Hi FSL experts,
I had a question regarding the use of fslcc. I wanted to compare two different processing methods for individual subject ICA (comparing Subjects 1's ICA maps (melodic) generated from preprocessing method A and B). I wanted to see how ICA components from method 1 compare to 2 (see which are the same). Secondly I wanted to classify which are noise and signal based on a template such as the Yeo 7 network template. I was not sure if the following is correct:
1) Create a binary mask from thresholded zthresh files in the stats folder for all 45 components in method A and 49 components in method B
3dcalc -a thresh_zstat${i}.nii.gz -expr 'bool(a)' -prefix mask/mask${i}.nii.gz #create a binary mask of each thresholded mask preserving positive and negative values
2) Concatenate all thresholded maps into a 4D file called THRESH.nii which contains all 49 ICAs from method B
3) The script below will use the binary mask of method A for each component as the mask for fslcc , and threshold results of abs (cc) at 0.2. It will compare the alternative hypothesis theory component 1, 2,3 ..... against all components found in method B to see how the components correspond to each other.
csh
set i=1
mkdir mask
mkdir fslcc
while ( $i < 46 )
fslcc -m ./mask/mask${i}.nii.gz -t 0.2 thresh_zstat${i}.nii.gz ../../../methodB/MELODIC/stats/THRESH.nii.gz >> fslcc_${i}
@ i++
end
4) I will take the Yeo 7 template and save each component into a separate subbrick as a binary mask such that subbrick 1 is component 1 etc. Then I will apply the inverse transform to get it from MNI space to subject space and use fslcc to compare against networks found in method A and B for noise classification.
fslcc --nodemean 7network.nii.gz THRESHA.nii.gz >> methodA_signalnoise.txt
fslcc --nodemean 7networks.nii.gz THRESHB.nii.gz >> methodB_signalnoise.txt
-----Is step 4 valid to do since the network masks are just binary files . I added --nodemean do to the fact that the binary file demeaned would go to zero.
-----For the cross correlation since the template is a series of masks of 0's and 1's and the THRESH.nii data is the thresholded zscores, will the correlation be sensitive enough or accurate to pick up that belongs as signal vs noise (ie not in template so uncorrelated) since one of the inputs is just a mask itself?
----Is there a better way I should be doing this last stage?
http://psych.colorado.edu/~anre8906/guides/01-ica.html (reference)
Thanks,
Ajay
|