Dear Jelle,
I wonder if I gave you the right advice as it involves tweaking the
SPM.mat file, which is not something I would usually recommend... If you
look at spm_graph.m in the "event-related responses" section, you will
see that what the code does is to first create and store a FIR model for
the condition of interest in variable X ("build a simple FIR model
subpartition (X); bin size = TR") then take the full design matrix
SPM.xX.X and replace the regressors that were initially encoding the
condition of interest with the variable X defined above. So you could do
the same: specify a GLM with the informed basis set, open SPM.mat,
modify SPM.xX.X and save it again before model estimation...
For your additional questions:
1) 32s is the default of the FIR time window and it reflects what you
expect from the duration of the impulse response function of the system
you are studying (the BOLD response here).
2) regressors from conditions with zero duration are rescaled according
to (microtime resolution) / TR. So you would indeed get a value of 16 if
you have a TR of 1s and use the default microtime resolution of 16.
Best regards,
Guillaume.
On 07/08/17 20:58, Jelle Zorn wrote:
> Dear Guillaume,
>
> Thank you for your feedback. I like very much your suggestion of using the informed basis set by default and modifying the design matrix to use a
> FIR model for some selected conditions. The possibility to use different convolution basis sets for different conditions with SPM is something that I have been looking for from the beginning, and I am happy to hear that a workaround to make this possible exists.
>
> However, after having looked at the code in spm_graph, I'm affraid that I don't quite understand how to implement this in practice. Could you elaborate on how to replace some selected conditions in a already existing design matrix by FIR model subpartitions (X) using this code?
>
> I also have two additional questions:
>
> 1) I have read that the time-period for which SPM estimates the PSTH using FIR is 32s by default (as defined by the value 32 in xBF.order in spm_graph). Is this indeed the case, and if so, what would be the implications when using a window length (and order) that exceed this value?
> 2) From having a look at my design matrix acquired by using the FIR basis set only, I see that the FIR events are coded by a value of 16 in each bin which seems to correspond to my microtime onset. Is this to be expected?
>
> Thanks a lot for your help.
>
> Kind regards,
>
> Jelle
>
>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|