help
help > RE: high-pass filter
Feb 14, 2021 01:02 AM | Alfonso Nieto-Castanon - Boston University
RE: high-pass filter
Dear Bram,
Difficult to tell, but my first guess would be that this is just some issue in the way the power spectra is being displayed. I would suggest perhaps to demean the timeseries before entering it into your power-spectrum-display function of choice to make sure that the baseline/constant values of the BOLD signal are not leaking and being displayed as those strange-looking low frequency plateaus in your figures. For example, if you have Matlab's signal processing toolbox you may use something like:
pspectrum(x-mean(x),1/TR);
to display the power spectrum of x disregarding the constant/DC term and acquired at 1/TR sampling frequency (in your case ~0.625?)
Hope this helps
Alfonso
Originally posted by Bram Beckers:
Difficult to tell, but my first guess would be that this is just some issue in the way the power spectra is being displayed. I would suggest perhaps to demean the timeseries before entering it into your power-spectrum-display function of choice to make sure that the baseline/constant values of the BOLD signal are not leaking and being displayed as those strange-looking low frequency plateaus in your figures. For example, if you have Matlab's signal processing toolbox you may use something like:
pspectrum(x-mean(x),1/TR);
to display the power spectrum of x disregarding the constant/DC term and acquired at 1/TR sampling frequency (in your case ~0.625?)
Hope this helps
Alfonso
Originally posted by Bram Beckers:
Dear Alfonso,
Thank you for your reply.
Some info regarding the data:
This is a single subject - single session - long functional run (36min, 1350 measurements). It is a optimally combined, but non-denoised multi-echo run (not just concatanated, but combined using T2* combination method). It has undergone default preprocessing with minimal smoothing, but deliberately not denoised as this dataset will be used as a reference to compare several denoising strategies.
The global signal I extracted for the whole timeseries (mean of voxel values) within a mask containing our regions of interest (mostly insula, cingulate cortex, thalamus and brainstem). For future purposes, I would extract global signal whole brain. I used the multitaper method in MATLAB to compute the power spectrum. I suspect the odd constant power spectrum below 0.01Hz might be caused by a too high number of tapers. If I use a more plug&play approach with a MRI quality toolbox I get a power spectrum with a very sharp peak below 0.01Hz.
I double checked the denoising tab and the bandpass setting was definitely at [0.008 inf] (you can even see the improved distribution of connectivity values after denoising). The 'denoising' output is the inputfile with prefix -d
Any idea what I've been doing wrong?
Best regards,
Bram
Thank you for your reply.
Some info regarding the data:
This is a single subject - single session - long functional run (36min, 1350 measurements). It is a optimally combined, but non-denoised multi-echo run (not just concatanated, but combined using T2* combination method). It has undergone default preprocessing with minimal smoothing, but deliberately not denoised as this dataset will be used as a reference to compare several denoising strategies.
The global signal I extracted for the whole timeseries (mean of voxel values) within a mask containing our regions of interest (mostly insula, cingulate cortex, thalamus and brainstem). For future purposes, I would extract global signal whole brain. I used the multitaper method in MATLAB to compute the power spectrum. I suspect the odd constant power spectrum below 0.01Hz might be caused by a too high number of tapers. If I use a more plug&play approach with a MRI quality toolbox I get a power spectrum with a very sharp peak below 0.01Hz.
I double checked the denoising tab and the bandpass setting was definitely at [0.008 inf] (you can even see the improved distribution of connectivity values after denoising). The 'denoising' output is the inputfile with prefix -d
Any idea what I've been doing wrong?
Best regards,
Bram
Threaded View
| Title | Author | Date |
|---|---|---|
| Bram Beckers | Feb 3, 2021 | |
| Alfonso Nieto-Castanon | Feb 4, 2021 | |
| Bram Beckers | Feb 8, 2021 | |
| Alfonso Nieto-Castanon | Feb 14, 2021 | |