[FieldTrip] on spectral analysis of MEG data

Jose joseluisblues at gmail.com
Wed Aug 24 18:04:07 CEST 2016


dear Fieldtrip community,

I'm performing analysis of CTF MEG data. I have two conditions of interest:
A and B, and I have performed spectral decomposition with the wavelet
method to obtain the power (frequencies from 6 to 50 Hz). I have some
questions regarding my analysis:

1. As a first approach I wanted to do analysis at the sensor level and then
to move at the source level. This is a rather exploratory analysis. I have
noted oscillatory activity when analysing ERFs, so I decided to perform
spectral analysis but I don't have any a priori hypothesis about any
frequency or sensor. I'm using a cluster-based permutation (CBP) test to
evaluate significance, but I'm not sure how to do a multi-sensor analysis.
Since I have computed the power from frequencies ranging from 6 to 50 Hz, I
could perform a CBP test for each frequency, but this would become a
multiple comparison problem isn't? Otherwise what I have see in some
publications is to average across sensors, and do a permutation test
similar to what is done in single-sensor analysis,

2. I have performed spectral analysis of my data for magnetometers and
gradiometers. I was wondering if I should expect the same results for both
magnetometers and gradiometers. Since the activity of the planar gradient
do not cover the same sensors than the magnetometers I wouldn't expect it.
However, perhaps the effects should fall within the same frequency
boundaries?

3. I was wondering why in the "Within subjects experiments" example of the
tutorial "Cluster-based permutation tests on time-frequency data" (
http://www.fieldtriptoolbox.org/tutorial/cluster_permutation_freq) the
cfg.alpha is set to 0.025. Since the data are planar gradients, thus
positive values, one would need to perform a one-side test, and thus an
alpha of 0.05? Maybe in this case the data was baseline corrected and thus
one would then have positive and negative values, might that be the case?
As a general question, if I'm working with planar gradient data, I should
need differentially set my alpha value regarding if is baseline-corrected
or not, is that correct?

4. I have seen some publications that include an additional correction to
the analysis after performing a permutation test. For instance, Busch et al
(J Neurosc 2009) use a resampling test and the FDR method to test
significance of differences in power in the Fz channel. The general
question is when one can apply only a resampling/permutation test, and when
one have to include a correction like FDR?

Thanks in advance,
Jose
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20160824/a352f01c/attachment.html>


More information about the fieldtrip mailing list