Hi Jason,
It is indeed a great question.
LOW-PASS FILTER
Application of a low-pass filter (e.g., retaining frequencies slower than 0.1 Hz) as a preprocessing step is (as the name indicates) aimed at retaining low-frequencies (=removal of higher frequencies). If you use MELODIC for ICA these higher frequencies are usually nicely captured in one or more of the components and separated from the other components. For an example of what high-frequency noise looks like, see slide #32 of the "MELODIC slideshow" for which you will find a link here: http://www.fmrib.ox.ac.uk/analysis/research/melodic/. I like slide #32 because it shows the high-frequency signal in the BOLD data which you clearly are not interested in and want to remove as best as you can. For seed based analysis it is thus necessary to apply some processing step that removes this high frequency noise. One example of how to do this is to apply a low-pass filter another example is using ICA to identify and remove the noise and then continue with your seed based analysis.
HIGH PASS FILTER
Removing low-frequency drift (e.g. scanner drift) is the aim of the a high-pass filter (e.g. retaining frequencies faster than 0.009 Hz) and is a step I would advise before ICA and before seed-based analysis.
Because the raw BOLD signal timecourses are noisy due to scanner artifacts, participant motion, and physiological
sources such as cardiac and respiratory cycles, applying a temporal bandpass filter with the above mentioned cut-offs is likely a good idea. Nonetheless, two quick remarks: 1) the exact frequencies that work best for removing noise are not know and might be different for differently acquired data (think: different TR), and 2) applying a bandpass filter does -of course- not mean that all ALL noise is removed.
Many of the topics mentioned above are discussed in more detail in one of our methodologically oriented papers: Van Dijk et al., JNeurophys, 2010 (http://jn.physiology.org/cgi/content/abstract/103/1/297), in particular the section of the introduction under the head "Physiological noise, anticorrelations, and optimization".
Hope this helps,
Koene
On Mon, 19 Mar 2012 16:11:51 +0000, Stephen Smith <[log in to unmask]> wrote:
>I often wonder about the same thing��
>
>We don't generally do (or recommend) lowpass filtering (the "0.08" part of this). If you have high-frequency noise then doing lowpass could help ameliorate this, but there's better ways to do that in general than simple lowpass filtering (like physiological regressors and/or ICA cleanup), and those methods shouldn't have the downside of losing potentially valuable higher-frequency signal.
>
>We do generally remove low frequency drift though - but the exact nature of how aggressive this should be will depend on your data.
>
>Cheers.
>
>
>
>
>On 19 Mar 2012, at 16:07, Jason Steffener wrote:
>
>> Hello all,
>> I know this is a simple question, which I should know the answer to
>> but I am going to put it out there.
>>
>> Can someone tell me why the study of the resting state temporally
>> filters the data to only those frequencies between ~0.01 Hz and 0.08
>> Hz?
>>
>> I know that this was done in Biswal et al. 1995, but I am not clear on
>> what the motivation is for this choice.
>>
>> If someone can point me to papers that I must have missed that would be great.
>>
>> Thank you,
>> Jason.
>>
>
>
>---------------------------------------------------------------------------
>Stephen M. Smith, Professor of Biomedical Engineering
>Associate Director, Oxford University FMRIB Centre
>
>FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>+44 (0) 1865 222726 (fax 222717)
>[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>---------------------------------------------------------------------------
>
>
>
>
|