[Eeglablist] ICA and dipfit: high residual variance

[Makoto Miyakoshi]

Dear Giovanni,

After (e) you can see your component scalp maps. Try 'Tools' -> 'Reject
data using ICA' -> 'Reject components by map'. As John gave you the
reference, good IC results are usually represented by dipolar scalp maps.
See below.

1) Decomposed time series are good but scalp maps are messy -> your channel
location file is wrong
2) Both time series and scalp maps are messy -> your data are wrong

By the way, what's the r.v. values? You should tell us what numbers you
have to give us a rough guess.

Makoto

On Wed, Sep 21, 2016 at 5:30 AM, Giovanni Vecchiato <
giovanni.vecchiato at gmail.com> wrote:

> Dear all,
>
>
>
> I’m performing an ICA followed by a dipole source localization (DIPFIT)
> and I’m not satisfied by the outcome of the analysis because it returns too
> high values of residual variance (RV%).
>
>
>
> Here the details of the EEG signal acquisition and processing (eeglab
> v13.6.5b):
>
>
>
> 1.    Data acquisition with EGI (srate = 500 Hz; 129 channel):
>
> a.    two blocks of an execution task (EX) + two blocks of motor imagery
> task (IM) (overall, around 17 minutes of recording)
>
>
>
> 2.    Signal processing
>
> a.    PREP pipeline with line frequency removed = [50 100 150 200]; the
> other parameters are the default ones
>
> b.    high pass filtering @ 0.1 Hz
>
> c.    data segmentation in EX and IM conditions ([-1, 6] seconds), 40
> trials per condition
>
> d.    PCA on the IM dataset (35 trials remained after trial rejection)
>
> e.    Extended ICA with 27 PCs to retain (which explain the 99% of the
> variance)
>
> f.     DIPFIT analysis using the MNI template and manual co-registration
> (the actual EGI channel positions are correctly loaded) returning an
> average residual variance of (0.5 +/- 0.2) which values seem too high to be
> taken into account for a following analysis of ICs.
>
>
>
> I have tried several modification to the above processing chain (e.g. low
> pass filtering @ 45 Hz; ICA on a larger number of PCs and on a larger
> number of trials (80); ICA performed with AMICA; ICA performed with Brain
> Vision Analyzer – Extended ICA) but with no improvement of the results. I
> have also tried on different datasets and tasks with the same results.
>
>
>
> Do you all have any suggestion to achieve lower DIPFIT residual variance
> which are plausible from the neurophysiological point of view?
>
> I can also share a sample dataset if you consider it useful.
>
>
>
> Thanks in advance,
>
> Best,
>
> Giovanni
>
>
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.
> ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20161017/d467e0fe/attachment.html>