[FieldTrip] Odd behavior in DICS

[Azeez Adebimpe]

Hi Alexander,

DICS also works similarly to eLORETA but you need to remove bias after  you
compute the source.

1. Estimate the noise from the data  with
cfg.dics.projectnoise='yes';  the default is no, so you need to set  it
while computing source.
2. Divide the pow by noise estimate
Sourcedb=source;
Sourcedb.avg.pow=sourcedb.avg.pow./sourcedb.avg.noise;
3.  plot the Sourcedb.avg.pow; to see
hope it help,
Best,
Azeez

On Mon, Aug 15, 2016 at 10:18 AM, Nakhnikian, Alexander <
Alexander_Nakhnikian at hms.harvard.edu> wrote:

> Hello All,
>
>
> I'm getting strange results from DICS applied to EEG data. I'm wondering
> if anyone else has run into these issues and/or can suggest a solution.
>
>
> I don't have individual sMRI so I'm using the standard BEM and MRI that
> come with FT. I constructed the lead fields as follows. First, I constrain
> the dipoles of interest to the cortical mantle using the AAL atlas. I then
> compute the lead fields using ft_prepare_leadfields; I don't have a
> baseline for these data so the lead fields are normalized. I feed the
> precomputed lead fields and data into ft_sourceanalysis to find the
> source-level power for each subject.
>
>
> 1) DICS, but not eLORETA, returns extremely large results at certain
> voxels. These extrema dominate the grand averages. In each subject, these
> voxels are always at the edge of the cortex, near the skull. The voxel
> locations are not exactly the same across subjects, but for any given
> subject they tend to appear in one of a few locations (i.e. over the right
> occipital cortex). The lead fields have the same magnitude (following
> normalization) across all voxels. I did notice, however, that voxels at
> which extreme values appear tend to have a larger condition number (~12)
> than voxels at which abnormal values do not appear in any subject (~2). I
> do not know whether this is relevant but I thought it might have some
> effect on adaptive but not non-adaptive filters. Since I'm constructing the
> forward model using FT templates, I thought someone else might have
> encountered this problem before.
>
>
> 2) As I mentioned above, eLORETA does not return such extreme results;
> furthermore, when we contrast controls with schizophrenia patients using an
> independent samples t-test with cluster control for multiple comparisons we
> find significant differences between groups using eLORETA but not DICS. The
> raw differences between the grand averages returned by DICS and eLORETA are
> similar, but when we run stats using DICS most of the correct p values are
> equal to 1 with one or two in the 0.8-0.9 range. I'm looking for converging
> evidence between different source analysis methods and it's not clear to me
> why DICS and eLORETA return such different results.
>
>
> 3) I thought that voxels are which unusually large values occur in
> controls and patients could be throwing off the stats; however, DICS does
> not return significant contrasts between control and schizophrenia even
> when I mask out the extreme voxels.
>
>
> Thanks in advance for any advice.
>
>
> Best,
>
>
> Alexander
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20160815/3430cfcb/attachment-0002.html>