Hello FSL experts,
I've been doing some work with residualized data (similar to that from the 1000FC pipeline) in FSL 4.1.9 and, first, I've been having trouble getting the GUI to successfully complete a Melodic on the data. If the Melodic does work, the components would always seem to not be registered properly, even if registration was not done in the alternative pre-processing, or if we tried to bypass the registration to Hi-Res in the GUI by doing that ourselves first.
I think this is mostly due to the fact that doing 1000FC-type preprocessing includes processing steps from AFNI and other non-FSL algorithms that might not be compatible. Also in some cases we were regressing out the global mean, white matter and CSF. Hence my first question is, do you think registration would fail if Melodic doesn't have the requisite internal structure/landmarks since they've been regressed out? Second, do you know if using other pre-processing, such as the 1000FC script will cause the Melodic to fail?
Also we've been told that Melodic cannot handle negative values. Is this true? We decided to add an offset to remove any negative values just in case.
Lastly, we've decided to do all the preprocessing and motion regression before hand and also register the images to the standard and MNI 2mm brain before sending them to the Melodic. Since the GUI always seems to want to do registration, we've only gotten a command-line Melodic to work. The most recent problem that we've had is that the dimensionality we request (-d 30) is not what is output (21 dimensions). I'm attaching the code and output for you to look at and I would appreciate your advice on all the issues I've raised and this one in particular:
Melodic Version 3.10
melodic -i ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_1kRep.nii.gz ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_ES.nii.gz ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_ESU.nii.gz ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_LS.nii.gz ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_LSU.nii.gz -o TEST_Std_Space_2mm_resolution --nobet --bgthreshold=3 --tr=2.0 --report --guireport=../../report.html --bgimage=bg_image -d 30 --mmthresh=0.5 -a concat
Melodic results will be in ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_1kRep.ica
Create mask ... done
Reading data file ADB91047_01_Y1_20080223_s005_a001_PIRATE_EPI_resid_1kRep ... done
Estimating data smoothness ... done
Removing mean image ... done
Normalising by voxel-wise variance ... done
Excluding voxels with constant value ... done
Data size : 182 x 274102
Starting PCA ... done
Start whitening using 21 dimensions ...
retaining 74.9544 percent of the variability
I think this may simply be a matter of adding something in the command-line code. Thanks in advance for your help with this!