Hello Dave,
I have three comments:
1) In general, one shouldn't simply orthogonalize the data with respect to a
nuisance variable prior to regression (because this would violate the error
covariance assumption and also the model validity assumption if your
covariates of interest are correlated with the nuisance effect) . Instead,
one should include it in the regression model. So, for example, if you take
Vince's advice, I would not orthogonalize the data with respect to the ICA
components deemed as artifacts. I would instead include those components as
(nuisance) regressors. Vince, any comment?
2) If you feel very confident that the effect was at .01 Hz, you could
create sine and cosine "user specified covariates" at that frequency, which
would take care of your problem.
3) I imagine you chose this design because of
psychological/physiological/pharmacological constraints. But, and you might
already know this, but for BOLD fMRI the design you described is poor in
terms of estimation efficiency (which is proportional to its relative
sensitivity). That is, unless you have figured out how to get rid of the
low-frequency noise.
Best,
Eric
----- Original Message -----
From: "David Brennan" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Thursday, August 10, 2006 9:05 AM
Subject: [SPM] Filtering of periodic noise
Hi,
As part of a study we have collected data which appears to be affacted by
periodic signal intensity variations (with a period of 100 seconds). We have
since discovered that this was caused by a faulty air-conditioning unit!
The study involved collecting data over 25 minutes with a baseline for the
first 5 mins followed by a challange for 5 minutes with 15 minutes of
proceeding baseline.
Would it be possible to either low pass filter this data (or to band pass
it) without affacting the SPM analysis? If so is it possible to do this
within SPM?
Thanks for your help
Dave Brennan
|