[FieldTrip] What kind of baseline correction to use for time-frequency analysis

"Jörn M. Horschig" jm.horschig at donders.ru.nl
Tue Jun 17 09:46:47 CEST 2014


Hi Esther,

>  1. If I understand it correctly, with the default config setting for
>     cfg.polyremoval the mean of the whole trial is subtracted (prior
>     to spectral analysis) to get rid of the offset in oscillations.
>
'Getting rid of the offset in oscillations' is not the purpose of 
demeaning prior to doing a frequency analysis. See here: 
http://fieldtrip.fcdonders.nl/faq/why_does_my_tfr_look_strange

>  1. Does it mean that since I’m doing analyses on a participant level
>     first, I only need to do this kind of baseline correction (from
>     stimulus onset on)?
>

No, this is not a baseline correction as you are thinking of here. 
Remember that a power spectrum is obtained by squaring the fourier 
coefficients, i.e. power is always positive. Also, you want to baseline 
correct per frequency, because of the 1/f characteristic.

>  1. I thought that I should /not/subtract the mean of a window prior
>     to stimulus onset, from the raw trial data before doing
>     time-frequency analysis (the kind of baseline correction you would
>     do with ERP analyses), but I see this happening in some articles.
>     Did I not get it right, or is this related to the distinction
>     between evoked and induced power, or something else?
>
As mentioned in the link above, you should demean the whole trial in 
FieldTrip because of the way the mtmconvol method is implemented. Any 
other "baseline correction" in the time-domain will only affect the DC 
bin (ie 0Hz), but not the estimatoin at individual frequencies. This is 
because baseline correction in the time-domain is a mere subtraction of 
a your data by a single number. Hence, the frequency content of your 
data stays unchanged.

>  1. Or is it an option (and should I) subtract the mean of a window
>     prior to stimulus onset only after decomposition of the data in
>     the power spectrum, for all trials?
>
Yes, in any plotting function, you can set cfg.baseline and 
cfg.baselinetype. You can also call ft_freqbaseline. Note baselining is 
also not a trivial thing here. You can serious harm (i.e. invalidate) 
your statistics if you're doing it wrong (but that depends on your 
experimental design and question). Usually, if you compare conditions 
you should not do a baseline correction at all (otherwise any 
statistical difference might show up due to a difference in baseline to 
start with).

Best,
Jörn

On 6/16/2014 4:35 PM, Esther Eijlers wrote:
> Dear community,
>
> Although I’ve seen other questions related to this topic, I still have 
> some questions regarding baseline correction when executing a 
> time-frequency analysis:
>
> After time-frequency analysis, I want to use the trial output either 
> to compare (within participants) trials from different conditions or 
> to put the datapoints of different trials into a regression (also on 
> participant level).
> The question for me is how to deal with a baseline correction in this 
> situation? The data before each trial of interest is in reaction to 
> viewing a fixation cross.
>
>  1. If I understand it correctly, with the default config setting for
>     cfg.polyremoval the mean of the whole trial is subtracted (prior
>     to spectral analysis) to get rid of the offset in oscillations.
>  2. Does it mean that since I’m doing analyses on a participant level
>     first, I only need to do this kind of baseline correction (from
>     stimulus onset on)?
>  3. I thought that I should /not/subtract the mean of a window prior
>     to stimulus onset, from the raw trial data before doing
>     time-frequency analysis (the kind of baseline correction you would
>     do with ERP analyses), but I see this happening in some articles.
>     Did I not get it right, or is this related to the distinction
>     between evoked and induced power, or something else?
>  4. Or is it an option (and should I) subtract the mean of a window
>     prior to stimulus onset only after decomposition of the data in
>     the power spectrum, for all trials?
>
> Thanks a lot!
>
> Best,
> Esther
>
>
> Disclaimer
> ------------------------------------------------------------------------
> De informatie verzonden in dit e-mail bericht is vertrouwelijk en is 
> uitsluitend bestemd voor de geadresseerde van dit bericht. Lees 
> verder: www.eur.nl/email-disclaimer <http://www.eur.nl/email-disclaimer>
> The information in this e-mail message is confidential and may be 
> legally privileged. Read more: www.eur.nl/english/email-disclaimer 
> <http://www.eur.nl/english/email-disclaimer>
> ------------------------------------------------------------------------
>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip


-- 
Jörn M. Horschig
PhD Student
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
The Netherlands

Contact:
E-Mail: jm.horschig at donders.ru.nl
Tel:    +31-(0)24-36-68493
Web: http://www.ru.nl/donders

Visiting address:
Trigon, room 2.30
Kapittelweg 29
NL-6525 EN Nijmegen
The Netherlands




More information about the fieldtrip mailing list