Hi Andreas,
I belive that SPM still uses a cosine-basis set for hp filtering. In
FSL, the hp filter is implemented via a Gaussian weighted least squares
line fit, see
Marchini, J. and Ripley, B. (2000). A new statistical approach to
detecting significant activation in functional MRI. NeuroImage,
12(4):366–380
for details. Essentially, a sliding window is moved across the time
series and within this window a straight line is fitted through the
data points using a least-squares criterion (but weighted by a Gaussian
wrt to distance from the centre of the window. The window itself is
then moved one data point and the process is repeated. As a final
result, this technique creates a low-frequency time series which
'follows' the main time series. When subtracted from the data, only
higher frequencies remain. Wrt translating between packages: a
hp-filter cutoff value should be comparable (i.e. the same hp filter
cutoff should actually remove similar frequencies form the
time-series).
cheers
christian
On 7 Oct 2004, at 08:22, Mark Jenkinson wrote:
> Hi Andreas,
>
> I'm afraid I don't know how the SPM filter works so I can't
> comment on that, but the use of -1 in the call from melodic
> just means that that filtering is to be ignored. In the
> help for ip it says:
>
> -t <hp_sigma> <lp_sigma> : Bandpass temporal filtering; nonlinear
> highpass and Gaussian linear lowpass (with sigmas in volumes, not
> seconds); set either sigma<0 to skip that filter
>
> So it is the use of sigma<0 to skip the filter which is being used.
>
> Hope that helps a little,
> Mark
>
>
> On 7 Oct 2004, at 07:38, Andreas Bartsch wrote:
>
>> Hi folks;),
>>
>> there seems to be a fundamental difference of the hp-filtering as
>> implemented in FSL & SPM. Could someone please elaborate a bit on the
>> details? Is there a simple rule of thumb how a given hp-filter cutoff
>> (in [secs]) translates into a corresponding value of the other
>> package? Or is there a more profound difference in the characteristics
>> of the filtered signal?
>> Also: ip offers the option to low-pass filter data. By default,
>> melodic seems call it at "-1". Is the kernel specified in negative
>> values?
>> Thanks + best regards-
>> Andreas
>>
>>
>
>
--
Christian F. Beckmann
Oxford University Centre for Functional
Magnetic Resonance Imaging of the Brain,
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
Email: [log in to unmask] - http://www.fmrib.ox.ac.uk/~beckmann/
Phone: +44(0)1865 222782 Fax: +44(0)1865 222717
|