used selxavg3-sess in freesurfer 5.0 to run a big GLM analysis on a monkey data set containing 147 runs. But When I check the results, I found that in the voxels where the t-values should be very high were turned out to have a value of zero. I checked the brainmask, the ces and the cesvar values, they all look fine. It seems that only the ces values in these voxels are a bit high compared to the nearby normal voxels (e.g. 1.17 (ces)/0.00048(cesvar) compared to 0.74/0.00040 in a nearby normal voxel). And it seems that not only me have this problem (similar problem posted in the mailinglist before. http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg13789.html).
Do you know what the problem would be and how to solve it? Thanks very much for your help.
Attached please find an image of the problem.
Best,
Qi Zhu, PostDoc Laboratorium voor Neuro- en Psychofysiologie K.U.Leuven Medical School Herestraat 49, B-3000 Leuven (Belgium)