Parameter source of mc parameters in 1stlvl model setup


I was wondering about the influence of the keyword parameter_source in SpecifyModel on the model fit.

The documentation says:

parameter_source: (‘SPM’ or ‘FSL’ or ‘AFNI’ or ‘FSFAST’ or ‘NIPY’,
nipype default value: SPM)
Source of motion parameters

The default value is SPM, so it seems obvious that if motion correction has been done with fsl’s mcflirt, you would want to change it to FSL. However, in the fsl example workflow, the model is specified with SPM as parameter source, even though motion correction was done with mcflirt. So, does it matter at all what you choose there?

The reasoning I am asking, is that for a couple of all the runs in my dataset, fitting the autoregressive model during film_gls crashes due to a singular design matrix, and I can’t find an obvious reasons for that. So, perhaps the parameter_source might play a part in that?

The translation code is here:

if source.upper() == 'FSL':
    params = params[[3, 4, 5, 0, 1, 2]]

If your mcflirt parameters show up in more than six columns, then there could be an issue, but otherwise it’s a simple reordering, which should have no effect on film_gls. That said, there’s no harm in specifying parameter_source='FSL'.

1 Like

Makes sense. Thanks!