Questions about ICA

✉️ Mailing List Archive (read-only)
1

Dear Maria,

I'm replying on mne_analysis so other people can join in or follow our
conversation. Please join if you haven't already (
Mne_analysis Info Page)

Hi Denis,

I would yet need noise covariance matrixes for roi analysis. I have

calculated them with EOG rejections using mne_process_raw as shown below. I
wonder if it causes problems if the noise covariance matrixes are
claculated with EOG rejections but averaged signals are calculated without
EOG rejections (and ica is applied on epochs).

I would not see why that would be a problem, it's a good ideat to not
estimate the covariance matrix on biological artefacts. Don't you use ICA
for handling EOG/ECG? Should be fine then.
What I normally do is to compute the noise covariance after applying ICA.
That makes things even more matching, e.g., also applying the same amount
of rank reduction.

I can't use rejections because in the sustained field analysis long

epochs cause too many rejections. For transient analysis whit shorter
epochs EOG rejections worked well. Should I calculate new covariance
matrixes for sustained field analysis using data where EOG rejections are
not used but ica is applied instead? Any advice would help a lot!

Would it make sense to compute your noise cov from preceeding baseline
segments?
Then you couls simplify things a bit.
Apply ICA to your epochs and then use `mne.compute_covariance` on your
epochs.
FYI in there's also `mne.compute_raw_data_covariance` for raw data.

This is how I have calculated covariance matrixes:

foreach file ( ah_sentences1b ah_sentences16b ah_sentences1b2 )
        mne_process_raw --raw "$file"_raw.fif \
        --digtrig "STI 014" --grad 0 --projon --events "$file"_merged.eve

\

        --cov

/scratch/braindata/mhhakone/intell/scripts/cov_desc_merged.cov --savecovtag
"-merged.cov"

end

where cov_desc_merged.cov:

cov {
        name "intell"

        gradReject 3000e-13
        magReject 4e-12
        eogReject 150e-6
        logfile log_averaging

        def {
                name "all"
                event 1
                tmin -0.1
                tmax 0
                bmin -0.1
                bmax 0
        }

}

Thanks,
Maria

Hope this helps,
Denis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140804/86bee43c/attachment.html