Hi Sima,

since in a longitudinal study the observed information in the time points are correlated, care needs to be taken in the statistical analysis. People usually use Linear Mixed Models for that purpose. I am not a statistician and therefore the wrong person to ask. You should contact your local bio-statistician as there are several design choices that can be made depending on your case.

For a simple analysis we usually recommend to compute a single number statistic for each subject (e.g. thickness difference or percent change) and then use that across groups.

Best, Martin

On Wed, 2011-01-19 at 12:55 +0000, sima chalavi wrote:

Dear Freesurfer experts

I have some longitudinal data from tp1 and tp2 for three different groups (A,B,C) and 2 different genders (G1,G2).

I followed the longitudinal stream and processed all the images in FS 4.5.

For further analysis, I saw the post from Martin and Doug at http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14266.html in which they suggested to first do the mris_preproc for each time point and then calculate the differences using mris_calc.

What I did was instead of doing mris_preproc for each tp separately I made a FSGD file and put all the information about our different classes including baseline (BL) and followup (FU), see below, and then run mris_preproc which gives me ?h_thickness_both_tp_fwhm10.mgh files.

GroupDescriptorFile 1

Title diagnosis_gender_glmdir

MeasurementName thickness

Class A_G1_BL

Class A_G1_FU

Class A_G2_BL

Class A_G2_FU

Class B_G1_BL

Class B_G1_FU

Class B_G2_BL

Class B_G2_FU

Class C_G1_BL

Class C_G1_FU

Class C_G2_BL

Class C_G2_FU

After mris_preproc for example I made the following contrast to compare the changes from BL to FU and then I ran mri_glmfit:

contrast to compare (A_BL vs C_BL) vs (A_FU vs C_FU):

contarst_AvCBLvAvCFU=[1 -1 1 -1 0 0 0 0 -1 1 -1 1]

...

...

mri_glmfit - - y ?h_thickness_both_tp_fwhm10.mgh - - fsgd

diagnosis_gender_BL_FU.fsgd - - C contarst_AvCBLvAvCFU.mtx - - surf fsaverage ?h - - cortex - - glmdir glmdir

So using this approach I wanted to compare the difference between the 2 tps using the contrasts and not calculating first the differences using mris_calc and then runnig glm. However, I am not sure if I am doing the right thing. Could you please send me your feedbacks.

Thanks in advance,

Sima Chalavi.

-- Sima Chalavi, Ph.D. student

King's College London Institute of Psychiatry (IoP) Box P040, De Crespigny Park London SE5 8AF. Tel:+44-2078485702

BCN Neuroimaging Center (NIC) University Medical Center Groningen Antonius Deusinglaan 2 P.O. Box 196 9713 AW Groningen www.neuroimaging-DID.com _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer