Hello,
I have used time courses data with 1000 ROIs.  I heard that the data were already high-pass filtered with the cut-off frequency 0.01Hz.  However, when I draw a power spectrum, I still see very high power spectrum below 0.01Hz. 

Previously Chengwei pointed out the same problem on time course data with a smaller ROIs.
I was wondering if the reason is also 3Dfourier function. Or am I doing something wrong?
 
I do not have much knowledge about preprocessing. So I am also wondering if this problem can simply be solved by refiltering.  
 
Thanks!

Dear M. Guo,

I believe the temporal filtering (high-pass filter, cut-off frequency 0.01 Hz) has been correctly applied in the 1000 ROI NIAK release. I attached the averaged normalized power spectrum for one random subject (9907452, run 2) of the NYU site. You will note the quasi-absence of power below 0.01 Hz, and the otherwise 1/f like behavior. The residual small power in that bandwidth is due to the fact that the temporal filtering was applied in native space, and the data was then resampled in stereotaxic space using a non-linear spatial transform with tricubic spatial interpolations. I think those effects are negligible and I would not apply an additional temporal filtering. 

Don't hesitate if you have further questions on the NIAK release. I hope this helps,

Pierre Bellec

Dear Pierre,
 
Thank you for your response. 
I applied the same subject (9907452) and got a similar plot as yours.
But, when I used a different subject (id=1050345, run2) from Peking site, it shows strong peak below 0.01Hz.  Could you look at the attached plot please?
 
The first plot is the averaged power spectrum for subject 1050345, the second plot shows power spectra of all ROIs.
The third and fourth plots are for subject 9907452.
 
Thank you again!
 

Dear M. Guo,

I don't know if it makes sense to pick up on that thread after 6 months, but I believe I (finally) have an explanation for your observations. In NIAK release 0.6.4.3, the data are resampled in the stereotaxic space (with a non-linear transformation) after high-pass filtering. It seems that this operation can induce slow time drifts back in the time series. I would expect these fluctutations to impact a minority of time series, but you may want to re-filter the data. In the next release of NIAK (0.7) the resampling will be done at the same time as motion correction (SPM-style), and the filtering operations will be performed in stereotaxic space following resampling. Other preprocessing steps will also be added (most notably the scrubbing of time frames showing excessive motion, as described in Power Neuroimage 2012). We will make a new release of the ADHD200 preprocessed data at this time. 

Sorry for the inconvenience. Best regards,

Pierre Bellec