[Freesurfer] glm results : opposite direction between whole-brain analysis and feature extraction in the significant clusters

Hi,

I am working on R1 maps. I ran a very simple glm on smoothed R1 maps, FWHM = 8 mm, population 1, population 2 , and age as nuisance factor.

After POS MC simulation CW correction, I had two significant clusters resulting from this in the planum temporale and Heschl’s sulcus. The contrast being population 1 - population 2, my interpretation was higher R1 values in these two clusters in Pop 1 as compared to Pop 2.

After that I wanted to run some post-hoc tests let say, in a statistical software similar to SPSS, JMP, on the mean R1 values from each of these clusters.

The strategy I picked was to stay in the common space, fsaverage, and apply mri_segstats:

#Extract mean R1 in montecarlo's significantcluster

for i in `cat subjects` do mri_segstats --mul 1000 --annot fsaverage lh PT_mc-z.pos.sig.ocn --i $SUBJECTS_DIR/${i}/surf/lh.R1.fsaverage.fwhm8.mgh --sum lh.PT.R1.${i}.stats done

for i in `cat subjects` do mri_segstats --mul 1000 --annot fsaverage lh HS_mc-z.pos.sig.ocn --i $SUBJECTS_DIR/${i}/surf/lh.R1.fsaverage.fwhm8.mgh --sum lh.HS.R1.${i}.stats done

Everything when fine but as compared to my glm analysis when looking at mean group differences in my stat software I have the opposite : larger R1 values in Pop 2 as compared to Pop 1 (after removing the effect of age or not, and removing one outlier or not).

I went back to my glm and checked everything, fsgd file, matrix and so on, everything seems ok I am not mistaken on the direction of my glm contrast (pop 1 - pop 2) so I do not get it at all.

Any advices?

For now I am thinking about going back in the individual space, registering my clusters in the individual surfaces (label2surf something like that) and then extract my mean R1 values, or extract values in non-smoothed maps, but I do not think it will change anything.

Damien