Hi,
Just to add to this - most FSL filtering operations apart from
the high-pass filtering use a convolution (in principle, although
possibly implemented in different ways) where the kernel is
restricted at the ends. That is, if the kernel is 7 voxels wide
but you are at the edge voxel it will only include the middle
voxel and the 3 points inside the valid Field Of View. It then
renormalises the kernel based on the sum of the valid points
used. This means that if you have an image (or timeseries)
that is constant throughout, the resulting filtered version
shows no edge effects.
Hope this is clear.
All the best,
Mark
On 13 Oct 2010, at 18:46, Eugene Duff wrote:
> Hey Catherine -
>
> Which filtering are you referring to? The high pass filtering in FEAT
> is performed by a running lines smoother - a Gaussian-weighted
> least-squares straight line fitting (uses a straight line fit through
> nearby voxels to estimate trend - see A New Statistical Approach to
> Detecting Significant Activation in Functional MRI - Jonathan L.
> Marchini and Brian D. Ripley). This nonlinear approach is not greatly
> affected by end effects, so no mirroring is used, as I understand it.
>
> All the best,
>
> Eugene
>
> --
>
> Centre for Functional MRI of the Brain (FMRIB) | University of Oxford
> John Radcliffe Hospital | Headington
> OX3 9DU | Oxford | UK
>
> Ph: +44 (0) 1865 222 523 | Mob: +44 (0) 7946 362 059 | Fax: +44 (0)
> 1865 222 717
>
> --
>
>
> On 12 October 2010 03:52, Catherine Davey <[log in to unmask]>
> wrote:
>>
>> Hi, I just wondered if the FSL filter implementation uses the
>> reflection method, in which an initial portion of the time series
>> is reflected (as a mirror image) around time point zero (and then
>> repeated at the end of the time series and filtered backwards), to
>> avoid losing time points to initialisation (this is the method used
>> by both the matlab filter function, and the spm spm_filter function)?
>>
>> If not, then how does FSL deal with initialising in IIR filter?
>> and, can my filter theoretical require as many coefficients as the
>> number of time points I have?
>>
>> I hope this makes sense!
>> Cheers,
>> Catherine
>>
>
|