Dear fsl users and Steve,
I used randomise to explore regional brain atrophy differences between patients and controls by using flow image from SIENA. The contrast was
1 -1
-1 1
1 0
0 1
, where the first EV was patients and the second EV was controls. The randomise command line was
randomise -i flow.nii.gz -o diagnosis -d design.mat -t design.con -n 5000 -T
I had 12 outputs: 1-4_tfce_corrp FWE corrected p maps; 1-4 _tfce_p uncorrected p maps; 1-4_tstat raw statistic maps. According to previous post in the list (e.g. https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;b38bddc0.0706), I firstly checked _tfce_corrp maps in fslview with range [0.95, 1], where only _tfce_corrp_tstat2 showed a significant tiny cluster in lingual gyrus. I think this result is weird as patients definitely lost greater brain volumes than controls based on PBVC results. So, I checked tstat2 suggested by Steve with both positive (red) and negative (blue) value on by turning on the second colourmap in (i) button in fslview. I tried to set up different range from [0.1, 3] to [0, 1]. There are only several cluster showed on the tstat2 map which does not cover the whole brain edge. I think even if there is no significant cluster in the _tfce_corrp map, there must be raw statistic data in each voxel in the tstat2 map. In order to prove this, I extracted brain edge movements from the flow image within a random voxel where no value showed on the tstat2. The t-test results showed significant group difference between patients and controls. So I think the regional result must be wrong, but I do not know what I did wrong. I have checked whether the matrix matches the sequence of flow image. There was no problem. According to the manual of randomise I think I do need to use -D option as there is no covariates in the matrix. I do no know where did I do wrong to cause this problem. I would greatly appreciate if anyone could provide any possible solution.
Joyce
|