Hi,
fsl_regfilt does not do band-pass filtering on the original data so does not look like the thing you're interested in. Instead, it filters on the basis of regressing out a set of time series (specified with the -d option).
Normally (using -f) you'd simply list the columns that you think should be removed from the data. With the -F option fsl_regfilt chooses these columns for you, using a frequency/spatial heuristic (that nope, we have not published ;) The first parameter effectively works like the high-frequency cutoff (i.e. 0.15 default means that one 'selection' criterion is how much power the columns in the design have within the 0-0.15 vs the 0.15 to Nyquist range). The second parameter (fthresh2) is _not_ a frequency threshold but a total (high freq. ration combined with spatial regression magnitude) heuristic threshold that empirically speaking 'works reasonably well' (your milage may vary)
On a related note: when filtering you should select components to remove on the basis of having a criterion for their removal, rather than remove everything on the basis of having a criterion for non-removal (as yous eem to be interested in doing). That is, the selection should be guided by 'this is an artefact' rather than by 'this is not signal'. Otherwise your downstream null-hypothesis test(s) will become invalid!
hth
Christian
On 9 Jul 2013, at 05:06, "Yang, Daniel" <[log in to unmask]> wrote:
> I am new to FSL and try to denoise my data using ICA/MELODIC.
>
> When I look at fsl_regfilt, I see that there are a few useful options.
>
> -F,--freqfilt
> filter out components based on high vs. low frequency content
> --freq_ic
> switch off IC Z-stats filtering as part of frequency filtering
> --freq_ic_smooth
> smoothing width for IC Z-stats filtering as part of frequency filtering
> --fthresh
> frequency threshold ratio - default: 0.15
> --fthresh2
> frequency filter score threshold - default: 0.02
>
> I'd like to use the "-F" option, and I know the frequency of my stimuli presentation, so I would like to filter out any frequency higher or lower than the stimuli presentation frequency.
>
> My questions are:
>
> a. How do I specify the high vs. low frequency content? Are they -fthres and -fthres2?
> b. What exactly are -fthres and -fthres2?
>
> Thanks!!
> Daniel
>
> --
> Yung-Jui "Daniel" Yang, PhD
> Postdoctoral Researcher
> Yale Child Study Center
> New Haven, CT
> (203) 737-5454
--
Christian F. Beckmann, DPhil
Professor for Neuroimaging Analysis | MIRA Institute for Biomedical Technology and Technical Medicine | University of Twente | The Netherlands
Principal Investigator: Statistical Imaging Neuroscience | Donders Centre for Cognitive Neuroimaging | Radboud University Nijmegen | The Netherlands
Senior Research Fellow | Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB) | University of Oxford | UK
Hon. Senior Lecturer in MRI Methodology | Computational, Cognitive and Clinical Neuroscience Lab (C3NL) | Imperial College London | UK
Mail: P.O. Box 9101 | NL-6500 HB Nijmegen | The Netherlands
Visiting: Kapittelweg 29 | NL-6525 EN Nijmegen | The Netherlands
Tel/Fax: +31 24 36 10880 / 10989 | e-mail: [log in to unmask]
|