Dear Hikaru
Modelling and removing known confounds, in particular movement and global signal changes (and, if you have physiological data, those signal variations that can be explained by these) is necessary independently of how you filter your data (high & low-pass or AR(1) & high-pass), simply because these confounds can (and usually will) introduce spurious correlations
The problem when applying suggestions on which pre-processing strategy is optimal for task-based fMRI to resting-state data is, that you are somewhat looking at different things. Whereas in task-based fMRI you want to maximize efficiency with respect to your stimulus-function (more precisely, the stimulus function convolved with the canonical hrf). Now in resting-state data, you don't have the situation of rapid event-related designs, where low-pass filtering could compromise your inference. Rather, you are dealing with (relatively) slowly evolving mental states. Think of it from a psychological perspective: The subjects are not thinking ""nothing" but rather will engage in a spontaneous train of thought (unconstrained cognition), in which the change of "topics" (so to speak) will be considerably slower than in a rapid ER design. Add to that the fact, that the signals you measure have passed through the BOLD-response acting like a low-pass filter itself, few (if any) higher-frequency changes in your data should correspond to the actual signal if interest, i.e., (co-) fluctuations of different regions evoked by the changing mental states in unconstrained cognition.
My practical advice would be to repeat your analysis using the "standard" setting, i.e., employing a band-pass filter. If you can replicate your findings, this would add a lot of confidence into them.
Best
Simon
________________________________________
Von: SPM (Statistical Parametric Mapping) [[log in to unmask]]" im Auftrag von "Hikaru Takeuchi [[log in to unmask]]
Gesendet: Montag, 14. Mai 2012 15:34
An: [log in to unmask]
Betreff: Re: [SPM] AW: [SPM] Autoregressive model or low-pass filter in RSfMRI analyses
Dear Prof. Torben Lund, and Prof. Eickhoff, Simon and other SPM users,
Thank you very much for the information and references.
I went through the information. I think the same problem remained (low pass filter + high pass filter + modeling other regressors V.S. autoregressive model + high pass filter + modeling other regressors?).
I understood the importance of modeling movement related factors and some other global variables, though unfortunately, we don't have physiological or respiratory related information in obtained data. They helped a lot. But though I may be wrong, probably modeling these things is important regardless of whether we use low pass filter or autoregressive model instead of low pass filter, isn't it?
I am using autoregressive model based on the reference of (Della-Maggiore et al., 2002, Neuroimage, An Empirical Comparison of SPM Preprocessing Parameters to the Analysis of fMRI Data). For example, abstract says "When dealing with fMRI time series with short interstimulus intervals (<8 s), the use of first-order autoregressive model is recommended over a low-pass filter (HRF) because it reduces the risk of inferential
bias while providing a relatively good power." and the discussion says "However, when dealing with fMRI time series where stimuli are presented close together, such as in rapid presentation eventrelated designs, the risk of inferential bias would increase. In those cases, the use of AR1 is recommend over the low-pass filter as it appears to decrease the number of false positives while maintaining a relatively
high power". We can use both low pass filter and autoregressive model, but it seems in this paper, the use of both tended to increase false positives when compared with autoregressive model only and it seems the autoregressive + no low pass filter is the best in terms of controlling false positive as well as power of the analyses. Doesn't these suggest we should use autoregressive model and not low-pass filter in resting state fMRI analyses, too, or things are different when other variables are put in models or rsfMRI analyses?
I am sorry if I am having a totally wrong question as I am not understanding the details of these statistics and models well, but I would be happy, if someone would help me as many reviewers are having same questions in the use of autoregressive model over low-pass fileter.
Hikaru Takeuchi
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Kennen Sie schon unseren neuen Film? http://www.fz-juelich.de/film
Kennen Sie schon unsere app? http://www.fz-juelich.de/app
|