Dear FreeSurfer experts!

I have a problem which relates to the ICV (eTIV) measure in different versions, and would be thankful for any suggestions:

I have previously examined associations between body mass index (BMI) and global cortical measures (e.g. cortical volume) in a sample of patients with severe mental illness using stats from FS 4.5.0. I then found a small but significant positive association between BMI and volume (p<0.05) (corrected for age, gender and diagnosis), which became more highly significant when covarying for ICV (p<0.0001). According to my hypothesis, I interpreted this as a tendency towards larger brains relative to skull size with increasing BMI (it should be noted that although the association was significant, a beta-value of about .1 indicated the effect was small). Now, when I updated the analyses with stats from FS 5.3.0, using the same data set, no significant results emerge: The positive association is now at trend-level (p=0.105), and when covarying for ICV (eTIV) it disappears altogether.

I’m struggling to understand this difference. The analyses are fairly similar up to the point where ICV (eTIV) is added as a covariate, but adjusting for ICV changes the results quite a lot. Both cortical volume and ICV correlates highly between the two versions (r = .96 and .94, respectively), but when I calculate cortical volume/ICV-ratios based on stats from each version, the correlation between these is lower (r = .75). I have so far ruled out (I believe) any differences in quality control or editing as a source of the difference. Both versions have used the talairach.xfm transform and not the older talairach_with_skull. The problem could be related to collinearity (correlations between cortical volume and ICV were r=.81 in FS 4.5.0 and r=.74 in FS 5.3.0), but VIF and tolerance statistics does not indicate a problem (and I suppose this would affect results equally much between versions (?)).

So, my questions are: - Are there any differences between FS 4.5.0 and FS 5.3.0 in how ICV (eTIV) is calculated? - Is it inappropriate to correct for intracranial volume when examining global cortical volume rather than regional volumes? - Do any of you have suggestions as to how I can figure out what is the ‘true’ result here?

Any help is appreciated!