Dear Antonia,
In EEG analysis we deal with artefact by rejecting trials. This is not
applicable to NIRS where the time series are modelled with a
convolution model of continuous data similarly to fMRI. It is possible
to suppress artefacts by manipulating the whitening matrix in GLM
estimation so that for artefactual segments you specify very high
variance which effectively nulls them out. This approach will be
implemented in some new MEEG functionality I'm developing for SPM12
and it might be possible then to also use it for NIRS (at least I'll
try to keep it in mind and to make things generic). At the moment it
is possible to do it by changing the low-level code of spm_spm.
Best,
Vladimir
On Wed, Apr 18, 2012 at 10:19 AM, Antonia Hamilton
<[log in to unmask]> wrote:
> sending again in the hope that people who know the answer are back from
> their holidays ...
>
> ---------------------
>
> Dear SPM people,
>
> I am trying to use SPM to analyse some infant NIRS data, using the SPM-NIRS
> package from Korea (http://bisp.kaist.ac.kr/NIRS-SPM.html) which calls SPM8
> for building and estimating the GLM.
>
> The problem I have concerns artefact rejection. As this is data from
> babies, there are parts of each trial where the baby didn't look or moved
> and I need to ignore this data. For every n x c dataset, I have an extra n
> x c matrix with 1s in all the locations with good data and 0s in all the
> locations with artefacts. So when I fit my GLM, I want to be able to
> include this artefact coding and to ignore the bad parts of the data. This
> currently isn't implemented in the SPM-NIRS package, but I know it must be
> implemented in the EEG bits of SPM8 where it is possible to reject blinks
> etc. So I am hoping that someone can tell me how SPM8 fits the GLM to EEG
> data while ignoring artefacts, and which bits of code I need to grab to do
> the same with my NIRS data.
>
> Thanks for your help,
>
> Antonia
|