--- Intensity in tbss_tstat1 should be your t values. 0-0.05 is too small to me. for t distribution, t>2.353 for p< 0.05 with d.f. 3. Are you sure that you look at the right image? The p values should be in 0.95-1 range for significance. (due to the FSL setting).
I have 30 controls and 32 patients and the way I defined up my matrices is as you can see in the attached files: design.mat and design.con (I saved them for you in a .txt format so you can open them). While checking with imglob *_FA.*
I saw first the controls files then the patients files so in the desing.mat I wrote first my controls with 0 in the first column and 1 in the second column and the I added the patients with 1 and the 0 in the second column.
--- Sorry, I did not see the design matrix that you mentioned in the mail. (no .txt file) You can send to me at [log in to unmask] instead of FSL mailing list. Maybe the attachments are not allowed here.
3) After I ran the randomise:
design_ttest2 design 30 32
randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -n 5000 --T2 -V
I don't understand why my matrix changed. If I open now after randomise my design. mat I see first patients with 1 and 0 the controls with 0 and 1. Can you please explain how this is possible?
--- It should not change. Something is weird.
4) If I visualize the UNTHRESHOLDED t-stats results what's the meaning of the values that comes after -b:
fslview $FSLDIR/data/standard/MNI152_T1_1mm mean_FA_skeleton -l Green -b 0.2,0.8 tbss_tstat1 -l Red-Yellow -b 3,6 tbss_tstat2 -l Blue-Lightblue -b 3,6
--- After -b, it is the threshold range that you use for your mean_FA_skeleton so that the messy small branches will go away only the main skeleton will be displayed. 0.2-0.8 According to your data, you can change that. If you patients have significantly lower FA, you can use 0.15-0.6 etc.
--- From this command, it seems to me that your tstat values are in the range of 3 to 6. That makes more sense than 0-0.05 you mentioned earlier.
I suppose is a threshold only for data visualization or is something else. I saw some people use 1.3,3 instead of 3, 6. Can you pleases explain? Which one is the best to be used?
--- This is really for display. Arbitrary numbers. That's why I wrote the commands in the former email for you to mask your t-stat image. Then you can use #fslstats tbss_tstat1_masked -R and see the new t value range. Use that for your display.
5) When I check my corrected p-value image:
fslview $FSLDIR/data/standard/MNI152_T1_1mm mean_FA_skeleton -l Green -b 0.2,0.7 tbss_tfce_corrp_tstat1 -l Red-Yellow -b 0.9,1 the only clusters which survived is for a 0.9 threshold. In this case I got two clusters with Z-max = 0.913.
Is this value equal with 1-p so in this case the p value for the two clusters will be 0.087? How is possible to get the same p value for both clusters? I suppose this clusters are not significant statistically since >0.05. Am I right?
--- Yes. it is not significant. p>0.05. Your corrected p value. You can report uncorrected p value. Personally, I am skeptical about this whole multiple comparison thing. Different p value for different individual due to the multiple comparison correction. You can use AFNI commands like 3dttest++ to do the statistics analysis. It is parametric method instead of non-parametric. 3dttest++ use FDR for multiple comparison correction. Or you can try to use fslstats to look at your raw data and have a basic feeling about the data. Then, you probably will know why it is not significant. Randomise is non-parametric (not sensitive) compared with parametric method. You have n>30 for each group. It should be safe to use parametric methods. Also, according to Dr. Smith's paper, the skeletonised images follow Gaussian distribution. read this paper: doi: 10.1016/j.eplepsyres.2011.02.001 It stated pros and cons of TBSS. They used it for epilepsy patients.
Thank you very much for your time and help.
--- No problem. You're very welcome. Have a nice weekend.
Best regards,
Antonella