Hello,
I would like to comment Burkhard's message that discusses the noise
properties of the transformation methods. A separate issue is the bug in
the MaxFilter program that causes noise problems not related to the
mathematical method itself. Jukka Nenonen will send information to this
list on how to circumvent the bug before the fix will be released.
Burkhard makes a good point about the fact that any transformation
method inherently increases noise when virtually moving the head
position over long distances (in the range of several centimeters). This
can be understood, for example, by considering the case where the head
is in a very low position during the measurement and then virtually
transformed to a clearly upper position. If we assume that the MEG
sensors detect the same spatial features of the field from both
positions, then the noise has to increase in the transformation because
the actual measurement has a lower signal-to-noise ratio than the upper
position (closer to the sensors) would produce. Otherwise, we would be
able to artificially increase the SNR just by transforming the head
position closer and closer to the sensors, which obviously cannot be the
case. Recently we have developed a noise weighting solution to prevent
noisy transformations, and we are currently testing the method.
Best regards,
Samu
Burkhard Maess wrote:
> Hallo,
>
> I have never used the -trans option so far. Until now, I have not seen
> the noisy stuff in my current data. But it reminds me to something I
> have seen with other software and another MEG device as well:
>
> It looks as if the distance between helmet and head is changed too
> heavily during the transformation. Our own home-made software has
> produced a similar output whenever the two head positions (before and
> after transformation) were too distant to each other.
> We have used the home-made software to allow for averaging across
> blocks measured from one subject in one or even several sessions.
> There is no problem to be expected if the head position stays the same
> within 1 to 2 centimeters. Nowadays, I use maxmove for the same
> purpose as well.
>
> Computation of a grand average is an interesting step in MEG analysis.
> If your are lucky you may already get a glimpse to your effects.
> Although the neuromag manual recommends the use of MaxMove for much
> crisper grandaverages, I think, this really overemphasizes the
> importance of the field grand average to the MEG community. We have
> the possibilities to do individual source localization. So our grand
> average should be based on those localization results and not magnetic
> field measurements. Therefore, I think, it is a waste of effort to
> recompute a complete MEG experiment to a virtual position at which
> none of the participants was placed during a recording. Sorry, if this
> statement was too harsh, but I stick to the opinion that maxmove has
> to fail if we try to recompute the field distribution for a very
> distant head position (say, e.g. more than 2 centimeters distance
> between measurement and transformed head position).
>
> I regularly use the option '-origin -2.4 1.9 50.4' in the command
> line. I always used this option '-frame head', too. But, I have not
> checked which position the output file stores.
>
> Finally, I would like to say, that Jukka Nenonen has recommended to
> work with manual bad channel detection whenever you analyze data from
> continuous HPI measurements (which I have done exclusively). Use
> '-autobad off -bad 533 2222' when you like to declare MEG0533 and
> MEG2222 bad.
> The manual bad channel detection results in a more stable behaviour
> when estimating the current head position from the HPI coil signals.
> With "autobad on" the head position had with some subjects/data sets a
> bistable appearance. The head appeared as moving between two close
> positions (approx. 2mm away from each other). Nothing serious all in
> all, but different from real behaviour of the subject, I guess.
>
> Cheers,
> Burkhard
>
>
>
--
-----------------------------------------------------------
Samu Taulu
Elekta Neuromag Oy
Street address: Elimäenkatu 22, Helsinki, Finland
Mailing address: P.O. Box 68, FIN-00511 HELSINKI, Finland
Tel: +358 9 756 240 83 (direct), +358 9 756 2400 (operator)
Fax: +358 9 756 240 11
E-mail: [log in to unmask]
|