Dear Laith,

The answers depend on what kind of analysis you want to do exactly. If you want to analyse averaged evoked responses then that's what you should invert and in that case there is no problem with data size. But for connectivity analysis one would usually need single trial data. 

A useful feature in SPM12 is virtual montage. If your data can be converted to SPM format then projecting it into source space can be seen as a particular case of montage and montage can be applied online (see 'Switch' option in the montage tool). Some inverse methods (e.g. minimum norm or dipole projection) do not depend on the data, but just on the forward model. So you could pre-compute the inverse projectors for these methods, put them in a montage and you're done. 

If your inverse method does depend on the data, it'd usually depend on some data summary (like covariance matrix) and then you need to find a way to compute this summary in an efficient way from your large dataset. An important rule of thumb to keep in mind is that if your inverse operator depends on the data then you should only project through it the data that were used to compute it. E.g. it'd be wrong to compute inverse projectors from the average data and use them to project raw data. 

When comparing conditions it is good practice to use the same inverse operator for all conditions, otherwise you could introduce artefactual differences. In principle you could compute the inverse operator separately for every epoch but this is not what's usually done. 

Best,

Vladimir


 


-------- Original Message --------
Subject:        [fil.spm] Source Analysis of Long-term Recordings
Date:   Thu, 18 Jun 2015 15:59:58 +0200
From:   Laith Hamid <[log in to unmask]>
To:     [log in to unmask] <[log in to unmask]>
CC:     [log in to unmask] <[log in to unmask]>,
[log in to unmask] <[log in to unmask]>



Dear SPM developers,

I am a user of SPM12 source analysis and I am currently working on a
project which aims to project long-term EEG recordings of sleep or motor
data into the source space. We are mainly interested in the projection
of the whole recording into the source space for further connectivity
analyses. Now since MATLAB cannot exceed a certain data length, how
should we invert the data? Should we invert the single conditions after
averaging or should we just invert the data epoch by epoch? What would
be a wise EEG segmentation approach? Would it be justifiable to use a
different inverse operator for every epoch/condition and then process
the whole source-space data set as if it were inverted in a single step?

Best regards and many thanks in advance,

Laith Hamid

M.Sc. Laith Hamid
Pediatric Brain Imaging (PedBI) group,
Department of Medical Psychology and Medical Sociology,
University Medical Complex of Schleswig-Holstein (UK-SH),
Campus Kiel,
Schwanenweg 20,
24105 Kiel,
Germany
Tel.:  0431 / 597-8742
Fax.: 0431 / 901 - 8745