36p confounds from FMRIPREP for FSL FEAT

Hi there,

This is my 1st post. I have these outputs from fMRIprep and it seems like there are 37 rather than 36 motion parameters including basic, derivative1, power2. Did I mess that up somehow?

global_signal
global_signal_derivative1
global_signal_derivative1_power2
global_signal_power2
csf
csf_derivative1
csf_power2
csf_derivative1_power2
white_matter
white_matter_derivative1
white_matter_power2
white_matter_derivative1_power2
csf_wm
trans_x
trans_x_derivative1
trans_x_derivative1_power2
trans_x_power2
trans_y
trans_y_derivative1
trans_y_power2
trans_y_derivative1_power2
trans_z
trans_z_derivative1
trans_z_derivative1_power2
trans_z_power2
rot_x
rot_x_derivative1
rot_x_derivative1_power2
rot_x_power2
rot_y
rot_y_derivative1
rot_y_derivative1_power2
rot_y_power2	rot_z
rot_z_derivative1
rot_z_power2
rot_z_derivative1_power2

I am getting a warning from FEAT about EVs being linear combo of others, too.

I also want to chop a couple of volumes for T1 equilibrium but FSL doesn’t seem to be clever enough to ignore that same number of rows in my confound file when I choose to Delete 2 volumes on the Data tab in the FEAT GUI. What’s the best way to match these?

Thanks a lot!
DJO

Hi @djo155 and welcome to neurostars!

Not all of these are motion parameters. Only those that begin with trans and rot. The other are compartment-averaged signal (and their derivatives/power expansions). The “37th” one for you is the combined csf_wm average. This isn’t typically included in 36P.

Hard to say how big of a deal this is without seeing something like a correlation heatmap between regressors. But GLMs should be well-equipped to handle multicollinearity, although it should be minimized if possible.

One possible solution, completely Python based, is to use Nilearn and Pandas to alter the images and crop the confounds, respectively. You can even do your GLM in Nilearn as well, obviating the need for FEAT, if that is more convenient.

Best,
Steven

Ah, that’s great. I’m pretty good with scripting unix + fsl commands for that as an alternative (`fslroi input output 2 -1’ to remove the 1st 2 volumes).

Thanks so much!