While I agree with Vladimir that there is no "correct way", and that inverting conditions separately before contrasting them is generally preferred (and is consistent with some of the assumptions behind aspects of SPM's spatiotemporal localisation), there may nonetheless be situations where inverting a contrast is preferable, particularly when the differences in evoked responses are small (and SNR low). Otherwise, when SPM performs a temporal SVD and takes only the dominant temporal modes, these might include only the variance in the evoked response (relative to the mean over the epoch) that is common to two or conditions, ie exclude that (smaller) part of the variance that differs between those conditions.
If you do decide to invert a contrast of conditions, one possible solution to the statistical problem of inferring a reliable localisation (eg across participants) that Vladimir raises (see also Henson et al, NI, 2007) might be to apply the inverse operator (calculated from inverting a contrast) back to the original conditions, and write out condition-specific localisations, and compare them directly? This is not possible within the SPM GUI, but might be worth exploring?
R
-------------------------------------------------------
Dr Richard Henson
Assistant Director, Neuroimaging
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
Cambridge, CB2 7EF, UK
Office: +44 (0)1223 355 294 x522
Mob: +44 (0)794 1377 345
Fax: +44 (0)1223 359 062
http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal
-------------------------------------------------------
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
> Behalf Of Vladimir Litvak
> Sent: 24 August 2010 23:52
> To: [log in to unmask]
> Subject: Re: [SPM] EEG, source localization in SPM8
>
> Dear Sun,
>
> There is no 'correct' way of doing things which always works. It
> depends on the data. In principle we recommend inverting the
> conditions you will later compare together. Since you are inverting
> contrasts, how were you planning to do statistics across subjects? If
> you have a way of analysing each contrast separately all the way
> through then you can also invert it separately. It can happen that the
> inverse solution will be dominated by a condition with strongest
> deflection and will 'ignore' the other conditions. This can be
> especially severe for contrasts where there is very low SNR. Perhaps
> you should try inverting the original waveforms and not the contrasts
> and compute the contrast on the inversion results.
>
> Best,
>
> Vladimir
>
>
>
> Sent from my iPad
>
> On 23 Aug 2010, at 03:17, Sun Delin <[log in to unmask]> wrote:
>
> > Dear Vladimir,
> >
> > I found that the solutions of source localization (SPM8) for grandmean
> data (not group inversion) were different according to the number of
> contrasts that should be inverted. In my 2*2 designed ERP experiment,
> signals of 4 conditions averaged across subjects were gotten by using
> 'Grand mean' in the 'other' option. I would like to investigate the source
> information of the difference between conditions, therefore, contrasts were
> made by using 'contrast' in the 'other' option. However, I found that the
> solution of contrast file containing only one contrast (e.g. a1b1 - a1b2)
> was different from the solutions containing several contrasts (e.g. a1b1 -
> a1b2; a1b1 - a2b1; ....). I am confused about how many and what contrasts
> should be inverted simultaneously. I am looking forward to your reply.
> >
> > Bests,
> > Sun Delin
|