Over the past few days, I’ve been learning about how to deal with orthogonalization of regressors in your fMRI level 1 GLMs. Consequently, I’ve fallen deep into a new rabbit hole and now I figured it would be useful to the universe to share my collated knowledge over the last 24 hours.
For more leisure learning, I would highly recommend Jeanette Mumford’s excellent video tutorial on orthogonalization as not ideal for dealing with collinearity of your regressors.
Anyway, it turns out (as of June 2021) orthogonalization of modulations are default turned ON in spm 12, in case you were wondering. So we’re going to want to turn this off in our Level 1 GLMs. If you set up your models via the batch editor, you can just do this manually. However, pragmatically this is incredibly tedious and you may want to just tweak existing batch scripts you’ve already generated… for example. 😅
After much googling for how to actually change these settings via batch scripting, I stumbled across this useful link on the spm forum, where some folks were discussing how to turn off orthogonalization of parametric modulator regressors in spm 12. In brief, there are two ways to do this.
Option 1: Set orthogonalization on and off flexibly within each condition
To accomplish this, you will need to define each condition separately in your matlabbatch script. This is a bit more tenable if you are entering your conditions manually on the GUI or even in script with few conditions, but this sort of gets unwieldy fast depending on the number of conditions you want to set.
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).name = 'ConditionName'
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).onset = yourOnsetVector
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).pmod(1).name = 'PmodName'
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).pmod(1).param = yourPmodVector
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).pmod(1).poly = 1
matlabbatch{2}.spm.stats.fmri_spec.sess(1).cond(1).orth = 0 or 1Here’s a screenshot of my whole batch, for reference, in case you were wondering where this code snippet sits with respect to the other lines of code and spm commands.
Option 2: Set orthogonalization on and off flexibly using the “multiple conditions” option
Here, this assumes that you’ve set nicely set up your stimulus onsets in a .mat file, a cell arrays for “names”, “onsets”, “durations”, and “pmod” variables on your model (note your pmod should be a struct). You will also want to define an “orth” cell array (True/False or 1/0), for which of these variables you would like to be orthogonalized. If you want to remove orthogonalization for all of them, this is pretty straightforward and you set all of your array variables to 0. One addition note is that “orth()” is technically a function in matlab, but you can assign values to it without any problem, but you cannot just call orth by itself otherwise you will run into errors.
Here is a screenshot of all of these arrays. Basically the important thing to keep in mind is that the number of columns should be the same across all of your arrays/struct variables, and they should correspond to your regressors. There’s a bit more coding that goes into how I’ve set this up, but I have not gone into that here.
Here is a snippet of code for the multiple conditions version, as well as a screenshot
matlabbatch{2}.spm.stats.fmri_spec.sess.cond = struct('name',{},'onset',{},'duration',{},'tmod',{}, 'pmod', {}, 'orth', {});
matlabbatch{2}.spm.stats.fmri_spec.sess.multi = {'../SOTS/TCB_Event_AllIntervPmod_RewPenRTacc_sots_allTcat.mat'};
matlabbatch{2}.spm.stats.fmri_spec.sess.regress = struct('name', {}, 'val', {});
Alright, that is pretty much it. Feel free to let me know if there’s anything I’ve missed, or if you have any other useful suggestions for implementing orthogonalization in your GLM models in spm12!
computational brain blog 