Hello!
I am very new to functional connectivity analysis and have two questions:
1) I am about to start resting state analysis and have read mixed inputs about whether to use or not use low-pass filter when analyzing resting state data. CONN has a default setting [0.008 0.09] for band pass filter that has been recommended to do denoising for resting state analysis. However, some publications such as from Smith et. al., 2013 (Resting-state fMRI in the Human Connectome Project) suggest not using low-pass filtering in cases where "multiple voxels'' timeseries are averaged together are less prone to the effects of thermal noise, and hence may be degraded, rather than improved, by any lowpass temporal filtering. Such analyses include: seed-based connectivity where the seed is an extended ROI and the data has been extensively spatially smoothed", which is the case in my condition. I was wondering if it would be a good idea to use [0.008 inf] for my resting state analysis?
2) My next question is about the role of the low pass filter in removing high frequency noise, such as cardiac effects and physiologic effects. So, if I dont use a low-pass filter, then I won''t be able to remove these physiological noise effects. So, I was wondering if I do choose to use [0.008 inf] as my band pass filter, how would I take care of these physiological effects? My TR is 0.662 for reference. Also would nyquist frequency play any role here in choosing the low-pass filter max value?
Thank you!
Hi
Regarding (2), the use of low-pass filter to remove physiological noise sources is only effective when your scanner acquisition Nyquist frequency (~0.75Hz for TR=0.66s) is above the main frequency range of those noise sources (e.g. respiratory ~ 0.3Hz and cardiac ~1Hz in adults), as otherwise noise ends up being aliased across the entire spectrum and filtering does not help much (in your case your TR sits somewhere in the middle there so low-pass could be helpful in removing slow respiratory-based fluctuations, but not cardiac fluctuations). In general (rather than, or in addition to, filtering) it is aCompCor, combined with motion regression and scrubbing, which I find helps the most in removing physiological noise sources from the BOLD signal.
And regarding (1), I see the main role of low-pass filtering not as denoising but rather as a way to focus your functional connectivity measures on the "slow" range of resting-state fluctuations that have been (historically) most studied. This general practice is changing, nevertheless, and many studies today are not only including higher frequencies but also explicitly studying how connectivity changes across the spectrum, so in my opinion the choice of whether to use low-pass or not should be taken in this larger context of how you want your study to compare and/or add to that larger literature. That said, you are also right that the process of spatial smoothing (or aggregating accross voxels within an ROI) heavily influences the spectral properties of the BOLD signal, with spatial-smoothing tending to also act to de-emphasize higher frequency signals, so the comparison of connectivity values across different studies with possibly different TRs, temporal filters, and/or spatial smoothing levels becomes very complicated.
Hope this helps
Alfonso
Originally posted by Nam Kap:
Hello!
I am very new to functional connectivity analysis and have two questions:
1) I am about to start resting state analysis and have read mixed inputs about whether to use or not use low-pass filter when analyzing resting state data. CONN has a default setting [0.008 0.09] for band pass filter that has been recommended to do denoising for resting state analysis. However, some publications such as from Smith et. al., 2013 (Resting-state fMRI in the Human Connectome Project) suggest not using low-pass filtering in cases where "multiple voxels'' timeseries are averaged together are less prone to the effects of thermal noise, and hence may be degraded, rather than improved, by any lowpass temporal filtering. Such analyses include: seed-based connectivity where the seed is an extended ROI and the data has been extensively spatially smoothed", which is the case in my condition. I was wondering if it would be a good idea to use [0.008 inf] for my resting state analysis?
2) My next question is about the role of the low pass filter in removing high frequency noise, such as cardiac effects and physiologic effects. So, if I dont use a low-pass filter, then I won''t be able to remove these physiological noise effects. So, I was wondering if I do choose to use [0.008 inf] as my band pass filter, how would I take care of these physiological effects? My TR is 0.662 for reference. Also would nyquist frequency play any role here in choosing the low-pass filter max value?
Thank you!
Thank you Alfonso! That really helped clear my doubts.