Hi, I have two groups (patients,controls) and one covariate (scan group: pre-scanner upgrade, post-scanner upgrade). I want to see if there are FA differences between groups adjusting for the scanner covariate. I have no interest of interaction between group and covariate.
Because I have a very large sample, I created the design files through a script instead of the GUI. I want to double check if my design is good because something weird is happening. I ran first a test randomise with just group (comparing FA differences between groups, no covariates), another randomise with just scanner covariate (comparing FA between scaners) and my final one. The first two take days to run (which is expected, as the sample is large). But the third one, in which I include all the variables, runs in minutes! which makes no sense to me.
* This is my first test (which takes days to run):
design.con:
/NumWaves 2
/ContrastName1 CON>PAT
/ContrastName2 PAT>CON
/NumContrasts 2
/Matrix
1 -1
-1 1
design.mat (here first column is 1 for controls, second column 1 for patients. controls and patients are mingled but are in the same order as the nifti files):
/NumWaves 2
/NumPoints 80
/Matrix
0 1
1 0
1 0
0 1
... etc (80 lines)
* my second test looks exactly the same. only the values of the matrix change depending if the subject was scanned pre or post upgrade.
* then actual final design which runs in minutes, looks like this:
design.con:
/NumWaves 3
/ContrastName1 CON>PAT
/ContrastName2 PAT>CON
/NumContrasts 2
/Matrix
1 -1 0
-1 1 0
design.mat (here first column has 1 for controls, second column 1 for patients, third column is the demeaned scanned group. So, pre upgrade was 1, post upgrade was 0, then i calculated the mean of all the subjects and subtracted that):
/NumWaves 3
/NumPoints 80
/Matrix
0 1 -0.323007...
1 0 -0.323007...
1 0 -0.323007...
0 1 -0.323007...
... etc (80 lines)
Am I doing something wrong in my last design? why does it run so fast? I am even running on less cores than the first two!
This is my randomise command, the same for the 3 test:
randomise_parallel -i all_FA_skeletonised -o ttest_grp_evscangrp -m mean_FA_skeleton_mask -d design.mat -t design.con -n 500 -D --T2 -x
I am using -D even if I demeaned the scanner group because my actual data is not demeaned. Also, is there a way to know what is the best number of iterations? How do I know if 500 is good?
Thanks!
Thanks!
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|