Dear FSL experts,
I ran a TBSS analysis on a group of 37 subjects (20 patients, 17 controls). The image preprocessing used was as recommended on the FSL website.
The statistics were performed with randomise using the design_ttest2 standard matrix and contrast with 10,000 permutations.
I chose to view the tfce_corrp_tstat for results. Nothing or only small anatomically irrelevant clusters "light up", with range of display 0-1, for one or the other contrast (controls>patients or patients>control). The strange thing is when I view the uncorrect maps, I cannot even see much noise in the controls>patients map whereas there seems to be normal noise in the patients>controls map. Nothing in the corrected maps reach statistical significance.
The same results were obtained, with very similar maximum 1-p values, in a preliminary analysis with only 22 of the present 37 subjects.
Our team tried to review the problems that could have possibly arisen. There was debate over low SNR and image saturation problems in the first B0 volumes which could have introduced noise in the results.
Could that be a plausible explanation? Considering these results (abnormally low noise in the uncorrected control>patients, no real increase in 1-p values with more subjects), do you identify a step in image preprocessing or randomise parameters that could explain them?
Thank you for you invaluable help!
Tim Coolen
MD student
Université de Liège
|