Did you map the networks into the native subject space and then compute the mean thickness for each ROI? If so, try to compute the subject/ROI means after all preprocessing (ie, on the argument to --y in mri_glmfit). You can do this with mri_segstats using --annot fsaverage hemi parc and specifying --avgwf as the output. Then do your ROI test on that table

On 9/25/2020 4:43 AM, Martin Juneja wrote:

External Email - Use Caution

Hello FreeSurfer experts,

I have this interesting situation. I understand its important to include ICV as a covariate while conducting cortical volume (CV) analysis.

So for my clinical sample (CL) vs healthy-controls (HCs):

1. If I do *network-wise* analysis (i.e., after parcellating the whole

brain into 17 networks), I get highly significant differences in CV (MANCOVA: CL < HCs) for a specific network (say N1) - *if I include ICV as a covariate*, but not otherwise. 2. If I do *whole-brain vertex-wise* analysis, I get highly significant differences in CV (CL < HCs) for a specific cluster (which is overlapping with an area of N1) - *if I do not include ICV as a covariate*, but not otherwise.

I am not sure how to interpret this i.e., ICV as a covariate for network-wise analysis plays an opposite role i.e., makes my findings stronger compared to ICV as a covariate for whole-brain vertex-wise analysis i.e., makes my findings weaker. Does ICV as a covariate add some kind of noise during whole-brain vertex-wise analysis?

I would really appreciate any help in understanding this.

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer