"Jörn M. Horschig"
jm.horschig at donders.ru.nl
Fri Aug 31 09:50:17 CEST 2012
the question you are asking does not deserve a definite 'yes' or 'no'.
In any case, note that when averaging over subjects, missing sensors for
one subject will be removed for all other subjects as well. So, if you
are talking about different missing sensors per subject, an
interpolation will definitely be a wise thing to do. In case you choose
for interpolating the missing channels, you can try to check out the new
spherical spline interpolation method (see help ft_channelrepair) -
theoretically that should result in a quite good reconstruction. But the
(standard) nearest neighbour interpolation is fine as well.
With respect to your second question: missing sensors will influence the
neighbour configuration depending on the method you chose for neighbour
selection. For the 'distance' and 'template' method, each neighbouring
channel of the missing will have one sensor less (i.e. the missing one).
The 'triangulation' method will create triangle neighbour no matter how
many channels you have, so it will just ignore the whole in space. In
any case, I'd propose that you check out how satisfied you personally
are with the neighbours by calling ft_neighbourplot. Just last week, I
added the new option that if you use cfg.enableedit='yes', you can
interactively change the neighbourhood structure within the
neighbourplot. If I were you, I would start from the template method. As
an additional piece of information: I am currently working on improving
the neighbour templates, and will upload them probably next week.
All that advertised, let me also say that having 4 out of 275 sensors
missing won't affect your statistics much. Here, it of course depends on
your region if interest, research question etc. E.g. if you are
interested in posterior alpha power, but miss four temporal channels,
there is no reason to worry. So, without knowing what you are looking
for and where the missing channels are, I cannot give you a general
advise what to do. As a last remark, the sensitiviy of the statistics
will of course be influenced by missing sensors, but I doubt that it
will matter much. Btw, I am not quite sure how much the spherical spline
interpolation will buy you here, you might give it a try. I bet that it
will increase your sensitivity more than the nearest neighbour approach,
although both might result in rather low increases. But this is just a
general gut feeling rather than anything I empirically validated.
On 8/30/2012 10:33 AM, Nenad Polomac wrote:
> I am analyzing data obtained on CTF MEG 275 system. However, 4 sensors
> are broken and data ended up with 271 sensor. So, my question is
> should I interpolate the missing sensors with FT_CHANNELREPAI
> <http://fieldtrip.fcdonders.nl/reference/ft_channelrepair>_R_? In the
> case I leave data with 271 sensor, will those missing sensors
> influence neighbors configuration in the statistic and the source
> localisation analysis?
> Thank you in advance!
> Kind regards!
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
Jörn M. Horschig
Donders Institute for Brain, Cognition and Behaviour
Centre for Cognitive Neuroimaging
Radboud University Nijmegen
Neuronal Oscillations Group
FieldTrip Development Team
P.O. Box 9101
NL-6500 HB Nijmegen
E-Mail: jm.horschig at donders.ru.nl
Trigon, room 2.30
NL-6525 EN Nijmegen
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip