please find my responses inline.

On Di, 2019-01-29 at 14:18 -0600, Getaneh Bayu wrote:

Dear Freesurfer Experts,

I have a question about how to interpret cortical thickness difference between groups using LME MATLAB tools.

I have three groups(g0, g1, and g2) and five time points(t= 0,0.5, 1,2,3).

I am trying to use LME model with random effects y-int and time from the base line.

Based on the LME tutorial and the questions and answers from Freesurfer support list:

Y=B1+B2*t+B3*g1+B4*g1*t+B5*g2+B6*g2*t+ ....

To see the difference between group g0 and group g1, I constructed the following contrast matrix using the tutorial provided in LME MATLAB tools

C = [zeros(1,3) [1 0 0] zeros(1,4)];

CM.C=[zeros(1,3) [1 0 0] zeros(1,4)];

Based on the tutorial I run the following functions.

rhstats = lme_mass_fit_Rgw(X,[1 2],rhY,ni,rhTh0,rhRgs,rhsphere); F_rhstats = lme_mass_F(rhstats,CM); fs_write_fstats(F_rhstats,rhmri,'/backup/cross/rh_AB_cross_sig.mgh','sig');

nv=length(rhstats);

Beta2 = zeros(1,nv);

for i=1:nv

if ~isempty(rhstats(i).Bhat)

Beta2(i) = rhstats(i).Bhat(2);

end;

rhmri1 = rhmri;

rhmri1.volsz(4) = 1;

fs_write_Y(Beta2,rhmri1,'rhBeta2.mgh');

[detvtx,sided_pval,pth] = lme_mass_FDR2(F_rhstats.pval,F_rhstats.sgn,rhcortex,0.05,0);

fs_write_Y(sided_pval,rhmri1,'spval.mgh');

When I view sig.mgh using tksurfer I see regions with red and blue color.

1) Does the result shows the oveall change for all the five time points?

Your contrast vector means that you will be evaluating B4, i.e. the interaction between group 1 and time, or - put differently - the difference in slopes for groups g0 and g1. More precisely, is there a greater (this can also mean: less negative) slope in group 1 than in group 0. All time points will contribute to this effect.

2) How do I know the difference between the two groups for each 5 time points?

In my eyes, this model cannot be used for testing group differences at each single time point, since time is treated as a continous variable. This is in contrast to an ANOVA, where we would treat timepoints as a categorical variable. You might want to consider conducting separate (non-LME) analyses for each time-point.

3) Does red means thickness of group A is lgreater than thickness of group B?

That depends on the settings in your GUI, on how the model was specified, and how the contrasts were set up, so there is no general answer in my opinion.

Note, however, that although the LME toolbox conducts F-tests, which are direction-less, the direction can be inferred from the F- or p-maps, since the LME toolbox attaches a sign to them when writing with the fs_write_stats.m function. This sign depends on how the contrast was formulated and also on the sign of the beta coefficient. Essentially, it will be the sign of the value resulting from the multiplication of the contrast vector with the beta coefficients vector.

For example, if the contrast vector contains a single positive value (i.e., +1) and zeros otherwise, and if the beta coefficient is also positive, then the sign will be positive. If the same contrast results in a negative beta coefficient, the sign will be negative. The sign will also be negative for the converse scenario of a negative contrast value (i.e., -1) and a positive beta coefficient. The sign will again be positive for a negative contrast value in combination with a negative beta coefficient.

4) what does rhBeta2.mgh and spval.mgh tell us? How can I extract the information from the file?

The first file will contain the second beta coefficient of the model, note the Bhat(2) indexing in your code. The second file contains the p-values as returned by lme_mass_FDR2. Both the beta coefficients and the p-values can be visualized by loading a surface in the 'freeview' GUI, and choosing these files as overlays (Overlay --> load generic).

With regards

Getaneh